[Mlir-commits] [mlir] [MLIR][XeGPU][VectorToXeGPU] Lower vector.load/store/transfer_read/transfer_write to new offsets syntax (PR #162095)

[Dmitry Chigarev]

================
@@ -392,6 +358,62 @@ static Value computeOffsets(PatternRewriter &rewriter, OpType gatScatOp,
       .getResult();
 }
 
+// Collapses shapes of a nD memref to the target rank while applying offsets for
+// the collapsed dimensions. Returns the new memref value and the remaining
+// offsets for the last targetRank dimensions. For example:
+//   input: %memref = memref<2x4x8x32xf32>, offsets=[%i0, %i1, %i2, %i3],
+//   output: %memref[%i0, %i1, 0, 0] -> memref<8x32xf32>, offsets: [%i2, %i3]
+static std::pair<Value, SmallVector<OpFoldResult>>
+convertMemrefAndOffsetsToTargetRank(PatternRewriter &rewriter, Location loc,
+                                    Value memref,
+                                    SmallVector<OpFoldResult> offsets,
+                                    int64_t targetRank = 2) {
----------------
dchigarev wrote:

it's not necessary. removed the default value

https://github.com/llvm/llvm-project/pull/162095