potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 07:13:02 +04:00
Code Packages Releases Wiki Activity

Merge pull request #24122 from fengyuentau:remove_tengine

Browse Source 
dnn: cleanup of tengine backend #24122

🚀 Cleanup for OpenCV 5.0. Tengine backend is added for convolution layer speedup on ARM CPUs, but it is not maintained and the convolution layer on our default backend has reached similar performance to that of Tengine.

Tengine backend related PRs:
- https://github.com/opencv/opencv/pull/16724
- https://github.com/opencv/opencv/pull/18323

### Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
      Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
This commit is contained in:
Yuantao Feng
2023-08-09 02:26:02 -04:00
committed by GitHub
parent 7d59db4ec4
commit ba70ec99b3
10 changed files with 3 additions and 696 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/dnn/src/layers/convolution_layer.cpp
-85
View File
@@ -62,9 +62,6 @@
#include "opencl_kernels_dnn.hpp"
using namespace cv::dnn::ocl4dnn;
#endif
#ifdef HAVE_TENGINE
#include "../tengine4dnn/include/tengine_graph_convolution.hpp"
#endif
#ifdef HAVE_CUDA
#include "../cuda4dnn/primitives/convolution.hpp"
@@ -267,10 +264,6 @@ public:
float power;
#endif
#ifdef HAVE_TENGINE
teng_graph_t tengine_graph;
#endif
#ifdef HAVE_CUDA
cuda4dnn::ConvolutionConfiguration::FusionMode cudaFusionMode;
cuda4dnn::ConvolutionConfiguration::ActivationType cudaActType;
@@ -289,20 +282,8 @@ public:
#ifdef HAVE_CUDA
cudaFusionMode = cuda4dnn::ConvolutionConfiguration::FusionMode::NONE;
cudaActType = cuda4dnn::ConvolutionConfiguration::ActivationType::IDENTITY;
#endif
#ifdef HAVE_TENGINE
tengine_graph=NULL;
#endif
}
#ifdef HAVE_TENGINE
~ConvolutionLayerImpl()
{
if(NULL != tengine_graph )
{
tengine_release(tengine_graph);
}
}
#endif
MatShape computeColRowShape(const MatShape &inpShape, const MatShape &outShape) const CV_OVERRIDE
{
@@ -466,13 +447,6 @@ public:
for(int i = 0; i < numOutput; i++ )
biasvec[i] = biasMat.at<float>(i);
}
#ifdef HAVE_TENGINE
if(NULL != tengine_graph )
{
tengine_release(tengine_graph);
tengine_graph = NULL ;
}
#endif
#ifdef HAVE_OPENCL
convolutionOp.release();
#endif
@@ -1305,65 +1279,6 @@ public:
}
}
#ifdef HAVE_TENGINE
bool tengine_ret = false;
std::vector<Mat> teng_in, teng_out;
inputs_arr.getMatVector(teng_in);
outputs_arr.getMatVector(teng_out);
int inch = teng_in[0].size[1]; // inch
int in_h = teng_in[0].size[2]; // in_h
int in_w = teng_in[0].size[3]; // in_w
int out_b = teng_out[0].size[0]; // out batch size
int outch = teng_out[0].size[1]; // outch
int out_h = teng_out[0].size[2]; // out_h
int out_w = teng_out[0].size[3]; // out_w
float *input_ = teng_in[0].ptr<float>();
float *output_ = teng_out[0].ptr<float>();
float *kernel_ = weightsMat.ptr<float>();
float *teg_bias = &biasvec[0];
int nstripes = std::max(getNumThreads(), 1);
/* tengine_init will run when first time. */
if(NULL == tengine_graph)
{
// pads_begin: 0 - pad_top, 1 - pad_left
// pads_end: 0 - pad_bottom, 1 - pad_right
// pad_h0: pad_top, pad_h1: pad_bottom
// pad_w0: pad_left, pad_w1: pad_right
tengine_graph = tengine_init(name.c_str(), input_, inch, ngroups, in_h, in_w,
output_, out_b, outch, out_h, out_w,
kernel_, kernel_size.size(), kernel.height, kernel.width,
teg_bias, stride.height, stride.width,
pads_begin[0], pads_end[0], pads_begin[1], pads_end[1], dilation.height, dilation.width,
weightsMat.step1(), padMode, tengine_graph, nstripes);
// printf("Init(%s): input=%p(%d %d %d %d ),output=%p(%d %d %d %d ),kernel=%p(%ld %d %d ), bias=%p ,"
// "stride(%d %d), pad(%d %d %d %d), dilation(%d %d) ,weightsMat=%ld, padMode=%s ,tengine_graph = %p \n",
// name.c_str(),input_, inch, ngroups, in_h, in_w,
// output_, out_b, outch, out_h, out_w,
// kernel_, kernel_size.size(), kernel.height, kernel.width,
// teg_bias, stride.height, stride.width,
// pads_begin[0], pads_end[0], pads_begin[1], pads_end[1], dilation.height, dilation.width,
// weightsMat.step1(), padMode.c_str() ,tengine_graph);
}
if(NULL != tengine_graph)
{
tengine_ret = tengine_forward(tengine_graph);
}
/* activation */
if((true == tengine_ret) && activ )
{
int out_cstep = out_h * out_w; // out_cstep
ActivationLayer* activ_ = activ.get();
activ_->forwardSlice(output_, output_, out_cstep, out_cstep, 0, outch);
}
if(false == tengine_ret)
#endif
{
int nstripes = std::max(getNumThreads(), 1);
int conv_dim = CONV_2D;