[llvm] [mlir] [MLIR][NVVM][NVPTX] Add collector_b operand to tcgen05.mma intrinsics (PR #214280)
Kirill Vedernikov via llvm-commits
llvm-commits at lists.llvm.org
Thu Aug 6 02:40:37 PDT 2026
https://github.com/kvederni updated https://github.com/llvm/llvm-project/pull/214280
>From 9dad587f83695add42f07fd4d8d72fcb455727c6 Mon Sep 17 00:00:00 2001
From: Kirill Vedernikov <kvedernikov at nvidia.com>
Date: Wed, 5 Aug 2026 18:33:24 +0200
Subject: [PATCH 1/5] [MLIR][NVVM][NVPTX] Add collector_b operand to
tcgen05.mma intrinsics
---
llvm/include/llvm/IR/IntrinsicsNVVM.td | 42 +-
llvm/lib/IR/AutoUpgrade.cpp | 84 ++
llvm/lib/Target/NVPTX/NVPTXIntrinsics.td | 143 ++-
llvm/lib/Target/NVPTX/NVPTXSubtarget.h | 2 +
.../Assembler/auto_upgrade_nvvm_intrinsics.ll | 68 ++
.../tcgen05-mma-collector-b-i8-invalid.ll | 10 +
.../CodeGen/NVPTX/tcgen05-mma-collector-b.ll | 1054 +++++++++++++++++
...n05-mma-disable-output-lane-collector-b.ll | 457 +++++++
.../tcgen05-mma-sp-collector-b-i8-invalid.ll | 10 +
.../NVPTX/tcgen05-mma-sp-collector-b.ll | 699 +++++++++++
...-mma-sp-disable-output-lane-collector-b.ll | 458 +++++++
.../NVPTX/tcgen05-mma-tensor-formatted.ll | 50 -
mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td | 18 +-
mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp | 10 +
.../nvvm/tcgen05-mma-block-scale-shared.mlir | 162 ++-
.../nvvm/tcgen05-mma-block-scale-tensor.mlir | 162 ++-
.../LLVMIR/nvvm/tcgen05-mma-shared.mlir | 258 ++--
.../tcgen05-mma-sp-block-scale-shared.mlir | 162 ++-
.../tcgen05-mma-sp-block-scale-tensor.mlir | 162 ++-
.../LLVMIR/nvvm/tcgen05-mma-sp-shared.mlir | 258 ++--
.../LLVMIR/nvvm/tcgen05-mma-sp-tensor.mlir | 354 +++---
.../LLVMIR/nvvm/tcgen05-mma-tensor.mlir | 354 +++---
22 files changed, 4183 insertions(+), 794 deletions(-)
create mode 100644 llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b-i8-invalid.ll
create mode 100644 llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b.ll
create mode 100644 llvm/test/CodeGen/NVPTX/tcgen05-mma-disable-output-lane-collector-b.ll
create mode 100644 llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b-i8-invalid.ll
create mode 100644 llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b.ll
create mode 100644 llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-disable-output-lane-collector-b.ll
delete mode 100644 llvm/test/CodeGen/NVPTX/tcgen05-mma-tensor-formatted.ll
diff --git a/llvm/include/llvm/IR/IntrinsicsNVVM.td b/llvm/include/llvm/IR/IntrinsicsNVVM.td
index fc04a5dc64e78..bc35373e5f026 100644
--- a/llvm/include/llvm/IR/IntrinsicsNVVM.td
+++ b/llvm/include/llvm/IR/IntrinsicsNVVM.td
@@ -3284,10 +3284,15 @@ foreach sparse = [0, 1] in {
);
defvar flags = [llvm_i32_ty, // kind
llvm_i32_ty, // cta_group
- llvm_i32_ty]; // collector_usage_a
+ llvm_i32_ty, // collector_usage_a
+ llvm_i32_ty]; // collector_usage_b
defvar nargs = !size(args);
defvar scale_d_imm = ArgIndex<!sub(nargs, 1)>;
defvar scale_d_imm_range = [ImmArg<scale_d_imm>, Range<scale_d_imm, 0, 16>];
+ defvar kind_idx = ArgIndex<nargs>;
+ defvar cta_group_idx = ArgIndex<!add(nargs, 1)>;
+ defvar collector_a_idx = ArgIndex<!add(nargs, 2)>;
+ defvar collector_b_idx = ArgIndex<!add(nargs, 3)>;
// Check if this is the specific llvm.nvvm.tcgen05.mma.tensor intrinsic.
defvar is_target_intrinsic = !and(!eq(sparse, 0),
@@ -3298,18 +3303,19 @@ foreach sparse = [0, 1] in {
defvar base_properties = !listconcat(
mma.common_intr_props,
!if(!eq(scale_d, 1), scale_d_imm_range, []),
- [Range<ArgIndex<nargs>, 0, !if(!eq(scale_d, 1), 2, 4)>, // kind
- Range<ArgIndex<!add(nargs, 1)>, 1, 3>, // cta_group
- Range<ArgIndex<!add(nargs, 2)>, 0,
- !if(!eq(ashift, 1), 2, 4)> // collector_usage
+ [Range<kind_idx, 0, !if(!eq(scale_d, 1), 2, 4)>,
+ Range<cta_group_idx, 1, 3>,
+ Range<collector_a_idx, 0, !if(!eq(ashift, 1), 2, 4)>,
+ Range<collector_b_idx, 0, 4>
]
);
defvar intrinsic_properties = !if(is_target_intrinsic,
!listconcat(base_properties,
- [ArgInfo<ArgIndex<nargs>, [ArgName<"kind">, ImmArgPrinter<"printTcgen05MMAKind">]>,
- ArgInfo<ArgIndex<!add(nargs, 1)>, [ArgName<"cta_group">]>,
- ArgInfo<ArgIndex<!add(nargs, 2)>, [ArgName<"collector">, ImmArgPrinter<"printTcgen05CollectorUsageOp">]>]),
+ [ArgInfo<kind_idx, [ArgName<"kind">, ImmArgPrinter<"printTcgen05MMAKind">]>,
+ ArgInfo<cta_group_idx, [ArgName<"cta_group">]>,
+ ArgInfo<collector_a_idx, [ArgName<"collector_a">, ImmArgPrinter<"printTcgen05CollectorUsageOp">]>,
+ ArgInfo<collector_b_idx, [ArgName<"collector_b">, ImmArgPrinter<"printTcgen05CollectorUsageOp">]>]),
base_properties);
def mma.record_name:
@@ -3338,15 +3344,20 @@ foreach sparse = [0, 1] in {
[disable_output_lane_type]
);
defvar flags = [llvm_i32_ty, // kind_flag
- llvm_i32_ty]; // collector_usage_a_flag
+ llvm_i32_ty, // collector_usage_a_flag
+ llvm_i32_ty]; // collector_usage_b_flag
defvar nargs = !size(args);
defvar scale_d_imm = ArgIndex<!sub(nargs, 2)>;
defvar scale_d_imm_range = [ImmArg<scale_d_imm>, Range<scale_d_imm, 0, 16>];
+ defvar kind_idx = ArgIndex<nargs>;
+ defvar collector_a_idx = ArgIndex<!add(nargs, 1)>;
+ defvar collector_b_idx = ArgIndex<!add(nargs, 2)>;
defvar intrinsic_properties = !listconcat(
mma.common_intr_props,
!if(!eq(scale_d, 1), scale_d_imm_range, []),
- [Range<ArgIndex<nargs>, 0, !if(!eq(scale_d, 1), 2, 4)>,
- Range<ArgIndex<!add(nargs, 1)>, 0, !if(!eq(ashift, 1), 2, 4)>]
+ [Range<kind_idx, 0, !if(!eq(scale_d, 1), 2, 4)>,
+ Range<collector_a_idx, 0, !if(!eq(ashift, 1), 2, 4)>,
+ Range<collector_b_idx, 0, 4>]
);
def mma.record_name : DefaultAttrsIntrinsicFlags<[], args, flags,
@@ -3369,16 +3380,19 @@ foreach sparse = [0, 1] in {
[llvm_tmem_ptr_ty, // scale_a
llvm_tmem_ptr_ty]); // scale_b
defvar flags = [llvm_i32_ty, // cta_group
- llvm_i32_ty]; // collector_usage_a
+ llvm_i32_ty, // collector_usage_a
+ llvm_i32_ty]; // collector_usage_b
defvar nargs = !size(args);
defvar cta_group = ArgIndex<nargs>;
- defvar collector_usage = ArgIndex<!add(nargs, 1)>;
+ defvar collector_usage_a = ArgIndex<!add(nargs, 1)>;
+ defvar collector_usage_b = ArgIndex<!add(nargs, 2)>;
if NVVM_TCGEN05_MMA_BLOCKSCALE_SUPPORTED<kind, scale_vec_size>.ret then {
def mma.record_name : DefaultAttrsIntrinsicFlags<[], args, flags,
!listconcat(mma.common_intr_props,
[Range<cta_group, 1, 3>,
- Range<collector_usage, 0, 4>]),
+ Range<collector_usage_a, 0, 4>,
+ Range<collector_usage_b, 0, 4>]),
mma.intr_name>;
}
} // scale_vec_size
diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index d2216e4072337..446d132074fd9 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -1320,6 +1320,18 @@ static Intrinsic::ID shouldUpgradeNVPTXBF16Intrinsic(StringRef Name) {
return Intrinsic::not_intrinsic;
}
+static Intrinsic::ID shouldUpgradeNVPTXTcgen05MMAIntrinsic(Function *F,
+ StringRef Name) {
+ if (!Name.consume_front("tcgen05.mma."))
+ return Intrinsic::not_intrinsic;
+
+ // tcgen05.mma.ws.* variants do not need collector-b appended.
+ if (Name.starts_with("ws"))
+ return Intrinsic::not_intrinsic;
+
+ return F->getIntrinsicID();
+}
+
static bool consumeNVVMPtrAddrSpace(StringRef &Name) {
return Name.consume_front("local") || Name.consume_front("shared") ||
Name.consume_front("global") || Name.consume_front("constant") ||
@@ -1772,6 +1784,13 @@ static bool upgradeIntrinsicFunction1(Function *F, Function *&NewFn,
return true;
}
+ // Upgrade tcgen05.mma intrinsics missing collector_usage_b.
+ IID = shouldUpgradeNVPTXTcgen05MMAIntrinsic(F, Name);
+ if (IID != Intrinsic::not_intrinsic) {
+ NewFn = Intrinsic::getOrInsertDeclaration(F->getParent(), IID);
+ return NewFn != F;
+ }
+
// The following nvvm intrinsics correspond exactly to an LLVM idiom, but
// not to an intrinsic alone. We expand them in UpgradeIntrinsicCall.
//
@@ -5680,6 +5699,71 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) {
NewCall = Builder.CreateCall(NewFn, Args);
break;
}
+ case Intrinsic::nvvm_tcgen05_mma_shared:
+ case Intrinsic::nvvm_tcgen05_mma_shared_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_shared_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_shared_mxf4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_shared_mxf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_shared_mxf4nvf4_block_scale_block16:
+ case Intrinsic::nvvm_tcgen05_mma_shared_mxf4nvf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_shared_mxf8f6f4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_shared_mxf8f6f4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_shared_scale_d:
+ case Intrinsic::nvvm_tcgen05_mma_shared_scale_d_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_shared_scale_d_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_mxf4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_mxf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_mxf4nvf4_block_scale_block16:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_mxf4nvf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_mxf8f6f4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_mxf8f6f4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_scale_d:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_scale_d_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_sp_shared_scale_d_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_disable_output_lane_cg1_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_disable_output_lane_cg2_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_mxf4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_mxf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_mxf4nvf4_block_scale_block16:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_mxf4nvf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_mxf8f6f4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_mxf8f6f4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg1_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg2_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_tensor:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_disable_output_lane_cg1_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_disable_output_lane_cg2_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_mxf4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_mxf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_mxf4nvf4_block_scale_block16:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_mxf4nvf4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_mxf8f6f4_block_scale:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_mxf8f6f4_block_scale_block32:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg1:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg1_ashift:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg2:
+ case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg2_ashift: {
+ SmallVector<Value *, 12> Args(CI->args());
+ Args.push_back(Builder.getInt32(0)); // collector_usage_b = discard(0)
+ NewCall = Builder.CreateCall(NewFn, Args);
+ break;
+ }
case Intrinsic::riscv_sha256sig0:
case Intrinsic::riscv_sha256sig1:
case Intrinsic::riscv_sha256sum0:
diff --git a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
index ea47ddcc02ae7..249aee4a09fa5 100644
--- a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
+++ b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -6191,7 +6191,7 @@ foreach dim = ["x", "y", "z"] in {
//
class Tcgen05MMABase<bit IsSparse, string ASpace, string Kind, int CtaGroup,
- string CollectorUsage> :
+ string CollectorUsage, string CollectorUsageB = ""> :
NVPTXInst<(outs), (ins), "?", []>,
Requires<[]> {
@@ -6201,6 +6201,7 @@ class Tcgen05MMABase<bit IsSparse, string ASpace, string Kind, int CtaGroup,
!eq(CollectorUsage, "fill"): 2,
!eq(CollectorUsage, "use"): 3
);
+ int CollectorUsageAVal = CollectorUsageVal;
int KindVal = !cond(
!eq(Kind, "f16"): 0,
@@ -6209,6 +6210,15 @@ class Tcgen05MMABase<bit IsSparse, string ASpace, string Kind, int CtaGroup,
!eq(Kind, "i8"): 3,
);
+ bit HasCollectorB = !ne(CollectorUsageB, "");
+ int CollectorUsageBVal = !cond(
+ !eq(CollectorUsageB, "discard"): 0,
+ !eq(CollectorUsageB, "lastuse"): 1,
+ !eq(CollectorUsageB, "fill"): 2,
+ !eq(CollectorUsageB, "use"): 3,
+ true: 0
+ );
+
dag SparseMetadataIns = !if(IsSparse, (ins B32:$spmetadata), (ins));
string SparseMetadataStr = !if(IsSparse, ", [$spmetadata]", "");
dag SparseMetadataIntr = !if(IsSparse, (ins i32:$spmetadata), (ins));
@@ -6231,11 +6241,11 @@ class Tcgen05MMABase<bit IsSparse, string ASpace, string Kind, int CtaGroup,
# ".cta_group::" # CtaGroup
# ".kind::" # Kind;
- string BaseOperandsStr = " [$dtmem], "
- # AOperandStr
- # ", $b"
- # SparseMetadataStr
- # ", $idesc";
+ string BaseOperandsCommonStr = " [$dtmem], "
+ # AOperandStr
+ # ", $b";
+ string IDescStr = ", $idesc";
+ string BaseOperandsStr = BaseOperandsCommonStr # SparseMetadataStr # IDescStr;
string InputDStr = ", $enable_inp_d";
}
@@ -6244,13 +6254,17 @@ class Tcgen05MMABase<bit IsSparse, string ASpace, string Kind, int CtaGroup,
//
class Tcgen05MMAInst<bit IsSparse, string ASpace, string Kind, int CtaGroup,
- string CollectorUsage, bit IsScaleInputD, bit IsAShift> :
- Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsage> {
-
- let Predicates = !cond(
- IsScaleInputD : [callSubtarget<"hasTcgen05MMAScaleInputDImm">],
- !eq(Kind, "i8") : [callSubtarget<"hasTcgen05MMAI8Kind">],
- true : [callSubtarget<"hasTcgen05InstSupport">]
+ string CollectorUsageA, string CollectorUsageB,
+ bit IsScaleInputD, bit IsAShift> :
+ Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsageA,
+ CollectorUsageB> {
+ let Predicates = !listconcat(
+ !if(HasCollectorB, [callSubtarget<"hasTcgen05RubinFamily">], []),
+ !cond(
+ IsScaleInputD : [callSubtarget<"hasTcgen05MMAScaleInputDImm">],
+ !eq(Kind, "i8") : [callSubtarget<"hasTcgen05MMAI8Kind">],
+ true : [callSubtarget<"hasTcgen05InstSupport">]
+ )
);
Intrinsic Intrin = !cast<Intrinsic>(
@@ -6266,7 +6280,8 @@ class Tcgen05MMAInst<bit IsSparse, string ASpace, string Kind, int CtaGroup,
let AsmString = Prefix
# SpCtaKindStr
# !if(IsAShift, ".ashift", "")
- # ".collector::a::" # CollectorUsage
+ # ".collector::a::" # CollectorUsageA
+ # !if(HasCollectorB, ".collector::b::" # CollectorUsageB, "")
# BaseOperandsStr
# InputDStr
# ScaleInpStr
@@ -6276,7 +6291,7 @@ class Tcgen05MMAInst<bit IsSparse, string ASpace, string Kind, int CtaGroup,
ScaleInpIntr);
dag FlagOperands = (Intrin (i32 KindVal), (i32 CtaGroup),
- (i32 CollectorUsageVal));
+ (i32 CollectorUsageAVal), (i32 CollectorUsageBVal));
let Pattern = [!con(IntrinsicPattern, FlagOperands)];
}
@@ -6286,16 +6301,18 @@ foreach sparse = [0, 1] in {
foreach space = ["tensor", "shared"] in {
foreach kind = ["f16", "tf32", "f8f6f4", "i8"] in {
foreach cta_group = [1, 2] in {
- foreach collector_usage = ["discard", "lastuse", "fill", "use"] in {
+ foreach collector_usage_a = ["discard", "lastuse", "fill", "use"] in {
foreach scale_input_d = !if(!or(!eq(kind, "f16"),
!eq(kind, "tf32")), [0, 1], [0]) in {
foreach ashift = !if(!eq(space, "tensor"), [0, 1], [0]) in {
-
- def : Tcgen05MMAInst<sparse, space, kind, cta_group,
- collector_usage, scale_input_d, ashift>;
+ foreach collector_usage_b = ["", "discard", "lastuse", "fill", "use"] in {
+ def : Tcgen05MMAInst<sparse, space, kind, cta_group,
+ collector_usage_a, collector_usage_b,
+ scale_input_d, ashift>;
+ } // collector_usage_b
} // ashift
} // scale_input_d
- } // collector_usage
+ } // collector_usage_a
} // cta_group
} // kind
} // space
@@ -6317,7 +6334,9 @@ class Tcgen05MMADisableOutputLaneTypeProfile<bit IsSparse, string ASpace,
!if(IsSparse, [i32], []), // spmetadata
!if(IsScaleInputD, [i64], []), // scale_input_d
!listsplat(i32, DisableOutputLaneVecSize), // disable_output_lane
- [i32, i32] // kind, collector_usage
+ [i32], // kind
+ [i32], // collector_usage_a
+ [i32] // collector_usage_b
);
let Constraints = !foreach(x, !range(!size(VTs)), SDTCisVT<x, VTs[x]>);
let NumOperands = !size(Constraints);
@@ -6340,14 +6359,18 @@ class Tcgen05MMADisableOutputLaneSDNode<bit IsSparse, string ASpace,
//
class Tcgen05MMADisableOutputLaneInst<bit IsSparse, string ASpace, string Kind,
- int CtaGroup, string CollectorUsage,
+ int CtaGroup, string CollectorUsageA,
+ string CollectorUsageB,
bit IsScaleInputD, bit IsAShift> :
- Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsage> {
-
- let Predicates = !cond(
- IsScaleInputD : [callSubtarget<"hasTcgen05MMAScaleInputDImm">],
- !eq(Kind, "i8") : [callSubtarget<"hasTcgen05MMAI8Kind">],
- true : [callSubtarget<"hasTcgen05InstSupport">]
+ Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsageA,
+ CollectorUsageB> {
+ let Predicates = !listconcat(
+ !if(HasCollectorB, [callSubtarget<"hasTcgen05RubinFamily">], []),
+ !cond(
+ IsScaleInputD : [callSubtarget<"hasTcgen05MMAScaleInputDImm">],
+ !eq(Kind, "i8") : [callSubtarget<"hasTcgen05MMAI8Kind">],
+ true : [callSubtarget<"hasTcgen05InstSupport">]
+ )
);
SDNode Opcode = Tcgen05MMADisableOutputLaneSDNode<IsSparse, ASpace, CtaGroup,
@@ -6387,7 +6410,8 @@ class Tcgen05MMADisableOutputLaneInst<bit IsSparse, string ASpace, string Kind,
let AsmString = Prefix
# SpCtaKindStr
# !if(IsAShift, ".ashift", "")
- # ".collector::a::" # CollectorUsage
+ # ".collector::a::" # CollectorUsageA
+ # !if(HasCollectorB, ".collector::b::" # CollectorUsageB, "")
# BaseOperandsStr
# ", " # DisableOutputLaneStr
# InputDStr
@@ -6398,7 +6422,8 @@ class Tcgen05MMADisableOutputLaneInst<bit IsSparse, string ASpace, string Kind,
ScaleInpIntr,
DisableOutputLaneInput);
- dag FlagOperands = (Opcode (i32 KindVal), (i32 CollectorUsageVal));
+ dag FlagOperands = (Opcode (i32 KindVal), (i32 CollectorUsageAVal),
+ (i32 CollectorUsageBVal));
let Pattern = [!con(IntrinsicPattern, FlagOperands)];
}
@@ -6408,17 +6433,19 @@ foreach sparse = [0, 1] in {
foreach space = ["tensor", "shared"] in {
foreach kind = ["f16", "tf32", "f8f6f4", "i8"] in {
foreach cta_group = [1, 2] in {
- foreach collector_usage = ["fill", "use", "lastuse", "discard"] in {
+ foreach collector_usage_a = ["fill", "use", "lastuse", "discard"] in {
foreach scale_input_d = !if(!or(!eq(kind, "f16"),
!eq(kind, "tf32")), [0, 1], [0]) in {
foreach ashift = !if(!eq(space, "tensor"), [0, 1], [0]) in {
- def :
- Tcgen05MMADisableOutputLaneInst<sparse, space, kind, cta_group,
- collector_usage, scale_input_d,
- ashift>;
+ foreach collector_usage_b = ["", "discard", "lastuse", "fill", "use"] in {
+ def :
+ Tcgen05MMADisableOutputLaneInst<sparse, space, kind, cta_group,
+ collector_usage_a, collector_usage_b,
+ scale_input_d, ashift>;
+ } // collector_usage_b
} // ashift
} // scale_input_d
- } // collector_usage
+ } // collector_usage_a
} // cta_group
} // kind
} // space
@@ -6430,16 +6457,19 @@ foreach sparse = [0, 1] in {
class Tcgen05MMABlockScaleInst<bit IsSparse, string ASpace, string Kind,
int CtaGroup, string ScaleVecSize,
- string CollectorUsage>:
- Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsage> {
-
- let Predicates = !cond(
- !and(IsSparse,
- !eq(Kind, "mxf4")) : [callSubtarget<"hasTcgen05MMASparseMxf4">],
- !and(IsSparse,
- !eq(Kind, "mxf4nvf4")) : [callSubtarget<"hasTcgen05MMASparseMxf4nvf4">],
- !ne(ScaleVecSize, "") : [callSubtarget<"hasTcgen05InstSupport">, hasPTX<88>],
- true : [callSubtarget<"hasTcgen05InstSupport">]
+ string CollectorUsageA, string CollectorUsageB>:
+ Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsageA,
+ CollectorUsageB> {
+ let Predicates = !listconcat(
+ !if(HasCollectorB, [callSubtarget<"hasTcgen05RubinFamily">], []),
+ !cond(
+ !and(IsSparse,
+ !eq(Kind, "mxf4")) : [callSubtarget<"hasTcgen05MMASparseMxf4">],
+ !and(IsSparse,
+ !eq(Kind, "mxf4nvf4")) : [callSubtarget<"hasTcgen05MMASparseMxf4nvf4">],
+ !ne(ScaleVecSize, "") : [callSubtarget<"hasTcgen05InstSupport">, hasPTX<88>],
+ true : [callSubtarget<"hasTcgen05InstSupport">]
+ )
);
Intrinsic Intrin = !cast<Intrinsic>(
@@ -6457,7 +6487,8 @@ class Tcgen05MMABlockScaleInst<bit IsSparse, string ASpace, string Kind,
let AsmString = Prefix
# SpCtaKindStr
# ".block_scale" # ScaleVecSize
- # ".collector::a::" # CollectorUsage
+ # ".collector::a::" # CollectorUsageA
+ # !if(HasCollectorB, ".collector::b::" # CollectorUsageB, "")
# BaseOperandsStr
# ", [$scale_a], [$scale_b]"
# InputDStr
@@ -6467,7 +6498,8 @@ class Tcgen05MMABlockScaleInst<bit IsSparse, string ASpace, string Kind,
(Intrin i32:$scale_a,
i32:$scale_b));
- dag FlagOperands = (Intrin (i32 CtaGroup), (i32 CollectorUsageVal));
+ dag FlagOperands = (Intrin (i32 CtaGroup), (i32 CollectorUsageAVal),
+ (i32 CollectorUsageBVal));
let Pattern = [!con(IntrinsicPattern, FlagOperands)];
}
@@ -6478,12 +6510,15 @@ foreach sparse = [0, 1] in {
foreach kind = ["mxf8f6f4", "mxf4", "mxf4nvf4"] in {
foreach scale_vec_size = ["", ".block16", ".block32"] in {
foreach cta_group = [1, 2] in {
- foreach collector_usage = ["fill", "use", "lastuse", "discard"] in {
- if NVVM_TCGEN05_MMA_BLOCKSCALE_SUPPORTED<kind, scale_vec_size>.ret then {
- def : Tcgen05MMABlockScaleInst<sparse, space, kind, cta_group,
- scale_vec_size, collector_usage>;
- }
- } // collector_usage
+ foreach collector_usage_a = ["fill", "use", "lastuse", "discard"] in {
+ foreach collector_usage_b = ["", "discard", "lastuse", "fill", "use"] in {
+ if NVVM_TCGEN05_MMA_BLOCKSCALE_SUPPORTED<kind, scale_vec_size>.ret then {
+ def : Tcgen05MMABlockScaleInst<sparse, space, kind, cta_group,
+ scale_vec_size, collector_usage_a,
+ collector_usage_b>;
+ }
+ } // collector_usage_b
+ } // collector_usage_a
} // cta_group
} // scale_vec_size
} // kind
diff --git a/llvm/lib/Target/NVPTX/NVPTXSubtarget.h b/llvm/lib/Target/NVPTX/NVPTXSubtarget.h
index 22affe4f40759..b91e538842170 100644
--- a/llvm/lib/Target/NVPTX/NVPTXSubtarget.h
+++ b/llvm/lib/Target/NVPTX/NVPTXSubtarget.h
@@ -73,6 +73,8 @@ class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
return TSInfo.get();
}
+ bool hasTcgen05RubinFamily() const { return hasPTXWithFamilySMs(94, {107}); }
+
// Checks PTX version and family-specific and architecture-specific SM
// versions. For example, sm_100{f/a} and any future variants in the same
// family will match for any PTX version greater than or equal to
diff --git a/llvm/test/Assembler/auto_upgrade_nvvm_intrinsics.ll b/llvm/test/Assembler/auto_upgrade_nvvm_intrinsics.ll
index cc66175fbe8b2..389a15180bd79 100644
--- a/llvm/test/Assembler/auto_upgrade_nvvm_intrinsics.ll
+++ b/llvm/test/Assembler/auto_upgrade_nvvm_intrinsics.ll
@@ -141,6 +141,15 @@ declare void @llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.3d(ptr addrspace(
declare void @llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.4d(ptr addrspace(3), ptr, i32, i32, i32, i32, i64, i1)
declare void @llvm.nvvm.cp.async.bulk.tensor.reduce.add.im2col.5d(ptr addrspace(3), ptr, i32, i32, i32, i32, i32, i64, i1)
+declare void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6), i64, i64, i32, i1, i32, i32, i32)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6), i64, i64, i32, i1, ptr addrspace(6), i32, i32, i32)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6), ptr addrspace(6), i64, i32, i1, ptr addrspace(6), i32, i32, i32)
+declare void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6), i64, i64, i32, i1, <4 x i32>, i32, i32)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6), ptr addrspace(6), i64, i32, i1, ptr addrspace(6), <8 x i32>, i32, i32)
+declare void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6), i64, i64, i32, i1, ptr addrspace(6), ptr addrspace(6), i32, i32)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6), i64, i64, i32, i1, ptr addrspace(6), ptr addrspace(6), ptr addrspace(6), i32, i32)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6), ptr addrspace(6), i64, i32, i1, ptr addrspace(6), ptr addrspace(6), ptr addrspace(6), i32, i32)
+
declare void @llvm.nvvm.barrier0()
declare void @llvm.nvvm.barrier.n(i32)
declare void @llvm.nvvm.bar.sync(i32)
@@ -538,6 +547,65 @@ define void @nvvm_cp_async_bulk_tensor_reduce_shapes(ptr addrspace(3) %src, ptr
ret void
}
+; CHECK-LABEL: @nvvm_tcgen05_mma_default_collector_b
+define void @nvvm_tcgen05_mma_default_collector_b(
+ ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor,
+ i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d,
+ ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a,
+ ptr addrspace(6) %scale_b, <4 x i32> %disable_output_lanev4,
+ <8 x i32> %disable_output_lanev8) {
+; CHECK: call void @llvm.nvvm.tcgen05.mma.shared({{.*}}i32 0, i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared(
+ ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc,
+ i1 %enable_inp_d, i32 0, i32 1, i32 0)
+
+; CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared({{.*}}i32 1, i32 2, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(
+ ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc,
+ i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 2,
+ i32 1)
+
+; CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift({{.*}}i32 2, i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(
+ ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b,
+ i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata,
+ i32 2, i32 1, i32 0)
+
+; CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1({{.*}}i32 0, i32 3, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(
+ ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc,
+ i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0,
+ i32 3)
+
+; CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift({{.*}}i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(
+ ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b,
+ i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata,
+ <8 x i32> %disable_output_lanev8, i32 1, i32 0)
+
+; CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale({{.*}}i32 2, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(
+ ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc,
+ i1 %enable_inp_d, ptr addrspace(6) %scale_a,
+ ptr addrspace(6) %scale_b, i32 2, i32 1)
+
+; CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32({{.*}}i32 1, i32 2, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(
+ ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc,
+ i1 %enable_inp_d, ptr addrspace(6) %spmetadata,
+ ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1,
+ i32 2)
+
+; CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16({{.*}}i32 1, i32 3, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(
+ ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b,
+ i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata,
+ ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1,
+ i32 3)
+
+ ret void
+}
+
define void @cta_barriers(i32 %x, i32 %y, i32 %z) {
; CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)
; CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 %x)
diff --git a/llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b-i8-invalid.ll b/llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b-i8-invalid.ll
new file mode 100644
index 0000000000000..7479f56d78851
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b-i8-invalid.ll
@@ -0,0 +1,10 @@
+; RUN: not --crash llc -o - -mcpu=sm_107f -mattr=+ptx94 -march=nvptx64 %s 2>&1 | FileCheck %s
+
+target triple = "nvptx64-nvidia-cuda"
+
+define void @tcgen05_mma_i8_collector_b_sm107f(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; CHECK: LLVM ERROR: Cannot select:
+ ; kind=i8(3), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 3, i32 1, i32 0, i32 1)
+ ret void
+}
diff --git a/llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b.ll b/llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b.ll
new file mode 100644
index 0000000000000..3bd32a9d4eadb
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/tcgen05-mma-collector-b.ll
@@ -0,0 +1,1054 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -o - -mcpu=sm_107f -mattr=+ptx94 -march=nvptx64 %s | FileCheck %s
+; RUN: llvm-as < %s | llvm-dis | FileCheck %s --check-prefixes=FORMAT
+; RUN: %if ptxas-sm_107f && ptxas-isa-9.4 %{ llc < %s -march=nvptx64 -mattr=+ptx94 -mcpu=sm_107f | %ptxas-verify -arch=sm_107f %}
+
+; ---------------------------------------------------------------------------
+; f16, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_f16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_f16_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=fill */ i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=use */ i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 2, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=fill */ i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_f16_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_f16_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_f16_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_f16_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_f16_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_f16_cta1_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_f16_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::lastuse.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::fill.collector::b::use [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::use.collector::b::fill [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::use [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 2, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 3, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 3, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f16, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_f16_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_f16_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 1, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_f16_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_f16_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_f16_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_f16_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_f16_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_f16_cta2_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_f16_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_tf32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_tf32_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=fill */ i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=use */ i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_tf32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_tf32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_tf32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_tf32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_tf32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_tf32_cta1_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_tf32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::lastuse.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f8f6f4, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_f8f6f4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_f8f6f4_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=fill */ i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=use */ i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_f8f6f4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_f8f6f4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_f8f6f4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_f8f6f4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_f8f6f4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_f8f6f4_cta1_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_f8f6f4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_tf32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_tf32_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 1, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_tf32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_tf32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_tf32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_tf32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_tf32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_tf32_cta2_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_tf32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 1, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f8f6f4, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_f8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_f8f6f4_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=fill */ i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=use */ i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=lastuse */ i32 1, /* collector_b=lastuse */ i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_f8f6f4_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_f8f6f4_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_f8f6f4_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_f8f6f4_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_f8f6f4_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_f8f6f4_cta2_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_f8f6f4_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::lastuse [%r1], [%r3], %rd2, %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, cta_group::1
+; Covers: shared+tensor, representative collector_a+collector_b combos
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 3, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::fill.collector::b::use [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::use.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::lastuse.collector::b::fill [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 3, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::fill.collector::b::use [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::fill [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.collector::a::fill.collector::b::use [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.collector::a::fill [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4nvf4 (.block16), cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block16.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block16.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block16.collector::a::fill.collector::b::use [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block16.collector::a::fill [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block16.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 3)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.collector::a::fill.collector::b::use [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.collector::a::fill [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 3)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4, .block32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.block32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.block32.collector::a::discard [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4nvf4, .block16, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block16_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block16.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block16.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block16.collector::a::discard [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block16.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4nvf4, .block32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block32.collector::a::discard [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf4nvf4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, .block32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, f16, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_scale_d_f16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_scale_d_f16_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_scale_d_f16_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_scale_d_f16_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_scale_d_f16_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_scale_d_f16_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_scale_d_f16_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_scale_d_f16_cta1_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_scale_d_f16_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::lastuse.collector::b::fill [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.ashift.collector::a::discard [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, f16, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_scale_d_f16_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_scale_d_f16_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 1, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_scale_d_f16_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_scale_d_f16_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_scale_d_f16_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_scale_d_f16_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_scale_d_f16_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_scale_d_f16_cta2_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_scale_d_f16_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.ashift.collector::a::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 0, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, tf32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_scale_d_tf32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_scale_d_tf32_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 1, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_scale_d_tf32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_scale_d_tf32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_scale_d_tf32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_scale_d_tf32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_scale_d_tf32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_scale_d_tf32_cta2_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_scale_d_tf32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.ashift.collector::a::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 1, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, tf32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_scale_d_tf32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_scale_d_tf32_cta1(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_scale_d_tf32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<4>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_scale_d_tf32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_scale_d_tf32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_scale_d_tf32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_scale_d_tf32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_scale_d_tf32_cta1_param_4];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_scale_d_tf32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.ashift.collector::a::discard [%r1], [%r3], %rd2, %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 0, i32 1, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, .block32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf8f6f4_block32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4, .block32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4_block32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.block32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.block32.collector::a::discard [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf4nvf4, .block32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; FORMAT-LABEL: define void @tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2(
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 2)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; FORMAT-NEXT: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+; CHECK-LABEL: tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<6>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_collector_b_block_scale_mxf4nvf4_block32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block32.collector::a::discard [%r1], [%r5], %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::mxf4nvf4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, %r2, [%r3], [%r4], %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 2)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
diff --git a/llvm/test/CodeGen/NVPTX/tcgen05-mma-disable-output-lane-collector-b.ll b/llvm/test/CodeGen/NVPTX/tcgen05-mma-disable-output-lane-collector-b.ll
new file mode 100644
index 0000000000000..886f7f7930276
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/tcgen05-mma-disable-output-lane-collector-b.ll
@@ -0,0 +1,457 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -o - -mcpu=sm_107f -mattr=+ptx94 -march=nvptx64 %s | FileCheck %s
+; RUN: %if ptxas-sm_107f && ptxas-isa-9.4 %{ llc < %s -march=nvptx64 -mattr=+ptx94 -mcpu=sm_107f | %ptxas-verify -arch=sm_107f %}
+
+; ---------------------------------------------------------------------------
+; f16, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_f16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4) {
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_f16_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<8>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_f16_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_f16_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_f16_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_f16_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_f16_cta1_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_f16_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ld.param.b32 %r7, [tcgen05_mma_disable_output_lane_collector_b_f16_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::fill.collector::b::use [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::use.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::use [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=lastuse(1), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 1, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=fill(2), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 2, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=use(3), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 3, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=use(3), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 3, i32 0)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f16, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_f16_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8) {
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_f16_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<12>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_6+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r7, %r8, %r9, %r10}, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ld.param.b32 %r11, [tcgen05_mma_disable_output_lane_collector_b_f16_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.ashift.collector::a::discard.collector::b::use [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::lastuse.collector::b::use [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::lastuse [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 2)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=lastuse(1), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 3)
+ ; kind=f16(0), cta_group=2, collector_a=lastuse(1), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 3)
+ ; kind=f16(0), cta_group=2, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 0)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_tf32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4) {
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_tf32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<8>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta1_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ld.param.b32 %r7, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.ashift.collector::a::discard.collector::b::use [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::use.collector::b::fill [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::use [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=use(3), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 1, i32 3, i32 2)
+ ; kind=tf32(1), cta_group=1, collector_a=use(3), collector_b=discard(0), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 1, i32 3, i32 0)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_tf32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8) {
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=fill(2)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_tf32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<12>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_6+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r7, %r8, %r9, %r10}, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ld.param.b32 %r11, [tcgen05_mma_disable_output_lane_collector_b_tf32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.ashift.collector::a::lastuse [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 0)
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f8f6f4, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4) {
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<8>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ld.param.b32 %r7, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.ashift.collector::a::discard.collector::b::use [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::fill.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::fill [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 2)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=fill(2), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 2, i32 2, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=fill(2), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 2, i32 2, i32 0)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f8f6f4, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8) {
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<12>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_6+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r7, %r8, %r9, %r10}, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ld.param.b32 %r11, [tcgen05_mma_disable_output_lane_collector_b_f8f6f4_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.ashift.collector::a::lastuse.collector::b::fill [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.ashift.collector::a::lastuse [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 2, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 2, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=lastuse(1), collector_b=fill(2), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 2, i32 1, i32 2)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 2, i32 1, i32 0)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f16 scale_d, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4) {
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<8>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 5;
+; CHECK-NEXT: ld.param.b32 %r7, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 3;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.ashift.collector::a::discard.collector::b::use [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 7;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 2;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::lastuse [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 2;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 5;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 5, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 3, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 2)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=discard(0), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 7, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 3)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 2, <4 x i32> %disable_output_lanev4, i32 0, i32 1, i32 1)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 2, <4 x i32> %disable_output_lanev4, i32 0, i32 1, i32 0)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 5, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32 scale_d, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lanev4) {
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=discard(0), collector_b=fill(2)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<8>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 4;
+; CHECK-NEXT: ld.param.b32 %r7, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 6;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.ashift.collector::a::lastuse [%r1], [%r7], %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 1;
+; CHECK-NEXT: tcgen05.mma.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r3, %r4, %r5, %r6}, %p1, 4;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 4, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 6, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 1, <4 x i32> %disable_output_lanev4, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 1, <4 x i32> %disable_output_lanev4, i32 1, i32 1, i32 0)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 4, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f16 scale_d, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8) {
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<12>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_6+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r7, %r8, %r9, %r10}, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 5;
+; CHECK-NEXT: ld.param.b32 %r11, [tcgen05_mma_disable_output_lane_collector_b_f16_scale_d_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 3;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.ashift.collector::a::lastuse.collector::b::use [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 7;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.ashift.collector::a::lastuse [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 7;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 5;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 5, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 1)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 3, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 2)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 7, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 3)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 7, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 0)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 5, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32 scale_d, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lanev8) {
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=discard(0), collector_b=fill(2)
+; CHECK-LABEL: tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<12>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_4];
+; CHECK-NEXT: ld.param.v4.b32 {%r3, %r4, %r5, %r6}, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_6+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r7, %r8, %r9, %r10}, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 4;
+; CHECK-NEXT: ld.param.b32 %r11, [tcgen05_mma_disable_output_lane_collector_b_tf32_scale_d_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 6;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.ashift.collector::a::lastuse [%r1], [%r11], %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 1;
+; CHECK-NEXT: tcgen05.mma.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, %r2, {%r7, %r8, %r9, %r10, %r3, %r4, %r5, %r6}, %p1, 4;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 4, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 6, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 1, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 1, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 0)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, i64 4, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 0)
+ ret void
+}
diff --git a/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b-i8-invalid.ll b/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b-i8-invalid.ll
new file mode 100644
index 0000000000000..567eaa8644e5e
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b-i8-invalid.ll
@@ -0,0 +1,10 @@
+; RUN: not --crash llc -o - -mcpu=sm_107f -mattr=+ptx94 -march=nvptx64 %s 2>&1 | FileCheck %s
+
+target triple = "nvptx64-nvidia-cuda"
+
+define void @tcgen05_mma_sp_i8_collector_b_sm107f(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK: LLVM ERROR: Cannot select:
+ ; kind=i8(3), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 3, i32 1, i32 0, i32 1)
+ ret void
+}
diff --git a/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b.ll b/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b.ll
new file mode 100644
index 0000000000000..0f1a84829e9d0
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-collector-b.ll
@@ -0,0 +1,699 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -o - -mcpu=sm_107f -mattr=+ptx94 -march=nvptx64 %s | FileCheck %s
+; RUN: %if ptxas-sm_107f && ptxas-isa-9.4 %{ llc < %s -march=nvptx64 -mattr=+ptx94 -mcpu=sm_107f | %ptxas-verify -arch=sm_107f %}
+
+; ---------------------------------------------------------------------------
+; f16, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_f16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_f16_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_f16_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_f16_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_f16_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_f16_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_f16_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_f16_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_f16_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::lastuse.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::fill.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::use.collector::b::fill [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::use [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ret;
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 1, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 1, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=fill(2), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 2, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=use(3), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 3, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=use(3), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 3, i32 0)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f16, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_f16_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_f16_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_f16_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_f16_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_f16_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_f16_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_f16_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_f16_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_f16_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.ashift.collector::a::discard [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ret;
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 2, i32 0, i32 1)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 2, i32 0, i32 1)
+ ; kind=f16(0), cta_group=2, collector_a=lastuse(1), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 2, i32 1, i32 2)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=lastuse(1), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 2, i32 0, i32 1)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=discard(0), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 2, i32 0, i32 0)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 0, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_tf32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_tf32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_tf32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_tf32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_tf32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_tf32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_tf32_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_tf32_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_tf32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::lastuse.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ret;
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=lastuse(1), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 1, i32 0)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f8f6f4, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_f8f6f4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_f8f6f4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_f8f6f4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_f8f6f4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_f8f6f4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_f8f6f4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_f8f6f4_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_f8f6f4_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_f8f6f4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.ashift.collector::a::discard [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ret;
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 2)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 2)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 1, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 1, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=lastuse(1), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=discard(0), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 0)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; tf32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_tf32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_tf32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_tf32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_tf32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_tf32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_tf32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_tf32_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_tf32_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_tf32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ret;
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 2, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 2, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=lastuse(1), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 2, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=2, collector_a=lastuse(1), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 2, i32 1, i32 2)
+ ; kind=tf32(1), cta_group=2, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 2, i32 1, i32 0)
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 1, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; f8f6f4, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_f8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_f8f6f4_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_f8f6f4_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_f8f6f4_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_f8f6f4_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_f8f6f4_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_f8f6f4_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_f8f6f4_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_f8f6f4_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::lastuse.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1;
+; CHECK-NEXT: ret;
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 2)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 2)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=lastuse(1), ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 1, i32 1)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 1, i32 1)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 1, i32 0)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 2, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, f16, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_scale_d_f16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_scale_d_f16_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_scale_d_f16_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_scale_d_f16_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_scale_d_f16_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_scale_d_f16_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_scale_d_f16_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_scale_d_f16_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_scale_d_f16_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::lastuse.collector::b::fill [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.ashift.collector::a::discard [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ ; kind=f16(0), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 0, i32 2)
+ ; kind=f16(0), cta_group=1, scale_d=0, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 1, i32 1)
+ ; tensor, kind=f16(0), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 0, i32 1)
+ ; tensor, kind=f16(0), cta_group=1, scale_d=0, collector_a=lastuse(1), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 1, i32 2)
+ ; tensor+ashift, kind=f16(0), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 0, i32 1)
+ ; tensor+ashift, kind=f16(0), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 0, i32 0)
+ ; kind=f16(0), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, f16, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_scale_d_f16_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_scale_d_f16_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_scale_d_f16_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_scale_d_f16_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_scale_d_f16_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_scale_d_f16_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_scale_d_f16_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_scale_d_f16_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_scale_d_f16_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.ashift.collector::a::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ ; kind=f16(0), cta_group=2, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 2, i32 0, i32 1)
+ ; tensor, kind=f16(0), cta_group=2, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 2, i32 0, i32 1)
+ ; tensor+ashift, kind=f16(0), cta_group=2, scale_d=0, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 2, i32 1, i32 1)
+ ; tensor+ashift, kind=f16(0), cta_group=2, scale_d=0, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 2, i32 1, i32 0)
+ ; kind=f16(0), cta_group=2, scale_d=0, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 0, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, tf32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_scale_d_tf32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_scale_d_tf32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta1_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.ashift.collector::a::discard [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ ; kind=tf32(1), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 1, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=1, scale_d=0, collector_a=lastuse(1), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 1, i32 1, i32 2)
+ ; tensor, kind=tf32(1), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 1, i32 0, i32 1)
+ ; tensor+ashift, kind=tf32(1), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 1, i32 0, i32 1)
+ ; tensor+ashift, kind=tf32(1), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 1, i32 0, i32 0)
+ ; kind=tf32(1), cta_group=1, scale_d=0, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; scale_input_d + collector_b, tf32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_scale_d_tf32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_scale_d_tf32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<5>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta2_param_6];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_scale_d_tf32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.ashift.collector::a::lastuse [%r1], [%r4], %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, %p1, 0;
+; CHECK-NEXT: ret;
+ ; kind=tf32(1), cta_group=2, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 2, i32 0, i32 1)
+ ; tensor, kind=tf32(1), cta_group=2, scale_d=0, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 2, i32 0, i32 1)
+ ; tensor+ashift, kind=tf32(1), cta_group=2, scale_d=0, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 2, i32 1, i32 1)
+ ; tensor+ashift, kind=tf32(1), cta_group=2, scale_d=0, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 2, i32 1, i32 0)
+ ; kind=tf32(1), cta_group=2, scale_d=0, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 0, i32 1, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<7>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_7];
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_8];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ld.param.b32 %r6, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::fill.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::use.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::lastuse.collector::b::fill [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::lastuse [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ret;
+ ; cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ ; cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 2)
+ ; cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 3)
+ ; cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 1)
+ ; cta_group=1, collector_a=fill(2), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 2, i32 3)
+ ; cta_group=1, collector_a=use(3), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 3, i32 2)
+ ; tensor, cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ ; tensor, cta_group=1, collector_a=lastuse(1), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 2)
+ ; tensor, cta_group=1, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 0)
+ ; cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<7>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_7];
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_8];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ld.param.b32 %r6, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::discard.collector::b::lastuse [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::fill.collector::b::use [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::fill [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ret;
+ ; cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ ; cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+ ; tensor, cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ ; tensor, cta_group=2, collector_a=fill(2), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 3)
+ ; tensor, cta_group=2, collector_a=fill(2), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 2, i32 0)
+ ; cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, .block32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<7>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_7];
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_8];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ld.param.b32 %r6, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ret;
+ ; cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ ; cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 1, i32 1)
+ ; tensor, cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 1)
+ ; tensor, cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ; cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; block_scale (section 4), mxf8f6f4, .block32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b) {
+; CHECK-LABEL: tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<7>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_6];
+; CHECK-NEXT: ld.param.b32 %r4, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_7];
+; CHECK-NEXT: ld.param.b32 %r5, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_8];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ld.param.b32 %r6, [tcgen05_mma_sp_collector_b_block_scale_mxf8f6f4_block32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::lastuse.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard.collector::b::lastuse [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], [%r6], %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::mxf8f6f4.block_scale.block32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, [%r4], [%r5], %p1;
+; CHECK-NEXT: ret;
+ ; cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ ; cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 1, i32 1)
+ ; tensor, cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 1)
+ ; tensor, cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ; cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 2, i32 0, i32 0)
+ ret void
+}
diff --git a/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-disable-output-lane-collector-b.ll b/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-disable-output-lane-collector-b.ll
new file mode 100644
index 0000000000000..2e8b45dcafe40
--- /dev/null
+++ b/llvm/test/CodeGen/NVPTX/tcgen05-mma-sp-disable-output-lane-collector-b.ll
@@ -0,0 +1,458 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -o - -mcpu=sm_107f -mattr=+ptx94 -march=nvptx64 %s | FileCheck %s
+; RUN: %if ptxas-sm_107f && ptxas-isa-9.4 %{ llc < %s -march=nvptx64 -mattr=+ptx94 -mcpu=sm_107f | %ptxas-verify -arch=sm_107f %}
+
+; ---------------------------------------------------------------------------
+; sp f16, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4) {
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<9>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ld.param.b32 %r8, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::use [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.ashift.collector::a::discard.collector::b::lastuse [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::lastuse.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::fill.collector::b::use [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::use.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::use [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=lastuse(1), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 1, i32 2)
+ ; kind=f16(0), cta_group=1, collector_a=fill(2), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 2, i32 3)
+ ; kind=f16(0), cta_group=1, collector_a=use(3), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 3, i32 1)
+ ; kind=f16(0), cta_group=1, collector_a=use(3), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 3, i32 0)
+ ; kind=f16(0), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp f16, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8) {
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<13>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_7+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r8, %r9, %r10, %r11}, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ld.param.b32 %r12, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.ashift.collector::a::discard.collector::b::use [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::lastuse.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 1)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 2)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 3)
+ ; kind=f16(0), cta_group=2, collector_a=lastuse(1), collector_b=use(3)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 3)
+ ; kind=f16(0), cta_group=2, collector_a=lastuse(1), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 0)
+ ; kind=f16(0), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp tf32, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4) {
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<9>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ld.param.b32 %r8, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.ashift.collector::a::discard.collector::b::use [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::use.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::use [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 1)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), cta_group=1, collector_a=use(3), collector_b=fill(2)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 1, i32 3, i32 2)
+ ; kind=tf32(1), cta_group=1, collector_a=use(3), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 1, i32 3, i32 0)
+ ; kind=tf32(1), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp tf32, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8) {
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=fill(2)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<13>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_7+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r8, %r9, %r10, %r11}, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ld.param.b32 %r12, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.ashift.collector::a::lastuse [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 0)
+ ; kind=tf32(1), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp f8f6f4, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4) {
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<9>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ld.param.b32 %r8, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.ashift.collector::a::discard.collector::b::use [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::fill.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::fill [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 2)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=fill(2), collector_b=lastuse(1)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 2, i32 2, i32 1)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=fill(2), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 2, i32 2, i32 0)
+ ; kind=f8f6f4(2), cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp f8f6f4, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8) {
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=use(3)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<13>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_7+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r8, %r9, %r10, %r11}, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::use [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ld.param.b32 %r12, [tcgen05_mma_sp_disable_output_lane_collector_b_f8f6f4_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard.collector::b::fill [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.ashift.collector::a::lastuse.collector::b::use [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.ashift.collector::a::lastuse [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f8f6f4.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 2, i32 0, i32 3)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 2, i32 0, i32 2)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=lastuse(1), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 2, i32 1, i32 3)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 2, i32 1, i32 0)
+ ; kind=f8f6f4(2), cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8, i32 2, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp f16 scale_d, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4) {
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<9>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 5;
+; CHECK-NEXT: ld.param.b32 %r8, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 3;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.ashift.collector::a::lastuse.collector::b::use [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 7;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.ashift.collector::a::lastuse [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 7;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 5;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 5, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 1)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 3, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 2)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 7, <4 x i32> %disable_output_lanev4, i32 0, i32 1, i32 3)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 7, <4 x i32> %disable_output_lanev4, i32 0, i32 1, i32 0)
+ ; kind=f16(0), scale_d, cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 5, <4 x i32> %disable_output_lanev4, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp tf32 scale_d, cta_group::1
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lanev4) {
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=discard(0), collector_b=fill(2)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<9>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 4;
+; CHECK-NEXT: ld.param.b32 %r8, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta1_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 6;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 2;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.ashift.collector::a::lastuse [%r1], [%r8], %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 2;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::1.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r4, %r5, %r6, %r7}, %p1, 4;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 4, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 6, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 2, <4 x i32> %disable_output_lanev4, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 2, <4 x i32> %disable_output_lanev4, i32 1, i32 1, i32 0)
+ ; kind=tf32(1), scale_d, cta_group=1, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 4, <4 x i32> %disable_output_lanev4, i32 1, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp f16 scale_d, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8) {
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=discard(0), collector_b=lastuse(1)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<13>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_7+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r8, %r9, %r10, %r11}, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::lastuse [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 5;
+; CHECK-NEXT: ld.param.b32 %r12, [tcgen05_mma_sp_disable_output_lane_collector_b_f16_scale_d_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard.collector::b::fill [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 3;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.ashift.collector::a::lastuse.collector::b::use [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 7;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.ashift.collector::a::lastuse [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 7;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::f16.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 5;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 5, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 1)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=discard(0), collector_b=fill(2), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 3, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 2)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=use(3), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 7, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 3)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 7, <8 x i32> %disable_output_lanev8, i32 0, i32 1, i32 0)
+ ; kind=f16(0), scale_d, cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 5, <8 x i32> %disable_output_lanev8, i32 0, i32 0, i32 0)
+ ret void
+}
+
+; ---------------------------------------------------------------------------
+; sp tf32 scale_d, cta_group::2
+; ---------------------------------------------------------------------------
+
+define void @tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lanev8) {
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=discard(0), collector_b=fill(2)
+; CHECK-LABEL: tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2(
+; CHECK: {
+; CHECK-NEXT: .reg .pred %p<2>;
+; CHECK-NEXT: .reg .b16 %rs<3>;
+; CHECK-NEXT: .reg .b32 %r<13>;
+; CHECK-NEXT: .reg .b64 %rd<3>;
+; CHECK-EMPTY:
+; CHECK-NEXT: // %bb.0:
+; CHECK-NEXT: ld.param.b8 %rs1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_5];
+; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
+; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
+; CHECK-NEXT: ld.param.b32 %r1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_0];
+; CHECK-NEXT: ld.param.b64 %rd1, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_2];
+; CHECK-NEXT: ld.param.b64 %rd2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_3];
+; CHECK-NEXT: ld.param.b32 %r2, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_4];
+; CHECK-NEXT: ld.param.b32 %r3, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_6];
+; CHECK-NEXT: ld.param.v4.b32 {%r4, %r5, %r6, %r7}, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_7+16];
+; CHECK-NEXT: ld.param.v4.b32 {%r8, %r9, %r10, %r11}, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_7];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::fill [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 4;
+; CHECK-NEXT: ld.param.b32 %r12, [tcgen05_mma_sp_disable_output_lane_collector_b_tf32_scale_d_cta2_param_1];
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard.collector::b::use [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 6;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.ashift.collector::a::lastuse.collector::b::lastuse [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 2;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.ashift.collector::a::lastuse [%r1], [%r12], %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 2;
+; CHECK-NEXT: tcgen05.mma.sp.cta_group::2.kind::tf32.collector::a::discard [%r1], %rd1, %rd2, [%r3], %r2, {%r8, %r9, %r10, %r11, %r4, %r5, %r6, %r7}, %p1, 4;
+; CHECK-NEXT: ret;
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 4, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 2)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=discard(0), collector_b=use(3), tensor
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 6, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 3)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=lastuse(1), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 2, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 1)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=lastuse(1), collector_b=discard(0), tensor.ashift
+ call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 2, <8 x i32> %disable_output_lanev8, i32 1, i32 1, i32 0)
+ ; kind=tf32(1), scale_d, cta_group=2, collector_a=discard(0), collector_b=discard(0)
+ call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %dtmem, i64 %ashared, i64 %b, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 4, <8 x i32> %disable_output_lanev8, i32 1, i32 0, i32 0)
+ ret void
+}
diff --git a/llvm/test/CodeGen/NVPTX/tcgen05-mma-tensor-formatted.ll b/llvm/test/CodeGen/NVPTX/tcgen05-mma-tensor-formatted.ll
deleted file mode 100644
index 479de53dd90f2..0000000000000
--- a/llvm/test/CodeGen/NVPTX/tcgen05-mma-tensor-formatted.ll
+++ /dev/null
@@ -1,50 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
-; NOTE: This sample test demonstrates the pretty print feature for NVPTX intrinsics
-; RUN: llvm-as < %s | llvm-dis | FileCheck %s
-
-target triple = "nvptx64-nvidia-cuda"
-
-define void @tcgen05_mma_fp16_cta1(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d) {
- ; CHECK-LABEL: define void @tcgen05_mma_fp16_cta1(
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 0)
-
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=lastuse */ i32 1)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 1)
-
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=fill */ i32 2)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 2)
-
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=use */ i32 3)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 0, i32 1, i32 3)
-
- ret void
-}
-
-define void @tcgen05_mma_f8f6f4_cta2(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d) {
- ; CHECK-LABEL: define void @tcgen05_mma_f8f6f4_cta2(
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=discard */ i32 0)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 0)
-
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=lastuse */ i32 1)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 1)
-
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=fill */ i32 2)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 2)
-
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=use */ i32 3)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 %b, i32 %idesc, i1 %enable_inp_d, i32 2, i32 2, i32 3)
-
- ret void
-}
-
-; This test verifies that printImmArg is safe to call on all constant arguments, but only prints comments for arguments that have pretty printing configured.
-define void @test_mixed_constants_edge_case(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor) {
- ; CHECK-LABEL: define void @test_mixed_constants_edge_case(
- ; CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 42, i32 100, i1 true, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
- call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) %dtmem, ptr addrspace(6) %atensor, i64 42, i32 100, i1 true, i32 3, i32 1, i32 0)
-
- ret void
-}
-
-declare void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6), ptr addrspace(6), i64, i32, i1, i32, i32, i32)
diff --git a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
index 75bc7527aba52..d524f8aa99039 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
@@ -6104,7 +6104,7 @@ defvar Tcgen05MMABlockScaleKindAttr =
[EnumAttrIsOneOf<Tcgen05MMAKindAttr, Tcgen05MMABlockScaleKindList>]>;
def NVVM_Tcgen05MMAOp : NVVM_Op<"tcgen05.mma",
- [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110]>]> {
+ [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110, 107]>]> {
let summary = "Performs MMA operation on 5th-gen tensor cores";
let description = [{
@@ -6144,6 +6144,7 @@ def NVVM_Tcgen05MMAOp : NVVM_Op<"tcgen05.mma",
Default Attributes:
- collectorOp is a Tcgen05MMACollectorOp attribute with matrix A as the collector buffer
+ - collectorOpB is a Tcgen05MMACollectorOp attribute with matrix B as the collector buffer
- `aShift` shifts the rows of the A matrix down by one row and can only be
applied if A is in tensor memory
@@ -6156,6 +6157,8 @@ def NVVM_Tcgen05MMAOp : NVVM_Op<"tcgen05.mma",
CTAGroupKindAttr:$ctaGroup,
DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
"Tcgen05MMACollectorOp::DISCARD">:$collectorOp,
+ DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
+ "Tcgen05MMACollectorOp::DISCARD">:$collectorOpB,
UnitAttr:$aShift,
LLVM_PointerTensor:$matrixD,
AnyTypeOf<[LLVM_PointerTensor, I64]>:$matrixA,
@@ -6187,7 +6190,7 @@ def NVVM_Tcgen05MMAOp : NVVM_Op<"tcgen05.mma",
}
def NVVM_Tcgen05MMASparseOp : NVVM_Op<"tcgen05.mma.sp",
- [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110]>]> {
+ [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110, 107]>]> {
let summary = "Performs MMA operation with sparse A matrix on 5th-gen tensor cores";
let description = [{
@@ -6219,6 +6222,8 @@ def NVVM_Tcgen05MMASparseOp : NVVM_Op<"tcgen05.mma.sp",
CTAGroupKindAttr:$ctaGroup,
DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
"Tcgen05MMACollectorOp::DISCARD">:$collectorOp,
+ DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
+ "Tcgen05MMACollectorOp::DISCARD">:$collectorOpB,
UnitAttr:$aShift,
LLVM_PointerTensor:$matrixD,
AnyTypeOf<[LLVM_PointerTensor, I64]>:$matrixA,
@@ -6268,7 +6273,7 @@ def Tcgen05MMABlockScaleAttr : EnumAttr<NVVM_Dialect, Tcgen05MMABlockScale,
}
def NVVM_Tcgen05MMABlockScaleOp : NVVM_Op<"tcgen05.mma.block_scale",
- [NVVMRequiresSMf<[100, 101, 110]>]> {
+ [NVVMRequiresSMf<[100, 101, 110, 107]>]> {
let summary = "Performs block scaled MMA operation on 5th-gen tensor cores";
let description = [{
@@ -6301,6 +6306,7 @@ def NVVM_Tcgen05MMABlockScaleOp : NVVM_Op<"tcgen05.mma.block_scale",
Default Attributes:
- collectorOp is a Tcgen05MMACollectorOp attribute with matrix A as the collector buffer
+ - collectorOpB is a Tcgen05MMACollectorOp attribute with matrix B as the collector buffer
[For more information, see PTX ISA](https://docs.nvidia.com/cuda/parallel-thread-execution/#tcgen05-mma-instructions-mma)
}];
@@ -6312,6 +6318,8 @@ def NVVM_Tcgen05MMABlockScaleOp : NVVM_Op<"tcgen05.mma.block_scale",
"Tcgen05MMABlockScale::DEFAULT">:$blockScale,
DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
"Tcgen05MMACollectorOp::DISCARD">:$collectorOp,
+ DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
+ "Tcgen05MMACollectorOp::DISCARD">:$collectorOpB,
LLVM_PointerTensor:$matrixD,
AnyTypeOf<[LLVM_PointerTensor, I64]>:$matrixA,
I64:$matrixB,
@@ -6341,7 +6349,7 @@ def NVVM_Tcgen05MMABlockScaleOp : NVVM_Op<"tcgen05.mma.block_scale",
}
def NVVM_Tcgen05MMASparseBlockScaleOp : NVVM_Op<"tcgen05.mma.sp.block_scale",
- [NVVMRequiresSMf<[100, 101, 110]>]> {
+ [NVVMRequiresSMf<[100, 101, 110, 107]>]> {
let summary = "Performs block scaled MMA operation with sparse A matrix on 5th-gen tensor cores";
let description = [{
@@ -6374,6 +6382,8 @@ def NVVM_Tcgen05MMASparseBlockScaleOp : NVVM_Op<"tcgen05.mma.sp.block_scale",
"Tcgen05MMABlockScale::DEFAULT">:$blockScale,
DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
"Tcgen05MMACollectorOp::DISCARD">:$collectorOp,
+ DefaultValuedAttr<Tcgen05MMACollectorOpAttr,
+ "Tcgen05MMACollectorOp::DISCARD">:$collectorOpB,
LLVM_PointerTensor:$matrixD,
AnyTypeOf<[LLVM_PointerTensor, I64]>:$matrixA,
I64:$matrixB,
diff --git a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
index a7793ddbaf7e8..83dbc33d4f618 100644
--- a/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
+++ b/mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
@@ -5694,6 +5694,9 @@ Tcgen05MMAOp::getIntrinsicIDAndArgs(Operation &op, LLVM::ModuleTranslation &mt,
args.push_back(
builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOp())));
+ args.push_back(
+ builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOpB())));
+
return {ID, args};
}
@@ -5879,6 +5882,9 @@ mlir::NVVM::IDArgPair Tcgen05MMASparseOp::getIntrinsicIDAndArgs(
args.push_back(
builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOp())));
+ args.push_back(
+ builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOpB())));
+
return {ID, args};
}
@@ -5913,6 +5919,8 @@ mlir::NVVM::IDArgPair Tcgen05MMABlockScaleOp::getIntrinsicIDAndArgs(
static_cast<unsigned>(getNVVMCtaGroupKind(thisOp.getCtaGroup()))));
args.push_back(
builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOp())));
+ args.push_back(
+ builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOpB())));
auto kind = thisOp.getKind();
auto blockScale = thisOp.getBlockScale();
@@ -6010,6 +6018,8 @@ mlir::NVVM::IDArgPair Tcgen05MMASparseBlockScaleOp::getIntrinsicIDAndArgs(
static_cast<unsigned>(getNVVMCtaGroupKind(thisOp.getCtaGroup()))));
args.push_back(
builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOp())));
+ args.push_back(
+ builder.getInt32(static_cast<unsigned>(thisOp.getCollectorOpB())));
auto kind = thisOp.getKind();
auto blockScale = thisOp.getBlockScale();
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-shared.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-shared.mlir
index 99e3fe45a51b7..620080f3fbad6 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-shared.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-shared.mlir
@@ -4,35 +4,35 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -42,35 +42,35 @@ llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -80,35 +80,35 @@ llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -118,35 +118,35 @@ llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_de
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -156,35 +156,35 @@ llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_de
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -194,37 +194,103 @@ llvm.func @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_block_scale_collector_b
+llvm.func @nvvm_tcgen05_mma_block_scale_collector_b(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-tensor.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-tensor.mlir
index 5138160201103..ff8c64e8d2cc0 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-tensor.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-block-scale-tensor.mlir
@@ -4,35 +4,35 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -42,35 +42,35 @@ llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -80,35 +80,35 @@ llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -118,35 +118,35 @@ llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_tm
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -156,35 +156,35 @@ llvm.func @nvvm_tcgen05_mma_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_tm
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -194,37 +194,103 @@ llvm.func @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %
// CHECK-LABEL: @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_mxf4nvf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_block_scale_collector_b
+llvm.func @nvvm_tcgen05_mma_block_scale_collector_b(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 1)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 2)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 3)
+ nvvm.tcgen05.mma.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-shared.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-shared.mlir
index 286df36730e77..7957fba879574 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-shared.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-shared.mlir
@@ -3,67 +3,67 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_cta_1
llvm.func @nvvm_tcgen05_mma_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
@@ -73,67 +73,67 @@ llvm.func @nvvm_tcgen05_mma_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc:
// CHECK-LABEL: @nvvm_tcgen05_mma_cta_2
llvm.func @nvvm_tcgen05_mma_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
@@ -146,35 +146,35 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
@@ -186,35 +186,35 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64)
@@ -224,67 +224,67 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i
// CHECK-LABEL: @nvvm_tcgen05_mma_disable_output_lane_cta_1
llvm.func @nvvm_tcgen05_mma_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane : vector<4 x i32>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<4 x i32>)
@@ -294,67 +294,67 @@ llvm.func @nvvm_tcgen05_mma_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>, %a
// CHECK-LABEL: @nvvm_tcgen05_mma_disable_output_lane_cta_2
llvm.func @nvvm_tcgen05_mma_disable_output_lane_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane: vector<8 x i32>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, vector<8 x i32>)
@@ -366,35 +366,35 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_disable_output_lane_cta_1(%d_tmem : !llv
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<4 x i32>)
@@ -406,37 +406,103 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_disable_output_lane_cta_2(%d_tmem : !llv
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, i64, vector<8 x i32>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_collector_b
+llvm.func @nvvm_tcgen05_mma_collector_b(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1)
+
+ llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-shared.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-shared.mlir
index a683ec96d11a9..f7f85b50b2325 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-shared.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-shared.mlir
@@ -4,35 +4,35 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -42,35 +42,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -80,35 +80,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -118,35 +118,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -156,35 +156,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -194,37 +194,103 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_sp_block_scale_collector_b
+llvm.func @nvvm_tcgen05_mma_sp_block_scale_collector_b(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-tensor.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-tensor.mlir
index a5db4f2bbac4b..b4c0efbf3f8ca 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-tensor.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-block-scale-tensor.mlir
@@ -4,35 +4,35 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -42,35 +42,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -80,35 +80,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf8f6f4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -118,35 +118,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -156,35 +156,35 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_1
llvm.func @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_1>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
@@ -194,37 +194,103 @@ llvm.func @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_2
llvm.func @nvvm_tcgen05_mma_sp_mxf4nvf4_block_scale_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, blockScale = #nvvm.tcgen05_mma_block_scale<block16>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
{kind = #nvvm.tcgen05_mma_kind<mxf4nvf4>, ctaGroup = #nvvm.cta_group<cta_2>, blockScale = #nvvm.tcgen05_mma_block_scale<block32>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_sp_block_scale_collector_b
+llvm.func @nvvm_tcgen05_mma_sp_block_scale_collector_b(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>, %spmetadata: !llvm.ptr<6>) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 1)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 2)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 3, i32 3)
+ nvvm.tcgen05.mma.sp.block_scale %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata, %scale_a, %scale_b
+ {kind = #nvvm.tcgen05_mma_kind<mxf8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, !llvm.ptr<6>, !llvm.ptr<6>)
+
+ llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-shared.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-shared.mlir
index 96044cf669d63..2df9160c16ee5 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-shared.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-shared.mlir
@@ -3,67 +3,67 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_cta_1
llvm.func @nvvm_tcgen05_mma_sp_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
@@ -73,67 +73,67 @@ llvm.func @nvvm_tcgen05_mma_sp_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %ades
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_cta_2
llvm.func @nvvm_tcgen05_mma_sp_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
@@ -146,35 +146,35 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
@@ -186,35 +186,35 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64)
@@ -224,67 +224,67 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_disable_output_lane_cta_1
llvm.func @nvvm_tcgen05_mma_sp_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane : vector<4 x i32>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
@@ -294,67 +294,67 @@ llvm.func @nvvm_tcgen05_mma_sp_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>,
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_disable_output_lane_cta_2
llvm.func @nvvm_tcgen05_mma_sp_disable_output_lane_cta_2(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane: vector<8 x i32>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
@@ -366,35 +366,35 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_disable_output_lane_cta_1(%d_tmem : !
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
@@ -406,37 +406,103 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_disable_output_lane_cta_2(%d_tmem : !
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_sp_collector_b
+llvm.func @nvvm_tcgen05_mma_sp_collector_b(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %spmetadata: !llvm.ptr<6>) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_desc, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, i64, i64, i32, i1, !llvm.ptr<6>)
+
+ llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-tensor.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-tensor.mlir
index 709beb0508bb8..cf8fc7628b172 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-tensor.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-sp-tensor.mlir
@@ -3,99 +3,99 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_cta_1
llvm.func @nvvm_tcgen05_mma_sp_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
@@ -105,99 +105,99 @@ llvm.func @nvvm_tcgen05_mma_sp_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_cta_2
llvm.func @nvvm_tcgen05_mma_sp_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 2, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 3, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
@@ -210,51 +210,51 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
@@ -266,51 +266,51 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64)
@@ -320,99 +320,99 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_disable_output_lane_cta_1
llvm.func @nvvm_tcgen05_mma_sp_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane : vector<4 x i32>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<4 x i32>)
@@ -422,99 +422,99 @@ llvm.func @nvvm_tcgen05_mma_sp_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>,
// CHECK-LABEL: @nvvm_tcgen05_mma_sp_disable_output_lane_cta_2
llvm.func @nvvm_tcgen05_mma_sp_disable_output_lane_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane: vector<8 x i32>, %spmetadata: !llvm.ptr<6>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, vector<8 x i32>)
@@ -526,51 +526,51 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_disable_output_lane_cta_1(%d_tmem : !
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<4 x i32>)
@@ -582,53 +582,119 @@ llvm.func @nvvm_tcgen05_mma_sp_scale_d_imm_disable_output_lane_cta_2(%d_tmem : !
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>, i64, vector<8 x i32>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_sp_collector_b
+llvm.func @nvvm_tcgen05_mma_sp_collector_b(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %spmetadata: !llvm.ptr<6>) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 1)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 2)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 2, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.sp.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 0, i32 1, i32 3, i32 3)
+ nvvm.tcgen05.mma.sp %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d, %spmetadata
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, !llvm.ptr<6>)
+
+ llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-tensor.mlir b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-tensor.mlir
index 798e311778beb..aae9f4696d990 100644
--- a/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-tensor.mlir
+++ b/mlir/test/Target/LLVMIR/nvvm/tcgen05-mma-tensor.mlir
@@ -3,99 +3,99 @@
// CHECK-LABEL: @nvvm_tcgen05_mma_cta_1
llvm.func @nvvm_tcgen05_mma_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
@@ -105,99 +105,99 @@ llvm.func @nvvm_tcgen05_mma_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>,
// CHECK-LABEL: @nvvm_tcgen05_mma_cta_2
llvm.func @nvvm_tcgen05_mma_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector=discard */ i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector_a=discard */ i32 0, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector=lastuse */ i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector_a=lastuse */ i32 1, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 2, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i32 3, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector=fill */ i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector_a=fill */ i32 2, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 2, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=tf32 */ i32 1, /* cta_group= */ i32 2, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f8f6f4 */ i32 2, /* cta_group= */ i32 2, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector=use */ i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=i8 */ i32 3, /* cta_group= */ i32 2, /* collector_a=use */ i32 3, /* collector_b=discard */ i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
@@ -209,51 +209,51 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
@@ -265,51 +265,51 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 0, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, i32 1, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64)
@@ -319,99 +319,99 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !
// CHECK-LABEL: @nvvm_tcgen05_mma_disable_output_lane_cta_1
llvm.func @nvvm_tcgen05_mma_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane : vector<4 x i32>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <4 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<4 x i32>)
@@ -421,99 +421,99 @@ llvm.func @nvvm_tcgen05_mma_disable_output_lane_cta_1(%d_tmem : !llvm.ptr<6>, %a
// CHECK-LABEL: @nvvm_tcgen05_mma_disable_output_lane_cta_2
llvm.func @nvvm_tcgen05_mma_disable_output_lane_cta_2(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %disableOutputLane: vector<8 x i32>) {
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 2, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f8f6f4>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, <8 x i32> {{%[0-9]+}}, i32 3, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<i8>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, vector<8 x i32>)
@@ -525,51 +525,51 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_disable_output_lane_cta_1(%d_tmem : !llv
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <4 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<4 x i32>)
@@ -581,53 +581,119 @@ llvm.func @nvvm_tcgen05_mma_scale_d_imm_disable_output_lane_cta_2(%d_tmem : !llv
%scale_d_imm = llvm.mlir.constant(0:i64) : i64
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 0, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2.ashift(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 1, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, aShift} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 2, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 0, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
- // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, i64 0, <8 x i32> {{%[0-9]+}}, i32 1, i32 3, i32 0)
nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d scale = %scale_d_imm mask = %disableOutputLane
{kind = #nvvm.tcgen05_mma_kind<tf32>, ctaGroup = #nvvm.cta_group<cta_2>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1, i64, vector<8 x i32>)
llvm.return
}
+
+// CHECK-LABEL: @nvvm_tcgen05_mma_collector_b
+llvm.func @nvvm_tcgen05_mma_collector_b(%d_tmem : !llvm.ptr<6>, %a_tmem: !llvm.ptr<6>, %b_desc: i64, %idesc: i32, %enable_input_d: i1) {
+
+ // collector_a=discard(0), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=lastuse */ i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=lastuse(1), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=lastuse */ i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=fill(2), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=fill */ i32 2, /* collector_b=lastuse */ i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=use(3), collector_b=lastuse(1)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=lastuse */ i32 1)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<lastuse>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=discard(0), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=fill */ i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=lastuse(1), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=fill */ i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=fill(2), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=fill */ i32 2, /* collector_b=fill */ i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=use(3), collector_b=fill(2)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=fill */ i32 2)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<fill>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=discard(0), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=discard */ i32 0, /* collector_b=use */ i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=lastuse(1), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=lastuse */ i32 1, /* collector_b=use */ i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<lastuse>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=fill(2), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=fill */ i32 2, /* collector_b=use */ i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<fill>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ // collector_a=use(3), collector_b=use(3)
+ // CHECK: call void @llvm.nvvm.tcgen05.mma.tensor(ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, /* kind=f16 */ i32 0, /* cta_group= */ i32 1, /* collector_a=use */ i32 3, /* collector_b=use */ i32 3)
+ nvvm.tcgen05.mma %d_tmem, %a_tmem, %b_desc, %idesc, %enable_input_d
+ {kind = #nvvm.tcgen05_mma_kind<f16>, ctaGroup = #nvvm.cta_group<cta_1>, collectorOp = #nvvm.tcgen05_mma_collectorop<use>, collectorOpB = #nvvm.tcgen05_mma_collectorop<use>} : (!llvm.ptr<6>, !llvm.ptr<6>, i64, i32, i1)
+
+ llvm.return
+}
>From 9e08d4ddaa3491766dc0bd9a1954f3cf33130b6b Mon Sep 17 00:00:00 2001
From: Kirill Vedernikov <kvedernikov at nvidia.com>
Date: Wed, 5 Aug 2026 23:15:25 +0200
Subject: [PATCH 2/5] [MLIR][NVPTX] Address tcgen05.mma collector_b feedback
---
llvm/docs/NVPTXUsage.md | 218 ++++++++++----------
mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td | 8 +-
2 files changed, 114 insertions(+), 112 deletions(-)
diff --git a/llvm/docs/NVPTXUsage.md b/llvm/docs/NVPTXUsage.md
index f0a1439b30334..543d78daae307 100644
--- a/llvm/docs/NVPTXUsage.md
+++ b/llvm/docs/NVPTXUsage.md
@@ -2940,10 +2940,14 @@ The sparsity metadata (`%spmetadata`) as well as the block-scale inputs for
`A / B` matrices (`%scale_a` and `%scale_b`) reside in Tensor Memory.
To facilitate opportunistic re-use of `A / B` matrix data across a sequence of
-MMA operations, the `A/B` matrices are loaded into a collector buffer
-(`%collector_usage_a_op_flag`, `%collector_usage_b_buffer_flag`, and
-`%collector_usage_b_op_flag`). The flag value of the collector_usage flag in the
-intrinsic specifies the nature of the re-use
+MMA operations, the `A/B` matrices are loaded into collector buffers. The
+`tcgen05.mma`, `tcgen05.mma.disable_output_lane`, and
+`tcgen05.mma.block_scale` intrinsics use `%collector_usage_a_op_flag` and
+`%collector_usage_b_op_flag` to describe the usage of matrix `A` and matrix `B`
+collector buffers. The `tcgen05.mma.ws` intrinsics use
+`%collector_usage_b_buffer_flag` and `%collector_usage_b_op_flag` to describe
+the matrix `B` collector buffer. The flag value of the collector usage flag in
+the intrinsic specifies the nature of the re-use.
There are three kinds of matrix descriptors used by the tcgen05 family of
instructions:
@@ -2985,7 +2989,8 @@ reuse a previously loaded `A` or `B` matrix; however reuse is opportunistic in
that the TensorCore may reload a matrix even when it has permission to reuse
that matrix. Thus, the source memory of an A or B matrix must not be modified
while the MMA instruction using those matrices has not completed - regardless of
-collector qualifier permissions.
+collector qualifier permissions. Values other than `DISCARD` for the matrix `B`
+collector usage require support for the `.collector::b::...` PTX qualifier.
The `cta_group::1` specifies that the operation is performed on the Tensor
Memory of the executing thread’s CTA only. The `cta_group::2` specifies that the
@@ -3007,19 +3012,19 @@ through intrinsics is highly complex, and the following table outlines the large
number of intrinsics required to fully support the `tcgen05.mma` instruction
set.
-| variant | Configuration | Total Variants |
-| ---------------------------------- | ------------------------------------------------------------------------------------------------- | -------------- |
-| tcgen05.mma.shared | 2 (space) x 2 (sp) x 4 (kind) x 2 (cta_group) x 4 (collector_usage) | 128 |
-| tcgen05.mma.tensor.ashift | 2 (sp) x 4 (kind) x 2 (cta_group) x 2 (collector_usage) | 32 |
-| tcgen05.mma.scale_d | 2 (space) x 2 (sp) x 2 (kind) x 2 (cta_group) x 4 (collector_usage) | 128 |
-| tcgen05.mma.scale_d.tensor.ashift | 2 (sp) x 2 (kind) x 2 (cta_group) x 2 (collector_usage) | 16 |
-| tcgen05.mma.disable_output_lane | 2 (space) x 2 (sp) x 4 (kind) x 2 (cta_group) x 4 (collector_usage) | 128 |
-| tcgen05.mma.disable_output_lane... | 2 (sp) x 4 (kind) x 2 (cta_group) x 2 (collector_usage) | 32 |
-| tcgen05.mma.block_scale | 2 (space) x 1 (mxf4nvf4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage) | 32 |
-| tcgen05.mma.block_scale | 2 (space) x 1 (mxf4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage) | 32 |
-| tcgen05.mma.block_scale | 2 (space) x 1 (mxf8f6f4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage) | 32 |
-| tcgen05.mma.ws | 2 (space) x 2 (sp) x 4 (kind) x 2 (zero_col_mask) x 4 (collector_usage_op) x 4 (collector_buffer) | 256 |
-| Total | | 816 |
+| variant | Configuration | Total Variants |
+| ---------------------------------- | ------------------------------------------------------------------------------------------------------------- | -------------- |
+| tcgen05.mma.shared | 2 (space) x 2 (sp) x 4 (kind) x 2 (cta_group) x 4 (collector_usage_a) x 4 (collector_usage_b) | 512 |
+| tcgen05.mma.tensor.ashift | 2 (sp) x 4 (kind) x 2 (cta_group) x 2 (collector_usage_a) x 4 (collector_usage_b) | 128 |
+| tcgen05.mma.scale_d | 2 (space) x 2 (sp) x 2 (kind) x 2 (cta_group) x 4 (collector_usage_a) x 4 (collector_usage_b) | 512 |
+| tcgen05.mma.scale_d.tensor.ashift | 2 (sp) x 2 (kind) x 2 (cta_group) x 2 (collector_usage_a) x 4 (collector_usage_b) | 64 |
+| tcgen05.mma.disable_output_lane | 2 (space) x 2 (sp) x 4 (kind) x 2 (cta_group) x 4 (collector_usage_a) x 4 (collector_usage_b) | 512 |
+| tcgen05.mma.disable_output_lane... | 2 (sp) x 4 (kind) x 2 (cta_group) x 2 (collector_usage_a) x 4 (collector_usage_b) | 128 |
+| tcgen05.mma.block_scale | 2 (space) x 1 (mxf4nvf4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage_a) x 4 (collector_usage_b) | 128 |
+| tcgen05.mma.block_scale | 2 (space) x 1 (mxf4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage_a) x 4 (collector_usage_b) | 128 |
+| tcgen05.mma.block_scale | 2 (space) x 1 (mxf8f6f4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage_a) x 4 (collector_usage_b) | 128 |
+| tcgen05.mma.ws | 2 (space) x 2 (sp) x 4 (kind) x 2 (zero_col_mask) x 4 (collector_usage_op) x 4 (collector_buffer) | 256 |
+| Total | | 2496 |
To reduce the number of possible intrinsic variations, we've modeled the
`tcgen05.mma` instructions using flag operands. We've added range checks to
@@ -3031,20 +3036,20 @@ intrinsic modifiers to avoid supporting invalid combinations of features.
##### Syntax:
```llvm
-declare void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; .sp variants
-declare void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; .scale_d variants
-declare void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, i32 %cta_group_flag, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.scale_d<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, i32 %cta_group_flag, i32 %kind_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.scale_d(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.scale_d<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; sp.scale_d variants
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i64 %scale_d_imm, i32 %cta_group_flag, i32 %collector_usage_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i64 %scale_d_imm, i32 %cta_group, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 %scale_d_imm, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 %scale_d_imm, i32 %kind_flag, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
```
##### Overview:
@@ -3071,9 +3076,10 @@ specifying an additional `%spmetadata` argument
`.ashift` shifts the rows of the A matrix down by one row, except for the last
row in the Tensor Memory. `.ashift` is only allowed with M = 128 or M = 256.
-The `%collector_usage_a_op_flag` flag specifies the usage of collector buffer
-for matrix `A`. It is illegal to specify either of `USE` or `FILL` for
-`%collector_usage_a_op_flag` along with `.ashift`
+The `%collector_usage_a_op_flag` and `%collector_usage_b_op_flag` flags specify
+the usage of collector buffers for matrix `A` and matrix `B`, respectively. It
+is illegal to specify either of `USE` or `FILL` for
+`%collector_usage_a_op_flag` along with `.ashift`.
For more information, refer to the
[PTX ISA](https://docs.nvidia.com/cuda/parallel-thread-execution/#tcgen05-mma-instructions-mma)
@@ -3096,14 +3102,14 @@ The following tables describe the possible values of the flag arguments
| CG1 | 1 |
| CG2 | 2 |
-`%collector_usage_a_op_flag` flag:
+`%collector_usage_a_op_flag` and `%collector_usage_b_op_flag` flags:
-| `collector_usage_a_op_flag` | value |
-| --------------------------- | ----- |
-| DISCARD | 0 |
-| LASTUSE | 1 |
-| USE | 2 |
-| FILL | 3 |
+| collector usage flag | value |
+| -------------------- | ----- |
+| DISCARD | 0 |
+| LASTUSE | 1 |
+| FILL | 2 |
+| USE | 3 |
#### '`llvm.nvvm.tcgen05.mma.block_scale*`'
@@ -3111,34 +3117,34 @@ The following tables describe the possible values of the flag arguments
```llvm
; mxf8f6f4
-declare void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf8f6f4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; mxf4
-declare void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.mxf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; mxf4nvf4
-declare void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 cta_group_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block16(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.mxf4nvf4.block_scale.block32(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, ptr addrspace(6) %scale_a, ptr addrspace(6) %scale_b, i32 %cta_group_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
```
##### Overview:
@@ -3161,57 +3167,57 @@ result in the initiation of the whole matrix multiply and accumulate operation.
When `.sp` is specified, the dimension of A matrix is `M x (K / 2)` and requires
specifying an additional `%spmetadata` argument
-The `%collector_usage_a_op_flag` flag specifies the usage of collector buffer
-for matrix `A`
+The `%collector_usage_a_op_flag` and `%collector_usage_b_op_flag` flags specify
+the usage of collector buffers for matrix `A` and matrix `B`, respectively.
For more information, refer to the
[PTX ISA](https://docs.nvidia.com/cuda/parallel-thread-execution/#tcgen05-mma-instructions-mma)
The following tables describe the possible values of the flag arguments
-`%cta_group`:
+`%cta_group_flag`:
-| `cta_group` | value |
-| ----------- | ----- |
-| CG1 | 1 |
-| CG2 | 2 |
+| `cta_group_flag` | value |
+| ---------------- | ----- |
+| CG1 | 1 |
+| CG2 | 2 |
-`%collector_usage_a_op_flag`:
+`%collector_usage_a_op_flag` and `%collector_usage_b_op_flag`:
-| `collector_usage_a_op_flag` | value |
-| --------------------------- | ----- |
-| DISCARD | 0 |
-| LASTUSE | 1 |
-| USE | 2 |
-| FILL | 3 |
+| collector usage flag | value |
+| -------------------- | ----- |
+| DISCARD | 0 |
+| LASTUSE | 1 |
+| FILL | 2 |
+| USE | 3 |
#### '`llvm.nvvm.tcgen05.mma.disable_output_lane*`'
##### Syntax:
```llvm
-declare void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; .sp variants
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; .scale_d variants
-declare void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.tensor.scale_d.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
; .sp.scale_d variants
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag)
-declare void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, ptr addrspace(6) %spmetadata, i1 %enable_inp_d, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg1(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.shared.scale_d.disable_output_lane.cg2(ptr addrspace(6) %d, i64 %adesc, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg1<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 %scale_d_imm, <4 x i32> %disable_output_lane_v4, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
+declare void @llvm.nvvm.tcgen05.mma.sp.tensor.scale_d.disable_output_lane.cg2<.ashift>(ptr addrspace(6) %d, ptr addrspace(6) %atensor, i64 %bdesc, i32 %idesc, i1 %enable_inp_d, ptr addrspace(6) %spmetadata, i64 %scale_d_imm, <8 x i32> %disable_output_lane_v8, i32 %kind_flag, i32 %collector_usage_a_op_flag, i32 %collector_usage_b_op_flag)
```
##### Overview:
@@ -3247,9 +3253,10 @@ specifiying an additional `%spmetadata` argument.
> `.ashift` shifts the rows of the A matrix down by one row, except for the last
> row in the Tensor Memory. `.ashift` is only allowed with M = 128 or M = 256.
-The `%collector_usage_a_op_flag` flag specifies the usage of collector buffer
-for matrix `A`. It is illegal to specify either of `USE` or `FILL` for
-`%collector_usage_a_op_flag` along with `.ashift`
+The `%collector_usage_a_op_flag` and `%collector_usage_b_op_flag` flags specify
+the usage of collector buffers for matrix `A` and matrix `B`, respectively. It
+is illegal to specify either of `USE` or `FILL` for
+`%collector_usage_a_op_flag` along with `.ashift`.
For more information, refer to the [PTX ISA](https://docs.nvidia.com/cuda/parallel-thread-execution/#tcgen05-mma-instructions-mma)
@@ -3264,21 +3271,16 @@ The following tables describes the possible values of the flag arguments
| F8F6F4 | 2 |
| I8 | 3 |
-`%cta_group_flag`:
-
-| `cta_group_flag` | value |
-| ---------------- | ----- |
-| CG1 | 1 |
-| CG2 | 2 |
+The `cg1` and `cg2` suffixes select CTA group 1 and CTA group 2, respectively.
-`%collector_usage_a_op_flag`:
+`%collector_usage_a_op_flag` and `%collector_usage_b_op_flag`:
-| `collector_usage_a_op_flag` | value |
-| --------------------------- | ----- |
-| DISCARD | 0 |
-| LASTUSE | 1 |
-| USE | 2 |
-| FILL | 3 |
+| collector usage flag | value |
+| -------------------- | ----- |
+| DISCARD | 0 |
+| LASTUSE | 1 |
+| FILL | 2 |
+| USE | 3 |
#### '`llvm.nvvm.tcgen05.mma.ws*`'
@@ -3358,8 +3360,8 @@ The following tables describes the possible values of the flag arguments
| --------------------------- | ----- |
| DISCARD | 0 |
| LASTUSE | 1 |
-| USE | 2 |
-| FILL | 3 |
+| FILL | 2 |
+| USE | 3 |
### Store Intrinsics
diff --git a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
index d524f8aa99039..0b308865e041a 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
@@ -6104,7 +6104,7 @@ defvar Tcgen05MMABlockScaleKindAttr =
[EnumAttrIsOneOf<Tcgen05MMAKindAttr, Tcgen05MMABlockScaleKindList>]>;
def NVVM_Tcgen05MMAOp : NVVM_Op<"tcgen05.mma",
- [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110, 107]>]> {
+ [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110]>]> {
let summary = "Performs MMA operation on 5th-gen tensor cores";
let description = [{
@@ -6190,7 +6190,7 @@ def NVVM_Tcgen05MMAOp : NVVM_Op<"tcgen05.mma",
}
def NVVM_Tcgen05MMASparseOp : NVVM_Op<"tcgen05.mma.sp",
- [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110, 107]>]> {
+ [AttrSizedOperandSegments, NVVMRequiresSMf<[100, 101, 110]>]> {
let summary = "Performs MMA operation with sparse A matrix on 5th-gen tensor cores";
let description = [{
@@ -6273,7 +6273,7 @@ def Tcgen05MMABlockScaleAttr : EnumAttr<NVVM_Dialect, Tcgen05MMABlockScale,
}
def NVVM_Tcgen05MMABlockScaleOp : NVVM_Op<"tcgen05.mma.block_scale",
- [NVVMRequiresSMf<[100, 101, 110, 107]>]> {
+ [NVVMRequiresSMf<[100, 101, 110]>]> {
let summary = "Performs block scaled MMA operation on 5th-gen tensor cores";
let description = [{
@@ -6349,7 +6349,7 @@ def NVVM_Tcgen05MMABlockScaleOp : NVVM_Op<"tcgen05.mma.block_scale",
}
def NVVM_Tcgen05MMASparseBlockScaleOp : NVVM_Op<"tcgen05.mma.sp.block_scale",
- [NVVMRequiresSMf<[100, 101, 110, 107]>]> {
+ [NVVMRequiresSMf<[100, 101, 110]>]> {
let summary = "Performs block scaled MMA operation with sparse A matrix on 5th-gen tensor cores";
let description = [{
>From 1358dd124cfc41504941209ed9e68d36a52bf7f0 Mon Sep 17 00:00:00 2001
From: Kirill Vedernikov <kvedernikov at nvidia.com>
Date: Thu, 6 Aug 2026 09:05:49 +0200
Subject: [PATCH 3/5] [NVVM] Fix formatting
---
llvm/lib/IR/AutoUpgrade.cpp | 12 ++++++++----
1 file changed, 8 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 446d132074fd9..b0e11904ace66 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -5738,9 +5738,11 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) {
case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d:
case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_ashift:
case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg1:
- case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg1_ashift:
+ case Intrinsic::
+ nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg1_ashift:
case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg2:
- case Intrinsic::nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg2_ashift:
+ case Intrinsic::
+ nvvm_tcgen05_mma_sp_tensor_scale_d_disable_output_lane_cg2_ashift:
case Intrinsic::nvvm_tcgen05_mma_tensor:
case Intrinsic::nvvm_tcgen05_mma_tensor_ashift:
case Intrinsic::nvvm_tcgen05_mma_tensor_disable_output_lane_cg1:
@@ -5756,9 +5758,11 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) {
case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d:
case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_ashift:
case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg1:
- case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg1_ashift:
+ case Intrinsic::
+ nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg1_ashift:
case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg2:
- case Intrinsic::nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg2_ashift: {
+ case Intrinsic::
+ nvvm_tcgen05_mma_tensor_scale_d_disable_output_lane_cg2_ashift: {
SmallVector<Value *, 12> Args(CI->args());
Args.push_back(Builder.getInt32(0)); // collector_usage_b = discard(0)
NewCall = Builder.CreateCall(NewFn, Args);
>From da41e9f544cb19e46061a3656b5ae421fd289882 Mon Sep 17 00:00:00 2001
From: Kirill Vedernikov <kvedernikov at nvidia.com>
Date: Thu, 6 Aug 2026 10:34:27 +0200
Subject: [PATCH 4/5] [NVPTX] Simplify collector_b choices
---
llvm/lib/Target/NVPTX/NVPTXIntrinsics.td | 25 ++++++++++++------------
1 file changed, 12 insertions(+), 13 deletions(-)
diff --git a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
index 249aee4a09fa5..4f08cc2da7f66 100644
--- a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
+++ b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -6191,7 +6191,7 @@ foreach dim = ["x", "y", "z"] in {
//
class Tcgen05MMABase<bit IsSparse, string ASpace, string Kind, int CtaGroup,
- string CollectorUsage, string CollectorUsageB = ""> :
+ string CollectorUsage, string CollectorUsageB = "discard"> :
NVPTXInst<(outs), (ins), "?", []>,
Requires<[]> {
@@ -6210,13 +6210,12 @@ class Tcgen05MMABase<bit IsSparse, string ASpace, string Kind, int CtaGroup,
!eq(Kind, "i8"): 3,
);
- bit HasCollectorB = !ne(CollectorUsageB, "");
+ bit HasCollectorBQualifier = !ne(CollectorUsageB, "discard");
int CollectorUsageBVal = !cond(
!eq(CollectorUsageB, "discard"): 0,
!eq(CollectorUsageB, "lastuse"): 1,
!eq(CollectorUsageB, "fill"): 2,
- !eq(CollectorUsageB, "use"): 3,
- true: 0
+ !eq(CollectorUsageB, "use"): 3
);
dag SparseMetadataIns = !if(IsSparse, (ins B32:$spmetadata), (ins));
@@ -6259,7 +6258,7 @@ class Tcgen05MMAInst<bit IsSparse, string ASpace, string Kind, int CtaGroup,
Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsageA,
CollectorUsageB> {
let Predicates = !listconcat(
- !if(HasCollectorB, [callSubtarget<"hasTcgen05RubinFamily">], []),
+ !if(HasCollectorBQualifier, [callSubtarget<"hasTcgen05RubinFamily">], []),
!cond(
IsScaleInputD : [callSubtarget<"hasTcgen05MMAScaleInputDImm">],
!eq(Kind, "i8") : [callSubtarget<"hasTcgen05MMAI8Kind">],
@@ -6281,7 +6280,7 @@ class Tcgen05MMAInst<bit IsSparse, string ASpace, string Kind, int CtaGroup,
# SpCtaKindStr
# !if(IsAShift, ".ashift", "")
# ".collector::a::" # CollectorUsageA
- # !if(HasCollectorB, ".collector::b::" # CollectorUsageB, "")
+ # !if(HasCollectorBQualifier, ".collector::b::" # CollectorUsageB, "")
# BaseOperandsStr
# InputDStr
# ScaleInpStr
@@ -6305,7 +6304,7 @@ foreach sparse = [0, 1] in {
foreach scale_input_d = !if(!or(!eq(kind, "f16"),
!eq(kind, "tf32")), [0, 1], [0]) in {
foreach ashift = !if(!eq(space, "tensor"), [0, 1], [0]) in {
- foreach collector_usage_b = ["", "discard", "lastuse", "fill", "use"] in {
+ foreach collector_usage_b = ["discard", "lastuse", "fill", "use"] in {
def : Tcgen05MMAInst<sparse, space, kind, cta_group,
collector_usage_a, collector_usage_b,
scale_input_d, ashift>;
@@ -6365,7 +6364,7 @@ class Tcgen05MMADisableOutputLaneInst<bit IsSparse, string ASpace, string Kind,
Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsageA,
CollectorUsageB> {
let Predicates = !listconcat(
- !if(HasCollectorB, [callSubtarget<"hasTcgen05RubinFamily">], []),
+ !if(HasCollectorBQualifier, [callSubtarget<"hasTcgen05RubinFamily">], []),
!cond(
IsScaleInputD : [callSubtarget<"hasTcgen05MMAScaleInputDImm">],
!eq(Kind, "i8") : [callSubtarget<"hasTcgen05MMAI8Kind">],
@@ -6411,7 +6410,7 @@ class Tcgen05MMADisableOutputLaneInst<bit IsSparse, string ASpace, string Kind,
# SpCtaKindStr
# !if(IsAShift, ".ashift", "")
# ".collector::a::" # CollectorUsageA
- # !if(HasCollectorB, ".collector::b::" # CollectorUsageB, "")
+ # !if(HasCollectorBQualifier, ".collector::b::" # CollectorUsageB, "")
# BaseOperandsStr
# ", " # DisableOutputLaneStr
# InputDStr
@@ -6437,7 +6436,7 @@ foreach sparse = [0, 1] in {
foreach scale_input_d = !if(!or(!eq(kind, "f16"),
!eq(kind, "tf32")), [0, 1], [0]) in {
foreach ashift = !if(!eq(space, "tensor"), [0, 1], [0]) in {
- foreach collector_usage_b = ["", "discard", "lastuse", "fill", "use"] in {
+ foreach collector_usage_b = ["discard", "lastuse", "fill", "use"] in {
def :
Tcgen05MMADisableOutputLaneInst<sparse, space, kind, cta_group,
collector_usage_a, collector_usage_b,
@@ -6461,7 +6460,7 @@ class Tcgen05MMABlockScaleInst<bit IsSparse, string ASpace, string Kind,
Tcgen05MMABase<IsSparse, ASpace, Kind, CtaGroup, CollectorUsageA,
CollectorUsageB> {
let Predicates = !listconcat(
- !if(HasCollectorB, [callSubtarget<"hasTcgen05RubinFamily">], []),
+ !if(HasCollectorBQualifier, [callSubtarget<"hasTcgen05RubinFamily">], []),
!cond(
!and(IsSparse,
!eq(Kind, "mxf4")) : [callSubtarget<"hasTcgen05MMASparseMxf4">],
@@ -6488,7 +6487,7 @@ class Tcgen05MMABlockScaleInst<bit IsSparse, string ASpace, string Kind,
# SpCtaKindStr
# ".block_scale" # ScaleVecSize
# ".collector::a::" # CollectorUsageA
- # !if(HasCollectorB, ".collector::b::" # CollectorUsageB, "")
+ # !if(HasCollectorBQualifier, ".collector::b::" # CollectorUsageB, "")
# BaseOperandsStr
# ", [$scale_a], [$scale_b]"
# InputDStr
@@ -6511,7 +6510,7 @@ foreach sparse = [0, 1] in {
foreach scale_vec_size = ["", ".block16", ".block32"] in {
foreach cta_group = [1, 2] in {
foreach collector_usage_a = ["fill", "use", "lastuse", "discard"] in {
- foreach collector_usage_b = ["", "discard", "lastuse", "fill", "use"] in {
+ foreach collector_usage_b = ["discard", "lastuse", "fill", "use"] in {
if NVVM_TCGEN05_MMA_BLOCKSCALE_SUPPORTED<kind, scale_vec_size>.ret then {
def : Tcgen05MMABlockScaleInst<sparse, space, kind, cta_group,
scale_vec_size, collector_usage_a,
>From f4862aff7dd340e11cf19ebda2dd83f6fc93fd5c Mon Sep 17 00:00:00 2001
From: Kirill Vedernikov <kvedernikov at nvidia.com>
Date: Thu, 6 Aug 2026 11:39:39 +0200
Subject: [PATCH 5/5] [NVPTX] Update tcgen05.mma docs
---
llvm/docs/NVPTXUsage.md | 27 +++++----------------------
1 file changed, 5 insertions(+), 22 deletions(-)
diff --git a/llvm/docs/NVPTXUsage.md b/llvm/docs/NVPTXUsage.md
index 543d78daae307..490a5b3b99abb 100644
--- a/llvm/docs/NVPTXUsage.md
+++ b/llvm/docs/NVPTXUsage.md
@@ -3008,28 +3008,11 @@ Memory for the resultant matrix D will not be updated
#### Intrinsic Design:
Given the broad feature set of `tcgen05.mma` instruction modeling these
-through intrinsics is highly complex, and the following table outlines the large
-number of intrinsics required to fully support the `tcgen05.mma` instruction
-set.
-
-| variant | Configuration | Total Variants |
-| ---------------------------------- | ------------------------------------------------------------------------------------------------------------- | -------------- |
-| tcgen05.mma.shared | 2 (space) x 2 (sp) x 4 (kind) x 2 (cta_group) x 4 (collector_usage_a) x 4 (collector_usage_b) | 512 |
-| tcgen05.mma.tensor.ashift | 2 (sp) x 4 (kind) x 2 (cta_group) x 2 (collector_usage_a) x 4 (collector_usage_b) | 128 |
-| tcgen05.mma.scale_d | 2 (space) x 2 (sp) x 2 (kind) x 2 (cta_group) x 4 (collector_usage_a) x 4 (collector_usage_b) | 512 |
-| tcgen05.mma.scale_d.tensor.ashift | 2 (sp) x 2 (kind) x 2 (cta_group) x 2 (collector_usage_a) x 4 (collector_usage_b) | 64 |
-| tcgen05.mma.disable_output_lane | 2 (space) x 2 (sp) x 4 (kind) x 2 (cta_group) x 4 (collector_usage_a) x 4 (collector_usage_b) | 512 |
-| tcgen05.mma.disable_output_lane... | 2 (sp) x 4 (kind) x 2 (cta_group) x 2 (collector_usage_a) x 4 (collector_usage_b) | 128 |
-| tcgen05.mma.block_scale | 2 (space) x 1 (mxf4nvf4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage_a) x 4 (collector_usage_b) | 128 |
-| tcgen05.mma.block_scale | 2 (space) x 1 (mxf4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage_a) x 4 (collector_usage_b) | 128 |
-| tcgen05.mma.block_scale | 2 (space) x 1 (mxf8f6f4) x 2 (cta_group) x 2 (scale_vec_size) x 4 (collector_usage_a) x 4 (collector_usage_b) | 128 |
-| tcgen05.mma.ws | 2 (space) x 2 (sp) x 4 (kind) x 2 (zero_col_mask) x 4 (collector_usage_op) x 4 (collector_buffer) | 256 |
-| Total | | 2496 |
-
-To reduce the number of possible intrinsic variations, we've modeled the
-`tcgen05.mma` instructions using flag operands. We've added range checks to
-these flags to prevent invalid values. We also expanded some flags back into
-intrinsic modifiers to avoid supporting invalid combinations of features.
+through intrinsics is highly complex, we've modeled the `tcgen05.mma`
+instructions using flag operands. We've added range checks to these flags to
+prevent invalid values. We also expanded some flags back into intrinsic
+modifiers to avoid supporting invalid combinations of features. These intrinsics
+capture ~2000+ individual PTX instruction combinations of this family.
#### '`llvm.nvvm.tcgen05.mma.*`'
More information about the llvm-commits
mailing list