[Mlir-commits] [mlir] [MLIR] VectorEmulateNarrowType to support loading of unaligned vectors (PR #113411)

[Han-Chung Wang]

================
@@ -474,26 +604,42 @@ struct ConvertVectorTransferRead final
         rewriter.create<memref::ExtractStridedMetadataOp>(loc, op.getSource());
 
     OpFoldResult linearizedIndices;
-    std::tie(std::ignore, linearizedIndices) =
+    memref::LinearizedMemRefInfo linearizedInfo;
+    std::tie(linearizedInfo, linearizedIndices) =
         memref::getLinearizedMemRefOffsetAndSize(
             rewriter, loc, srcBits, dstBits,
             stridedMetadata.getConstifiedMixedOffset(),
             stridedMetadata.getConstifiedMixedSizes(),
             stridedMetadata.getConstifiedMixedStrides(),
             getAsOpFoldResult(adaptor.getIndices()));
 
-    auto numElements = (origElements + scale - 1) / scale;
-    auto newReadType = VectorType::get(numElements, newElementType);
+    auto foldedFrontPaddingSize = getFrontPaddingSize(
+        rewriter, loc, linearizedInfo, isUnalignedEmulation);
+
+    if (!foldedFrontPaddingSize) {
+      // unimplemented case for dynamic front padding size
+      return failure();
+    }
+
+    auto numElements =
+        llvm::alignTo(*foldedFrontPaddingSize + origElements, scale) / scale;
 
     auto newRead = rewriter.create<vector::TransferReadOp>(
-        loc, newReadType, adaptor.getSource(),
+        loc, VectorType::get(numElements, newElementType), adaptor.getSource(),
         getValueOrCreateConstantIndexOp(rewriter, loc, linearizedIndices),
         newPadding);
 
-    auto bitCast =
-        rewriter.create<vector::BitCastOp>(loc, op.getType(), newRead);
+    auto bitCast = rewriter.create<vector::BitCastOp>(
+        loc, VectorType::get(numElements * scale, oldElementType), newRead);
+
+    Value bitCastResult = bitCast->getResult(0);
+    if (isUnalignedEmulation) {
+      bitCastResult =
+          extractSubvectorFrom(rewriter, loc, op.getType(), bitCastResult,
+                               *foldedFrontPaddingSize, origElements);
+    }
+    rewriter.replaceOp(op, bitCastResult);
----------------
hanhanW wrote:

I'd use `Value result = ...` which looks cleaner; it's clearer in the unaligned cases.

https://github.com/llvm/llvm-project/pull/113411