Skip to content

[AMDGPU] Add a debug option -amdgpu-snop-padding for GCNHazardRecognizer #146587

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

Merged
merged 3 commits into from
Jul 2, 2025
Merged

[AMDGPU] Add a debug option -amdgpu-snop-padding for GCNHazardRecognizer #146587

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
975e492
[AMDGPU] Add a debug option `-amdgpu-snop-padding` for `GCNHazardReco…
shiltian Jul 1, 2025
18cd251
resolve review comments
shiltian Jul 2, 2025
54bace4
fix comments
shiltian Jul 2, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion llvm/lib/Target/AMDGPU/GCNHazardRecognizer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,11 @@ static cl::opt<unsigned, false, MFMAPaddingRatioParser>
cl::desc("Fill a percentage of the latency between "
"neighboring MFMA with s_nops."));

// This is intended for debugging purposes only.
static cl::opt<unsigned>
NopPadding("amdgpu-snop-padding", cl::init(0), cl::Hidden,
cl::desc("Insert a s_nop x before every instruction"));

//===----------------------------------------------------------------------===//
// Hazard Recognizer Implementation
//===----------------------------------------------------------------------===//
Expand Down Expand Up @@ -300,7 +305,7 @@ unsigned GCNHazardRecognizer::PreEmitNoops(MachineInstr *MI) {
unsigned W = PreEmitNoopsCommon(MI);
fixHazards(MI);
CurrCycleInstr = nullptr;
return W;
return std::max(W, NopPadding.getValue());
}

unsigned GCNHazardRecognizer::PreEmitNoopsCommon(MachineInstr *MI) {
Expand Down
132 changes: 132 additions & 0 deletions llvm/test/CodeGen/AMDGPU/amdgpu-snop-padding.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,132 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -amdgpu-snop-padding=8 -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=GCN8 %s
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx942 -amdgpu-snop-padding=16 -run-pass post-RA-hazard-rec %s -o - | FileCheck -check-prefix=GCN16 %s

---
name: test_snop_padding
tracksRegLiveness: true
frameInfo:
maxAlignment: 4
stack:
- { id: 0, type: spill-slot, size: 4, alignment: 4, stack-id: sgpr-spill }
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: '$sgpr0_sgpr1_sgpr2_sgpr3'
stackPtrOffsetReg: '$sgpr32'
frameOffsetReg: '$sgpr33'
hasSpilledSGPRs: true
body: |
; GCN8-LABEL: name: test_snop_padding
; GCN8: bb.0:
; GCN8-NEXT: successors: %bb.1(0x80000000)
; GCN8-NEXT: liveins: $sgpr6, $sgpr10_sgpr11
; GCN8-NEXT: {{ $}}
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: S_BRANCH %bb.1
; GCN8-NEXT: {{ $}}
; GCN8-NEXT: bb.1:
; GCN8-NEXT: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; GCN8-NEXT: liveins: $sgpr6, $sgpr10_sgpr11
; GCN8-NEXT: {{ $}}
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 10, implicit $exec
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: S_CBRANCH_EXECZ %bb.3, implicit $exec
; GCN8-NEXT: {{ $}}
; GCN8-NEXT: bb.2:
; GCN8-NEXT: successors: %bb.3(0x80000000)
; GCN8-NEXT: liveins: $sgpr6, $sgpr10_sgpr11
; GCN8-NEXT: {{ $}}
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: SI_SPILL_S32_SAVE killed $sgpr6, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: S_NOP 0
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: renamable $sgpr6 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 20, implicit $exec
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: S_BRANCH %bb.3
; GCN8-NEXT: {{ $}}
; GCN8-NEXT: bb.3:
; GCN8-NEXT: liveins: $sgpr10_sgpr11
; GCN8-NEXT: {{ $}}
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: $sgpr5 = V_READFIRSTLANE_B32 [[V_MOV_B32_e32_]], implicit $exec
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: S_STORE_DWORD_IMM $sgpr5, $sgpr10_sgpr11, 0, 0
; GCN8-NEXT: S_NOP 7
; GCN8-NEXT: SI_RETURN
;
; GCN16-LABEL: name: test_snop_padding
; GCN16: bb.0:
; GCN16-NEXT: successors: %bb.1(0x80000000)
; GCN16-NEXT: liveins: $sgpr6, $sgpr10_sgpr11
; GCN16-NEXT: {{ $}}
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_BRANCH %bb.1
; GCN16-NEXT: {{ $}}
; GCN16-NEXT: bb.1:
; GCN16-NEXT: successors: %bb.3(0x40000000), %bb.2(0x40000000)
; GCN16-NEXT: liveins: $sgpr6, $sgpr10_sgpr11
; GCN16-NEXT: {{ $}}
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 10, implicit $exec
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_CBRANCH_EXECZ %bb.3, implicit $exec
; GCN16-NEXT: {{ $}}
; GCN16-NEXT: bb.2:
; GCN16-NEXT: successors: %bb.3(0x80000000)
; GCN16-NEXT: liveins: $sgpr6, $sgpr10_sgpr11
; GCN16-NEXT: {{ $}}
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: SI_SPILL_S32_SAVE killed $sgpr6, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 0
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: renamable $sgpr6 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 20, implicit $exec
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_BRANCH %bb.3
; GCN16-NEXT: {{ $}}
; GCN16-NEXT: bb.3:
; GCN16-NEXT: liveins: $sgpr10_sgpr11
; GCN16-NEXT: {{ $}}
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: $sgpr5 = V_READFIRSTLANE_B32 [[V_MOV_B32_e32_]], implicit $exec
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_STORE_DWORD_IMM $sgpr5, $sgpr10_sgpr11, 0, 0
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: S_NOP 7
; GCN16-NEXT: SI_RETURN
bb.0:
liveins: $sgpr6, $sgpr10_sgpr11
S_BRANCH %bb.1
bb.1:
liveins: $sgpr6, $sgpr10_sgpr11
%0:vgpr_32 = V_MOV_B32_e32 10, implicit $exec
S_CBRANCH_EXECZ %bb.3, implicit $exec
bb.2:
liveins: $sgpr6, $sgpr10_sgpr11
SI_SPILL_S32_SAVE killed $sgpr6, %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
S_NOP 0
renamable $sgpr6 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $sgpr32
%0:vgpr_32 = V_MOV_B32_e32 20, implicit $exec
S_BRANCH %bb.3
bb.3:
liveins: $sgpr10_sgpr11
$sgpr5 = V_READFIRSTLANE_B32 %0:vgpr_32, implicit $exec
S_STORE_DWORD_IMM $sgpr5, $sgpr10_sgpr11, 0, 0
SI_RETURN
...
Loading