[llvm] AMDGPU: Add some more mfma hazard recognizer tests (PR #84727)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Mon Mar 11 23:36:19 PDT 2024
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/84727
>From dacb3e4a33cfdbbfd328daf0bf1265d3a54238a6 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Mon, 11 Mar 2024 12:03:27 +0530
Subject: [PATCH 1/2] AMDGPU: Add some more mfma hazard recognizer tests
---
.../CodeGen/AMDGPU/mai-hazards-gfx940.mir | 399 ++++++++++++++++++
1 file changed, 399 insertions(+)
diff --git a/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir b/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
index 4d307a444b19c6..7fa73fc8f74c8b 100644
--- a/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
+++ b/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
@@ -2028,3 +2028,402 @@ body: |
$agpr0_agpr1 = V_MFMA_F64_4X4X4F64_e64 $agpr0_agpr1, $agpr0_agpr1, $agpr0_agpr1, 0, 0, 0, implicit $mode, implicit $exec
BUFFER_STORE_DWORDX2_OFFEN_exact $vgpr2_vgpr3, $vgpr0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec
...
+
+...
+# 2 pass source
+# GCN-LABEL: name: xdl_mfma_2pass_write_vgpr_xdl_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_2pass_write_vgpr_xdl_mfma_read_overlap_srcc
+body: |
+ bb.0:
+
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr2_vgpr3_vgpr4_vgpr5, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 2 pass source
+# GCN-LABEL: name: xdl_mfma_2pass_write_vgpr_xdl_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 4
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_2pass_write_vgpr_xdl_mfma_read_overlap_srca
+body: |
+ bb.0:
+
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr0_vgpr1, $vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 2 pass source
+# GCN-LABEL: name: xdl_mfma_2pass_write_vgpr_xdl_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 4
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_2pass_write_vgpr_xdl_mfma_read_overlap_srcb
+body: |
+ bb.0:
+
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr6_vgpr7, $vgpr2_vgpr3, $vgpr8_vgpr9_vgpr10_vgpr11, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 4 pass source
+# GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 4
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srcc
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr8_vgpr9, $vgpr10_vgpr11, $vgpr2_vgpr3_vgpr4_vgpr5, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 4 pass source
+# GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 6
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srca
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr2_vgpr3, $vgpr10_vgpr11, $vgpr6_vgpr7_vgpr8_vgpr9, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 4 pass source
+# GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 6
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srcb
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr10_vgpr11, $vgpr2_vgpr3, $vgpr6_vgpr7_vgpr8_vgpr9, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 2 pass source
+# GCN-LABEL: name: xdl_mfma_2pass_write_vgpr_sgemm_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_2pass_write_vgpr_sgemm_mfma_read_overlap_srcc
+body: |
+ bb.0:
+
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr6, $vgpr8, $vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, 0, implicit $mode, implicit $exec
+
+...
+
+...
+# 2 pass source
+# GCN-LABEL: name: xdl_mfma_2pass_write_vgpr_sgemm_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 4
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_2pass_write_vgpr_sgemm_mfma_read_overlap_srca
+body: |
+ bb.0:
+
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr1, $vgpr8, $vgpr6_vgpr7_vgpr8_vgpr9, 0, 0, 0, implicit $mode, implicit $exec
+
+...
+
+...
+# 2 pass source
+# GCN-LABEL: name: xdl_mfma_2pass_write_vgpr_sgemm_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 4
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_2pass_write_vgpr_sgemm_mfma_read_overlap_srcb
+body: |
+ bb.0:
+
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X4F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr8, $vgpr1, $vgpr6_vgpr7_vgpr8_vgpr9, 0, 0, 0, implicit $mode, implicit $exec
+
+...
+
+...
+# 4 pass source
+# GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 6
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srcc
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr8, $vgpr1, $vgpr6_vgpr7_vgpr8_vgpr9, 0, 0, 0, implicit $mode, implicit $exec
+
+...
+
+...
+# 4 pass source
+# GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 6
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srca
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr1, $vgpr8, $vgpr6_vgpr7_vgpr8_vgpr9, 0, 0, 0, implicit $mode, implicit $exec
+
+...
+
+...
+# 4 pass source
+# GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 6
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srcb
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr8, $vgpr1, $vgpr6_vgpr7_vgpr8_vgpr9, 0, 0, 0, implicit $mode, implicit $exec
+
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: xdl_mfma_8pass__write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 0
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_8pass__write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcc
+body: |
+ bb.0:
+ renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr0_vgpr1, killed $vgpr2_vgpr3, 1065353216, 0, 0, 0, implicit $mode, implicit $exec
+
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr18, $vgpr19, $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: xdl_mfma_8pass__write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_8pass__write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
+body: |
+ bb.0:
+ renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr0_vgpr1, killed $vgpr2_vgpr3, 1065353216, 0, 0, 0, implicit $mode, implicit $exec
+
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr0, $vgpr33, $vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: xdl_mfma_8pass__write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_8pass__write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
+body: |
+ bb.0:
+ renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr0_vgpr1, killed $vgpr2_vgpr3, 1065353216, 0, 0, 0, implicit $mode, implicit $exec
+
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr33, $vgpr1, $vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 16 pass source
+# GCN-LABEL: name: xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 0
+# GCN-NEXT: V_MFMA
+name: xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcc
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr126_vgpr127, $vgpr128_vgpr129, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X2F32_vgprcd_e64 killed $vgpr32, killed $vgpr33, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 16 pass source
+# GCN-LABEL: name: xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr126_vgpr127, $vgpr128_vgpr129, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X2F32_vgprcd_e64 killed $vgpr0, killed $vgpr33, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 16 pass source
+# GCN-LABEL: name: xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr126_vgpr127, $vgpr128_vgpr129, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X2F32_vgprcd_e64 killed $vgpr33, killed $vgpr0, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47, 1, 2, 3, implicit $mode, implicit $exec
+
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: nonxdl_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: V_MFMA
+name: nonxdl_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcc
+body: |
+ bb.0:
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr18, $vgpr19, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr18, $vgpr19, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: nonxdl_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 1
+# GCN-NEXT: V_MFMA
+name: nonxdl_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
+body: |
+ bb.0:
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr18, $vgpr19, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr3, $vgpr19, $vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: nonxdl_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 1
+# GCN-NEXT: V_MFMA
+name: nonxdl_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
+body: |
+ bb.0:
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr18, $vgpr19, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = V_MFMA_F32_16X16X1F32_vgprcd_e64 $vgpr19, $vgpr3, $vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, implicit $mode, implicit $exec
+...
+...
+# 8 pass source
+# GCN-LABEL: name: xdl_mfma_8pass__write_vgpr_xdl_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 0
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_8pass__write_vgpr_xdl_mfma_read_overlap_srcc
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr0_vgpr1, killed $vgpr2_vgpr3, 1065353216, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr18_vgpr19, killed $vgpr20_vgpr21, $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: xdl_mfma_8pass__write_vgpr_xdl_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_8pass__write_vgpr_xdl_mfma_read_overlap_srca
+body: |
+ bb.0:
+ renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr0_vgpr1, killed $vgpr2_vgpr3, 1065353216, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr2_vgpr3, killed $vgpr36_vgpr37, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 8 pass source
+# GCN-LABEL: name: xdl_mfma_8pass__write_vgpr_xdl_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_mfma_8pass__write_vgpr_xdl_mfma_read_overlap_srcb
+body: |
+ bb.0:
+ renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr0_vgpr1, killed $vgpr2_vgpr3, 1065353216, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = V_MFMA_F32_32X32X8F16_vgprcd_e64 killed $vgpr36_vgpr37, killed $vgpr2_vgpr3, $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+...
+
+...
+# 16 pass source
+# GCN-LABEL: name: xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srcc
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 0
+# GCN-NEXT: V_MFMA
+name: xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srcc
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr126_vgpr127, $vgpr128_vgpr129, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr126_vgpr127, $vgpr128_vgpr129, $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31_vgpr32_vgpr33, 0, 0, 0, implicit $mode, implicit $exec
+
+...
+
+...
+# 16 pass source
+# GCN-LABEL: name: xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srca
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srca
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr126_vgpr127, $vgpr128_vgpr129, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr2_vgpr3, $vgpr128_vgpr129, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, 0, 0, 0, implicit $mode, implicit $exec
+
+
+...
+
+...
+# 16 pass source
+# GCN-LABEL: name: xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srcb
+# GCN: V_MFMA
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 7
+# GCN-NEXT: S_NOP 2
+# GCN-NEXT: V_MFMA
+name: xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srcb
+body: |
+ bb.0:
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr126_vgpr127, $vgpr128_vgpr129, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 = V_MFMA_F32_32X32X4F16_vgprcd_e64 $vgpr128_vgpr129, $vgpr2_vgpr3, $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47_vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63, 0, 0, 0, implicit $mode, implicit $exec
+
+...
>From 3841f97ed10383dde2e23dcbb773a7ef608f6ff3 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Tue, 12 Mar 2024 12:04:35 +0530
Subject: [PATCH 2/2] Fix not overlapping in overlap tests
---
llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir b/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
index 7fa73fc8f74c8b..44a0280bdacd1a 100644
--- a/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
+++ b/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
@@ -2171,7 +2171,7 @@ name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srcc
body: |
bb.0:
$vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_16X16X16F16_vgprcd_e64 $vgpr4_vgpr5, $vgpr6_vgpr7, $vgpr0_vgpr1_vgpr2_vgpr3, 1, 2, 3, implicit $mode, implicit $exec
- $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr8, $vgpr1, $vgpr6_vgpr7_vgpr8_vgpr9, 0, 0, 0, implicit $mode, implicit $exec
+ $vgpr0_vgpr1_vgpr2_vgpr3 = V_MFMA_F32_4X4X1F32_vgprcd_e64 $vgpr8, $vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, 0, implicit $mode, implicit $exec
...
More information about the llvm-commits
mailing list