AMDGPU/SI: Enable MemorySSA for EarlyCSE

[• tstellarAMD]

Summary

With MemorySSA this pass does a much better job of CSE'ing intrinsic
with pointer arguments.

[arsenm]

Can you add a testcase where this changes something? I assume this works with the GVN path?

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

return TargetIRAnalysis([this](const Function &F) {
     return TargetTransformInfo(AMDGPUTTIImpl(this, F));
   });
 }
 
 void AMDGPUPassConfig::addEarlyCSEOrGVNPass() {
   if (getOptLevel() == CodeGenOpt::Aggressive)
     addPass(createGVNPass());
   else
-    addPass(createEarlyCSEPass());
+    addPass(createEarlyCSEPass(true));
 }
 
 void AMDGPUPassConfig::addStraightLineScalarOptimizationPasses() {
   addPass(createSeparateConstOffsetFromGEPPass());
   addPass(createSpeculativeExecutionPass());
   // ReassociateGEPs exposes more opportunites for SLSR. See
   // the example in reassociate-geps-and-slsr.ll.
   addPass(createStraightLineStrengthReducePass());