R600: Use StructurizeCFGPass for non SI targets

StructurizeCFG pass allows to make complex cfg reducible ; it allows a lot of
shader from shadertoy (which exhibits complex control flow constructs) to works
correctly with respect to CFG handling (and allow us to detect potential bug in
other part of the backend).

We provide a cmd line argument to disable the pass for debug purpose.

Patch by: Vincent Lejeune

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 192363

lib/Target/R600/AMDGPU.td

@@ -21,6 +21,11 @@ def FeatureDumpCode : SubtargetFeature <"DumpCode",
         "true",
         "Dump MachineInstrs in the CodeEmitter">;
 
+def FeatureIRStructurizer : SubtargetFeature <"EnableIRStructurizer",
+        "EnableIRStructurizer",
+        "true",
+        "Enable IR Structurizer">;
+
 // Target features
 
 def FeatureFP64     : SubtargetFeature<"fp64",

lib/Target/R600/AMDGPUSubtarget.cpp

@@ -36,6 +36,7 @@ AMDGPUSubtarget::AMDGPUSubtarget(StringRef TT, StringRef CPU, StringRef FS) :
   Gen = AMDGPUSubtarget::R600;
   FP64 = false;
   CaymanISA = false;
+  EnableIRStructurizer = false;
   ParseSubtargetFeatures(GPU, FS);
   DevName = GPU;
 }
@@ -65,6 +66,10 @@ AMDGPUSubtarget::hasCaymanISA() const {
   return CaymanISA;
 }
 bool
+AMDGPUSubtarget::IsIRStructurizerEnabled() const {
+  return EnableIRStructurizer;
+}
+bool
 AMDGPUSubtarget::isTargetELF() const {
   return false;
 }

lib/Target/R600/AMDGPUSubtarget.h

@@ -48,6 +48,7 @@ private:
   enum Generation Gen;
   bool FP64;
   bool CaymanISA;
+  bool EnableIRStructurizer;
 
   InstrItineraryData InstrItins;
 
@@ -63,6 +64,7 @@ public:
   enum Generation getGeneration() const;
   bool hasHWFP64() const;
   bool hasCaymanISA() const;
+  bool IsIRStructurizerEnabled() const;
 
   virtual bool enableMachineScheduler() const {
     return getGeneration() <= NORTHERN_ISLANDS;

lib/Target/R600/AMDGPUTargetMachine.cpp

@@ -33,6 +33,7 @@
 #include "llvm/Transforms/Scalar.h"
 #include <llvm/CodeGen/Passes.h>
 
+
 using namespace llvm;
 
 extern "C" void LLVMInitializeR600Target() {
@@ -123,9 +124,11 @@ bool
 AMDGPUPassConfig::addPreISel() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
   addPass(createFlattenCFGPass());
+  if (ST.IsIRStructurizerEnabled() ||
+      ST.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS)
+    addPass(createStructurizeCFGPass());
   if (ST.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS) {
     addPass(createSITypeRewriter());
-    addPass(createStructurizeCFGPass());
     addPass(createSIAnnotateControlFlowPass());
   } else {
     addPass(createR600TextureIntrinsicsReplacer());

lib/Target/R600/R600EmitClauseMarkers.cpp

@@ -84,6 +84,7 @@ private:
     switch (MI->getOpcode()) {
     case AMDGPU::KILL:
     case AMDGPU::RETURN:
+    case AMDGPU::IMPLICIT_DEF:
       return true;
     default:
       return false;

lib/Target/R600/R600Packetizer.cpp

@@ -340,7 +340,7 @@ bool R600Packetizer::runOnMachineFunction(MachineFunction &Fn) {
     MachineBasicBlock::iterator End = MBB->end();
     MachineBasicBlock::iterator MI = MBB->begin();
     while (MI != End) {
-      if (MI->isKill() ||
+      if (MI->isKill() || MI->getOpcode() == AMDGPU::IMPLICIT_DEF ||
           (MI->getOpcode() == AMDGPU::CF_ALU && !MI->getOperand(8).getImm())) {
         MachineBasicBlock::iterator DeleteMI = MI;
         ++MI;