[Mlir-commits] [mlir] [mlir][ROCDL] Refactor wmma intrinsics to use attributes not operands where possible (PR #167041)

[Muzammiluddin Syed]

================
@@ -1311,11 +1300,33 @@ struct WMMAOpLowering : public ConvertOpToLLVMPattern<WMMAOp> {
     if (chipset.majorVersion != 11 && chipset.majorVersion != 12)
       return op->emitOpError("WMMA only supported on gfx11 and gfx12");
 
-    // The WMMA operations represent vectors of bf16s as vectors of i16s, so we
-    // need to bitcast bfloats to i16 and then bitcast them back.
+    bool isGFX1250 = chipset == Chipset(12, 5, 0);
+
+    // The WMMA operations represent vectors of bf16s as vectors of i16s
+    // (except on gfx1250), so we need to bitcast bfloats to i16 and then
+    // bitcast them back.
+    auto aType = dyn_cast<VectorType>(adaptor.getSourceA().getType());
+    auto bType = dyn_cast<VectorType>(adaptor.getSourceA().getType());
+    auto destCType = dyn_cast<VectorType>(adaptor.getDestC().getType());
----------------
Muzammiluddin-Syed-ECE wrote:

Not that I can see in .cpp.inc files

https://github.com/llvm/llvm-project/pull/167041