[llvm] [AMDGPU] Fix gfx1250 WMMA latencies (PR #202522)
Austin Kerbow via llvm-commits
llvm-commits at lists.llvm.org
Sat Jul 4 21:54:23 PDT 2026
https://github.com/kerbowa updated https://github.com/llvm/llvm-project/pull/202522
>From b5936a06dc9e86e188c38cedcb0a8cb266776fd0 Mon Sep 17 00:00:00 2001
From: Austin Kerbow <Austin.Kerbow at amd.com>
Date: Mon, 15 Jun 2026 23:13:22 -0500
Subject: [PATCH] [AMDGPU] Fix gfx1250 WMMA latencies
- 16x16x64 FP8/BF8 WMMA run in 4 cycles, not 8. Add a 4-cycle
WriteXDL1PassWMMA write and split the FP8/BF8 InstRW by shape so that
16x16x128 FP8/BF8 (and F16/BF16) keep the 8-cycle latency.
- f8f6f4 WMMA run in 4 cycles when both matrix inputs are f4 and 8 cycles
when any input is f6 or f8, instead of the previous 8/16. Update the
scheduling predicate accordingly.
The co-execution hazard category is derived from the WMMA latency. The
4-cycle WMMAs have a single co-execution slot, so give them their own
category (2 wait states before a dependent WMMA, 1 before a dependent VALU)
rather than bucketing them with the 8-cycle WMMAs.
---
.../lib/Target/AMDGPU/GCNHazardRecognizer.cpp | 22 ++--
llvm/lib/Target/AMDGPU/SISchedule.td | 27 ++--
.../misched-into-wmma-hazard-shadow.mir | 4 -
.../AMDGPU/wmma-coexecution-valu-hazards.mir | 67 ----------
.../AMDGPU/wmma-hazards-gfx1250-w32.mir | 117 ------------------
.../llvm-mca/AMDGPU/gfx1250-wmma-cycles.s | 66 ++++++++++
6 files changed, 97 insertions(+), 206 deletions(-)
create mode 100644 llvm/test/tools/llvm-mca/AMDGPU/gfx1250-wmma-cycles.s
diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
index 722ddb5d0c4dd..84aeb80a798a0 100644
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -2146,15 +2146,11 @@ static bool isCoexecutableVALUInst(const MachineInstr &MI) {
//
// Category 0: WMMA with Latency 8
// WMMA_*F16, WMMA_*BF16
-// WMMA_*FP8FP8
-// WMMA_*FP8BF8
-// WMMA_*BF8FP8
-// WMMA_*BF8BF8
-// WMMA_*F8F6F4 if SRCA & SRCB != F8
+// WMMA_*_16X16X128_{FP8,BF8}
+// WMMA_*F8F6F4 if SRCA & SRCB are not both F4
//
// Category 1: WMMA Latency 16
// WMMA_IU8
-// WMMA_*F8F6F4 if SRCA OR SRCB == F8
//
// Category 2: SWMMAC with Latency 8
// SWMMAC_*F16, SWMMAC_*BF16,
@@ -2179,6 +2175,10 @@ static bool isCoexecutableVALUInst(const MachineInstr &MI) {
// V_WMMA_F32_32X16X128_F4
// V_WMMA_I32_16X16X64_IU8
// V_WMMA_I32_16X16X64_IU8
+//
+// Category 6: gfx1250 WMMA with Latency 4 (one co-execution slot)
+// WMMA_*_16X16X64_{FP8,BF8}
+// WMMA_*F8F6F4 if SRCA & SRCB are both F4
static unsigned getWMMAHazardInstInCategory(const MachineInstr &MI,
const SIInstrInfo *TII,
const TargetSchedModel &SchedModel,
@@ -2190,6 +2190,12 @@ static unsigned getWMMAHazardInstInCategory(const MachineInstr &MI,
unsigned Latency = SchedModel.computeInstrLatency(&MI);
switch (Latency) {
+ case 4:
+ // Dense 4-cycle WMMA (gfx1250 16x16x64 FP8/BF8 and f8f6f4 with both
+ // inputs F4). One co-execution slot; there is no 4-cycle SWMMAC.
+ assert(!IsSWMMAC && "no 4-cycle SWMMAC expected");
+ Category = 6;
+ break;
case 8:
Category = IsSWMMAC ? 2 : 0;
break;
@@ -2220,8 +2226,8 @@ int GCNHazardRecognizer::checkWMMACoexecutionHazards(MachineInstr *MI) const {
// (WMMAWaitStates if the second is also a WMMA, VALUWaitStates if the second
// is a VALU). Refer to SPG 4.6.12.1. "Requirements for WMMA data hazards" for
// numbers, which depends on the category of the first WMMA.
- const int WMMAWaitStates[] = {5, 9, 3, 5, 9, 17};
- const int VALUWaitStates[] = {4, 8, 2, 4, 8, 16};
+ const int WMMAWaitStates[] = {5, 9, 3, 5, 9, 17, 2};
+ const int VALUWaitStates[] = {4, 8, 2, 4, 8, 16, 1};
unsigned Category = 0;
auto IsWMMAHazardFn = [MI, TII, &Category, this](const MachineInstr &I) {
diff --git a/llvm/lib/Target/AMDGPU/SISchedule.td b/llvm/lib/Target/AMDGPU/SISchedule.td
index d542d0780edcc..8dbe4384eb6f1 100644
--- a/llvm/lib/Target/AMDGPU/SISchedule.td
+++ b/llvm/lib/Target/AMDGPU/SISchedule.td
@@ -67,6 +67,7 @@ def Write8PassDGEMM : SchedWrite;
def Write16PassDGEMM : SchedWrite;
// WMMA/SWMMA instructions
+def WriteXDL1PassWMMA : SchedWrite;
def WriteXDL2PassWMMA : SchedWrite;
def WriteXDL4PassWMMA : SchedWrite;
def WriteXDL8PassWMMA : SchedWrite;
@@ -497,18 +498,18 @@ def : HWWriteRes<WriteSALUDummy, [HWSALU], 2>;
} // End SchedModel = GFX12SpeedModel
-// Check if any matrix inputs are interpreted as f8 in an f8f6f4
-// wmma instruction.
-def PredIsF8_WMMA_SCALE : SchedPredicate<[{
- TII->getNamedOperand(*MI, AMDGPU::OpName::matrix_a_fmt)->getImm() <= AMDGPU::WMMA::MATRIX_FMT_BF8 ||
- TII->getNamedOperand(*MI, AMDGPU::OpName::matrix_b_fmt)->getImm() <= AMDGPU::WMMA::MATRIX_FMT_BF8
+// Check whether the two matrix inputs of an f8f6f4 wmma instruction are not
+// both interpreted as f4. On gfx1250 an f8f6f4 wmma runs in 4 cycles when both
+// inputs are f4, and in 8 cycles when any input is f6 or f8.
+// TODO: This isn't reflected in MCA.
+def PredIsNotBothF4_WMMA_SCALE : SchedPredicate<[{
+ TII->getNamedOperand(*MI, AMDGPU::OpName::matrix_a_fmt)->getImm() != AMDGPU::WMMA::MATRIX_FMT_FP4 ||
+ TII->getNamedOperand(*MI, AMDGPU::OpName::matrix_b_fmt)->getImm() != AMDGPU::WMMA::MATRIX_FMT_FP4
}]>;
-// If either matrix format is f8, the instruction takes 2x as many
-// cycles. TODO: This isn't reflected in MCA.
def WriteWMMAScale_16X16X128_F8F6F4 : SchedWriteVariant<[
- SchedVar<PredIsF8_WMMA_SCALE, [WriteXDL4PassWMMA]>,
- SchedVar<NoSchedPred, [WriteXDL2PassWMMA]>
+ SchedVar<PredIsNotBothF4_WMMA_SCALE, [WriteXDL2PassWMMA]>,
+ SchedVar<NoSchedPred, [WriteXDL1PassWMMA]>
]>;
// Check whether matrix_a_fmt and matrix_a_fmt are both MATRIX_FMT_FP4.
@@ -526,6 +527,8 @@ def WriteWMMAScaleFP4_16X16X128_F8F6F4 : SchedWriteVariant<[
multiclass GFX125xCommonWriteRes {
+let ReleaseAtCycles = [4] in
+def : HWWriteRes<WriteXDL1PassWMMA, [HWXDL], 4>;
let ReleaseAtCycles = [8] in
def : HWWriteRes<WriteXDL2PassWMMA, [HWXDL], 8>;
let ReleaseAtCycles = [16] in
@@ -573,7 +576,11 @@ def : HWWriteRes<WriteDoubleAdd, [HWVALU, HWRC], 37>;
def : HWWriteRes<WriteDoubleCvt, [HWVALU, HWRC], 37>;
def : HWWriteRes<WriteTrans64, [HWVALU, HWTransVALU, HWRC], 38>;
-def : InstRW<[WriteXDL2PassWMMA], (instregex "^V_[S]*WMMA[C]*_.*_(FP8|BF8|BF16|F16)_w32")>;
+// gfx1250 WMMA/XDL latencies. 16x16x64 FP8/BF8 run in 4 cycles; 16x16x128
+// FP8/BF8 and F16/BF16 run in 8 cycles; IU8/IU4 run in 16 cycles.
+def : InstRW<[WriteXDL1PassWMMA], (instregex "^V_[S]*WMMA[C]*_.*_16X16X64_(FP8|BF8)_(FP8|BF8)_w32")>;
+def : InstRW<[WriteXDL2PassWMMA], (instregex "^V_[S]*WMMA[C]*_.*_(BF16|F16)_w32")>;
+def : InstRW<[WriteXDL2PassWMMA], (instregex "^V_[S]*WMMA[C]*_.*_16X16X128_(FP8|BF8)_(FP8|BF8)_w32")>;
def : InstRW<[WriteXDL4PassWMMA], (instregex "^V_[S]*WMMA[C]*_.*_(IU8|IU4)_w32")>;
def : InstRW<[WriteWMMAScale_16X16X128_F8F6F4], (instregex "^V_WMMA_.*_16X16X128_F8F6F4.*_w32")>;
def : InstRW<[WriteXDL2PassWMMA], (instregex "^V_WMMA.*_F32_32X16X128_F4")>;
diff --git a/llvm/test/CodeGen/AMDGPU/misched-into-wmma-hazard-shadow.mir b/llvm/test/CodeGen/AMDGPU/misched-into-wmma-hazard-shadow.mir
index e2082ec9f2baf..13c2b340d2d86 100644
--- a/llvm/test/CodeGen/AMDGPU/misched-into-wmma-hazard-shadow.mir
+++ b/llvm/test/CodeGen/AMDGPU/misched-into-wmma-hazard-shadow.mir
@@ -20,10 +20,6 @@ body: |
; GCN-NEXT: $vgpr47 = V_LSHL_ADD_U32_e64 killed $vgpr42, 1, $vgpr42, implicit $exec
; GCN-NEXT: $vgpr48 = V_LSHL_ADD_U32_e64 killed $vgpr41, 1, $vgpr41, implicit $exec
; GCN-NEXT: $vgpr49 = V_LSHL_ADD_U32_e64 killed $vgpr40, 1, $vgpr40, implicit $exec
- ; GCN-NEXT: V_NOP_e32 implicit $exec
- ; GCN-NEXT: V_NOP_e32 implicit $exec
- ; GCN-NEXT: V_NOP_e32 implicit $exec
- ; GCN-NEXT: V_NOP_e32 implicit $exec
; GCN-NEXT: $vgpr4 = V_ADD_F32_e32 1065353216, killed $vgpr4, implicit $mode, implicit $exec
; GCN-NEXT: $vgpr5 = V_ADD_F32_e32 1065353216, killed $vgpr5, implicit $mode, implicit $exec
; GCN-NEXT: $vgpr6 = V_ADD_F32_e32 1065353216, killed $vgpr6, implicit $mode, implicit $exec
diff --git a/llvm/test/CodeGen/AMDGPU/wmma-coexecution-valu-hazards.mir b/llvm/test/CodeGen/AMDGPU/wmma-coexecution-valu-hazards.mir
index 0c09229d4d9c2..246afd986986e 100644
--- a/llvm/test/CodeGen/AMDGPU/wmma-coexecution-valu-hazards.mir
+++ b/llvm/test/CodeGen/AMDGPU/wmma-coexecution-valu-hazards.mir
@@ -291,9 +291,6 @@ body: |
; GFX1250-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_Use1
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr27 = V_ADD_F32_e32 $vgpr22, $vgpr26, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_Use1
@@ -365,9 +362,6 @@ body: |
; GFX1250-NEXT: $sgpr2 = S_MOV_B32 2
; GFX1250-NEXT: $sgpr3 = S_MOV_B32 3
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr27 = V_ADD_F32_e32 $vgpr22, $vgpr26, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_Use1_with_4_salus_in_between
@@ -408,9 +402,6 @@ body: |
; GFX1250-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_D1
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr22 = V_ADD_F32_e32 $vgpr26, $vgpr27, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_D1
@@ -439,9 +430,6 @@ body: |
; GFX1250-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_A0_overlaps_D1
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr0 = V_ADD_F32_e32 $vgpr26, $vgpr27, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_A0_overlaps_D1
@@ -470,9 +458,6 @@ body: |
; GFX1250-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_B0_overlaps_D1
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr8 = V_ADD_F32_e32 $vgpr26, $vgpr27, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_B0_overlaps_D1
@@ -762,10 +747,6 @@ body: |
; GFX1250-NEXT: $vgpr41 = V_MOV_B32_e32 41, implicit $exec
; GFX1250-NEXT: $vgpr42 = V_MOV_B32_e32 42, implicit $exec
; GFX1250-NEXT: $vgpr43 = V_MOV_B32_e32 43, implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr48 = V_ADD_F32_e32 $vgpr32, $vgpr47, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F4_D0_overlaps_Use1_with_4_valus_in_between
@@ -817,10 +798,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr48 = V_ADD_F32_e32 $vgpr32, $vgpr47, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F4_D0_overlaps_Use1_with_4_salus_in_between
@@ -872,10 +849,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr32 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F4_D0_overlaps_D1
@@ -915,10 +888,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr0 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F4_A0_overlaps_D1
@@ -958,10 +927,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr16 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F4_B0_overlaps_D1
@@ -1041,9 +1006,6 @@ body: |
; GFX1250-NEXT: $sgpr2 = S_MOV_B32 2
; GFX1250-NEXT: $sgpr3 = S_MOV_B32 3
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr48 = V_ADD_F32_e32 $vgpr32, $vgpr47, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_D0_overlaps_Use1_with_4_salus_in_between
@@ -1084,9 +1046,6 @@ body: |
; GFX1250-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_D0_overlaps_D1
; GFX1250: early-clobber $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39 = V_WMMA_F32_16X16X128_F8F6F4_f4_f4_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 4, 4, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr32 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_D0_overlaps_D1
@@ -1115,9 +1074,6 @@ body: |
; GFX1250-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_A0_overlaps_D1
; GFX1250: early-clobber $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39 = V_WMMA_F32_16X16X128_F8F6F4_f4_f4_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 4, 4, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr0 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_A0_overlaps_D1
@@ -1146,9 +1102,6 @@ body: |
; GFX1250-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_B0_overlaps_D1
; GFX1250: early-clobber $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39 = V_WMMA_F32_16X16X128_F8F6F4_f4_f4_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 4, 4, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr16 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_B0_overlaps_D1
@@ -1486,10 +1439,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr48 = V_ADD_F32_e32 $vgpr32, $vgpr47, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_D0_overlaps_Use1
@@ -1596,10 +1545,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr48 = V_ADD_F32_e32 $vgpr32, $vgpr47, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_D0_overlaps_Use1_with_8_salus_in_between
@@ -1663,10 +1608,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr32 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_D0_overlaps_D1
@@ -1706,10 +1647,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr0 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_A0_overlaps_D1
@@ -1749,10 +1686,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: $vgpr16 = V_ADD_F32_e32 $vgpr47, $vgpr48, implicit $mode, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_B0_overlaps_D1
diff --git a/llvm/test/CodeGen/AMDGPU/wmma-hazards-gfx1250-w32.mir b/llvm/test/CodeGen/AMDGPU/wmma-hazards-gfx1250-w32.mir
index 1d8cc667d526e..71e117972dc55 100644
--- a/llvm/test/CodeGen/AMDGPU/wmma-hazards-gfx1250-w32.mir
+++ b/llvm/test/CodeGen/AMDGPU/wmma-hazards-gfx1250-w32.mir
@@ -389,9 +389,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, killed $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_fp8_fp8_D0_overlaps_A1
@@ -423,9 +420,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_fp8_fp8_D0_overlaps_B1
@@ -457,9 +451,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr16_vgpr17, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_fp8_fp8_D0_overlaps_Index1
@@ -491,9 +482,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, killed $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_fp8_bf8_D0_overlaps_A1
@@ -525,9 +513,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_fp8_bf8_D0_overlaps_B1
@@ -559,9 +544,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr16_vgpr17, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_fp8_bf8_D0_overlaps_Index1
@@ -593,9 +575,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, killed $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_bf8_fp8_D0_overlaps_A1
@@ -627,9 +606,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_bf8_fp8_D0_overlaps_B1
@@ -661,9 +637,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr16_vgpr17, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_bf8_fp8_D0_overlaps_Index1
@@ -695,9 +668,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, killed $vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_bf8_bf8_D0_overlaps_A1
@@ -729,9 +699,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49 = V_WMMA_F32_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_bf8_bf8_D0_overlaps_B1
@@ -763,9 +730,6 @@ body: |
; GFX1250: early-clobber $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23 = V_WMMA_F32_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr16_vgpr17, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f32_16x16x64_bf8_bf8_D0_overlaps_Index1
@@ -797,9 +761,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_A1
@@ -831,9 +792,6 @@ body: |
; GFX1250: early-clobber $vgpr30_vgpr31_vgpr32_vgpr33 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_B1
@@ -865,9 +823,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr22_vgpr23, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_fp8_D0_overlaps_Index1
@@ -899,9 +854,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_bf8_D0_overlaps_A1
@@ -933,9 +885,6 @@ body: |
; GFX1250: early-clobber $vgpr30_vgpr31_vgpr32_vgpr33 = V_WMMA_F16_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_bf8_D0_overlaps_B1
@@ -967,9 +916,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_FP8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr22_vgpr23, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fp8_bf8_D0_overlaps_Index1
@@ -1001,9 +947,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fb8_fp8_D0_overlaps_A1
@@ -1035,9 +978,6 @@ body: |
; GFX1250: early-clobber $vgpr30_vgpr31_vgpr32_vgpr33 = V_WMMA_F16_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_bf8_fp8_D0_overlaps_B1
@@ -1069,9 +1009,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_BF8_FP8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr22_vgpr23, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_fb8_fp8_D0_overlaps_Index1
@@ -1103,9 +1040,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_bf8_bf8_D0_overlaps_A1
@@ -1137,9 +1071,6 @@ body: |
; GFX1250: early-clobber $vgpr30_vgpr31_vgpr32_vgpr33 = V_WMMA_F16_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr38_vgpr39_vgpr40_vgpr41 = V_WMMA_F16_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29, killed $vgpr30_vgpr31_vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37, 8, killed $vgpr38_vgpr39_vgpr40_vgpr41, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_bf8_bf8_D0_overlaps_B1
@@ -1171,9 +1102,6 @@ body: |
; GFX1250: early-clobber $vgpr22_vgpr23_vgpr24_vgpr25 = V_WMMA_F16_16X16X64_BF8_BF8_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 8, killed $vgpr22_vgpr23_vgpr24_vgpr25, 0, 0, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr22_vgpr23, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_f16_16x16x64_bf8_bf8_D0_overlaps_Index1
@@ -1550,10 +1478,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_A1
@@ -1596,10 +1520,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_B1
@@ -1642,10 +1562,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_Index1
@@ -1812,9 +1728,6 @@ body: |
; GFX1250: early-clobber $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39 = V_WMMA_F32_16X16X128_F8F6F4_f4_f4_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 4, 4, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_D0_overlaps_A1
@@ -1846,9 +1759,6 @@ body: |
; GFX1250: early-clobber $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39 = V_WMMA_F32_16X16X128_F8F6F4_f4_f4_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 4, 4, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_D0_overlaps_B1
@@ -1880,9 +1790,6 @@ body: |
; GFX1250: early-clobber $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39 = V_WMMA_F32_16X16X128_F8F6F4_f4_f4_w32_twoaddr killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23, 8, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39, 4, 4, 0, 0, implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_F32_16x16x128_F8F6F4_F4_F4_D0_overlaps_Index1
@@ -1917,10 +1824,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_scale_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_A1
@@ -1963,10 +1866,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_scale_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_B1
@@ -2009,10 +1908,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_scale_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_Index1
@@ -2181,10 +2076,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_scale16_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_A1
@@ -2227,10 +2118,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71 = V_WMMA_F32_16X16X128_F8F6F4_f8_f8_w32_twoaddr killed $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, 8, killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, 0, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_scale16_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_B1
@@ -2273,10 +2160,6 @@ body: |
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
- ; GFX1250-NEXT: V_NOP_e32 implicit $exec
; GFX1250-NEXT: early-clobber $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 = V_SWMMAC_F32_16X16X128_FP8_FP8_w32_twoaddr killed $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71, killed $vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55, killed $vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, killed $vgpr32_vgpr33, 0, 0, 0, implicit $exec
;
; GFX1251-LABEL: name: test_wmma_scale16_F32_16x16x128_F8F6F4_F8_F8_D0_overlaps_Index1
diff --git a/llvm/test/tools/llvm-mca/AMDGPU/gfx1250-wmma-cycles.s b/llvm/test/tools/llvm-mca/AMDGPU/gfx1250-wmma-cycles.s
new file mode 100644
index 0000000000000..0b16e8f90e094
--- /dev/null
+++ b/llvm/test/tools/llvm-mca/AMDGPU/gfx1250-wmma-cycles.s
@@ -0,0 +1,66 @@
+# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
+# RUN: llvm-mca -mtriple=amdgcn -mcpu=gfx1250 --instruction-info < %s | FileCheck %s
+
+# gfx1250 WMMA/XDL latencies. The 16x16x64 FP8/BF8 forms run in 4 cycles, while
+# the wider 16x16x128 FP8/BF8 and the 16x16x32 F16/BF16 forms run in 8, and IU8
+# runs in 16.
+
+v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23]
+v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23]
+v_wmma_f32_16x16x128_fp8_fp8 v[16:23], v[0:15], v[8:23], v[16:23]
+v_wmma_f32_16x16x128_bf8_bf8 v[16:23], v[0:15], v[8:23], v[16:23]
+v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23]
+
+# CHECK: Iterations: 100
+# CHECK-NEXT: Instructions: 700
+# CHECK-NEXT: Total Cycles: 5601
+# CHECK-NEXT: Total uOps: 700
+
+# CHECK: Dispatch Width: 1
+# CHECK-NEXT: uOps Per Cycle: 0.12
+# CHECK-NEXT: IPC: 0.12
+# CHECK-NEXT: Block RThroughput: 56.0
+
+# CHECK: Instruction Info:
+# CHECK-NEXT: [1]: #uOps
+# CHECK-NEXT: [2]: Latency
+# CHECK-NEXT: [3]: RThroughput
+# CHECK-NEXT: [4]: MayLoad
+# CHECK-NEXT: [5]: MayStore
+# CHECK-NEXT: [6]: HasSideEffects (U)
+
+# CHECK: [1] [2] [3] [4] [5] [6] Instructions:
+# CHECK-NEXT: 1 4 4.00 U v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: 1 4 4.00 U v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: 1 8 8.00 U v_wmma_f32_16x16x128_fp8_fp8 v[16:23], v[0:15], v[8:23], v[16:23]
+# CHECK-NEXT: 1 8 8.00 U v_wmma_f32_16x16x128_bf8_bf8 v[16:23], v[0:15], v[8:23], v[16:23]
+# CHECK-NEXT: 1 8 8.00 U v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: 1 8 8.00 U v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: 1 16 16.00 U v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23]
+
+# CHECK: Resources:
+# CHECK-NEXT: [0] - HWBranch
+# CHECK-NEXT: [1] - HWExport
+# CHECK-NEXT: [2] - HWLGKM
+# CHECK-NEXT: [3] - HWRC
+# CHECK-NEXT: [4] - HWSALU
+# CHECK-NEXT: [5] - HWTransVALU
+# CHECK-NEXT: [6] - HWVALU
+# CHECK-NEXT: [7] - HWVMEM
+# CHECK-NEXT: [8] - HWXDL
+
+# CHECK: Resource pressure per iteration:
+# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8]
+# CHECK-NEXT: - - - - - - - - 56.00
+
+# CHECK: Resource pressure by instruction:
+# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] Instructions:
+# CHECK-NEXT: - - - - - - - - 4.00 v_wmma_f32_16x16x64_fp8_fp8 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: - - - - - - - - 4.00 v_wmma_f32_16x16x64_bf8_bf8 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: - - - - - - - - 8.00 v_wmma_f32_16x16x128_fp8_fp8 v[16:23], v[0:15], v[8:23], v[16:23]
+# CHECK-NEXT: - - - - - - - - 8.00 v_wmma_f32_16x16x128_bf8_bf8 v[16:23], v[0:15], v[8:23], v[16:23]
+# CHECK-NEXT: - - - - - - - - 8.00 v_wmma_f32_16x16x32_f16 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: - - - - - - - - 8.00 v_wmma_f32_16x16x32_bf16 v[16:23], v[0:7], v[8:15], v[16:23]
+# CHECK-NEXT: - - - - - - - - 16.00 v_wmma_i32_16x16x64_iu8 v[16:23], v[0:7], v[8:15], v[16:23]
More information about the llvm-commits
mailing list