[Mlir-commits] [mlir] [MLIR][XeGPU] Decompose unsupported 'vector.transfer_read'-transpose-permutations (PR #182875)

[Jianhui Li]

================
@@ -841,6 +862,54 @@ struct ContractionLowering : public OpRewritePattern<vector::ContractionOp> {
   }
 };
 
+// Splits 'vector.transfer_read' with unsupported transpose-permutations
+// into 'transfer_read() + transpose()'.
+struct TransferReadDecomposeUnsupportedTranspose
+    : public OpRewritePattern<vector::TransferReadOp> {
+  using Base::Base;
+
+  LogicalResult matchAndRewrite(vector::TransferReadOp readOp,
+                                PatternRewriter &rewriter) const override {
+    Location loc = readOp.getLoc();
+    AffineMap readMap = readOp.getPermutationMap();
+
+    bool isTransposeLoad = !readMap.isMinorIdentity();
+    if (!isTransposeLoad)
+      return failure();
+
+    bool isTransposeSupported = isTransferReadTransposeSupported(readOp);
+    if (isTransposeSupported)
+      return failure();
+
+    auto resultType = cast<VectorType>(readOp.getResult().getType());
+    if (!resultType)
+      return failure();
+
+    // 'Revert' permutation for the transfer_read result shape to make it
+    // 'untransposed'
+    auto newShape = applyPermutationMap(readMap, resultType.getShape());
+    auto newTransferReadRes =
+        VectorType::get(newShape, resultType.getElementType());
+
+    // Step 1. Create 'plain' transfer_read without transpose
+    auto newReadOp = vector::TransferReadOp::create(
+        rewriter, loc, newTransferReadRes, readOp.getBase(),
+        readOp.getIndices(), AffineMap::get(loc.getContext()),
+        readOp.getPadding(), readOp.getMask(), readOp.getInBoundsAttr());
+
+    // Step 2. Transpose the result of the 'plain' transfer_read
+    auto range = llvm::seq<int64_t>(0, readMap.getResults().size());
----------------
Jianhui-Li wrote:

>I'm questioning myself whether there are other use-cases for the transpose-attr at vector-to-xegpu level or we can always generate xegpu.load_nd (any_type, no transpose attr) + vector.transpose and suppose they will be merged automatically further down the pipeline? 

We plan to deprecate the transpose attr at the workgroup level. It is simpler for user to use just load_nd+transpose, no need to learn and use a xegpu-specific transpose attr. 

https://github.com/llvm/llvm-project/pull/182875