AMDGPU: Add an option -disable-promote-alloca-to-lds

[yaxunl]

Summary

Add this option for debugging and providing workaround.

By default it is off so no behavior change in backend.

[arsenm]

We already have a flag disable-promote-alloca-to-vector inside the pass itself. It would probably be more useful to add another corresponding one for the LDS promotion part if that's useful

[yaxunl]

Revised by Matt's comments.

[arsenm]

Needs test

[yaxunl]

add a test

[arsenm]

LGTM

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

   cl::ReallyHidden,
   cl::init(true));
 
+static cl::opt<bool> EnablePromoteAlloca(
+  "amdgpu-enable-promote-alloca",
+  cl::desc("Run promote alloca pass"),
+  cl::ReallyHidden,
+  cl::init(true));
+
 static cl::opt<bool>
 EnableEarlyIfConversion("amdgpu-early-ifcvt", cl::Hidden,
                         cl::desc("Run early if-conversion"),

 
   if (TM.getOptLevel() > CodeGenOpt::None) {
     addPass(createInferAddressSpacesPass());
-    addPass(createAMDGPUPromoteAlloca());
+    if (EnablePromoteAlloca)
+      addPass(createAMDGPUPromoteAlloca());
 
     if (EnableSROA)
       addPass(createSROAPass());