[AMDGPU] Add a new pass to insert waitcnts. Leave under an option for…

… testing. Based on comments in https://reviews.llvm.org/D31161. llvm-svn: 300023
llvm · Apr 12, 2017 · acb089e · acb089e
1 parent 04300b0
commit acb089e
Show file tree

Hide file tree

Showing 7 changed files with 1,881 additions and 6 deletions.
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.h b/llvm/lib/Target/AMDGPU/AMDGPU.h
@@ -48,6 +48,7 @@ FunctionPass *createSIFixControlFlowLiveIntervalsPass();
 FunctionPass *createSIFixSGPRCopiesPass();
 FunctionPass *createSIDebuggerInsertNopsPass();
 FunctionPass *createSIInsertWaitsPass();
+FunctionPass *createSIInsertWaitcntsPass();
 FunctionPass *createAMDGPUCodeGenPreparePass(const GCNTargetMachine *TM = nullptr);
 
 ModulePass *createAMDGPUAnnotateKernelFeaturesPass(const TargetMachine *TM = nullptr);
@@ -125,6 +126,9 @@ extern char &SIDebuggerInsertNopsID;
 void initializeSIInsertWaitsPass(PassRegistry&);
 extern char &SIInsertWaitsID;
 
+void initializeSIInsertWaitcntsPass(PassRegistry&);
+extern char &SIInsertWaitcntsID;
+
 void initializeAMDGPUUnifyDivergentExitNodesPass(PassRegistry&);
 extern char &AMDGPUUnifyDivergentExitNodesID;
 

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -112,6 +112,12 @@ static cl::opt<bool> EnableAMDGPUAliasAnalysis("enable-amdgpu-aa", cl::Hidden,
   cl::desc("Enable AMDGPU Alias Analysis"),
   cl::init(true));
 
+// Option to enable new waitcnt insertion pass.
+static cl::opt<bool> EnableSIInsertWaitcntsPass(
+  "enable-si-insert-waitcnts",
+  cl::desc("Use new waitcnt insertion pass"),
+  cl::init(false));
+
 extern "C" void LLVMInitializeAMDGPUTarget() {
   // Register the target
   RegisterTargetMachine<R600TargetMachine> X(getTheAMDGPUTarget());
@@ -134,6 +140,7 @@ extern "C" void LLVMInitializeAMDGPUTarget() {
   initializeAMDGPUUnifyMetadataPass(*PR);
   initializeSIAnnotateControlFlowPass(*PR);
   initializeSIInsertWaitsPass(*PR);
+  initializeSIInsertWaitcntsPass(*PR);
   initializeSIWholeQuadModePass(*PR);
   initializeSILowerControlFlowPass(*PR);
   initializeSIInsertSkipsPass(*PR);
@@ -810,7 +817,10 @@ void GCNPassConfig::addPreEmitPass() {
   // cases.
   addPass(&PostRAHazardRecognizerID);
 
-  addPass(createSIInsertWaitsPass());
+  if (EnableSIInsertWaitcntsPass)
+    addPass(createSIInsertWaitcntsPass());
+  else
+    addPass(createSIInsertWaitsPass());
   addPass(createSIShrinkInstructionsPass());
   addPass(&SIInsertSkipsPassID);
   addPass(createSIDebuggerInsertNopsPass());

diff --git a/llvm/lib/Target/AMDGPU/CMakeLists.txt b/llvm/lib/Target/AMDGPU/CMakeLists.txt
@@ -82,6 +82,7 @@ add_llvm_target(AMDGPUCodeGen
   SIFrameLowering.cpp
   SIInsertSkips.cpp
   SIInsertWaits.cpp
+  SIInsertWaitcnts.cpp
   SIInstrInfo.cpp
   SIISelLowering.cpp
   SILoadStoreOptimizer.cpp

diff --git a/llvm/lib/Target/AMDGPU/DSInstructions.td b/llvm/lib/Target/AMDGPU/DSInstructions.td
@@ -253,6 +253,8 @@ class DS_1A1D_PERMUTE <string opName, SDPatternOperator node = null_frag>
   [(set i32:$vdst,
    (node (DS1Addr1Offset i32:$addr, i16:$offset), i32:$data0))] > {
 
+  let LGKM_CNT = 0;
+
   let mayLoad = 0;
   let mayStore = 0;
   let isConvergent = 1;