[AMDGPU] Do not put memory instructions in *ALU SchedGroups #162560
Merged
+67
−4
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-backend-amdgpu Author: Jeffrey Byrnes (jrbyrnes) ChangesClassifying some memory instructions as VALU leads to unexpected behavior from the sched*barrier intrinsics. Full diff: https://github.com/llvm/llvm-project/pull/162560.diff 2 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp b/llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp
index dbe74b1b08f8c..a161c21f679e5 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp
@@ -2394,15 +2394,18 @@ bool SchedGroup::canAddMI(const MachineInstr &MI) const {
else if (((SGMask & SchedGroupMask::ALU) != SchedGroupMask::NONE) &&
(TII->isVALU(MI) || TII->isMFMAorWMMA(MI) || TII->isSALU(MI) ||
TII->isTRANS(MI)))
- Result = true;
+ Result = !(MI.mayLoad() || MI.mayLoad());
else if (((SGMask & SchedGroupMask::VALU) != SchedGroupMask::NONE) &&
TII->isVALU(MI) && !TII->isMFMAorWMMA(MI) && !TII->isTRANS(MI))
- Result = true;
+ // Some memory instructions may be marked as VALU (e.g. BUFFER_LOAD_*_LDS).
+ // For our purposes, these shall not be classified as VALU as this results
+ // in unexpected behavior.
+ Result = !(MI.mayLoad() || MI.mayLoad());
else if (((SGMask & SchedGroupMask::SALU) != SchedGroupMask::NONE) &&
TII->isSALU(MI))
- Result = true;
+ Result = !(MI.mayLoad() || MI.mayLoad());
else if (((SGMask & SchedGroupMask::MFMA) != SchedGroupMask::NONE) &&
TII->isMFMAorWMMA(MI))
diff --git a/llvm/test/CodeGen/AMDGPU/sched.group.classification.mir b/llvm/test/CodeGen/AMDGPU/sched.group.classification.mir
new file mode 100644
index 0000000000000..a4aad574aaaf4
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/sched.group.classification.mir
@@ -0,0 +1,59 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx950 -run-pass=machine-scheduler -o - %s | FileCheck %s
+
+---
+name: buffer_load_lds_not_valu
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1
+ ; CHECK-LABEL: name: buffer_load_lds_not_valu
+ ; CHECK: liveins: $vgpr0_vgpr1
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $exec = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF1:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF2:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF3:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_ADD_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF2]], [[DEF3]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[DEF3]], [[V_ADD_U32_e32_]], implicit $exec
+ ; CHECK-NEXT: $m0 = S_MOV_B32 0
+ ; CHECK-NEXT: BUFFER_LOAD_DWORDX4_LDS_OFFEN [[DEF]], [[DEF1]], 0, 0, 0, 0, implicit $exec, implicit $m0
+ ; CHECK-NEXT: [[V_ADD_U32_e32_2:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[V_ADD_U32_e32_]], [[V_ADD_U32_e32_1]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_3:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[V_ADD_U32_e32_1]], [[V_ADD_U32_e32_2]], implicit $exec
+ ; CHECK-NEXT: $m0 = S_MOV_B32 1
+ ; CHECK-NEXT: BUFFER_LOAD_DWORDX4_LDS_OFFEN [[DEF]], [[DEF1]], 0, 0, 0, 0, implicit $exec, implicit $m0
+ ; CHECK-NEXT: [[V_ADD_U32_e32_4:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[V_ADD_U32_e32_2]], [[V_ADD_U32_e32_3]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_5:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[V_ADD_U32_e32_3]], [[V_ADD_U32_e32_4]], implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U32_e32_6:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[V_ADD_U32_e32_4]], [[V_ADD_U32_e32_5]], implicit $exec
+ ; CHECK-NEXT: dead [[V_ADD_U32_e32_7:%[0-9]+]]:vgpr_32 = V_ADD_U32_e32 [[V_ADD_U32_e32_5]], [[V_ADD_U32_e32_6]], implicit $exec
+ ; CHECK-NEXT: SCHED_GROUP_BARRIER 2, 2, 0
+ ; CHECK-NEXT: SCHED_GROUP_BARRIER 4, 1, 0
+ ; CHECK-NEXT: SCHED_GROUP_BARRIER 2, 2, 0
+ ; CHECK-NEXT: SCHED_GROUP_BARRIER 4, 1, 0
+ ; CHECK-NEXT: SCHED_GROUP_BARRIER 2, 4, 0
+ ; CHECK-NEXT: S_ENDPGM 0
+ $exec = IMPLICIT_DEF
+ %0:vgpr_32 = IMPLICIT_DEF
+ %1:sgpr_128 = IMPLICIT_DEF
+ %2:vgpr_32 = IMPLICIT_DEF
+ %3:vgpr_32 = IMPLICIT_DEF
+ %4:vgpr_32 = V_ADD_U32_e32 %2, %3, implicit $exec
+ %5:vgpr_32 = V_ADD_U32_e32 %3, %4, implicit $exec
+ $m0 = S_MOV_B32 0
+ BUFFER_LOAD_DWORDX4_LDS_OFFEN %0, %1, 0, 0, 0, 0, implicit $exec, implicit $m0
+ $m0 = S_MOV_B32 1
+ BUFFER_LOAD_DWORDX4_LDS_OFFEN %0, %1, 0, 0, 0, 0, implicit $exec, implicit $m0
+ %6:vgpr_32 = V_ADD_U32_e32 %4, %5, implicit $exec
+ %7:vgpr_32 = V_ADD_U32_e32 %5, %6, implicit $exec
+ %8:vgpr_32 = V_ADD_U32_e32 %6, %7, implicit $exec
+ %9:vgpr_32 = V_ADD_U32_e32 %7, %8, implicit $exec
+ %10:vgpr_32 = V_ADD_U32_e32 %8, %9, implicit $exec
+ %11:vgpr_32 = V_ADD_U32_e32 %9, %10, implicit $exec
+ SCHED_GROUP_BARRIER 2, 2, 0
+ SCHED_GROUP_BARRIER 4, 1 ,0
+ SCHED_GROUP_BARRIER 2, 2, 0
+ SCHED_GROUP_BARRIER 4, 1 ,0
+ SCHED_GROUP_BARRIER 2, 4, 0
+ S_ENDPGM 0
+...
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
Change-Id: I35069ec1bc6bd47984b288f727cf8df6d6797e47
jrbyrnes
force-pushed
the
VALUSchedGroup
branch
from
6ea495f to
9f380e9
Compare
arsenm
reviewed
Oct 10, 2025
Change-Id: I82460e1b50d785833a46e9c17722382cfa845923
arsenm
reviewed
Oct 10, 2025
Change-Id: I2c65dd255ad37b3dab4449ccf94f972e78ebcced
arsenm
approved these changes
Oct 11, 2025
Change-Id: I8a16952a2f1007e470e86d3c985b92291c27a6be
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/11/builds/25917 Here is the relevant piece of the build log for the reference |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Classifying some memory instructions as VALU leads to unexpected behavior from the sched*barrier intrinsics.