Skip to content

release/20.x: AMDGPU: Handle gfx950 XDL-write-VGPR-Overlap-Src-AB wait state (#126732) #126744

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Feb 11, 2025
Merged

release/20.x: AMDGPU: Handle gfx950 XDL-write-VGPR-Overlap-Src-AB wait state (#126732) #126744

merged 1 commit into from
Feb 11, 2025

Conversation

llvmbot
Copy link
Member

@llvmbot llvmbot commented Feb 11, 2025

Backport c837f57

Requested by: @arsenm

@llvmbot llvmbot added this to the LLVM 20.X Release milestone Feb 11, 2025
@llvmbot
Copy link
Member Author

llvmbot commented Feb 11, 2025

@arsenm What do you think about merging this PR to the release branch?

@llvmbot llvmbot requested a review from arsenm February 11, 2025 15:37
@llvmbot llvmbot mentioned this pull request Feb 11, 2025
Merged
@llvmbot
Copy link
Member Author

llvmbot commented Feb 11, 2025

@llvm/pr-subscribers-backend-amdgpu

Author: None (llvmbot)

Changes

Backport c837f57

Requested by: @arsenm

Full diff: https://github.com/llvm/llvm-project/pull/126744.diff

3 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp (+9-7)
  • (modified) llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir (+40-20)
  • (modified) llvm/test/CodeGen/AMDGPU/mai-hazards-mfma-scale.gfx950.mir (+2-2)
diff --git a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
index 873d18e30a430a..844441308275fc 100644
--- a/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
+++ b/llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
@@ -2297,12 +2297,14 @@ GFX940_SMFMA_N_PassWritesVGPROverlappedSrcABWaitStates(int NumPasses) {
   return NumPasses + 2;
 }
 
-static int GFX940_XDL_N_PassWritesVGPROverlappedSrcABWaitStates(int NumPasses) {
-  // 2 pass -> 5
-  // 4 pass -> 7
-  // 8 pass -> 11
-  // 16 pass -> 19
-  return NumPasses + 3;
+static int GFX940_XDL_N_PassWritesVGPROverlappedSrcABWaitStates(int NumPasses,
+                                                                bool IsGFX950) {
+  // xdl def cycles | gfx940 | gfx950
+  // 2 pass         |  5        5
+  // 4 pass         |  7        8
+  // 8 pass         |  11       12
+  // 16 pass        |  19       20
+  return NumPasses + 3 + (NumPasses != 2 && IsGFX950);
 }
 
 int GCNHazardRecognizer::checkMAIHazards90A(MachineInstr *MI) {
@@ -2471,7 +2473,7 @@ int GCNHazardRecognizer::checkMAIHazards90A(MachineInstr *MI) {
           NeedWaitStates =
               isXDL(ST, *MI1)
                   ? GFX940_XDL_N_PassWritesVGPROverlappedSrcABWaitStates(
-                        NumPasses)
+                        NumPasses, ST.hasGFX950Insts())
                   : GFX940_SMFMA_N_PassWritesVGPROverlappedSrcABWaitStates(
                         NumPasses);
           break;
diff --git a/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir b/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
index 52891989b88fbd..1eb7ec4c142f20 100644
--- a/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
+++ b/llvm/test/CodeGen/AMDGPU/mai-hazards-gfx940.mir
@@ -417,7 +417,8 @@ body:             |
 # GCN-LABEL: name: xdl_sgemm16x16_mfma_write_agpr_mfma_srca_read_overlap
 # GCN:      V_MFMA
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_sgemm16x16_mfma_write_agpr_mfma_srca_read_overlap
 body:             |
@@ -439,7 +440,8 @@ body:             |
 # GCN-LABEL: name: smfmac32x32_write_agpr_mfma_srca_read_overlap
 # GCN:      V_SMFMAC
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            smfmac32x32_write_agpr_mfma_srca_read_overlap
 body:             |
@@ -450,7 +452,8 @@ body:             |
 # GCN-LABEL: name: smfmac32x32_write_agpr_smfmac_srcc_read_overlap
 # GCN:      V_SMFMAC
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_SMFMAC
 name:            smfmac32x32_write_agpr_smfmac_srcc_read_overlap
 body:             |
@@ -462,7 +465,8 @@ body:             |
 # GCN:      V_MFMA
 # GCN-NEXT: S_NOP 7
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_sgemm32x32_mfma_write_agpr_mfma_srca_read_overlap
 body:             |
@@ -1715,7 +1719,8 @@ body:             |
 ...
 # GCN-LABEL: name: xdl_sgemm16X16X16_mfma_write_agpr_mfma_srca_read_overlap
 # GCN:      V_MFMA
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_MFMA
 name:            xdl_sgemm16X16X16_mfma_write_agpr_mfma_srca_read_overlap
 body:             |
@@ -1725,7 +1730,8 @@ body:             |
 ...
 # GCN-LABEL: name: xdl_sgemm16X16X32_mfma_write_agpr_mfma_srcb_read_overlap
 # GCN:      V_MFMA
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_MFMA
 name:            xdl_sgemm16X16X32_mfma_write_agpr_mfma_srcb_read_overlap
 body:             |
@@ -1735,7 +1741,8 @@ body:             |
 ...
 # GCN-LABEL: name: xdl_sgemm16X16X16_mfma_write_vgpr_dmfma16x16_srca_read_overlap
 # GCN:      V_MFMA
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_MFMA
 name:            xdl_sgemm16X16X16_mfma_write_vgpr_dmfma16x16_srca_read_overlap
 body:             |
@@ -1826,7 +1833,8 @@ body:             |
 ...
 # GCN-LABEL: name: smfmac16x16x32_mfma_write_vgpr_smfmac_read_idx
 # GCN:      V_SMFMAC
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_SMFMAC
 name:            smfmac16x16x32_mfma_write_vgpr_smfmac_read_idx
 body:             |
@@ -2188,7 +2196,8 @@ body:             |
 # 4 pass source
 # GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srca
 # GCN:      V_MFMA
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_MFMA
 name:            xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srca
 body:             |
@@ -2202,7 +2211,8 @@ body:             |
 # 4 pass source
 # GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srcb
 # GCN:      V_MFMA
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_MFMA
 name:            xdl_mfma_4pass_write_vgpr_xdl_mfma_read_overlap_srcb
 body:             |
@@ -2276,7 +2286,8 @@ body:             |
 # 4 pass source
 # GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srca
 # GCN:      V_MFMA
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_MFMA
 name:            xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srca
 body:             |
@@ -2290,7 +2301,8 @@ body:             |
 # 4 pass source
 # GCN-LABEL: name: xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srcb
 # GCN:      V_MFMA
-# GCN-NEXT: S_NOP 6
+# GFX940-NEXT: S_NOP 6
+# GFX950-NEXT: S_NOP 7
 # GCN-NEXT: V_MFMA
 name:            xdl_mfma_4pass_write_vgpr_sgemm_mfma_read_overlap_srcb
 body:             |
@@ -2321,7 +2333,8 @@ body:             |
 # 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
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name: xdl_mfma_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
 body:             |
@@ -2336,7 +2349,8 @@ body:             |
 # 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
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name: xdl_mfma_8pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
 body:             |
@@ -2370,7 +2384,8 @@ body:             |
 # GCN:      V_MFMA
 # GCN-NEXT: S_NOP 7
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srca
 body:             |
@@ -2386,7 +2401,8 @@ body:             |
 # GCN:      V_MFMA
 # GCN-NEXT: S_NOP 7
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_16pass_write_vgpr_nonxdl_sgemm_mfma_read_overlap_srcb
 body:             |
@@ -2456,7 +2472,8 @@ body:             |
 # 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
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_mfma_8pass_write_vgpr_xdl_mfma_read_overlap_srca
 body:             |
@@ -2470,7 +2487,8 @@ body:             |
 # 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
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_mfma_8pass_write_vgpr_xdl_mfma_read_overlap_srcb
 body:             |
@@ -2502,7 +2520,8 @@ body:             |
 # GCN:      V_MFMA
 # GCN-NEXT: S_NOP 7
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srca
 body:             |
@@ -2519,7 +2538,8 @@ body:             |
 # GCN:      V_MFMA
 # GCN-NEXT: S_NOP 7
 # GCN-NEXT: S_NOP 7
-# GCN-NEXT: S_NOP 2
+# GFX940-NEXT: S_NOP 2
+# GFX950-NEXT: S_NOP 3
 # GCN-NEXT: V_MFMA
 name:            xdl_16pass_write_vgpr_xdl_mfma_read_overlap_srcb
 body:             |
diff --git a/llvm/test/CodeGen/AMDGPU/mai-hazards-mfma-scale.gfx950.mir b/llvm/test/CodeGen/AMDGPU/mai-hazards-mfma-scale.gfx950.mir
index 433236180b1375..4585eca8fe894a 100644
--- a/llvm/test/CodeGen/AMDGPU/mai-hazards-mfma-scale.gfx950.mir
+++ b/llvm/test/CodeGen/AMDGPU/mai-hazards-mfma-scale.gfx950.mir
@@ -254,7 +254,7 @@ body:             |
     ; GCN-NEXT: {{  $}}
     ; GCN-NEXT: renamable $vgpr0_vgpr1_vgpr2_vgpr3 = nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f8_f8_vgprcd_e64 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19, 0, 0, $sgpr4, $vgpr21, 12, 4, implicit $mode, implicit $exec
     ; GCN-NEXT: S_NOP 7
-    ; GCN-NEXT: S_NOP 2
+    ; GCN-NEXT: S_NOP 3
     ; GCN-NEXT: renamable $vgpr0_vgpr1_vgpr2_vgpr3 = nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f8_f8_vgprcd_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, killed $vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, killed $sgpr4, killed $vgpr21, 12, 4, implicit $mode, implicit $exec
     ; GCN-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
     renamable $vgpr0_vgpr1_vgpr2_vgpr3 = nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f8_f8_vgprcd_e64 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19, 0, 0, $sgpr4, $vgpr21, 12, 4, implicit $mode, implicit $exec
@@ -275,7 +275,7 @@ body:             |
     ; GCN: liveins: $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, $vgpr32, $sgpr4
     ; GCN-NEXT: {{  $}}
     ; GCN-NEXT: renamable $vgpr0_vgpr1_vgpr2_vgpr3 = nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f8_f8_vgprcd_e64 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19, 2, 2, $sgpr4, $vgpr21, 12, 4, implicit $mode, implicit $exec
-    ; GCN-NEXT: S_NOP 6
+    ; GCN-NEXT: S_NOP 7
     ; GCN-NEXT: renamable $vgpr0_vgpr1_vgpr2_vgpr3 = nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f8_f8_vgprcd_e64 killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, killed $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, killed $vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, killed $sgpr4, killed $vgpr21, 12, 4, implicit $mode, implicit $exec
     ; GCN-NEXT: S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
     renamable $vgpr0_vgpr1_vgpr2_vgpr3 = nofpexcept V_MFMA_SCALE_F32_16X16X128_F8F6F4_f8_f8_vgprcd_e64 $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, $vgpr16_vgpr17_vgpr18_vgpr19, 2, 2, $sgpr4, $vgpr21, 12, 4, implicit $mode, implicit $exec

@VigneshwarJ @tstellar
AMDGPU: Handle gfx950 XDL-write-VGPR-Overlap-Src-AB wait state (llvm#…
203cd18 
…126732)

gfx950 needs more additional waitstates from gfx940

(cherry picked from commit c837f57)
@tstellar tstellar merged commit 203cd18 into llvm:release/20.x Feb 11, 2025
7 of 10 checks passed
@github-actions GitHub Actions
Copy link

@arsenm (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@arsenm arsenm Awaiting requested review from arsenm

Assignees
No one assigned
Labels
backend:AMDGPU
Projects
LLVM Release Status
Status: Done
Milestone
LLVM 20.X Release
Development

Successfully merging this pull request may close these issues.
3 participants
@llvmbot @tstellar @VigneshwarJ