OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-07 12:21:27 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[caffe2][dnnlowp] Remove openmp usage in quantize dnnlowp op

Browse Source 
Summary: It creates cpu overload issues when openmp gets enabled and OMP_NUM_THREADS=1 is not set.

Test Plan: buck test //caffe2/caffe2/quantization/server:quantize_dnnlowp_op_test

Reviewed By: jspark1105

Differential Revision: D24437305

fbshipit-source-id: 426209fc33ce0d4680c478f584716837ee62cb5e
This commit is contained in:
Daya Khudia 2020-10-20 19:31:48 -07:00 committed by Facebook GitHub Bot
parent 6cd8b5e9a7
commit f47231bf0e
1 changed files with 1 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
caffe2/quantization/server/quantize_dnnlowp_op.cc
View File

@ -1,10 +1,6 @@
#include "quantize_dnnlowp_op.h"
#include "dnnlowp_op.h"
#ifdef _OPENMP
#include <omp.h>
#endif
#include "caffe2/core/tensor_int8.h"
#include "caffe2/quantization/server/int8_gen_quant_params.h"
#include "caffe2_dnnlowp_utils.h"
@ -67,16 +63,7 @@ bool QuantizeDNNLowPOp<T>::RunOnDevice() {
const float* in_data = Input(0).template data<float>();
T* out_data = output->t.template mutable_data<T>();
#ifdef _OPENMP
#pragma omp parallel
#endif
{
int i_begin, i_end;
tie(i_begin, i_end) = Get1DPartition(
Input(0).numel(), dnnlowp_get_num_threads(), dnnlowp_get_thread_num());
fbgemm::Quantize<T>(
in_data + i_begin, out_data + i_begin, i_end - i_begin, in_qparams);
}
fbgemm::Quantize<T>(in_data, out_data, Input(0).numel(), in_qparams);
PropagateOutputTensorQuantizationParams(this, 0, in_qparams);