Add comments

Author: yzhang93

mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td

@@ -1191,6 +1191,7 @@ def PadTilingInterfaceOp : Op<Transform_Dialect, "structured.pad_tiling_interfac
     iteration domain induces a padding of the operands that is consistent
     across the op semantics and, unlike for simple elementwise ops, may not be
     trivially deducible or specifiable on operands only (e.g. convolutions).
+    Currently, only a limited set of projected permutation map is supported.
 
     The specification of `padding_sizes` follows that of `tile_sizes` during
     tiling: the value "0" on a particular iterator encode "no padding". Like in

mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h

@@ -601,6 +601,13 @@ LogicalResult rewriteAsPaddedOp(RewriterBase &rewriter, LinalgOp opToPad,
 /// affine.apply operations.
 /// The `indexingMap` + `indexingSizes` encoding suits StructuredOps and
 /// provides a gentle portability path for Linalg-like ops with affine maps.
+/// The padded shape is computed by evaluating the maximum accessed index per
+/// dimension, which may involve multiplying by constant factors derived from
+/// the affine indexing expressions. Currently, only a limited set of projected
+/// permuation indexing map is supported, such as
+/// - affine_map<(d0, d1, d2) -> (d0, d1)>
+/// - affine_map<(d0, d1, d2) -> (d0, d1 + d2)>
+/// - affine_map<(d0, d1) -> (d0 * 3 + d1)>
 /// In the future, more general interfaces can be devised to encode similar
 /// shape evolutions and map between an op and its operands.
 SmallVector<OpFoldResult>

mlir/lib/Dialect/Linalg/Transforms/PadTilingInterface.cpp

@@ -85,6 +85,13 @@ static int64_t extractConstantMultiplier(AffineExpr expr) {
 /// The `indexingMap` + `indexingSizes` encoding suits StructuredOps.
 /// The implementaiton below iteratively combines increases from contributing
 /// dimensions using affine.apply operations.
+/// The padded shape is computed by evaluating the maximum accessed index per
+/// dimension, which may involve multiplying by constant factors derived from
+/// the affine indexing expressions. Currently, only a limited set of projected
+/// permutation indexing map is supported, such as
+/// - affine_map<(d0, d1, d2) -> (d0, d1)>
+/// - affine_map<(d0, d1, d2) -> (d0, d1 + d2)>
+/// - affine_map<(d0, d1) -> (d0 * 3 + d1)>
 /// In the future, more general interfaces can be devised to encode similar
 /// shape evolutions and map between an op and its operands.
 SmallVector<OpFoldResult> linalg::computePaddedShape(