[Mlir-commits] [mlir] [mlir][vector] Add support for vector extract/insert_strided_slice in vector distribution. (PR #145421)

[Jianhui Li]

================
@@ -1076,6 +1094,195 @@ struct WarpOpCreateMask : public WarpDistributionPattern {
   }
 };
 
+/// Sink out insert_strided_slice op feeding into a warp op yield.
+/// ```
+/// %0 = gpu.warp_execute_on_lane_0(%arg0) -> (vector<8x1xf32>) {
+///   ...
+///   %src = ... : vector<4x16xf32>
----------------
Jianhui-Li wrote:

nit: consider subgroup size as 32 to be consistent with test. 

https://github.com/llvm/llvm-project/pull/145421