[llvm] [AMDGPU] Correctly insert s_nops for dst forwarding hazard (PR #100276)

Jeffrey Byrnes via llvm-commits llvm-commits at lists.llvm.org
Mon Aug 5 09:45:55 PDT 2024


================
@@ -2904,6 +2905,15 @@ def getVCMPXOpFromVCMP : InstrMapping {
   let ValueCols = [["1"]];
 }
 
+def FP8DstByteSelTable : GenericTable {
+  let FilterClass = "VOP3_Pseudo";
----------------
jrbyrnes wrote:

We need to resolve hazard for mi300 and mi350 (at least). For gfx12 I think there is no issue, but `HasFP8DstByteSel` should be consistent -- it's behavior shouldn't be reliant upon the current users of it.

I've resolved the inconsistency by modifying VOP3_CVT_SR_F8_ByteSel_Profile

https://github.com/llvm/llvm-project/pull/100276


More information about the llvm-commits mailing list