Comments

Author: adam-smnk

mlir/lib/Conversion/VectorToXeGPU/VectorToXeGPU.cpp

@@ -330,7 +330,7 @@ struct ContractionLowering : public OpRewritePattern<vector::ContractionOp> {
       return rewriter.notifyMatchFailure(contractOp, "Expects acc 2D vector");
 
     // Accept only plain 2D data layout.
-    // VNNI packing is left to later lowering.
+    // VNNI packing is applied to DPAS as a separate lowering step.
     TypedValue<VectorType> lhs = contractOp.getLhs();
     TypedValue<VectorType> rhs = contractOp.getRhs();
     if (lhs.getType().getRank() != 2 || rhs.getType().getRank() != 2)

mlir/test/Conversion/VectorToXeGPU/contract-to-xegpu.mlir

@@ -48,8 +48,8 @@ func.func @dpas_gemm_i8(%lhs: vector<8x32xi8>, %rhs: vector<32x16xi8>,
 
 // -----
 
-// VNNI packing is added automatically by later XeGPU consumer.
 // For simplicity, only plain data layouts are currently supported.
+// VNNI packing is applied later as a separate lowering step.
 
 #map = affine_map<(d0, d1, d2, d3) -> (d0, d2, d3)>
 #map1 = affine_map<(d0, d1, d2, d3) -> (d2, d1, d3)>