[Mlir-commits] [mlir] [MLIR][XeGPU] Remove create tdesc op from xegpu dialect (PR #182804)

[Jianhui Li]

================
@@ -307,53 +218,6 @@ func.func @prefetch_vc_1(%src: memref<24x32xf16>) {
   return
 }
 
-// -----
-func.func @prefetch_vc_2(%src: ui64) {
-  %0 = arith.constant dense<[0, 8, 16, 24]> : vector<4xindex>
-  %1 = xegpu.create_tdesc %src, %0 : ui64, vector<4xindex>
-          -> !xegpu.tensor_desc<4x2xf32, #xegpu.scatter_tdesc_attr<chunk_size = 2>>
-  // expected-error at +1 {{invalid l1_hint: #xegpu.cache_hint<write_back>}}
----------------
Jianhui-Li wrote:

all the checks below should not be lost. We need to add these check in regular load/store/prefetch op as a common verification function, and modify these tests to use memref instead of tensor_desc. 

https://github.com/llvm/llvm-project/pull/182804