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

Jay Foad via llvm-commits llvm-commits at lists.llvm.org
Thu Jan 11 08:40:48 PST 2024


================
@@ -5813,6 +5813,18 @@ in table :ref:`amdgpu-amdhsa-memory-model-code-sequences-gfx6-gfx9-table`.
                                                               be reordered by
                                                               hardware.
 
+     load         *none*       *none*         - constant - !volatile & !nontemporal
----------------
jayfoad wrote:

Maybe this should say "constant address space **and** only if the compiler can prove that the address is uniform", but I'm not sure where to fit that into the table.

Should I mention scalar stores in the table? The text above says "Since the constant address space contents do not change during the execution of a kernel dispatch it is not legal to perform
stores".

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


More information about the llvm-commits mailing list