AMDGPU/GlobalISel: add AMDGPUGlobalISelDivergenceLowering pass (llvm#75340)

Add empty AMDGPUGlobalISelDivergenceLowering pass. This pass will
implement
- selection of divergent i1 phis as lane mask phis, requires lane mask
merging in some cases
- lower uses of divergent i1 values outside of the cycle using lane mask
merging
- lowering of all cases of temporal divergence:
- lower uses of uniform i1 values outside of the cycle using lane mask
merging
- lower uses of uniform non-i1 values outside of the cycle using a copy
to vgpr inside of the cycle

Add very detailed set of regression tests for cases mentioned above.

patch 1 from: llvm#73337

Author: petar-avramovic

llvm/lib/Target/AMDGPU/AMDGPU.h

@@ -36,6 +36,7 @@ FunctionPass *createSIAnnotateControlFlowPass();
 FunctionPass *createSIFoldOperandsPass();
 FunctionPass *createSIPeepholeSDWAPass();
 FunctionPass *createSILowerI1CopiesPass();
+FunctionPass *createAMDGPUGlobalISelDivergenceLoweringPass();
 FunctionPass *createSIShrinkInstructionsPass();
 FunctionPass *createSILoadStoreOptimizerPass();
 FunctionPass *createSIWholeQuadModePass();
@@ -162,6 +163,9 @@ extern char &SILowerWWMCopiesID;
 void initializeSILowerI1CopiesPass(PassRegistry &);
 extern char &SILowerI1CopiesID;
 
+void initializeAMDGPUGlobalISelDivergenceLoweringPass(PassRegistry &);
+extern char &AMDGPUGlobalISelDivergenceLoweringID;
+
 void initializeSILowerSGPRSpillsPass(PassRegistry &);
 extern char &SILowerSGPRSpillsID;
 

llvm/lib/Target/AMDGPU/AMDGPUGlobalISelDivergenceLowering.cpp

@@ -0,0 +1,68 @@
+//===-- AMDGPUGlobalISelDivergenceLowering.cpp ----------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// GlobalISel pass that selects divergent i1 phis as lane mask phis.
+/// Lane mask merging uses same algorithm as SDAG in SILowerI1Copies.
+/// Handles all cases of temporal divergence.
+/// For divergent non-phi i1 and uniform i1 uses outside of the cycle this pass
+/// currently depends on LCSSA to insert phis with one incoming.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+
+#define DEBUG_TYPE "amdgpu-global-isel-divergence-lowering"
+
+using namespace llvm;
+
+namespace {
+
+class AMDGPUGlobalISelDivergenceLowering : public MachineFunctionPass {
+public:
+  static char ID;
+
+public:
+  AMDGPUGlobalISelDivergenceLowering() : MachineFunctionPass(ID) {
+    initializeAMDGPUGlobalISelDivergenceLoweringPass(
+        *PassRegistry::getPassRegistry());
+  }
+
+  bool runOnMachineFunction(MachineFunction &MF) override;
+
+  StringRef getPassName() const override {
+    return "AMDGPU GlobalISel divergence lowering";
+  }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesCFG();
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+};
+
+} // End anonymous namespace.
+
+INITIALIZE_PASS_BEGIN(AMDGPUGlobalISelDivergenceLowering, DEBUG_TYPE,
+                      "AMDGPU GlobalISel divergence lowering", false, false)
+INITIALIZE_PASS_END(AMDGPUGlobalISelDivergenceLowering, DEBUG_TYPE,
+                    "AMDGPU GlobalISel divergence lowering", false, false)
+
+char AMDGPUGlobalISelDivergenceLowering::ID = 0;
+
+char &llvm::AMDGPUGlobalISelDivergenceLoweringID =
+    AMDGPUGlobalISelDivergenceLowering::ID;
+
+FunctionPass *llvm::createAMDGPUGlobalISelDivergenceLoweringPass() {
+  return new AMDGPUGlobalISelDivergenceLowering();
+}
+
+bool AMDGPUGlobalISelDivergenceLowering::runOnMachineFunction(
+    MachineFunction &MF) {
+  return false;
+}

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -375,6 +375,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
   initializeAMDGPUDAGToDAGISelPass(*PR);
   initializeGCNDPPCombinePass(*PR);
   initializeSILowerI1CopiesPass(*PR);
+  initializeAMDGPUGlobalISelDivergenceLoweringPass(*PR);
   initializeSILowerWWMCopiesPass(*PR);
   initializeSILowerSGPRSpillsPass(*PR);
   initializeSIFixSGPRCopiesPass(*PR);
@@ -1255,6 +1256,7 @@ bool GCNPassConfig::addLegalizeMachineIR() {
 void GCNPassConfig::addPreRegBankSelect() {
   bool IsOptNone = getOptLevel() == CodeGenOptLevel::None;
   addPass(createAMDGPUPostLegalizeCombiner(IsOptNone));
+  addPass(createAMDGPUGlobalISelDivergenceLoweringPass());
 }
 
 bool GCNPassConfig::addRegBankSelect() {

llvm/lib/Target/AMDGPU/CMakeLists.txt

@@ -55,6 +55,7 @@ add_llvm_target(AMDGPUCodeGen
   AMDGPUCtorDtorLowering.cpp
   AMDGPUExportClustering.cpp
   AMDGPUFrameLowering.cpp
+  AMDGPUGlobalISelDivergenceLowering.cpp
   AMDGPUGlobalISelUtils.cpp
   AMDGPUHSAMetadataStreamer.cpp
   AMDGPUInsertDelayAlu.cpp