[Mlir-commits] [mlir] [MLIR][AMDGPU] Add support for fp8 ops on gfx12 (PR #106388)

[Giuseppe Rossini]

================
@@ -385,6 +385,7 @@ static void wmmaPushInputOperand(ConversionPatternRewriter &rewriter,
                                  Location loc,
                                  const TypeConverter *typeConverter,
                                  bool isUnsigned, Value llvmInput,
+                                 Value mlirInput,
----------------
giuseros wrote:

This is because LLVM does not have an fp8 data type. fp8 gets converted to int8. So I need to have both the value before conversion (`mlirInput`) and the value after conversion(`llvmInput`)

https://github.com/llvm/llvm-project/pull/106388