cuda_host_mem.cpp

00001 /*M///////////////////////////////////////////////////////////////////////////////////////
00002 //
00003 //  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
00004 //
00005 //  By downloading, copying, installing or using the software you agree to this license.
00006 //  If you do not agree to this license, do not download, install,
00007 //  copy or use the software.
00008 //
00009 //
00010 //                          License Agreement
00011 //                For Open Source Computer Vision Library
00012 //
00013 // Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
00014 // Copyright (C) 2009, Willow Garage Inc., all rights reserved.
00015 // Copyright (C) 2013, OpenCV Foundation, all rights reserved.
00016 // Third party copyrights are property of their respective owners.
00017 //
00018 // Redistribution and use in source and binary forms, with or without modification,
00019 // are permitted provided that the following conditions are met:
00020 //
00021 //   * Redistribution's of source code must retain the above copyright notice,
00022 //     this list of conditions and the following disclaimer.
00023 //
00024 //   * Redistribution's in binary form must reproduce the above copyright notice,
00025 //     this list of conditions and the following disclaimer in the documentation
00026 //     and/or other materials provided with the distribution.
00027 //
00028 //   * The name of the copyright holders may not be used to endorse or promote products
00029 //     derived from this software without specific prior written permission.
00030 //
00031 // This software is provided by the copyright holders and contributors "as is" and
00032 // any express or implied warranties, including, but not limited to, the implied
00033 // warranties of merchantability and fitness for a particular purpose are disclaimed.
00034 // In no event shall the Intel Corporation or contributors be liable for any direct,
00035 // indirect, incidental, special, exemplary, or consequential damages
00036 // (including, but not limited to, procurement of substitute goods or services;
00037 // loss of use, data, or profits; or business interruption) however caused
00038 // and on any theory of liability, whether in contract, strict liability,
00039 // or tort (including negligence or otherwise) arising in any way out of
00040 // the use of this software, even if advised of the possibility of such damage.
00041 //
00042 //M*/
00043 
00044 #include "precomp.hpp"
00045 #include <map>
00046 
00047 using namespace cv;
00048 using namespace cv::cuda;
00049 
00050 #ifdef HAVE_CUDA
00051 
00052 namespace {
00053 
00054 class HostMemAllocator : public MatAllocator
00055 {
00056 public:
00057     explicit HostMemAllocator(unsigned int flags) : flags_(flags)
00058     {
00059     }
00060 
00061     UMatData* allocate(int dims, const int* sizes, int type,
00062                        void* data0, size_t* step,
00063                        int /*flags*/, UMatUsageFlags /*usageFlags*/) const
00064     {
00065         size_t total = CV_ELEM_SIZE(type);
00066         for (int i = dims-1; i >= 0; i--)
00067         {
00068             if (step)
00069             {
00070                 if (data0 && step[i] != CV_AUTOSTEP)
00071                 {
00072                     CV_Assert(total <= step[i]);
00073                     total = step[i];
00074                 }
00075                 else
00076                 {
00077                     step[i] = total;
00078                 }
00079             }
00080 
00081             total *= sizes[i];
00082         }
00083 
00084         UMatData* u = new UMatData(this);
00085         u->size = total;
00086 
00087         if (data0)
00088         {
00089             u->data = u->origdata = static_cast<uchar*>(data0);
00090             u->flags |= UMatData::USER_ALLOCATED;
00091         }
00092         else
00093         {
00094             void* ptr = 0;
00095             cudaSafeCall( cudaHostAlloc(&ptr, total, flags_) );
00096 
00097             u->data = u->origdata = static_cast<uchar*>(ptr);
00098         }
00099 
00100         return u;
00101     }
00102 
00103     bool allocate(UMatData* u, int /*accessFlags*/, UMatUsageFlags /*usageFlags*/) const
00104     {
00105         return (u != NULL);
00106     }
00107 
00108     void deallocate(UMatData* u) const
00109     {
00110         if (!u)
00111             return;
00112 
00113         CV_Assert(u->urefcount >= 0);
00114         CV_Assert(u->refcount >= 0);
00115 
00116         if (u->refcount == 0)
00117         {
00118             if ( !(u->flags & UMatData::USER_ALLOCATED) )
00119             {
00120                 cudaFreeHost(u->origdata);
00121                 u->origdata = 0;
00122             }
00123 
00124             delete u;
00125         }
00126     }
00127 
00128 private:
00129     unsigned int flags_;
00130 };
00131 
00132 } // namespace
00133 
00134 #endif
00135 
00136 MatAllocator* cv::cuda::HostMem::getAllocator(AllocType alloc_type)
00137 {
00138 #ifndef HAVE_CUDA
00139     (void) alloc_type;
00140     throw_no_cuda();
00141     return NULL;
00142 #else
00143     static std::map<unsigned int, Ptr<MatAllocator> > allocators;
00144 
00145     unsigned int flag = cudaHostAllocDefault;
00146 
00147     switch (alloc_type)
00148     {
00149     case PAGE_LOCKED:    flag = cudaHostAllocDefault; break;
00150     case SHARED:         flag = cudaHostAllocMapped;  break;
00151     case WRITE_COMBINED: flag = cudaHostAllocWriteCombined; break;
00152     default:             CV_Error(cv::Error::StsBadFlag, "Invalid alloc type");
00153     }
00154 
00155     Ptr<MatAllocator>& a = allocators[flag];
00156 
00157     if (a.empty())
00158     {
00159         a = makePtr<HostMemAllocator>(flag);
00160     }
00161 
00162     return a.get();
00163 #endif
00164 }
00165 
00166 #ifdef HAVE_CUDA
00167 namespace
00168 {
00169     size_t alignUpStep(size_t what, size_t alignment)
00170     {
00171         size_t alignMask = alignment - 1;
00172         size_t inverseAlignMask = ~alignMask;
00173         size_t res = (what + alignMask) & inverseAlignMask;
00174         return res;
00175     }
00176 }
00177 #endif
00178 
00179 void cv::cuda::HostMem::create(int rows_, int cols_, int type_)
00180 {
00181 #ifndef HAVE_CUDA
00182     (void) rows_;
00183     (void) cols_;
00184     (void) type_;
00185     throw_no_cuda();
00186 #else
00187     if (alloc_type == SHARED)
00188     {
00189         DeviceInfo devInfo;
00190         CV_Assert( devInfo.canMapHostMemory() );
00191     }
00192 
00193     type_ &= Mat::TYPE_MASK;
00194 
00195     if (rows == rows_ && cols == cols_ && type() == type_ && data)
00196         return;
00197 
00198     if (data)
00199         release();
00200 
00201     CV_DbgAssert( rows_ >= 0 && cols_ >= 0 );
00202 
00203     if (rows_ > 0 && cols_ > 0)
00204     {
00205         flags = Mat::MAGIC_VAL + Mat::CONTINUOUS_FLAG + type_;
00206         rows = rows_;
00207         cols = cols_;
00208         step = elemSize() * cols;
00209 
00210         if (alloc_type == SHARED)
00211         {
00212             DeviceInfo devInfo;
00213             step = alignUpStep(step, devInfo.textureAlignment());
00214         }
00215 
00216         int64 _nettosize = (int64)step*rows;
00217         size_t nettosize = (size_t)_nettosize;
00218 
00219         if (_nettosize != (int64)nettosize)
00220             CV_Error(cv::Error::StsNoMem, "Too big buffer is allocated");
00221 
00222         size_t datasize = alignSize(nettosize, (int)sizeof(*refcount));
00223 
00224         void* ptr = 0;
00225 
00226         switch (alloc_type)
00227         {
00228         case PAGE_LOCKED:    cudaSafeCall( cudaHostAlloc(&ptr, datasize, cudaHostAllocDefault) ); break;
00229         case SHARED:         cudaSafeCall( cudaHostAlloc(&ptr, datasize, cudaHostAllocMapped) );  break;
00230         case WRITE_COMBINED: cudaSafeCall( cudaHostAlloc(&ptr, datasize, cudaHostAllocWriteCombined) ); break;
00231         default:             CV_Error(cv::Error::StsBadFlag, "Invalid alloc type");
00232         }
00233 
00234         datastart = data =  (uchar*)ptr;
00235         dataend = data + nettosize;
00236 
00237         refcount = (int*)cv::fastMalloc(sizeof(*refcount));
00238         *refcount = 1;
00239     }
00240 #endif
00241 }
00242 
00243 HostMem cv::cuda::HostMem::reshape(int new_cn, int new_rows) const
00244 {
00245     HostMem hdr = *this;
00246 
00247     int cn = channels();
00248     if (new_cn == 0)
00249         new_cn = cn;
00250 
00251     int total_width = cols * cn;
00252 
00253     if ((new_cn > total_width || total_width % new_cn != 0) && new_rows == 0)
00254         new_rows = rows * total_width / new_cn;
00255 
00256     if (new_rows != 0 && new_rows != rows)
00257     {
00258         int total_size = total_width * rows;
00259 
00260         if (!isContinuous())
00261             CV_Error(cv::Error::BadStep, "The matrix is not continuous, thus its number of rows can not be changed");
00262 
00263         if ((unsigned)new_rows > (unsigned)total_size)
00264             CV_Error(cv::Error::StsOutOfRange, "Bad new number of rows");
00265 
00266         total_width = total_size / new_rows;
00267 
00268         if (total_width * new_rows != total_size)
00269             CV_Error(cv::Error::StsBadArg, "The total number of matrix elements is not divisible by the new number of rows");
00270 
00271         hdr.rows = new_rows;
00272         hdr.step = total_width * elemSize1();
00273     }
00274 
00275     int new_width = total_width / new_cn;
00276 
00277     if (new_width * new_cn != total_width)
00278         CV_Error(cv::Error::BadNumChannels, "The total width is not divisible by the new number of channels");
00279 
00280     hdr.cols = new_width;
00281     hdr.flags = (hdr.flags & ~CV_MAT_CN_MASK) | ((new_cn - 1) << CV_CN_SHIFT);
00282 
00283     return hdr;
00284 }
00285 
00286 void cv::cuda::HostMem::release()
00287 {
00288 #ifdef HAVE_CUDA
00289     if (refcount && CV_XADD(refcount, -1) == 1)
00290     {
00291         cudaFreeHost(datastart);
00292         fastFree(refcount);
00293     }
00294 
00295     dataend = data = datastart = 0;
00296     step = rows = cols = 0;
00297     refcount = 0;
00298 #endif
00299 }
00300 
00301 GpuMat cv::cuda::HostMem::createGpuMatHeader() const
00302 {
00303 #ifndef HAVE_CUDA
00304     throw_no_cuda();
00305     return GpuMat();
00306 #else
00307     CV_Assert( alloc_type == SHARED );
00308 
00309     void *pdev;
00310     cudaSafeCall( cudaHostGetDevicePointer(&pdev, data, 0) );
00311 
00312     return GpuMat(rows, cols, type(), pdev, step);
00313 #endif
00314 }
00315 
00316 void cv::cuda::registerPageLocked(Mat& m)
00317 {
00318 #ifndef HAVE_CUDA
00319     (void) m;
00320     throw_no_cuda();
00321 #else
00322     CV_Assert( m.isContinuous() );
00323     cudaSafeCall( cudaHostRegister(m.data, m.step * m.rows, cudaHostRegisterPortable) );
00324 #endif
00325 }
00326 
00327 void cv::cuda::unregisterPageLocked(Mat& m)
00328 {
00329 #ifndef HAVE_CUDA
00330     (void) m;
00331 #else
00332     cudaSafeCall( cudaHostUnregister(m.data) );
00333 #endif
00334 }
00335