[MLIR][Shape] Lower `shape.get_extent` to `std.dim` when possible

When the shape is derived from a tensor argument the shape extent can be derived
directly from that tensor with `std.dim`.
This lowering pattern circumvents the necessity to materialize the shape in
memory.

Differential Revision: https://reviews.llvm.org/D82644
This commit is contained in:
Frederik Gossen 2020-06-29 08:17:00 +00:00
parent 37cc4fa2ea
commit 76d72c941d
2 changed files with 23 additions and 0 deletions

View File

@ -19,3 +19,10 @@ def SizeToIndexOpConversion : Pat<
(Shape_SizeToIndexOp $arg),
(replaceWithValue $arg)>;
// Derive shape extent directly from shape origin if possible.
// This circumvents the necessity to materialize the shape in memory.
def GetExtentShapeOfConversion : Pat<
(Shape_GetExtentOp (Shape_ShapeOfOp $arg), $idx),
(Shape_IndexToSizeOp (DimOp $arg, (Shape_SizeToIndexOp $idx))),
[],
(addBenefit 10)>;

View File

@ -127,3 +127,19 @@ func @rank(%shape : !shape.shape) -> !shape.size {
%rank = shape.rank %shape
return %rank : !shape.size
}
// -----
// Express `get_extent` as `std.dim` when it relies directly on the outcome of a
// `shape_of` operation.
// CHECK-LABEL: @get_extent_shape_of
// CHECK-SAME: (%[[ARG:.*]]: tensor<2x3xf32>, %[[IDX:.*]]: index) -> index
func @get_extent_shape_of(%arg : tensor<2x3xf32>, %idx : !shape.size)
-> !shape.size {
// CHECK: %[[RESULT:.*]] = dim %[[ARG]], %[[IDX]] : tensor<2x3xf32>
// CHECK: return %[[RESULT]] : index
%shape = shape.shape_of %arg : tensor<2x3xf32>
%result = shape.get_extent %shape, %idx
return %result : !shape.size
}