[clang] [flang] [llvm] [AMDGPU] Use a target feature to enable __builtin_amdgcn_global_load_lds on gfx9/10 (PR #133055)

[Juan Manuel Martinez Caamaño via llvm-commits]

================
@@ -260,7 +260,7 @@ AMDGPUTargetInfo::AMDGPUTargetInfo(const llvm::Triple &Triple,
 
   MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64;
   CUMode = !(GPUFeatures & llvm::AMDGPU::FEATURE_WGP);
-  for (auto F : {"image-insts", "gws"})
+  for (auto F : {"image-insts", "gws", "mem-to-lds-load-insts"})
----------------
jmmartinez wrote:

I wasn't aware there were `buffer_store_lds` instructions. They seem to be removed after gfx940 and not available on gfx10.

I've renamed the attribute to match "vmem-to-lds-load-insts".

https://github.com/llvm/llvm-project/pull/133055