[Mlir-commits] [mlir] [mlir][linalg] Implement TilingInterface for winograd operators (PR #96184)

[llvmlistbot at llvm.org]

================
@@ -2813,6 +2813,100 @@ LogicalResult WinogradInputTransformOp::verify() {
   return success();
 }
 
+SmallVector<Range>
+WinogradInputTransformOp::getIterationDomain(OpBuilder &builder) {
+  Location loc = getLoc();
+  IndexType indexType = builder.getIndexType();
+  IntegerAttr zeroAttr = builder.getIntegerAttr(indexType, 0);
+  IntegerAttr oneAttr = builder.getIntegerAttr(indexType, 1);
+  Value output = getOutput();
+  SmallVector<Range> loopBounds(6);
+  for (unsigned dim = 0; dim < 6; ++dim) {
+    loopBounds[dim].offset = zeroAttr;
+    loopBounds[dim].size = getDimValue(builder, loc, output, dim);
+    loopBounds[dim].stride = oneAttr;
+  }
+  return loopBounds;
+}
+
+SmallVector<utils::IteratorType>
+WinogradInputTransformOp::getLoopIteratorTypes() {
+  SmallVector<utils::IteratorType> iteratorTypes(6,
+                                                 utils::IteratorType::parallel);
+  return iteratorTypes;
+}
+
+LogicalResult WinogradInputTransformOp::getResultTilePosition(
+    OpBuilder &builder, unsigned resultNumber, ArrayRef<OpFoldResult> offsets,
+    ArrayRef<OpFoldResult> sizes, SmallVector<OpFoldResult> &resultOffsets,
+    SmallVector<OpFoldResult> &resultSizes) {
+  IntegerAttr zeroAttr = builder.getI64IntegerAttr(0);
+  IntegerAttr oneAttr = builder.getI64IntegerAttr(1);
+
+  resultOffsets.append(
+      {zeroAttr, zeroAttr, offsets[2], offsets[3], zeroAttr, zeroAttr});
+  resultSizes.append(
+      {sizes[0], sizes[1], oneAttr, oneAttr, sizes[4], sizes[5]});
+
+  return success();
+}
+
+FailureOr<TilingResult>
+WinogradInputTransformOp::getTiledImplementation(OpBuilder &builder,
+                                                 ArrayRef<OpFoldResult> offsets,
+                                                 ArrayRef<OpFoldResult> sizes) {
+  IntegerAttr oneAttr = builder.getI64IntegerAttr(1);
+  IntegerAttr zeroAttr = builder.getI64IntegerAttr(0);
+  Value input = getInput();
+  auto inputType = cast<ShapedType>(input.getType());
+  ArrayRef<int64_t> inputShape = inputType.getShape();
+  int64_t inputH = inputShape[1];
+  int64_t inputW = inputShape[2];
+  int64_t m = getM();
+  int64_t r = getR();
+  int64_t alpha = m + r - 1;
+  int64_t alphaH = inputH != 1 ? alpha : 1;
+  int64_t alphaW = inputW != 1 ? alpha : 1;
+  IntegerAttr alphaHAttr = builder.getI64IntegerAttr(alphaH);
+  IntegerAttr alphaWAttr = builder.getI64IntegerAttr(alphaW);
+
+  Location loc = getLoc();
+  SmallVector<Value> tiledOperands;
+  SmallVector<OpFoldResult> sliceOffsets, sliceSizes;
+
+  MLIRContext *context = builder.getContext();
+  auto affineMap =
+      AffineMap::get(1, 0, {builder.getAffineDimExpr(0) * m}, context);
+  Value mappedOffset1 = builder.create<affine::AffineApplyOp>(
+      loc, affineMap,
+      getValueOrCreateConstantIndexOp(builder, loc, offsets[2]));
+  Value mappedOffset2 = builder.create<affine::AffineApplyOp>(
+      loc, affineMap,
+      getValueOrCreateConstantIndexOp(builder, loc, offsets[3]));
+
+  sliceOffsets.append({zeroAttr, mappedOffset1, mappedOffset2, zeroAttr});
+  sliceSizes.append({sizes[4], alphaHAttr, alphaWAttr, sizes[5]});
----------------
Max191 wrote:

It is pretty restrictive to have constant sizes in a tiled implementation. Unless it is absolutely required by the operation (for example, the outer dimensions of the transformed type need to be `alphaHxalphaW`), then it is generally a good idea to not have constant sizes in the tiled implementation. In this case `sizes[2]` and `sizes[3]` should be allowed to be greater than 1, which would mean instead of `alpha` the sizes should be `sizes[2] * m + r - 1`.

https://github.com/llvm/llvm-project/pull/96184