[llvm] [AMDGPU] Set glc/slc on volatile/nontemporal SMEM loads (PR #77443)

Tony Tye via llvm-commits llvm-commits at lists.llvm.org
Wed Jan 10 18:36:52 PST 2024


================
@@ -29,7 +29,6 @@ class SM_Pseudo <string opName, dag outs, dag ins, string asmOps, list<dag> patt
   let mayStore = 0;
   let mayLoad = 1;
   let hasSideEffects = 0;
-  let maybeAtomic = 0;
----------------
t-tye wrote:

Yes, I think if we change the implementation of the memory model we should ensure the ABI documentation is updated to match. I think only the non-atomic rows need to be updated to include the SMEM instructions since we do not support scalar atomics.

https://github.com/llvm/llvm-project/pull/77443


More information about the llvm-commits mailing list