[Mlir-commits] [mlir] [mlir][x86vector] AVX512-BF16 Dot op (PR #124800)

[Adam Siemieniuk]

================
@@ -271,6 +271,94 @@ def Vp2IntersectQIntrOp : AVX512_IntrOp<"vp2intersect.q.512", 2, [
                    VectorOfLengthAndType<[8], [I64]>:$b);
 }
 
+//===----------------------------------------------------------------------===//
+// AVX512-BF16 op definitions
+//===----------------------------------------------------------------------===//
+
+// Operation that is part of the input dialect.
+class AVX512BF16_Op<string mnemonic, list<Trait> traits = []> :
+  Op<X86Vector_Dialect, "avx512bf16." # mnemonic, traits> {}
----------------
adam-smnk wrote:

AFAIK, most intrinsics already follow naming convention of `x86_avx512_`. However, for `fp16` and `bf16` types, intrinsics have the type explicitly as a part of the name `x86_avx512(fp16|bf16)_`.
This new definition follows the current convention and mirrors the explicit naming style.

I think we could generalize it to have these ops under single `x86vector.avx512.` but honestly either way seems fine to me.

https://github.com/llvm/llvm-project/pull/124800