[llvm] [AMDGPU][LSV] Restrict large vectors in graphics (PR #92540)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Fri May 17 06:30:38 PDT 2024


================
@@ -364,14 +364,15 @@ unsigned GCNTTIImpl::getStoreVectorFactor(unsigned VF, unsigned StoreSize,
 }
 
 unsigned GCNTTIImpl::getLoadStoreVecRegBitWidth(unsigned AddrSpace) const {
-  if (AddrSpace == AMDGPUAS::GLOBAL_ADDRESS ||
-      AddrSpace == AMDGPUAS::CONSTANT_ADDRESS ||
+  if (AddrSpace == AMDGPUAS::GLOBAL_ADDRESS)
+    return 512;
----------------
arsenm wrote:

128 would be more correct on average for global than constant 

https://github.com/llvm/llvm-project/pull/92540


More information about the llvm-commits mailing list