[llvm] [AMDGPU] Use fake16 load/store with +real-true16 and sram-ecc (PR #161256)

Jay Foad via llvm-commits llvm-commits at lists.llvm.org
Tue Sep 30 02:52:49 PDT 2025


================
@@ -969,7 +969,7 @@ multiclass DSReadPat_t16<DS_Pseudo inst, ValueType vt, string frag> {
   }
 
   let OtherPredicates = [NotLDSRequiresM0Init] in {
-    foreach p = [NotHasTrue16BitInsts, UseFakeTrue16Insts] in
+    foreach p = [NotHasTrue16BitInsts, UseFakeTrue16Insts, UseLoadStoreTrue16WithSramECC] in
----------------
jayfoad wrote:

I really don't understand why you would have multiple copies of the same pattern with different preditcates. Surely it is better to have a single pattern with a single predicate equivalent to "NotHasTrue16BitInsts **OR** UseFakeTrue16Insts **OR** UseLoadStoreTrue16WithSramECC"?

(Pre-existing problem, but this patch makes it worse.)

https://github.com/llvm/llvm-project/pull/161256


More information about the llvm-commits mailing list