opencv/modules/dnn/test/test_torch_importer.cpp

591 lines
21 KiB
C++
Raw Normal View History

/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "test_precomp.hpp"
#include "npy_blob.hpp"
#include <opencv2/dnn/shape_utils.hpp>
#include <opencv2/dnn/layer.details.hpp> // CV_DNN_REGISTER_LAYER_CLASS
namespace opencv_test
{
using namespace std;
using namespace testing;
using namespace cv;
using namespace cv::dnn;
template<typename TStr>
static std::string _tf(TStr filename, bool inTorchDir = true, bool required = true)
{
String path = "dnn/";
if (inTorchDir)
path += "torch/";
path += filename;
return findDataFile(path, required);
}
TEST(Torch_Importer, simple_read)
{
Net net;
ASSERT_NO_THROW(net = readNetFromTorch(_tf("net_simple_net.txt"), false));
ASSERT_FALSE(net.empty());
}
class Test_Torch_layers : public DNNTestLayer
{
public:
void runTorchNet(const String& prefix, String outLayerName = "",
bool check2ndBlob = false, bool isBinary = false, bool evaluate = true,
double l1 = 0.0, double lInf = 0.0)
{
String suffix = (isBinary) ? ".dat" : ".txt";
Mat inp, outRef;
ASSERT_NO_THROW( inp = readTorchBlob(_tf(prefix + "_input" + suffix), isBinary) );
ASSERT_NO_THROW( outRef = readTorchBlob(_tf(prefix + "_output" + suffix), isBinary) );
checkBackend(backend, target, &inp, &outRef);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
Net net = readNetFromTorch(_tf(prefix + "_net" + suffix), isBinary, evaluate);
ASSERT_FALSE(net.empty());
net.setPreferableBackend(backend);
net.setPreferableTarget(target);
if (outLayerName.empty())
outLayerName = net.getLayerNames().back();
net.setInput(inp);
std::vector<Mat> outBlobs;
net.forward(outBlobs, outLayerName);
l1 = l1 ? l1 : default_l1;
lInf = lInf ? lInf : default_lInf;
normAssert(outRef, outBlobs[0], "", l1, lInf);
if (check2ndBlob && backend == DNN_BACKEND_OPENCV)
{
Mat out2 = outBlobs[1];
Mat ref2 = readTorchBlob(_tf(prefix + "_output_2" + suffix), isBinary);
normAssert(out2, ref2, "", l1, lInf);
}
}
};
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, run_convolution)
{
// Output reference values are in range [23.4018, 72.0181]
double l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.08 : default_l1;
2020-09-26 23:37:52 +08:00
double lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.43 : default_lInf;
runTorchNet("net_conv", "", false, true, true, l1, lInf);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, run_pool_max)
{
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
runTorchNet("net_pool_max", "", true);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, run_pool_ave)
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
{
runTorchNet("net_pool_ave");
}
TEST_P(Test_Torch_layers, run_reshape_change_batch_size)
{
runTorchNet("net_reshape");
}
TEST_P(Test_Torch_layers, run_reshape)
{
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
runTorchNet("net_reshape_batch");
runTorchNet("net_reshape_channels", "", false, true);
}
TEST_P(Test_Torch_layers, run_reshape_single_sample)
{
// Reference output values in range [14.4586, 18.4492].
runTorchNet("net_reshape_single_sample", "", false, false, true,
(target == DNN_TARGET_MYRIAD || target == DNN_TARGET_OPENCL_FP16) ? 0.033 : default_l1,
(target == DNN_TARGET_MYRIAD || target == DNN_TARGET_OPENCL_FP16) ? 0.05 : default_lInf);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, run_linear)
{
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_OPENCL_FP16);
runTorchNet("net_linear_2d");
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, run_concat)
{
runTorchNet("net_concat", "l5_torchMerge");
}
TEST_P(Test_Torch_layers, run_depth_concat)
{
runTorchNet("net_depth_concat", "", false, true, true, 0.0,
2020-09-26 23:37:52 +08:00
target == DNN_TARGET_OPENCL_FP16 ? 0.032 : 0.0);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, run_deconv)
{
runTorchNet("net_deconv");
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, run_batch_norm)
{
runTorchNet("net_batch_norm", "", false, true);
runTorchNet("net_batch_norm_train", "", false, true, false);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_prelu)
{
runTorchNet("net_prelu");
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_cadd_table)
{
runTorchNet("net_cadd_table");
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_softmax)
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
{
runTorchNet("net_softmax");
runTorchNet("net_softmax_spatial");
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_logsoftmax)
{
runTorchNet("net_logsoftmax");
runTorchNet("net_logsoftmax_spatial");
}
2020-02-07 21:40:50 +08:00
TEST_P(Test_Torch_layers, net_lp_pooling_square)
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
{
runTorchNet("net_lp_pooling_square", "", false, true);
2020-02-07 21:40:50 +08:00
}
TEST_P(Test_Torch_layers, net_lp_pooling_power)
{
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
runTorchNet("net_lp_pooling_power", "", false, true);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_conv_gemm_lrn)
{
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
double l1 = 0.0, lInf = 0.0;
if (target == DNN_TARGET_OPENCL_FP16)
{
l1 = 0.046;
lInf = 0.023;
}
// The OpenCL kernels use the native_ math functions which have
// implementation defined accuracy, so we use relaxed thresholds. See
// https://github.com/opencv/opencv/issues/9821 for more details.
else if (target == DNN_TARGET_OPENCL)
{
l1 = 0.02;
lInf = 0.02;
}
runTorchNet("net_conv_gemm_lrn", "", false, true, true, l1, lInf);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_inception_block)
{
runTorchNet("net_inception_block", "", false, true);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_normalize)
{
runTorchNet("net_normalize", "", false, true);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_padding)
{
runTorchNet("net_padding", "", false, true);
runTorchNet("net_spatial_zero_padding", "", false, true);
runTorchNet("net_spatial_reflection_padding", "", false, true);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_layers, net_non_spatial)
{
#if defined(INF_ENGINE_RELEASE) && ( \
INF_ENGINE_VER_MAJOR_EQ(2021030000) || \
INF_ENGINE_VER_MAJOR_EQ(2021040000) \
)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
// 2021.3: crash
// 2021.4: [ GENERAL_ERROR ] AssertionFailed: !out.networkInputs.empty()
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH); // exception
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_OPENCL_FP16)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH); // exception
#endif
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 &&
(target == DNN_TARGET_OPENCL || target == DNN_TARGET_OPENCL_FP16))
applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
runTorchNet("net_non_spatial", "", false, true);
}
TEST_P(Test_Torch_layers, run_paralel)
{
if (backend != DNN_BACKEND_OPENCV || target != DNN_TARGET_CPU)
throw SkipTestException(""); // TODO: Check this
runTorchNet("net_parallel", "l5_torchMerge");
}
TEST_P(Test_Torch_layers, net_residual)
{
2019-01-14 14:55:44 +08:00
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_RELEASE == 2018050000
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && (target == DNN_TARGET_OPENCL ||
2019-01-14 14:55:44 +08:00
target == DNN_TARGET_OPENCL_FP16))
applyTestTag(target == DNN_TARGET_OPENCL ? CV_TEST_TAG_DNN_SKIP_IE_OPENCL : CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16,
CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
2019-01-14 14:55:44 +08:00
#endif
runTorchNet("net_residual", "", false, true);
}
class Test_Torch_nets : public DNNTestLayer {};
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_nets, OpenFace_accuracy)
{
2019-08-06 23:41:30 +08:00
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2018050000)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
#endif
checkBackend();
const string model = findDataFile("dnn/openface_nn4.small2.v1.t7", false);
Net net = readNetFromTorch(model);
net.setPreferableBackend(backend);
net.setPreferableTarget(target);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
Mat sample = imread(findDataFile("cv/shared/lena.png"));
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
Mat sampleF32(sample.size(), CV_32FC3);
sample.convertTo(sampleF32, sampleF32.type());
sampleF32 /= 255;
resize(sampleF32, sampleF32, Size(96, 96), 0, 0, INTER_NEAREST);
Mat inputBlob = blobFromImage(sampleF32, 1.0, Size(), Scalar(), /*swapRB*/true);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
net.setInput(inputBlob);
Mat out = net.forward();
// Reference output values are in range [-0.17212, 0.263492]
// on Myriad problem layer: l4_Pooling - does not use pads_begin
2019-08-06 23:41:30 +08:00
float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 2e-3 : 1e-5;
float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 5e-3 : 1e-3;
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
Mat outRef = readTorchBlob(_tf("net_openface_output.dat"), true);
normAssert(out, outRef, "", l1, lInf);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
}
2018-10-19 22:43:26 +08:00
static Mat getSegmMask(const Mat& scores)
{
const int rows = scores.size[2];
const int cols = scores.size[3];
const int numClasses = scores.size[1];
Mat maxCl = Mat::zeros(rows, cols, CV_8UC1);
Mat maxVal(rows, cols, CV_32FC1, Scalar(0));
for (int ch = 0; ch < numClasses; ch++)
{
for (int row = 0; row < rows; row++)
{
const float *ptrScore = scores.ptr<float>(0, ch, row);
uint8_t *ptrMaxCl = maxCl.ptr<uint8_t>(row);
float *ptrMaxVal = maxVal.ptr<float>(row);
for (int col = 0; col < cols; col++)
{
if (ptrScore[col] > ptrMaxVal[col])
{
ptrMaxVal[col] = ptrScore[col];
ptrMaxCl[col] = (uchar)ch;
}
}
}
}
return maxCl;
}
// Computer per-class intersection over union metric.
static void normAssertSegmentation(const Mat& ref, const Mat& test)
{
CV_Assert_N(ref.dims == 4, test.dims == 4);
const int numClasses = ref.size[1];
CV_Assert(numClasses == test.size[1]);
Mat refMask = getSegmMask(ref);
Mat testMask = getSegmMask(test);
EXPECT_EQ(countNonZero(refMask != testMask), 0);
}
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_nets, ENet_accuracy)
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
{
2018-10-09 06:38:06 +08:00
applyTestTag(target == DNN_TARGET_CPU ? "" : CV_TEST_TAG_MEMORY_512MB);
checkBackend();
if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
throw SkipTestException("");
2020-02-07 21:40:50 +08:00
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2020010000)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
#else
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target != DNN_TARGET_CPU)
{
if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
if (target == DNN_TARGET_OPENCL) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
if (target == DNN_TARGET_MYRIAD) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
throw SkipTestException("");
}
#endif
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_GE(2021010000)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
2020-02-07 21:40:50 +08:00
#endif
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target != DNN_TARGET_CPU)
{
if (target == DNN_TARGET_OPENCL_FP16) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL_FP16, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
if (target == DNN_TARGET_OPENCL) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
if (target == DNN_TARGET_MYRIAD) applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
throw SkipTestException("");
}
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
Net net;
{
const string model = findDataFile("dnn/Enet-model-best.net", false);
2017-12-18 22:22:57 +08:00
net = readNetFromTorch(model, true);
ASSERT_TRUE(!net.empty());
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
}
net.setPreferableBackend(backend);
net.setPreferableTarget(target);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
Mat sample = imread(_tf("street.png", false));
Mat inputBlob = blobFromImage(sample, 1./255, Size(), Scalar(), /*swapRB*/true);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
net.setInput(inputBlob, "");
Mat out = net.forward();
Mat ref = blobFromNPY(_tf("torch_enet_prob.npy", false));
// Due to numerical instability in Pooling-Unpooling layers (indexes jittering)
2018-06-03 07:21:08 +08:00
// thresholds for ENet must be changed. Accuracy of results was checked on
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
// Cityscapes dataset and difference in mIOU with Torch is 10E-4%
2018-10-19 22:43:26 +08:00
normAssert(ref, out, "", 0.00044, /*target == DNN_TARGET_CPU ? 0.453 : */0.552);
normAssertSegmentation(ref, out);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
const int N = 3;
for (int i = 0; i < N; i++)
{
net.setInput(inputBlob, "");
Mat out = net.forward();
2018-10-19 22:43:26 +08:00
normAssert(ref, out, "", 0.00044, /*target == DNN_TARGET_CPU ? 0.453 : */0.552);
normAssertSegmentation(ref, out);
Merge pull request #9114 from pengli:dnn_rebase add libdnn acceleration to dnn module (#9114) * import libdnn code Signed-off-by: Li Peng <peng.li@intel.com> * add convolution layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add pooling layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add softmax layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add lrn layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add innerproduct layer ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * add HAVE_OPENCL macro Signed-off-by: Li Peng <peng.li@intel.com> * fix for convolution ocl Signed-off-by: Li Peng <peng.li@intel.com> * enable getUMat() for multi-dimension Mat Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat for ocl acceleration Signed-off-by: Li Peng <peng.li@intel.com> * use CV_OCL_RUN macro Signed-off-by: Li Peng <peng.li@intel.com> * set OPENCL target when it is available and disable fuseLayer for OCL target for the time being Signed-off-by: Li Peng <peng.li@intel.com> * fix innerproduct accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * remove trailing space Signed-off-by: Li Peng <peng.li@intel.com> * Fixed tensorflow demo bug. Root cause is that tensorflow has different algorithm with libdnn to calculate convolution output dimension. libdnn don't calculate output dimension anymore and just use one passed in by config. * split gemm ocl file split it into gemm_buffer.cl and gemm_image.cl Signed-off-by: Li Peng <peng.li@intel.com> * Fix compile failure Signed-off-by: Li Peng <peng.li@intel.com> * check env flag for auto tuning Signed-off-by: Li Peng <peng.li@intel.com> * switch to new ocl kernels for softmax layer Signed-off-by: Li Peng <peng.li@intel.com> * update softmax layer on some platform subgroup extension may not work well, fallback to non subgroup ocl acceleration. Signed-off-by: Li Peng <peng.li@intel.com> * fallback to cpu path for fc layer with multi output Signed-off-by: Li Peng <peng.li@intel.com> * update output message Signed-off-by: Li Peng <peng.li@intel.com> * update fully connected layer fallback to gemm API if libdnn return false Signed-off-by: Li Peng <peng.li@intel.com> * Add ReLU OCL implementation * disable layer fusion for now Signed-off-by: Li Peng <peng.li@intel.com> * Add OCL implementation for concat layer Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> * libdnn: update license and copyrights Also refine libdnn coding style Signed-off-by: Wu Zhiwen <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * DNN: Don't link OpenCL library explicitly * DNN: Make default preferableTarget to DNN_TARGET_CPU User should set it to DNN_TARGET_OPENCL explicitly if want to use OpenCL acceleration. Also don't fusion when using DNN_TARGET_OPENCL * DNN: refine coding style * Add getOpenCLErrorString * DNN: Use int32_t/uint32_t instread of alias * Use namespace ocl4dnn to include libdnn things * remove extra copyTo in softmax ocl path Signed-off-by: Li Peng <peng.li@intel.com> * update ReLU layer ocl path Signed-off-by: Li Peng <peng.li@intel.com> * Add prefer target property for layer class It is used to indicate the target for layer forwarding, either the default CPU target or OCL target. Signed-off-by: Li Peng <peng.li@intel.com> * Add cl_event based timer for cv::ocl * Rename libdnn to ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * use UMat for ocl4dnn internal buffer Remove allocateMemory which use clCreateBuffer directly Signed-off-by: Li Peng <peng.li@intel.com> Signed-off-by: wzw <zhiwen.wu@intel.com> * enable buffer gemm in ocl4dnn innerproduct Signed-off-by: Li Peng <peng.li@intel.com> * replace int_tp globally for ocl4dnn kernels. Signed-off-by: wzw <zhiwen.wu@intel.com> Signed-off-by: Li Peng <peng.li@intel.com> * create UMat for layer params Signed-off-by: Li Peng <peng.li@intel.com> * update sign ocl kernel Signed-off-by: Li Peng <peng.li@intel.com> * update image based gemm of inner product layer Signed-off-by: Li Peng <peng.li@intel.com> * remove buffer gemm of inner product layer call cv::gemm API instead Signed-off-by: Li Peng <peng.li@intel.com> * change ocl4dnn forward parameter to UMat Signed-off-by: Li Peng <peng.li@intel.com> * Refine auto-tuning mechanism. - Use OPENCV_OCL4DNN_KERNEL_CONFIG_PATH to set cache directory for fine-tuned kernel configuration. e.g. export OPENCV_OCL4DNN_KERNEL_CONFIG_PATH=/home/tmp, the cache directory will be /home/tmp/spatialkernels/ on Linux. - Define environment OPENCV_OCL4DNN_ENABLE_AUTO_TUNING to enable auto-tuning. - OPENCV_OPENCL_ENABLE_PROFILING is only used to enable profiling for OpenCL command queue. This fix basic kernel get wrong running time, i.e. 0ms. - If creating cache directory failed, disable auto-tuning. * Detect and create cache dir on windows Signed-off-by: Li Peng <peng.li@intel.com> * Refine gemm like convolution kernel. Signed-off-by: Li Peng <peng.li@intel.com> * Fix redundant swizzleWeights calling when use cached kernel config. * Fix "out of resource" bug when auto-tuning too many kernels. * replace cl_mem with UMat in ocl4dnnConvSpatial class * OCL4DNN: reduce the tuning kernel candidate. This patch could reduce 75% of the tuning candidates with less than 2% performance impact for the final result. Signed-off-by: Zhigang Gong <zhigang.gong@intel.com> * replace cl_mem with umat in ocl4dnn convolution Signed-off-by: Li Peng <peng.li@intel.com> * remove weight_image_ of ocl4dnn inner product Actually it is unused in the computation Signed-off-by: Li Peng <peng.li@intel.com> * Various fixes for ocl4dnn 1. OCL_PERFORMANCE_CHECK(ocl::Device::getDefault().isIntel()) 2. Ptr<OCL4DNNInnerProduct<float> > innerProductOp 3. Code comments cleanup 4. ignore check on OCL cpu device Signed-off-by: Li Peng <peng.li@intel.com> * add build option for log softmax Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ocl kernels in ocl4dnn Signed-off-by: Li Peng <peng.li@intel.com> * replace ocl4dnnSet with opencv setTo Signed-off-by: Li Peng <peng.li@intel.com> * replace ALIGN with cv::alignSize Signed-off-by: Li Peng <peng.li@intel.com> * check kernel build options Signed-off-by: Li Peng <peng.li@intel.com> * Handle program compilation fail properly. * Use std::numeric_limits<float>::infinity() for large float number * check ocl4dnn kernel compilation result Signed-off-by: Li Peng <peng.li@intel.com> * remove unused ctx_id Signed-off-by: Li Peng <peng.li@intel.com> * change clEnqueueNDRangeKernel to kernel.run() Signed-off-by: Li Peng <peng.li@intel.com> * change cl_mem to UMat in image based gemm Signed-off-by: Li Peng <peng.li@intel.com> * check intel subgroup support for lrn and pooling layer Signed-off-by: Li Peng <peng.li@intel.com> * Fix convolution bug if group is greater than 1 Signed-off-by: Li Peng <peng.li@intel.com> * Set default layer preferableTarget to be DNN_TARGET_CPU Signed-off-by: Li Peng <peng.li@intel.com> * Add ocl perf test for convolution Signed-off-by: Li Peng <peng.li@intel.com> * Add more ocl accuracy test Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_image with ocl::Image2D Signed-off-by: Li Peng <peng.li@intel.com> * Fix build failure in elementwise layer Signed-off-by: Li Peng <peng.li@intel.com> * use getUMat() to get blob data Signed-off-by: Li Peng <peng.li@intel.com> * replace cl_mem handle with ocl::KernelArg Signed-off-by: Li Peng <peng.li@intel.com> * dnn(build): don't use C++11, OPENCL_LIBRARIES fix * dnn(ocl4dnn): remove unused OpenCL kernels * dnn(ocl4dnn): extract OpenCL code into .cl files * dnn(ocl4dnn): refine auto-tuning Defaultly disable auto-tuning, set OPENCV_OCL4DNN_ENABLE_AUTO_TUNING environment variable to enable it. Use a set of pre-tuned configs as default config if auto-tuning is disabled. These configs are tuned for Intel GPU with 48/72 EUs, and for googlenet, AlexNet, ResNet-50 If default config is not suitable, use the first available kernel config from the candidates. Candidate priority from high to low is gemm like kernel, IDLF kernel, basick kernel. * dnn(ocl4dnn): pooling doesn't use OpenCL subgroups * dnn(ocl4dnn): fix perf test OpenCV has default 3sec time limit for each performance test. Warmup OpenCL backend outside of perf measurement loop. * use ocl::KernelArg as much as possible Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): fix bias bug for gemm like kernel * dnn(ocl4dnn): wrap cl_mem into UMat Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): Refine signature of kernel config - Use more readable string as signture of kernel config - Don't count device name and vendor in signature string - Default kernel configurations are tuned for Intel GPU with 24/48/72 EUs, and for googlenet, AlexNet, ResNet-50 net model. * dnn(ocl4dnn): swap width/height in configuration * dnn(ocl4dnn): enable configs for Intel OpenCL runtime only * core: make configuration helper functions accessible from non-core modules * dnn(ocl4dnn): update kernel auto-tuning behavior Avoid unwanted creation of directories * dnn(ocl4dnn): simplify kernel to workaround OpenCL compiler crash * dnn(ocl4dnn): remove redundant code * dnn(ocl4dnn): Add more clear message for simd size dismatch. * dnn(ocl4dnn): add const to const argument Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): force compiler use a specific SIMD size for IDLF kernel * dnn(ocl4dnn): drop unused tuneLocalSize() * dnn(ocl4dnn): specify OpenCL queue for Timer and convolve() method * dnn(ocl4dnn): sanitize file names used for cache * dnn(perf): enable Network tests with OpenCL * dnn(ocl4dnn/conv): drop computeGlobalSize() * dnn(ocl4dnn/conv): drop unused fields * dnn(ocl4dnn/conv): simplify ctor * dnn(ocl4dnn/conv): refactor kernelConfig localSize=NULL * dnn(ocl4dnn/conv): drop unsupported double / untested half types * dnn(ocl4dnn/conv): drop unused variable * dnn(ocl4dnn/conv): alignSize/divUp * dnn(ocl4dnn/conv): use enum values * dnn(ocl4dnn): drop unused innerproduct variable Signed-off-by: Li Peng <peng.li@intel.com> * dnn(ocl4dnn): add an generic function to check cl option support * dnn(ocl4dnn): run softmax subgroup version kernel first Signed-off-by: Li Peng <peng.li@intel.com>
2017-10-02 20:38:00 +08:00
}
}
// Check accuracy of style transfer models from https://github.com/jcjohnson/fast-neural-style
// th fast_neural_style.lua \
// -input_image ~/opencv_extra/testdata/dnn/googlenet_1.png \
// -output_image lena.png \
// -median_filter 0 \
// -image_size 0 \
// -model models/eccv16/starry_night.t7
// th fast_neural_style.lua \
// -input_image ~/opencv_extra/testdata/dnn/googlenet_1.png \
// -output_image lena.png \
// -median_filter 0 \
// -image_size 0 \
// -model models/instance_norm/feathers.t7
2018-03-05 23:21:19 +08:00
TEST_P(Test_Torch_nets, FastNeuralStyle_accuracy)
{
#if defined INF_ENGINE_RELEASE
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_MYRIAD
&& getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER);
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD
&& getInferenceEngineVPUType() == CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD_X, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
#endif
checkBackend();
2019-02-08 22:12:33 +08:00
#if defined(INF_ENGINE_RELEASE)
#if INF_ENGINE_RELEASE <= 2018050000
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_OPENCL)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_OPENCL, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
2019-02-08 22:12:33 +08:00
#endif
#endif
std::string models[] = {"dnn/fast_neural_style_eccv16_starry_night.t7",
"dnn/fast_neural_style_instance_norm_feathers.t7"};
std::string targets[] = {"dnn/lena_starry_night.png", "dnn/lena_feathers.png"};
for (int i = 0; i < 2; ++i)
{
const string model = findDataFile(models[i], false);
Net net = readNetFromTorch(model);
net.setPreferableBackend(backend);
net.setPreferableTarget(target);
2018-03-05 23:21:19 +08:00
Mat img = imread(findDataFile("dnn/googlenet_1.png"));
Mat inputBlob = blobFromImage(img, 1.0, Size(), Scalar(103.939, 116.779, 123.68), false);
net.setInput(inputBlob);
Mat out = net.forward();
// Deprocessing.
getPlane(out, 0, 0) += 103.939;
getPlane(out, 0, 1) += 116.779;
getPlane(out, 0, 2) += 123.68;
out = cv::min(cv::max(0, out), 255);
Mat ref = imread(findDataFile(targets[i]));
Mat refBlob = blobFromImage(ref, 1.0, Size(), Scalar(), false);
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{
double normL1 = cvtest::norm(refBlob, out, cv::NORM_L1) / refBlob.total();
if (target == DNN_TARGET_MYRIAD)
EXPECT_LE(normL1, 4.0f);
else
EXPECT_LE(normL1, 0.6f);
}
else
normAssert(out, refBlob, "", 0.5, 1.1);
}
}
INSTANTIATE_TEST_CASE_P(/**/, Test_Torch_nets, dnnBackendsAndTargets());
2018-03-31 16:11:10 +08:00
// Test a custom layer
// https://github.com/torch/nn/blob/master/doc/convolution.md#nn.SpatialUpSamplingNearest
class SpatialUpSamplingNearestLayer CV_FINAL : public Layer
{
public:
SpatialUpSamplingNearestLayer(const LayerParams &params) : Layer(params)
{
scale = params.get<int>("scale_factor");
}
static Ptr<Layer> create(LayerParams& params)
{
return Ptr<Layer>(new SpatialUpSamplingNearestLayer(params));
}
virtual bool getMemoryShapes(const std::vector<std::vector<int> > &inputs,
const int requiredOutputs,
std::vector<std::vector<int> > &outputs,
std::vector<std::vector<int> > &internals) const CV_OVERRIDE
{
std::vector<int> outShape(4);
outShape[0] = inputs[0][0]; // batch size
outShape[1] = inputs[0][1]; // number of channels
outShape[2] = scale * inputs[0][2];
outShape[3] = scale * inputs[0][3];
outputs.assign(1, outShape);
return false;
}
void forward(InputArrayOfArrays inputs_arr, OutputArrayOfArrays outputs_arr, OutputArrayOfArrays) CV_OVERRIDE
{
CV_TRACE_FUNCTION();
CV_TRACE_ARG_VALUE(name, "name", name.c_str());
std::vector<Mat> inputs, outputs;
inputs_arr.getMatVector(inputs);
outputs_arr.getMatVector(outputs);
Mat& inp = inputs[0];
Mat& out = outputs[0];
const int outHeight = out.size[2];
const int outWidth = out.size[3];
for (size_t n = 0; n < inp.size[0]; ++n)
{
for (size_t ch = 0; ch < inp.size[1]; ++ch)
{
resize(getPlane(inp, n, ch), getPlane(out, n, ch),
Size(outWidth, outHeight), 0, 0, INTER_NEAREST);
}
}
}
private:
int scale;
};
TEST_P(Test_Torch_layers, upsampling_nearest)
{
#if defined(INF_ENGINE_RELEASE) && INF_ENGINE_VER_MAJOR_EQ(2021030000)
if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_MYRIAD)
applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH); // TODO
#endif
// Test a custom layer.
CV_DNN_REGISTER_LAYER_CLASS(SpatialUpSamplingNearest, SpatialUpSamplingNearestLayer);
try
{
runTorchNet("net_spatial_upsampling_nearest", "", false, true);
}
catch (...)
{
LayerFactory::unregisterLayer("SpatialUpSamplingNearest");
throw;
}
LayerFactory::unregisterLayer("SpatialUpSamplingNearest");
// Test an implemented layer.
runTorchNet("net_spatial_upsampling_nearest", "", false, true);
}
INSTANTIATE_TEST_CASE_P(/**/, Test_Torch_layers, dnnBackendsAndTargets());
}