[llvm-branch-commits] [clang] [llvm] AMDGPU: Add v_smfmac_f32_16x16x64_f16 for gfx950 (PR #117202)
Matt Arsenault via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Thu Nov 21 10:13:04 PST 2024
https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/117202
None
>From 8f21ee1c353df309defdc980d9a84ee89e70ddf8 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Thu, 1 Feb 2024 15:01:47 +0530
Subject: [PATCH] AMDGPU: Add v_smfmac_f32_16x16x64_f16 for gfx950
---
clang/include/clang/Basic/BuiltinsAMDGPU.def | 1 +
.../CodeGenOpenCL/builtins-amdgcn-mfma.cl | 7 +
.../builtins-amdgcn-error-gfx950-param.cl | 7 +
.../builtins-amdgcn-error-gfx950.cl | 1 +
llvm/include/llvm/IR/IntrinsicsAMDGPU.td | 1 +
.../AMDGPU/AMDGPUInstructionSelector.cpp | 6 +
.../Target/AMDGPU/AMDGPURegisterBankInfo.cpp | 3 +-
llvm/lib/Target/AMDGPU/SIInstrInfo.td | 1 +
llvm/lib/Target/AMDGPU/VOP3PInstructions.td | 7 +
.../UniformityAnalysis/AMDGPU/intrinsics.ll | 9 +
.../AMDGPU/llvm.amdgcn.smfmac.gfx950.ll | 218 ++++++++++++++++++
llvm/test/MC/AMDGPU/mai-gfx950.s | 42 ++++
.../MC/Disassembler/AMDGPU/gfx950_mai.txt | 22 ++
13 files changed, 324 insertions(+), 1 deletion(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
diff --git a/clang/include/clang/Basic/BuiltinsAMDGPU.def b/clang/include/clang/Basic/BuiltinsAMDGPU.def
index 3b7cc559e88b29..f013714798cc54 100644
--- a/clang/include/clang/Basic/BuiltinsAMDGPU.def
+++ b/clang/include/clang/Basic/BuiltinsAMDGPU.def
@@ -444,6 +444,7 @@ TARGET_BUILTIN(__builtin_amdgcn_mfma_f32_32x32x16_bf16, "V16fV8yV8yV16fIiIiIi",
TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_16x16x64_i8, "V4iV4iV4iV4iIiIiIi", "nc", "gfx950-insts")
TARGET_BUILTIN(__builtin_amdgcn_mfma_i32_32x32x32_i8, "V16iV4iV4iV16iIiIiIi", "nc", "gfx950-insts")
+TARGET_BUILTIN(__builtin_amdgcn_smfmac_f32_16x16x64_f16, "V4fV8hV16hV4fiIiIi", "nc", "gfx950-insts")
//===----------------------------------------------------------------------===//
// GFX12+ only builtins.
//===----------------------------------------------------------------------===//
diff --git a/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl b/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
index 345f05f463bf44..e63d89a28de44d 100644
--- a/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
+++ b/clang/test/CodeGenOpenCL/builtins-amdgcn-mfma.cl
@@ -467,4 +467,11 @@ v4f test_mfma_f32_16x16x32_bf16(v8bf16 a, v8bf16 b, v4f c)
return __builtin_amdgcn_mfma_f32_16x16x32_bf16(a, b, c, 1, 2, 3);
}
+// CHECK-GFX950-LABEL: @test_smfmac_f32_16x16x64_f16
+// CHECK-GFX950: call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %a, <16 x half> %b, <4 x float> %c, i32 %idx, i32 0, i32 0)
+void test_smfmac_f32_16x16x64_f16(global v4f* out, v8h a, v16h b, v4f c, int idx)
+{
+ *out = __builtin_amdgcn_smfmac_f32_16x16x64_f16(a, b, c, idx, 0, 0);
+}
+
#endif
diff --git a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
index acaa20090dfcba..6366997465aeff 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950-param.cl
@@ -4,6 +4,7 @@
typedef float float4 __attribute__((ext_vector_type(4)));
typedef float float16 __attribute__((ext_vector_type(16)));
typedef half half8 __attribute__((ext_vector_type(8)));
+typedef half half16 __attribute__((ext_vector_type(16)));
typedef __bf16 bfloat8 __attribute__((ext_vector_type(8)));
typedef int int4 __attribute__((ext_vector_type(4)));
typedef int int8 __attribute__((ext_vector_type(8)));
@@ -62,3 +63,9 @@ void test_mfma_f32_16x16x32_bf16(__global float4* out, bfloat8 a, bfloat8 b, flo
*out = __builtin_amdgcn_mfma_f32_16x16x32_bf16(a, b, c, 0, X, 0); // expected-error{{argument to '__builtin_amdgcn_mfma_f32_16x16x32_bf16' must be a constant integer}}
*out = __builtin_amdgcn_mfma_f32_16x16x32_bf16(a, b, c, 0, 0, X); // expected-error{{argument to '__builtin_amdgcn_mfma_f32_16x16x32_bf16' must be a constant integer}}
}
+
+void test_smfmac_f32_16x16x64_f16(global float4* out, half8 a, half16 b, float4 c, int idx, int d)
+{
+ *out = __builtin_amdgcn_smfmac_f32_16x16x64_f16(a, b, c, idx, d, 0); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_16x16x64_f16' must be a constant integer}}
+ *out = __builtin_amdgcn_smfmac_f32_16x16x64_f16(a, b, c, idx, 0, d); // expected-error{{argument to '__builtin_amdgcn_smfmac_f32_16x16x64_f16' must be a constant integer}}
+}
diff --git a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
index 6bf76b3cba0f59..1e924e86f3b897 100644
--- a/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
+++ b/clang/test/SemaOpenCL/builtins-amdgcn-error-gfx950.cl
@@ -34,6 +34,7 @@ void test(__global float4* out0, half8 a0, half8 b0, float4 c0,
*out3 = __builtin_amdgcn_mfma_i32_16x16x64_i8(a3, b3, c3, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_i32_16x16x64_i8' needs target feature gfx950-insts}}
*out4 = __builtin_amdgcn_mfma_i32_32x32x32_i8(a4, b4, c4, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_i32_32x32x32_i8' needs target feature gfx950-insts}}
*out5 = __builtin_amdgcn_mfma_f32_16x16x32_bf16(a5, b5, c5, 0, 0, 0); // expected-error{{'__builtin_amdgcn_mfma_f32_16x16x32_bf16' needs target feature gfx950-insts}}
+ *out6 = __builtin_amdgcn_smfmac_f32_16x16x64_f16(a6, b6, c6, 0, 0, 0); // expected-error{{'__builtin_amdgcn_smfmac_f32_16x16x64_f16' needs target feature gfx950-insts}}
*out14 = __builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4(a14, b14, c14, 0, 0, 0, d14, 0, e14); // expected-error{{'__builtin_amdgcn_mfma_scale_f32_16x16x128_f8f6f4' needs target feature gfx950-insts}}
*out15 = __builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4(a15, b15, c15, 0, 0, 0, d15, 0, e15); // expected-error{{'__builtin_amdgcn_mfma_scale_f32_32x32x64_f8f6f4' needs target feature gfx950-insts}}
}
diff --git a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
index 479120f9c202bf..05a22588f50f9f 100644
--- a/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
+++ b/llvm/include/llvm/IR/IntrinsicsAMDGPU.td
@@ -3152,6 +3152,7 @@ def int_amdgcn_mfma_f32_16x16x32_bf16 : AMDGPUMfmaIntrinsic<llvm_v4f32_ty, llvm_
def int_amdgcn_mfma_f32_32x32x16_bf16 : AMDGPUMfmaIntrinsic<llvm_v16f32_ty, llvm_v8bf16_ty>;
def int_amdgcn_mfma_scale_f32_16x16x128_f8f6f4 : AMDGPUMfmaScaleIntrinsic<llvm_v4f32_ty>;
def int_amdgcn_mfma_scale_f32_32x32x64_f8f6f4 : AMDGPUMfmaScaleIntrinsic<llvm_v16f32_ty>;
+def int_amdgcn_smfmac_f32_16x16x64_f16 : AMDGPUMSmfmacIntrinsic<llvm_v4f32_ty, llvm_v8f16_ty, llvm_v16f16_ty>;
}
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index f7cba2f13c77ac..6b03d571abea68 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -1090,6 +1090,7 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC(MachineInstr &I) const {
case Intrinsic::amdgcn_smfmac_f32_32x32x32_bf8_fp8:
case Intrinsic::amdgcn_smfmac_f32_32x32x32_fp8_bf8:
case Intrinsic::amdgcn_smfmac_f32_32x32x32_fp8_fp8:
+ case Intrinsic::amdgcn_smfmac_f32_16x16x64_f16:
return selectSMFMACIntrin(I);
default:
return selectImpl(I, *CoverageInfo);
@@ -3435,6 +3436,8 @@ bool AMDGPUInstructionSelector::selectBVHIntrinsic(MachineInstr &MI) const{
return true;
}
+// FIXME: This should be removed and let the patterns select. We just need the
+// AGPR/VGPR combination versions.
bool AMDGPUInstructionSelector::selectSMFMACIntrin(MachineInstr &MI) const {
unsigned Opc;
switch (cast<GIntrinsic>(MI).getIntrinsicID()) {
@@ -3480,6 +3483,9 @@ bool AMDGPUInstructionSelector::selectSMFMACIntrin(MachineInstr &MI) const {
case Intrinsic::amdgcn_smfmac_f32_32x32x32_fp8_fp8:
Opc = AMDGPU::V_SMFMAC_F32_32X32X32_FP8_FP8_e64;
break;
+ case Intrinsic::amdgcn_smfmac_f32_16x16x64_f16:
+ Opc = AMDGPU::V_SMFMAC_F32_16X16X64_F16_e64;
+ break;
default:
llvm_unreachable("unhandled smfmac intrinsic");
}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 4b7ba4effc6857..7442b782aba5c2 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -4804,7 +4804,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
case Intrinsic::amdgcn_smfmac_f32_32x32x32_bf8_bf8:
case Intrinsic::amdgcn_smfmac_f32_32x32x32_bf8_fp8:
case Intrinsic::amdgcn_smfmac_f32_32x32x32_fp8_bf8:
- case Intrinsic::amdgcn_smfmac_f32_32x32x32_fp8_fp8: {
+ case Intrinsic::amdgcn_smfmac_f32_32x32x32_fp8_fp8:
+ case Intrinsic::amdgcn_smfmac_f32_16x16x64_f16: {
// vdst, srcA, srcB, srcC, idx
OpdsMapping[0] = getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI);
OpdsMapping[2] = getVGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI);
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index 05568b4f9ad1ec..8f85bd437781e6 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -2869,6 +2869,7 @@ def VOP_V4F32_I64_I64_V4F32 : VOPProfile <[v4f32, i64, i64, v4f32]>;
def VOP_V16F32_I64_I64_V16F32 : VOPProfile <[v16f32, i64, i64, v16f32]>;
def VOP_V4F32_V4F16_V8F16_I32 : VOPProfile <[v4f32, v4f16, v8f16, i32]>;
+def VOP_V4F32_V8F16_V16F16_I32 : VOPProfile <[v4f32, v8f16, v16f16, i32]>;
def VOP_V16F32_V4F16_V8F16_I32 : VOPProfile <[v16f32, v4f16, v8f16, i32]>;
def VOP_V4F32_V4I16_V8I16_I32 : VOPProfile <[v4f32, v4i16, v8i16, i32]>;
def VOP_V16F32_V4I16_V8I16_I32 : VOPProfile <[v16f32, v4i16, v8i16, i32]>;
diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index 9bd2b124f122dc..75ce6db2bfa7a8 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -629,6 +629,7 @@ def VOPProfileMAI_F32_I64_X32_VCD : VOPProfileMAI<VOP_V4F32_I64_I64_V4F32,
def VOPProfileMAI_F32_I64_X16_VCD : VOPProfileMAI<VOP_V16F32_I64_I64_V16F32, VISrc_512_b32, VDst_512, AVSrc_64>;
def VOPProfileSMFMAC_F32_16X16X32_F16 : VOPProfileSMFMAC<VOP_V4F32_V4F16_V8F16_I32, AVDst_128, AVSrc_64, AVSrc_128>;
+def VOPProfileSMFMAC_F32_16X16X64_F16 : VOPProfileSMFMAC<VOP_V4F32_V8F16_V16F16_I32, AVDst_128, AVSrc_128, AVSrc_256>;
def VOPProfileSMFMAC_F32_32X32X16_F16 : VOPProfileSMFMAC<VOP_V16F32_V4F16_V8F16_I32, AVDst_512, AVSrc_64, AVSrc_128>;
def VOPProfileSMFMAC_F32_16X16X32_I16 : VOPProfileSMFMAC<VOP_V4F32_V4I16_V8I16_I32, AVDst_128, AVSrc_64, AVSrc_128>;
def VOPProfileSMFMAC_F32_32X32X16_I16 : VOPProfileSMFMAC<VOP_V16F32_V4I16_V8I16_I32, AVDst_512, AVSrc_64, AVSrc_128>;
@@ -1042,6 +1043,10 @@ defm V_SMFMAC_F32_32X32X32_FP8_BF8 : SMFMACInst<"v_smfmac_f32_32x32x32_fp8_bf8",
defm V_SMFMAC_F32_32X32X32_FP8_FP8 : SMFMACInst<"v_smfmac_f32_32x32x32_fp8_fp8", "F32_32X32X32_F8", int_amdgcn_smfmac_f32_32x32x32_fp8_fp8>;
} // End SubtargetPredicate = HasFP8Insts, is_gfx940_xdl = 1
+let SubtargetPredicate = HasGFX950Insts in {
+defm V_SMFMAC_F32_16X16X64_F16 : SMFMACInst<"v_smfmac_f32_16x16x64_f16", "F32_16X16X64_F16", int_amdgcn_smfmac_f32_16x16x64_f16>;
+}
+
def MAIInstInfoTable : GenericTable {
let FilterClass = "MAIInst";
let CppTypeName = "MAIInstInfo";
@@ -2131,6 +2136,8 @@ defm V_SMFMAC_F32_32X32X32_BF8_FP8 : VOP3P_Real_SMFMAC <0x7d, "v_smfmac_f32_32x3
defm V_SMFMAC_F32_32X32X32_FP8_BF8 : VOP3P_Real_SMFMAC <0x7e, "v_smfmac_f32_32x32x32fp8bf8">;
defm V_SMFMAC_F32_32X32X32_FP8_FP8 : VOP3P_Real_SMFMAC <0x7f, "v_smfmac_f32_32x32x32fp8fp8">;
+defm V_SMFMAC_F32_16X16X64_F16 : VOP3P_Real_SMFMAC <0x5a, "v_smfmac_f32_16x16x64f16">;
+
defm V_PK_FMA_F32 : VOP3P_Real_vi <0x30>;
defm V_PK_MUL_F32 : VOP3P_Real_vi <0x31>;
defm V_PK_ADD_F32 : VOP3P_Real_vi <0x32>;
diff --git a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
index 2d7983ebafd653..edaa724834297c 100644
--- a/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
+++ b/llvm/test/Analysis/UniformityAnalysis/AMDGPU/intrinsics.ll
@@ -332,6 +332,15 @@ define amdgpu_kernel void @mfma_f32_16x16x32_bf16(<8 x bfloat> %arg0, <8 x bfloa
ret void
}
+declare <4 x float> @llvm.amdgcn.smmfmac.f32.16x16x64.f16(<8 x half>, <16 x half>, <4 x float>, i32, i32 immarg, i32 immarg)
+
+; CHECK: DIVERGENT: %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0)
+define amdgpu_kernel void @smfmac_f32_16x16x64_f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3, ptr addrspace(1) %out) {
+ %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0)
+ store <4 x float> %result, ptr addrspace(1) %out
+ ret void
+}
+
declare i32 @llvm.amdgcn.ds.swizzle(i32, i32) #1
declare i32 @llvm.amdgcn.permlane16.i32(i32, i32, i32, i32, i1, i1) #1
declare i32 @llvm.amdgcn.permlanex16.i32(i32, i32, i32, i32, i1, i1) #1
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
new file mode 100644
index 00000000000000..c5ca845d532414
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll
@@ -0,0 +1,218 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -march=amdgcn -mcpu=gfx950 -global-isel=0 < %s | FileCheck -enable-var-scope --check-prefixes=GCN,SDAG %s
+; RUN: llc -march=amdgcn -mcpu=gfx950 -global-isel=1 < %s | FileCheck -enable-var-scope --check-prefixes=GCN,GISEL %s
+
+declare i32 @llvm.amdgcn.workitem.id.x()
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.smfmac.f32.16x16x64.f16
+; --------------------------------------------------------------------
+
+declare <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half>, <16 x half>, <4 x float>, i32, i32 immarg, i32 immarg)
+
+define amdgpu_kernel void @test_smfmac_f32_16x16x64_f16__vgpr(ptr addrspace(1) %arg, <8 x half> %a, <16 x half> %b, i32 %idx) #0 {
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16__vgpr:
+; SDAG: ; %bb.0: ; %bb
+; SDAG-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24
+; SDAG-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x34
+; SDAG-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; SDAG-NEXT: v_lshlrev_b32_e32 v0, 4, v0
+; SDAG-NEXT: v_mov_b32_e32 v16, 0
+; SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; SDAG-NEXT: global_load_dwordx4 v[8:11], v0, s[6:7]
+; SDAG-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x44
+; SDAG-NEXT: s_load_dword s16, s[4:5], 0x64
+; SDAG-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
+; SDAG-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
+; SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; SDAG-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
+; SDAG-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
+; SDAG-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
+; SDAG-NEXT: v_mov_b64_e32 v[6:7], s[14:15]
+; SDAG-NEXT: v_mov_b32_e32 v17, s16
+; SDAG-NEXT: s_waitcnt vmcnt(0)
+; SDAG-NEXT: s_nop 0
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 v[8:11], v[12:15], v[0:7], v17 cbsz:1 abid:2
+; SDAG-NEXT: s_nop 6
+; SDAG-NEXT: global_store_dwordx4 v16, v[8:11], s[6:7]
+; SDAG-NEXT: s_endpgm
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__vgpr:
+; GISEL: ; %bb.0: ; %bb
+; GISEL-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x24
+; GISEL-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x34
+; GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; GISEL-NEXT: v_lshlrev_b32_e32 v0, 4, v0
+; GISEL-NEXT: s_waitcnt lgkmcnt(0)
+; GISEL-NEXT: global_load_dwordx4 v[8:11], v0, s[6:7]
+; GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x44
+; GISEL-NEXT: s_load_dword s16, s[4:5], 0x64
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
+; GISEL-NEXT: s_waitcnt lgkmcnt(0)
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[8:9]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[10:11]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[12:13]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[14:15]
+; GISEL-NEXT: v_mov_b32_e32 v16, s16
+; GISEL-NEXT: s_waitcnt vmcnt(0)
+; GISEL-NEXT: s_nop 0
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2
+; GISEL-NEXT: v_mov_b32_e32 v0, 0
+; GISEL-NEXT: s_nop 5
+; GISEL-NEXT: global_store_dwordx4 v0, v[8:11], s[6:7]
+; GISEL-NEXT: s_endpgm
+bb:
+ %id = call i32 @llvm.amdgcn.workitem.id.x()
+ %gep = getelementptr <4 x float>, ptr addrspace(1) %arg, i32 %id
+ %in.1 = load <4 x float>, ptr addrspace(1) %gep
+ %mai.1 = tail call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %a, <16 x half> %b, <4 x float> %in.1, i32 %idx, i32 1, i32 2)
+ store <4 x float> %mai.1, ptr addrspace(1) %arg
+ ret void
+}
+
+define <4 x float> @test_smfmac_f32_16x16x64_f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_accvgpr_write_b32 a0, v12
+; SDAG-NEXT: v_accvgpr_write_b32 a1, v13
+; SDAG-NEXT: v_accvgpr_write_b32 a2, v14
+; SDAG-NEXT: v_accvgpr_write_b32 a3, v15
+; SDAG-NEXT: s_nop 1
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 a[0:3], v[0:3], v[4:11], v16
+; SDAG-NEXT: s_nop 6
+; SDAG-NEXT: v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT: v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT: v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT: v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16
+; GISEL-NEXT: s_nop 6
+; GISEL-NEXT: v_mov_b32_e32 v0, v12
+; GISEL-NEXT: v_mov_b32_e32 v1, v13
+; GISEL-NEXT: v_mov_b32_e32 v2, v14
+; GISEL-NEXT: v_mov_b32_e32 v3, v15
+; GISEL-NEXT: s_setpc_b64 s[30:31]
+ %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0)
+ ret <4 x float> %result
+}
+
+define <4 x float> @test_smfmac_f32_16x16x64_f16__flags0(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16__flags0:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_accvgpr_write_b32 a0, v12
+; SDAG-NEXT: v_accvgpr_write_b32 a1, v13
+; SDAG-NEXT: v_accvgpr_write_b32 a2, v14
+; SDAG-NEXT: v_accvgpr_write_b32 a3, v15
+; SDAG-NEXT: s_nop 1
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 a[0:3], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; SDAG-NEXT: s_nop 6
+; SDAG-NEXT: v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT: v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT: v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT: v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__flags0:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:1 abid:3
+; GISEL-NEXT: s_nop 6
+; GISEL-NEXT: v_mov_b32_e32 v0, v12
+; GISEL-NEXT: v_mov_b32_e32 v1, v13
+; GISEL-NEXT: v_mov_b32_e32 v2, v14
+; GISEL-NEXT: v_mov_b32_e32 v3, v15
+; GISEL-NEXT: s_setpc_b64 s[30:31]
+ %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 1, i32 immarg 3)
+ ret <4 x float> %result
+}
+
+define <4 x float> @test_smfmac_f32_16x16x64_f16__flags1(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16__flags1:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_accvgpr_write_b32 a0, v12
+; SDAG-NEXT: v_accvgpr_write_b32 a1, v13
+; SDAG-NEXT: v_accvgpr_write_b32 a2, v14
+; SDAG-NEXT: v_accvgpr_write_b32 a3, v15
+; SDAG-NEXT: s_nop 1
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 a[0:3], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; SDAG-NEXT: s_nop 6
+; SDAG-NEXT: v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT: v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT: v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT: v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__flags1:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[12:15], v[0:3], v[4:11], v16 cbsz:3 abid:1
+; GISEL-NEXT: s_nop 6
+; GISEL-NEXT: v_mov_b32_e32 v0, v12
+; GISEL-NEXT: v_mov_b32_e32 v1, v13
+; GISEL-NEXT: v_mov_b32_e32 v2, v14
+; GISEL-NEXT: v_mov_b32_e32 v3, v15
+; GISEL-NEXT: s_setpc_b64 s[30:31]
+ %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 3, i32 immarg 1)
+ ret <4 x float> %result
+}
+
+define <4 x float> @test_smfmac_f32_16x16x64_f16__sgpr(<8 x half> inreg %arg0, <16 x half> inreg %arg1, <4 x float> inreg %arg2, i32 inreg %arg3) {
+; SDAG-LABEL: test_smfmac_f32_16x16x64_f16__sgpr:
+; SDAG: ; %bb.0:
+; SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT: v_mov_b32_e32 v8, s0
+; SDAG-NEXT: v_mov_b32_e32 v9, s1
+; SDAG-NEXT: v_mov_b32_e32 v10, s2
+; SDAG-NEXT: v_mov_b32_e32 v11, s3
+; SDAG-NEXT: v_mov_b32_e32 v0, s16
+; SDAG-NEXT: v_mov_b32_e32 v1, s17
+; SDAG-NEXT: v_mov_b32_e32 v2, s18
+; SDAG-NEXT: v_mov_b32_e32 v3, s19
+; SDAG-NEXT: v_mov_b32_e32 v4, s20
+; SDAG-NEXT: v_mov_b32_e32 v5, s21
+; SDAG-NEXT: v_mov_b32_e32 v6, s22
+; SDAG-NEXT: v_mov_b32_e32 v7, s23
+; SDAG-NEXT: v_accvgpr_write_b32 a0, s24
+; SDAG-NEXT: v_accvgpr_write_b32 a1, s25
+; SDAG-NEXT: v_accvgpr_write_b32 a2, s26
+; SDAG-NEXT: v_accvgpr_write_b32 a3, s27
+; SDAG-NEXT: v_mov_b32_e32 v12, s28
+; SDAG-NEXT: s_nop 1
+; SDAG-NEXT: v_smfmac_f32_16x16x64_f16 a[0:3], v[8:11], v[0:7], v12
+; SDAG-NEXT: s_nop 6
+; SDAG-NEXT: v_accvgpr_read_b32 v0, a0
+; SDAG-NEXT: v_accvgpr_read_b32 v1, a1
+; SDAG-NEXT: v_accvgpr_read_b32 v2, a2
+; SDAG-NEXT: v_accvgpr_read_b32 v3, a3
+; SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: test_smfmac_f32_16x16x64_f16__sgpr:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3]
+; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1]
+; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17]
+; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[24:25]
+; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19]
+; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[20:21]
+; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[22:23]
+; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[26:27]
+; GISEL-NEXT: v_mov_b32_e32 v16, s28
+; GISEL-NEXT: s_nop 1
+; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[0:3], v[12:15], v[4:11], v16
+; GISEL-NEXT: s_setpc_b64 s[30:31]
+ %result = call <4 x float> @llvm.amdgcn.smfmac.f32.16x16x64.f16(<8 x half> %arg0, <16 x half> %arg1, <4 x float> %arg2, i32 %arg3, i32 immarg 0, i32 immarg 0)
+ ret <4 x float> %result
+}
+
+attributes #0 = { "amdgpu-flat-work-group-size"="1,256" }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GCN: {{.*}}
diff --git a/llvm/test/MC/AMDGPU/mai-gfx950.s b/llvm/test/MC/AMDGPU/mai-gfx950.s
index 919e0a296f7bb2..722c91743da2fe 100644
--- a/llvm/test/MC/AMDGPU/mai-gfx950.s
+++ b/llvm/test/MC/AMDGPU/mai-gfx950.s
@@ -1,6 +1,7 @@
// RUN: llvm-mc -triple=amdgcn -mcpu=gfx950 -show-encoding %s | FileCheck -check-prefix=GFX950 %s
// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx940 %s 2>&1 | FileCheck -check-prefix=ERR %s
// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx90a %s 2>&1 | FileCheck -check-prefix=ERR %s
+// RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1010 %s 2>&1 | FileCheck -check-prefix=ERR %s
// cbsz = SrcA
// blgp = SrcB
@@ -1050,3 +1051,44 @@ v_mfma_f32_16x16x32_bf16 a[0:3], v[0:3], v[0:3], a[4:7]
// GFX950: v_mfma_f32_16x16x32_bf16 v[0:3], a[0:3], a[0:3], v[4:7] ; encoding: [0x00,0x00,0xb5,0xd3,0x00,0x01,0x12,0x1c]
// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
v_mfma_f32_16x16x32_bf16 v[0:3], a[0:3], a[0:3], v[4:7]
+
+
+//===----------------------------------------------------------------------===//
+// SMFMAC opcodes.
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// v_smfmac_f32_16x16x64_f16
+//===----------------------------------------------------------------------===//
+
+// GFX950: v_smfmac_f32_16x16x64_f16 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xda,0xd3,0x02,0x09,0x0e,0x0c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64_f16 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x64_f16 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xda,0xd3,0x02,0x09,0x0e,0x0c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64f16 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], v[2:5], a[4:11], v1 ; encoding: [0x0a,0x80,0xda,0xd3,0x02,0x09,0x06,0x14]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64_f16 a[10:13], v[2:5], a[4:11], v1
+
+// GFX950: v_smfmac_f32_16x16x64_f16 v[10:13], a[2:5], v[4:11], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xda,0xd3,0x02,0x09,0x0a,0x0c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64_f16 v[10:13], a[2:5], v[4:11], v2 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], v[2:5], a[4:11], v3 ; encoding: [0x0a,0x80,0xda,0xd3,0x02,0x09,0x0e,0x14]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64_f16 a[10:13], v[2:5], a[4:11], v3
+
+// GFX950: v_smfmac_f32_16x16x64_f16 v[10:13], v[2:5], v[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xda,0xd3,0x02,0x0d,0x0a,0x04]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64_f16 v[10:13], v[2:5], v[6:13], v2 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], a[2:5], a[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x8b,0xda,0xd3,0x02,0x0d,0x0a,0x1c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64_f16 a[10:13], a[2:5], a[6:13], v2 cbsz:3 abid:1
+
+// GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3 ; encoding: [0x0a,0x99,0xda,0xd3,0x02,0x0d,0x0e,0x1c]
+// ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+v_smfmac_f32_16x16x64_f16 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
index f564c2ffb29a05..f1e517bfb5f07d 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx950_mai.txt
@@ -664,3 +664,25 @@
# GFX950: v_mfma_f32_16x16x32_bf16 v[0:3], v[0:3], v[0:3], v[0:3] blgp:5 ; encoding: [0x00,0x00,0xb5,0xd3,0x00,0x01,0x02,0xa4]
0x00,0x00,0xb5,0xd3,0x00,0x01,0x02,0xa4
+
+
+# GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], a[2:5], a[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x8b,0xda,0xd3,0x02,0x0d,0x0a,0x1c]
+0x0a,0x8b,0xda,0xd3,0x02,0x0d,0x0a,0x1c
+
+# GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], a[2:5], a[6:13], v3 cbsz:1 abid:3 ; encoding: [0x0a,0x99,0xda,0xd3,0x02,0x0d,0x0e,0x1c]
+0x0a,0x99,0xda,0xd3,0x02,0x0d,0x0e,0x1c
+
+# GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], v[2:5], a[4:11], v1 ; encoding: [0x0a,0x80,0xda,0xd3,0x02,0x09,0x06,0x14]
+0x0a,0x80,0xda,0xd3,0x02,0x09,0x06,0x14
+
+# GFX950: v_smfmac_f32_16x16x64_f16 a[10:13], v[2:5], a[4:11], v3 ; encoding: [0x0a,0x80,0xda,0xd3,0x02,0x09,0x0e,0x14]
+0x0a,0x80,0xda,0xd3,0x02,0x09,0x0e,0x14
+
+# GFX950: v_smfmac_f32_16x16x64_f16 v[10:13], a[2:5], v[4:11], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xda,0xd3,0x02,0x09,0x0a,0x0c]
+0x0a,0x0b,0xda,0xd3,0x02,0x09,0x0a,0x0c
+
+# GFX950: v_smfmac_f32_16x16x64_f16 v[10:13], a[2:5], v[4:11], v3 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xda,0xd3,0x02,0x09,0x0e,0x0c]
+0x0a,0x0b,0xda,0xd3,0x02,0x09,0x0e,0x0c
+
+# GFX950: v_smfmac_f32_16x16x64_f16 v[10:13], v[2:5], v[6:13], v2 cbsz:3 abid:1 ; encoding: [0x0a,0x0b,0xda,0xd3,0x02,0x0d,0x0a,0x04]
+0x0a,0x0b,0xda,0xd3,0x02,0x0d,0x0a,0x04
More information about the llvm-branch-commits
mailing list