AMDGPU: Change stack alignment

[arsenm]

Summary

While the incoming stack for a kernel is 256-byte aligned,
this refers to the base address of the entire wave. This isn't
useful information for most of codegen. Fixes unnecessarily
aligning stack objects in callees.

[arsenm]

ping

[kzhuravl]

LGTM.

[arsenm]

r300481

lib/Target/AMDGPU/AMDGPUSubtarget.h

public:
   unsigned getImplicitArgNumBytes(const MachineFunction &MF) const {
     if (isMesaKernel(MF))
       return 16;
     if (isAmdHsaOS() && isOpenCLEnv())
       return 32;
     return 0;
   }
 
+  // Scratch is allocated in 256 dword per wave blocks for the entire
+  // wavefront. When viewed from the perspecive of an arbitrary workitem, this
+  // is 4-byte aligned.
   unsigned getStackAlignment() const {
-    // Scratch is allocated in 256 dword per wave blocks.
-    return 4 * 256 / getWavefrontSize();
+    return 4;
   }
 
   bool enableMachineScheduler() const override {
     return true;
   }
 
   bool enableSubRegLiveness() const override {
     return true;