[PATCH] D92483: AMDGPU - Use MUBUF instructions for global address space access

[Tony Tye via Phabricator via llvm-commits]

t-tye added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp:93
   if (isAmdHsaOS()) // Turn on FlatForGlobal for HSA.
     FullFS += "+flat-for-global,+unaligned-access-mode,+trap-handler,";
 
----------------
Seems +flat-for-global should not be added if SI. Suggest pulling up later logic to here and combine.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp:115-116
+  if (Gen == AMDGPUSubtarget::INVALID) {
+     Gen = TT.getOS() == Triple::AMDHSA ? AMDGPUSubtarget::SEA_ISLANDS
+                                        : AMDGPUSubtarget::SOUTHERN_ISLANDS;
+  }
----------------
Should we consider using AMDGPUSubtarget::SEA_ISLANDS as the "generic" processor for all OSs? That can be a separate patch.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp:125-132
   if (!hasAddr64() && !FS.contains("flat-for-global")) {
     FlatForGlobal = true;
   }
 
+  // Use MUBUF instructions for global address space access in GFX60x
+  if (getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS) {
+    FlatForGlobal = false;
----------------
Would seem better to combine these? Perhaps move this up to be adjacent or combined with code above that sets flat-for-global.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D92483/new/

https://reviews.llvm.org/D92483