[AMDGPU][NPM] Port SILateBranchLowering to NPM #130063

optimisan · 2025-03-06T10:58:35Z

No description provided.

optimisan · 2025-03-06T10:58:56Z

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-03-10T05:11:26Z

@llvm/pr-subscribers-backend-amdgpu

Author: Akshat Oke (optimisan)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/130063.diff

6 Files Affected:

(modified) llvm/lib/Target/AMDGPU/AMDGPU.h (+9-1)
(modified) llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def (+1-1)
(modified) llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp (+3-2)
(modified) llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp (+30-10)
(modified) llvm/test/CodeGen/AMDGPU/early-term.mir (+2)
(modified) llvm/test/CodeGen/AMDGPU/readlane_exec0.mir (+1)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.h b/llvm/lib/Target/AMDGPU/AMDGPU.h
index b434676f85581..d1dc62e9cc526 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.h
@@ -213,7 +213,7 @@ extern char &SILowerControlFlowLegacyID;
 void initializeSIPreEmitPeepholePass(PassRegistry &);
 extern char &SIPreEmitPeepholeID;
 
-void initializeSILateBranchLoweringPass(PassRegistry &);
+void initializeSILateBranchLoweringLegacyPass(PassRegistry &);
 extern char &SILateBranchLoweringPassID;
 
 void initializeSIOptimizeExecMaskingLegacyPass(PassRegistry &);
@@ -384,6 +384,14 @@ class SIInsertHardClausesPass : public PassInfoMixin<SIInsertHardClausesPass> {
                         MachineFunctionAnalysisManager &MFAM);
 };
 
+class SILateBranchLoweringPass
+    : public PassInfoMixin<SILateBranchLoweringPass> {
+public:
+  PreservedAnalyses run(MachineFunction &MF,
+                        MachineFunctionAnalysisManager &MFAM);
+  static bool isRequired() { return true; }
+};
+
 FunctionPass *createAMDGPUAnnotateUniformValuesLegacy();
 
 ModulePass *createAMDGPUPrintfRuntimeBinding();
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def b/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def
index 3eabe087a8a33..318aad5590cda 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def
@@ -111,6 +111,7 @@ MACHINE_FUNCTION_PASS("si-form-memory-clauses", SIFormMemoryClausesPass())
 MACHINE_FUNCTION_PASS("si-i1-copies", SILowerI1CopiesPass())
 MACHINE_FUNCTION_PASS("si-insert-hard-clauses", SIInsertHardClausesPass())
 MACHINE_FUNCTION_PASS("si-insert-waitcnts", SIInsertWaitcntsPass())
+MACHINE_FUNCTION_PASS("si-late-branch-lowering", SILateBranchLoweringPass())
 MACHINE_FUNCTION_PASS("si-load-store-opt", SILoadStoreOptimizerPass())
 MACHINE_FUNCTION_PASS("si-lower-control-flow", SILowerControlFlowPass())
 MACHINE_FUNCTION_PASS("si-lower-sgpr-spills", SILowerSGPRSpillsPass())
@@ -132,7 +133,6 @@ DUMMY_MACHINE_FUNCTION_PASS("amdgpu-pre-ra-optimizations", GCNPreRAOptimizations
 DUMMY_MACHINE_FUNCTION_PASS("amdgpu-rewrite-partial-reg-uses", GCNRewritePartialRegUsesPass())
 DUMMY_MACHINE_FUNCTION_PASS("amdgpu-set-wave-priority", AMDGPUSetWavePriorityPass())
 
-DUMMY_MACHINE_FUNCTION_PASS("si-late-branch-lowering", SILateBranchLoweringPass())
 DUMMY_MACHINE_FUNCTION_PASS("si-pre-emit-peephole", SIPreEmitPeepholePass())
 // TODO: Move amdgpu-preload-kern-arg-prolog to MACHINE_FUNCTION_PASS since it
 // already exists.
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 6c24fe5f1441a..b9d62cc9e4b63 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -540,7 +540,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
   initializeSIWholeQuadModeLegacyPass(*PR);
   initializeSILowerControlFlowLegacyPass(*PR);
   initializeSIPreEmitPeepholePass(*PR);
-  initializeSILateBranchLoweringPass(*PR);
+  initializeSILateBranchLoweringLegacyPass(*PR);
   initializeSIMemoryLegalizerLegacyPass(*PR);
   initializeSIOptimizeExecMaskingLegacyPass(*PR);
   initializeSIPreAllocateWWMRegsLegacyPass(*PR);
@@ -2161,7 +2161,8 @@ void AMDGPUCodeGenPassBuilder::addPreEmitPass(AddMachinePass &addPass) const {
     // TODO: addPass(SIInsertHardClausesPass());
   }
 
-  // addPass(SILateBranchLoweringPass());
+  addPass(SILateBranchLoweringPass());
+
   if (isPassEnabled(EnableSetWavePriority, CodeGenOptLevel::Less)) {
     // TODO: addPass(AMDGPUSetWavePriorityPass());
   }
diff --git a/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp b/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp
index d02173f57ee37..0f5b6bd9374b0 100644
--- a/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp
@@ -16,6 +16,7 @@
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
 #include "SIMachineFunctionInfo.h"
 #include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachinePassManager.h"
 
 using namespace llvm;
 
@@ -23,7 +24,7 @@ using namespace llvm;
 
 namespace {
 
-class SILateBranchLowering : public MachineFunctionPass {
+class SILateBranchLowering {
 private:
   const SIRegisterInfo *TRI = nullptr;
   const SIInstrInfo *TII = nullptr;
@@ -33,14 +34,23 @@ class SILateBranchLowering : public MachineFunctionPass {
   void earlyTerm(MachineInstr &MI, MachineBasicBlock *EarlyExitBlock);
 
 public:
-  static char ID;
+  SILateBranchLowering(MachineDominatorTree *MDT) : MDT(MDT) {}
+
+  bool run(MachineFunction &MF);
 
   unsigned MovOpc;
   Register ExecReg;
+};
 
-  SILateBranchLowering() : MachineFunctionPass(ID) {}
+class SILateBranchLoweringLegacy : public MachineFunctionPass {
+public:
+  static char ID;
+  SILateBranchLoweringLegacy() : MachineFunctionPass(ID) {}
 
-  bool runOnMachineFunction(MachineFunction &MF) override;
+  bool runOnMachineFunction(MachineFunction &MF) override {
+    auto *MDT = &getAnalysis<MachineDominatorTreeWrapperPass>().getDomTree();
+    return SILateBranchLowering(MDT).run(MF);
+  }
 
   StringRef getPassName() const override {
     return "SI Final Branch Preparation";
@@ -55,15 +65,15 @@ class SILateBranchLowering : public MachineFunctionPass {
 
 } // end anonymous namespace
 
-char SILateBranchLowering::ID = 0;
+char SILateBranchLoweringLegacy::ID = 0;
 
-INITIALIZE_PASS_BEGIN(SILateBranchLowering, DEBUG_TYPE,
+INITIALIZE_PASS_BEGIN(SILateBranchLoweringLegacy, DEBUG_TYPE,
                       "SI insert s_cbranch_execz instructions", false, false)
 INITIALIZE_PASS_DEPENDENCY(MachineDominatorTreeWrapperPass)
-INITIALIZE_PASS_END(SILateBranchLowering, DEBUG_TYPE,
+INITIALIZE_PASS_END(SILateBranchLoweringLegacy, DEBUG_TYPE,
                     "SI insert s_cbranch_execz instructions", false, false)
 
-char &llvm::SILateBranchLoweringPassID = SILateBranchLowering::ID;
+char &llvm::SILateBranchLoweringPassID = SILateBranchLoweringLegacy::ID;
 
 static void generateEndPgm(MachineBasicBlock &MBB,
                            MachineBasicBlock::iterator I, DebugLoc DL,
@@ -144,11 +154,21 @@ void SILateBranchLowering::earlyTerm(MachineInstr &MI,
   MDT->insertEdge(&MBB, EarlyExitBlock);
 }
 
-bool SILateBranchLowering::runOnMachineFunction(MachineFunction &MF) {
+PreservedAnalyses
+llvm::SILateBranchLoweringPass::run(MachineFunction &MF,
+                                    MachineFunctionAnalysisManager &MFAM) {
+  auto *MDT = &MFAM.getResult<MachineDominatorTreeAnalysis>(MF);
+  if (!SILateBranchLowering(MDT).run(MF))
+    return PreservedAnalyses::all();
+
+  return getMachineFunctionPassPreservedAnalyses()
+      .preserve<MachineDominatorTreeAnalysis>();
+}
+
+bool SILateBranchLowering::run(MachineFunction &MF) {
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   TII = ST.getInstrInfo();
   TRI = &TII->getRegisterInfo();
-  MDT = &getAnalysis<MachineDominatorTreeWrapperPass>().getDomTree();
 
   MovOpc = ST.isWave32() ? AMDGPU::S_MOV_B32 : AMDGPU::S_MOV_B64;
   ExecReg = ST.isWave32() ? AMDGPU::EXEC_LO : AMDGPU::EXEC;
diff --git a/llvm/test/CodeGen/AMDGPU/early-term.mir b/llvm/test/CodeGen/AMDGPU/early-term.mir
index 77bc9729ee845..3d75d405a46d3 100644
--- a/llvm/test/CodeGen/AMDGPU/early-term.mir
+++ b/llvm/test/CodeGen/AMDGPU/early-term.mir
@@ -2,6 +2,8 @@
 # RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize64 -run-pass=si-late-branch-lowering -verify-machineinstrs  %s -o - | FileCheck -check-prefixes=GCN,GFX10 %s
 # RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize64 -run-pass=si-late-branch-lowering -verify-machineinstrs  %s -o - | FileCheck -check-prefixes=GCN,GFX11 %s
 
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize64 -passes=si-late-branch-lowering -verify-machineinstrs  %s -o - | FileCheck -check-prefixes=GCN,GFX11 %s
+
 --- |
   define amdgpu_ps void @early_term_scc0_end_block() {
     ret void
diff --git a/llvm/test/CodeGen/AMDGPU/readlane_exec0.mir b/llvm/test/CodeGen/AMDGPU/readlane_exec0.mir
index 6a286eafa6d58..a4c05aa781df7 100644
--- a/llvm/test/CodeGen/AMDGPU/readlane_exec0.mir
+++ b/llvm/test/CodeGen/AMDGPU/readlane_exec0.mir
@@ -1,4 +1,5 @@
 # RUN: llc -o - %s -mtriple=amdgcn -mcpu=fiji -run-pass=si-late-branch-lowering -verify-machineinstrs | FileCheck -check-prefix=GCN %s
+# RUN: llc -o - %s -mtriple=amdgcn -mcpu=fiji -passes=si-late-branch-lowering -verify-machineinstrs | FileCheck -check-prefix=GCN %s
 
 # GCN-LABEL: readlane_exec0
 # GCN: bb.0

llvm/test/CodeGen/AMDGPU/readlane_exec0.mir

llvm/test/CodeGen/AMDGPU/early-term.mir

optimisan force-pushed the users/optimisan/preemit/port-si-insert-hard-clauses branch from 88d2174 to 39768ea Compare March 7, 2025 10:11

optimisan force-pushed the users/optimisan/preemit/port-si-late-branch-lowering branch from 5f050b8 to 33a5201 Compare March 7, 2025 10:12

optimisan force-pushed the users/optimisan/preemit/port-si-insert-hard-clauses branch from 39768ea to 79a590f Compare March 10, 2025 04:44

optimisan force-pushed the users/optimisan/preemit/port-si-late-branch-lowering branch from 33a5201 to 70a3b58 Compare March 10, 2025 04:44

optimisan mentioned this pull request Mar 10, 2025

[AMDGPU][NFC] Format GCNCreateVOPD.cpp #130548

Merged

optimisan marked this pull request as ready for review March 10, 2025 05:10

llvmbot added the backend:AMDGPU label Mar 10, 2025

optimisan requested review from arsenm, cdevadas and vikramRH March 10, 2025 05:11

arsenm approved these changes Mar 10, 2025

View reviewed changes

optimisan force-pushed the users/optimisan/preemit/port-si-insert-hard-clauses branch from 79a590f to 48ec9e7 Compare March 10, 2025 05:21

optimisan force-pushed the users/optimisan/preemit/port-si-late-branch-lowering branch from 70a3b58 to 03b1e8e Compare March 10, 2025 05:22

cdevadas approved these changes Mar 10, 2025

View reviewed changes

llvm/test/CodeGen/AMDGPU/readlane_exec0.mir Outdated Show resolved Hide resolved

llvm/test/CodeGen/AMDGPU/early-term.mir Outdated Show resolved Hide resolved

optimisan force-pushed the users/optimisan/preemit/port-si-insert-hard-clauses branch from 48ec9e7 to 78cee57 Compare March 11, 2025 09:32

optimisan force-pushed the users/optimisan/preemit/port-si-late-branch-lowering branch from 03b1e8e to f095da2 Compare March 11, 2025 09:39

optimisan force-pushed the users/optimisan/preemit/port-si-insert-hard-clauses branch from 78cee57 to 3a314ee Compare March 25, 2025 06:36

Base automatically changed from users/optimisan/preemit/port-si-insert-hard-clauses to main March 25, 2025 10:03

optimisan added 2 commits March 26, 2025 08:12

[AMDGPU][NPM] Port SILateBranchLowering to NPM

d2e75f5

AS

0d5f939

optimisan force-pushed the users/optimisan/preemit/port-si-late-branch-lowering branch from f095da2 to 0d5f939 Compare March 26, 2025 08:14

arsenm approved these changes Mar 26, 2025

View reviewed changes

optimisan merged commit 719b029 into main Mar 26, 2025
11 checks passed

optimisan deleted the users/optimisan/preemit/port-si-late-branch-lowering branch March 26, 2025 13:58

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[AMDGPU][NPM] Port SILateBranchLowering to NPM #130063

[AMDGPU][NPM] Port SILateBranchLowering to NPM #130063

optimisan commented Mar 6, 2025

Uh oh!

optimisan commented Mar 6, 2025 •

edited

Loading

Uh oh!

llvmbot commented Mar 10, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

[AMDGPU][NPM] Port SILateBranchLowering to NPM #130063

[AMDGPU][NPM] Port SILateBranchLowering to NPM #130063

Conversation

optimisan commented Mar 6, 2025

Uh oh!

optimisan commented Mar 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Mar 10, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

optimisan commented Mar 6, 2025 •

edited

Loading