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

Fixed alignment for uniform buffer.

Browse Source 
PiperOrigin-RevId: 242064978
This commit is contained in:
A. Unique TensorFlower 2019-04-04 21:06:16 -07:00 committed by TensorFlower Gardener
parent bc1700a6f0
commit f3c82c4646
1 changed files with 11 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
tensorflow/lite/delegates/gpu/metal/kernels/reshape.cc
View File

@ -104,6 +104,8 @@ struct uniforms {
int4 src_size;
int4 dst_size;
int2 plane_xz;
int2 dummy0; // dummy, for alignment
int4 dummy1; // dummy, for alignment
};
$0
@ -211,20 +213,22 @@ std::vector<ComputeTaskDescriptorPtr> Reshapex4(int id, ValueId input_id,
[input_id, output_id](const std::map<ValueId, BHWC>& buffers) {
const auto& src_dim = buffers.find(input_id)->second;
const auto& dst_dim = buffers.find(output_id)->second;
std::vector<int> uniform_params{
std::vector<int32_t> uniform_params{
// int4 src_size
src_dim.w,
src_dim.h,
IntegralDivideRoundUp(src_dim.c, 4),
src_dim.w, src_dim.h, IntegralDivideRoundUp(src_dim.c, 4),
src_dim.w * src_dim.h,
// int4 dst_size
dst_dim.w,
dst_dim.h,
IntegralDivideRoundUp(dst_dim.c, 4),
dst_dim.w, dst_dim.h, IntegralDivideRoundUp(dst_dim.c, 4),
dst_dim.w * dst_dim.h,
// int2 plane_xz
src_dim.w * IntegralDivideRoundUp(src_dim.c, 4),
dst_dim.w * IntegralDivideRoundUp(dst_dim.c, 4),
0, // dummy, for alignment
0, // dummy, for alignment
0, // dummy, for alignment
0, // dummy, for alignment
0, // dummy, for alignment
0 // dummy, for alignment
};
return VectorToUint8Vector(uniform_params);
}},