added resetDevice function, removed MultiGpuManager
This commit is contained in:
parent
b6c195d44c
commit
f906c9b259
@ -64,6 +64,10 @@ namespace cv
|
||||
CV_EXPORTS void setDevice(int device);
|
||||
CV_EXPORTS int getDevice();
|
||||
|
||||
//! Explicitly destroys and cleans up all resources associated with the current device in the current process.
|
||||
//! Any subsequent API call to this device will reinitialize the device.
|
||||
CV_EXPORTS void resetDevice();
|
||||
|
||||
enum FeatureSet
|
||||
{
|
||||
FEATURE_SET_COMPUTE_10 = 10,
|
||||
@ -132,34 +136,6 @@ namespace cv
|
||||
int minorVersion_;
|
||||
};
|
||||
|
||||
/////////////////////////// Multi GPU Manager //////////////////////////////
|
||||
|
||||
// Provides functionality for working with many GPUs
|
||||
class CV_EXPORTS MultiGpuManager
|
||||
{
|
||||
public:
|
||||
MultiGpuManager();
|
||||
~MultiGpuManager();
|
||||
|
||||
// Must be called before any other GPU calls
|
||||
void init();
|
||||
|
||||
// Makes the given GPU active
|
||||
void gpuOn(int gpu_id);
|
||||
|
||||
// Finishes the piece of work on the current GPU
|
||||
void gpuOff();
|
||||
|
||||
static const int BAD_GPU_ID = -1;
|
||||
|
||||
private:
|
||||
void operator=(const MultiGpuManager&);
|
||||
MultiGpuManager(const MultiGpuManager&);
|
||||
|
||||
class Impl;
|
||||
Ptr<Impl> impl_;
|
||||
};
|
||||
|
||||
//////////////////////////////// Error handling ////////////////////////
|
||||
|
||||
CV_EXPORTS void error(const char *error_string, const char *file, const int line, const char *func);
|
||||
|
@ -72,7 +72,7 @@ namespace
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::builtWith(cv::gpu::FeatureSet feature_set)
|
||||
bool cv::gpu::TargetArchs::builtWith(cv::gpu::FeatureSet feature_set)
|
||||
{
|
||||
#if defined (HAVE_CUDA)
|
||||
return ::compareToSet(CUDA_ARCH_FEATURES, feature_set, std::greater_equal<int>());
|
||||
@ -83,13 +83,13 @@ CV_EXPORTS bool cv::gpu::TargetArchs::builtWith(cv::gpu::FeatureSet feature_set)
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::has(int major, int minor)
|
||||
bool cv::gpu::TargetArchs::has(int major, int minor)
|
||||
{
|
||||
return hasPtx(major, minor) || hasBin(major, minor);
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::hasPtx(int major, int minor)
|
||||
bool cv::gpu::TargetArchs::hasPtx(int major, int minor)
|
||||
{
|
||||
#if defined (HAVE_CUDA)
|
||||
return ::compareToSet(CUDA_ARCH_PTX, major * 10 + minor, std::equal_to<int>());
|
||||
@ -101,7 +101,7 @@ CV_EXPORTS bool cv::gpu::TargetArchs::hasPtx(int major, int minor)
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::hasBin(int major, int minor)
|
||||
bool cv::gpu::TargetArchs::hasBin(int major, int minor)
|
||||
{
|
||||
#if defined (HAVE_CUDA)
|
||||
return ::compareToSet(CUDA_ARCH_BIN, major * 10 + minor, std::equal_to<int>());
|
||||
@ -113,7 +113,7 @@ CV_EXPORTS bool cv::gpu::TargetArchs::hasBin(int major, int minor)
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrLessPtx(int major, int minor)
|
||||
bool cv::gpu::TargetArchs::hasEqualOrLessPtx(int major, int minor)
|
||||
{
|
||||
#if defined (HAVE_CUDA)
|
||||
return ::compareToSet(CUDA_ARCH_PTX, major * 10 + minor,
|
||||
@ -126,14 +126,14 @@ CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrLessPtx(int major, int minor)
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreater(int major, int minor)
|
||||
bool cv::gpu::TargetArchs::hasEqualOrGreater(int major, int minor)
|
||||
{
|
||||
return hasEqualOrGreaterPtx(major, minor) ||
|
||||
hasEqualOrGreaterBin(major, minor);
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreaterPtx(int major, int minor)
|
||||
bool cv::gpu::TargetArchs::hasEqualOrGreaterPtx(int major, int minor)
|
||||
{
|
||||
#if defined (HAVE_CUDA)
|
||||
return ::compareToSet(CUDA_ARCH_PTX, major * 10 + minor,
|
||||
@ -146,7 +146,7 @@ CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreaterPtx(int major, int minor)
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreaterBin(int major, int minor)
|
||||
bool cv::gpu::TargetArchs::hasEqualOrGreaterBin(int major, int minor)
|
||||
{
|
||||
#if defined (HAVE_CUDA)
|
||||
return ::compareToSet(CUDA_ARCH_BIN, major * 10 + minor,
|
||||
@ -161,9 +161,10 @@ CV_EXPORTS bool cv::gpu::TargetArchs::hasEqualOrGreaterBin(int major, int minor)
|
||||
|
||||
#if !defined (HAVE_CUDA)
|
||||
|
||||
CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount() { return 0; }
|
||||
CV_EXPORTS void cv::gpu::setDevice(int) { throw_nogpu(); }
|
||||
CV_EXPORTS int cv::gpu::getDevice() { throw_nogpu(); return 0; }
|
||||
int cv::gpu::getCudaEnabledDeviceCount() { return 0; }
|
||||
void cv::gpu::setDevice(int) { throw_nogpu(); }
|
||||
int cv::gpu::getDevice() { throw_nogpu(); return 0; }
|
||||
void cv::gpu::resetDevice() { throw_nogpu(); }
|
||||
size_t cv::gpu::DeviceInfo::freeMemory() const { throw_nogpu(); return 0; }
|
||||
size_t cv::gpu::DeviceInfo::totalMemory() const { throw_nogpu(); return 0; }
|
||||
bool cv::gpu::DeviceInfo::supports(cv::gpu::FeatureSet) const { throw_nogpu(); return false; }
|
||||
@ -173,7 +174,7 @@ void cv::gpu::DeviceInfo::queryMemory(size_t&, size_t&) const { throw_nogpu(); }
|
||||
|
||||
#else /* !defined (HAVE_CUDA) */
|
||||
|
||||
CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount()
|
||||
int cv::gpu::getCudaEnabledDeviceCount()
|
||||
{
|
||||
int count;
|
||||
cudaSafeCall( cudaGetDeviceCount( &count ) );
|
||||
@ -181,13 +182,13 @@ CV_EXPORTS int cv::gpu::getCudaEnabledDeviceCount()
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS void cv::gpu::setDevice(int device)
|
||||
void cv::gpu::setDevice(int device)
|
||||
{
|
||||
cudaSafeCall( cudaSetDevice( device ) );
|
||||
}
|
||||
|
||||
|
||||
CV_EXPORTS int cv::gpu::getDevice()
|
||||
int cv::gpu::getDevice()
|
||||
{
|
||||
int device;
|
||||
cudaSafeCall( cudaGetDevice( &device ) );
|
||||
@ -195,6 +196,12 @@ CV_EXPORTS int cv::gpu::getDevice()
|
||||
}
|
||||
|
||||
|
||||
void cv::gpu::resetDevice()
|
||||
{
|
||||
cudaSafeCall( cudaDeviceReset() );
|
||||
}
|
||||
|
||||
|
||||
size_t cv::gpu::DeviceInfo::freeMemory() const
|
||||
{
|
||||
size_t free_memory, total_memory;
|
||||
|
@ -1,151 +0,0 @@
|
||||
/*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) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., 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 "precomp.hpp"
|
||||
|
||||
#if !defined(HAVE_CUDA)
|
||||
|
||||
namespace cv { namespace gpu {
|
||||
|
||||
class MultiGpuManager::Impl {};
|
||||
MultiGpuManager::MultiGpuManager() { throw_nogpu(); }
|
||||
MultiGpuManager::~MultiGpuManager() {}
|
||||
void MultiGpuManager::init() { throw_nogpu(); }
|
||||
void MultiGpuManager::gpuOn(int) { throw_nogpu(); }
|
||||
void MultiGpuManager::gpuOff() { throw_nogpu(); }
|
||||
|
||||
}}
|
||||
|
||||
#else
|
||||
|
||||
#include <vector>
|
||||
#include <cuda.h>
|
||||
|
||||
#define cuSafeCall(expr) safeCall(expr, #expr, __FILE__, __LINE__)
|
||||
|
||||
using namespace std;
|
||||
|
||||
namespace cv { namespace gpu {
|
||||
|
||||
|
||||
class MultiGpuManager::Impl
|
||||
{
|
||||
public:
|
||||
Impl();
|
||||
|
||||
~Impl()
|
||||
{
|
||||
for (int i = 0; i < num_devices_; ++i)
|
||||
cuSafeCall(cuCtxDestroy(contexts_[i]));
|
||||
}
|
||||
|
||||
void gpuOn(int gpu_id)
|
||||
{
|
||||
if (gpu_id < 0 || gpu_id >= num_devices_)
|
||||
CV_Error(CV_StsBadArg, "MultiGpuManager::gpuOn: GPU ID is out of range");
|
||||
cuSafeCall(cuCtxPushCurrent(contexts_[gpu_id]));
|
||||
}
|
||||
|
||||
void gpuOff()
|
||||
{
|
||||
CUcontext prev_context;
|
||||
cuSafeCall(cuCtxPopCurrent(&prev_context));
|
||||
}
|
||||
|
||||
private:
|
||||
void safeCall(CUresult code, const char* expr, const char* file, int line)
|
||||
{
|
||||
if (code != CUDA_SUCCESS)
|
||||
error(expr, file, line, "");
|
||||
}
|
||||
|
||||
int num_devices_;
|
||||
vector<CUcontext> contexts_;
|
||||
};
|
||||
|
||||
|
||||
MultiGpuManager::Impl::Impl(): num_devices_(0)
|
||||
{
|
||||
num_devices_ = getCudaEnabledDeviceCount();
|
||||
contexts_.resize(num_devices_);
|
||||
|
||||
cuSafeCall(cuInit(0));
|
||||
|
||||
CUdevice device;
|
||||
CUcontext prev_context;
|
||||
for (int i = 0; i < num_devices_; ++i)
|
||||
{
|
||||
cuSafeCall(cuDeviceGet(&device, i));
|
||||
cuSafeCall(cuCtxCreate(&contexts_[i], 0, device));
|
||||
cuSafeCall(cuCtxPopCurrent(&prev_context));
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
MultiGpuManager::MultiGpuManager() {}
|
||||
MultiGpuManager::~MultiGpuManager() {}
|
||||
|
||||
|
||||
void MultiGpuManager::init()
|
||||
{
|
||||
impl_ = Ptr<Impl>(new Impl());
|
||||
}
|
||||
|
||||
|
||||
void MultiGpuManager::gpuOn(int gpu_id)
|
||||
{
|
||||
if (impl_.empty())
|
||||
CV_Error(CV_StsNullPtr, "MultiGpuManager::gpuOn: must be initialized before any calls");
|
||||
impl_->gpuOn(gpu_id);
|
||||
}
|
||||
|
||||
|
||||
void MultiGpuManager::gpuOff()
|
||||
{
|
||||
if (impl_.empty())
|
||||
CV_Error(CV_StsNullPtr, "MultiGpuManager::gpuOff: must be initialized before any calls");
|
||||
impl_->gpuOff();
|
||||
}
|
||||
|
||||
}}
|
||||
|
||||
#endif
|
@ -36,8 +36,6 @@ using namespace cv::gpu;
|
||||
|
||||
struct Worker { void operator()(int device_id) const; };
|
||||
|
||||
MultiGpuManager multi_gpu_mgr;
|
||||
|
||||
int main()
|
||||
{
|
||||
int num_devices = getCudaEnabledDeviceCount();
|
||||
@ -58,8 +56,6 @@ int main()
|
||||
}
|
||||
}
|
||||
|
||||
multi_gpu_mgr.init();
|
||||
|
||||
// Execute calculation in two threads using two GPUs
|
||||
int devices[] = {0, 1};
|
||||
parallel_do(devices, devices + 2, Worker());
|
||||
@ -70,7 +66,7 @@ int main()
|
||||
|
||||
void Worker::operator()(int device_id) const
|
||||
{
|
||||
multi_gpu_mgr.gpuOn(device_id);
|
||||
setDevice(device_id);
|
||||
|
||||
Mat src(1000, 1000, CV_32F);
|
||||
Mat dst;
|
||||
@ -95,8 +91,6 @@ void Worker::operator()(int device_id) const
|
||||
// after context is extracted from the stack
|
||||
d_src.release();
|
||||
d_dst.release();
|
||||
|
||||
multi_gpu_mgr.gpuOff();
|
||||
}
|
||||
|
||||
#endif
|
||||
|
@ -38,8 +38,6 @@ using namespace cv::gpu;
|
||||
|
||||
struct Worker { void operator()(int device_id) const; };
|
||||
|
||||
MultiGpuManager multi_gpu_mgr;
|
||||
|
||||
// GPUs data
|
||||
GpuMat d_left[2];
|
||||
GpuMat d_right[2];
|
||||
@ -89,43 +87,37 @@ int main(int argc, char** argv)
|
||||
return -1;
|
||||
}
|
||||
|
||||
multi_gpu_mgr.init();
|
||||
|
||||
// Split source images for processing on the GPU #0
|
||||
multi_gpu_mgr.gpuOn(0);
|
||||
setDevice(0);
|
||||
d_left[0].upload(left.rowRange(0, left.rows / 2));
|
||||
d_right[0].upload(right.rowRange(0, right.rows / 2));
|
||||
bm[0] = new StereoBM_GPU();
|
||||
multi_gpu_mgr.gpuOff();
|
||||
|
||||
// Split source images for processing on the GPU #1
|
||||
multi_gpu_mgr.gpuOn(1);
|
||||
setDevice(1);
|
||||
d_left[1].upload(left.rowRange(left.rows / 2, left.rows));
|
||||
d_right[1].upload(right.rowRange(right.rows / 2, right.rows));
|
||||
bm[1] = new StereoBM_GPU();
|
||||
multi_gpu_mgr.gpuOff();
|
||||
|
||||
// Execute calculation in two threads using two GPUs
|
||||
int devices[] = {0, 1};
|
||||
parallel_do(devices, devices + 2, Worker());
|
||||
|
||||
// Release the first GPU resources
|
||||
multi_gpu_mgr.gpuOn(0);
|
||||
setDevice(0);
|
||||
imshow("GPU #0 result", Mat(d_result[0]));
|
||||
d_left[0].release();
|
||||
d_right[0].release();
|
||||
d_result[0].release();
|
||||
delete bm[0];
|
||||
multi_gpu_mgr.gpuOff();
|
||||
|
||||
// Release the second GPU resources
|
||||
multi_gpu_mgr.gpuOn(1);
|
||||
setDevice(1);
|
||||
imshow("GPU #1 result", Mat(d_result[1]));
|
||||
d_left[1].release();
|
||||
d_right[1].release();
|
||||
d_result[1].release();
|
||||
delete bm[1];
|
||||
multi_gpu_mgr.gpuOff();
|
||||
|
||||
waitKey();
|
||||
return 0;
|
||||
@ -134,15 +126,13 @@ int main(int argc, char** argv)
|
||||
|
||||
void Worker::operator()(int device_id) const
|
||||
{
|
||||
multi_gpu_mgr.gpuOn(device_id);
|
||||
setDevice(device_id);
|
||||
|
||||
bm[device_id]->operator()(d_left[device_id], d_right[device_id],
|
||||
d_result[device_id]);
|
||||
|
||||
std::cout << "GPU #" << device_id << " (" << DeviceInfo().name()
|
||||
<< "): finished\n";
|
||||
|
||||
multi_gpu_mgr.gpuOff();
|
||||
}
|
||||
|
||||
#endif
|
||||
|
Loading…
Reference in New Issue
Block a user