[Mlir-commits] [mlir] [MLIR] [AMX] Utilize x86_amx type for AMX dialect in MLIR. (PR #111197)

[Ilya Enkovich]

================
@@ -4,45 +4,45 @@
 
 func.func @rowheight() {
   // expected-error at +1 {{'amx.tile_zero' op bad row height: 17}}
-  %0 = amx.tile_zero : vector<17x16xbf16>
+  %0 = amx.tile_zero : <17x16xbf16>
----------------
ienkovich wrote:

That looked odd to me as well. The reason appears to be in printers/parsers generated for operations. Here we have `amx::TileType` type that uses `amx.tile` mnemonic and is printed as `amx.tile<17x6xbf16>`. But operation's printers use `printStrippedAttrOrType` to print argument/result types. I suppose that's because they are known to always be `amx::TileType`. So, e.g. for block arguments we have `amx.tile<17x6xbf16>` but in AMX operation we have simply `<17x6xbf16>`. 

https://github.com/llvm/llvm-project/pull/111197