[Mlir-commits] [mlir] [MLIR][XeGPU][VectorToXeGPU] Fix transfer_read/write cases with non-contiguous memrefs (PR #158126)

[Dmitry Chigarev]

================
@@ -256,8 +266,8 @@ static SmallVector<Value> computeStrides(VectorTransferOpInterface xferOp,
 //   %orig_offset = %block_id_y * 4x2x6x32 // consider using affine map
 //   %offsets =  orig_offset + local_offsets
 static Value computeOffsets(VectorTransferOpInterface xferOp,
-                            PatternRewriter &rewriter,
-                            ArrayRef<Value> strides) {
+                            PatternRewriter &rewriter, ArrayRef<Value> strides,
+                            Value baseOffset) {
----------------
dchigarev wrote:

`computeOffsets` now takes a `baseOffset` obtained from `memref.extract_strided_metadata` in `computeMemrefMeta` (ex `computeStrides`)

https://github.com/llvm/llvm-project/pull/158126