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

Add profiling label to multithreaded conv.

Browse Source 
PiperOrigin-RevId: 266992949
This commit is contained in:
Alex Stark 2019-09-03 12:53:28 -07:00 committed by TensorFlower Gardener
parent bd76d3a802
commit 97f7450eff
1 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
tensorflow/lite/kernels/internal/optimized/multithreaded_conv.h
View File

@ -139,6 +139,10 @@ inline void Conv(const Eigen::ThreadPoolDevice& device,
const float* bias_data, const RuntimeShape& output_shape, const float* bias_data, const RuntimeShape& output_shape,
float* output_data, const RuntimeShape& im2col_shape, float* output_data, const RuntimeShape& im2col_shape,
float* im2col_data) { float* im2col_data) {
// Nest profiling under "Conv", to aggregate with other kernels.
gemmlowp::ScopedProfilingLabel label("Conv");
gemmlowp::ScopedProfilingLabel inner_label("Multithreaded EigenTensor");
// im2col data should not be generated for the multi-thread supporting case. // im2col data should not be generated for the multi-thread supporting case.
TFLITE_DCHECK(!im2col_data); TFLITE_DCHECK(!im2col_data);
(void)im2col_shape; (void)im2col_shape;