experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Improve BroadcastTo() to also support trivially broadcasting 1 -> n (as well as n -> n). Then, remove special casing in tile_ops.cc.

Browse Source 
In all cases that would be triggered by the special case, it now will build a BroadcastInDim() such that broadcast_shape == output_dims and the xla::Reshape will not be triggered. Thus, the special case is not needed.

PiperOrigin-RevId: 273015433
This commit is contained in:
A. Unique TensorFlower 2019-10-04 21:47:59 -07:00 committed by TensorFlower Gardener
parent c5c69cd6a8
commit 441d8667ac
2 changed files with 1 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
tensorflow/compiler/tf2xla/kernels/tile_ops.cc
View File

@ -79,29 +79,6 @@ class TileOp : public XlaOpKernel {
return;
}
bool can_tile_with_implicit_broadcast = true;
for (int i = 0; i < input_dims; ++i) {
int64 multiple = multiples[i];
// If the multiple and input dimension are not 1, then tile cannot be
// implemented with a single hlo broadcast.
if (multiple != 1 && input_shape.dim_size(i) != 1) {
can_tile_with_implicit_broadcast = false;
}
}
if (can_tile_with_implicit_broadcast) {
// Create a constant Zero the size of the output shape to leverage binary
// operation broadcast semantics.
auto broadcasted_zero = xla::Broadcast(
XlaHelpers::Zero(ctx->builder(), ctx->input_type(0)), output_dims);
if (ctx->input_type(0) == DT_BOOL) {
ctx->SetOutput(0, xla::Or(broadcasted_zero, input));
} else {
ctx->SetOutput(0, xla::Add(broadcasted_zero, input));
}
return;
}
auto result = BroadcastTo(ctx->Input("input"), output_dims);
OP_REQUIRES_OK(ctx, result.status());
ctx->SetOutput(0, result.ValueOrDie());

2
tensorflow/compiler/tf2xla/lib/broadcast.cc
View File

@ -61,7 +61,7 @@ xla::StatusOr<xla::XlaOp> BroadcastTo(xla::XlaOp input,
}
broadcast_dims.push_back(broadcast_shape.size());
if (*output_it == *input_it) {
if (*output_it == *input_it || *input_it == 1) {
broadcast_shape.push_back(*output_it);
} else if (*output_it != *input_it) {
// Add dimensions [I, O/I], which we will later flatten to just