cuda: opengl interop

vbirds · Sep 17, 2014 · 346c9b3 · 346c9b3
1 parent dc210db
commit 346c9b3
Show file tree

Hide file tree

Showing 6 changed files with 128 additions and 3 deletions.
diff --git a/src/codec/video/VideoDecoderCUDA.cpp b/src/codec/video/VideoDecoderCUDA.cpp
@@ -420,6  420,7 @@ bool VideoDecoderCUDAPrivate::initCuda()
     description = QString("CUDA device: %1 %2.%3 %4 MHz").arg(devname).arg(major).arg(minor).arg(clockRate/1000);
 
     //TODO: cuD3DCtxCreate > cuGLCtxCreate > cuCtxCreate
     // check user option "readBack". if not "readBack" and ctx version < 5.0, check gl or angle and run cuGLCtxCreate or cuD3DCtxCreate
     checkCudaErrors(cuCtxCreate(&cuctx, CU_CTX_SCHED_BLOCKING_SYNC, cudev)); //CU_CTX_SCHED_AUTO?
 
     unsigned int ctx_ver = 0;

diff --git a/src/cuda/SurfaceInterop_CUDA_GL.cpp b/src/cuda/SurfaceInterop_CUDA_GL.cpp
@@ -0,0  1,75 @@
 #include "SurfaceInterop_CUDA_GL.h"
 #include "cuda_gl_interop.h"
 namespace QtAV {
 namespace cuda {
 
 //cuda<5.0: cuGLCtxCreate to get max perf
 
 SurfaceInterop_CUDA_GL::SurfaceInterop_CUDA_GL()
 {
     memset(tex_res, 0, sizeof(tex_res));
 }
 
 SurfaceInterop_CUDA_GL::~SurfaceInterop_CUDA_GL()
 {
     for (int i = 0; i < 2;   i) {
         if (tex_res[i].res)
             cuGraphicsUnregisterResource(tex_res[i].res);
     }
 }
 
 bool SurfaceInterop_CUDA_GL::registerTextureResource(GLuint tex, int plane)
 {
     TextureRes &t_r(tex_res[plane]);
     if (t_r.tex == tex)
         return true;
     // register only once
     if (t_r.tex) {
         checkCudaErrors(cuGraphicsUnregisterResource(t_r.res));
     }
     checkCudaErrors(cuGraphicsGLRegisterImage(&t_r.res, tex, GL_TEXTURE_2D, CU_GRAPHICS_REGISTER_FLAGS_NONE)); //WO?
     t_r.tex = tex;
     return true;
 }
 
 void* SurfaceInterop_CUDA_GL::map(SurfaceType type, const VideoFormat& fmt, void* handle, int plane)
 {
     if (type != GLTextureSurface)
         return 0;
     Q_ASSERT(plane <= 1); //NV12
     Q_ASSERT(handle); //
     if (!registerTextureResource(*((GLuint*)handle), plane))
         return 0;
             //cuGraphicsSubResourceGetMappedPointer(CUdevicePtr) //>=0302
     CUarray array;
     CUresult r = cuGraphicsSubResourceGetMappedArray(&array, tex_res[plane].res, 0/*arrayIndex*/, 0 /*mipLevel*/);
     if (r != CUDA_SUCCESS)
         return 0;
     CUDA_MEMCPY2D cp2d;
     memset(&cp2d, 0, sizeof(cp2d));
 
     r = cuMemcpy2D(&cp2d);
     if (r != CUDA_SUCCESS)
         return 0;
     //cuGraphicsUnmapResources(1, tex_res[plane].res, 0);
     return handle;
 }
 
 void SurfaceInterop_CUDA_GL::unmap(void *handle)
 {
     CUgraphicsResource *res = 0;
     for (int i = 0; i < 2;   i) {
         if (tex_res[i].tex == *((GLuint*)handle)) {
             res = &tex_res[i].res;
             break;
         }
     }
     if (!res)
         return;
     cuGraphicsUnmapResources(1, res, 0);
 }
 
 } //namespace cuda
 } //namespace QtAV
 
 
diff --git a/src/cuda/SurfaceInterop_CUDA_GL.h b/src/cuda/SurfaceInterop_CUDA_GL.h
@@ -0,0  1,36 @@
 #ifndef QTAV_SURFACEINTEROP_CUDA_GL_H
 #define QTAV_SURFACEINTEROP_CUDA_GL_H
 
 #include <QtCore/QMap>
 #include <QtCore/QSharedPointer>
 #include "helper_cuda.h"
 #include "cuda_api.h"
 #include "QtAV/SurfaceInterop.h"
 
 namespace QtAV {
 namespace cuda {
 
 class SurfaceInterop_CUDA_GL : public VideoSurfaceInterop, public cuda_api
 {
 public:
     SurfaceInterop_CUDA_GL();
     ~SurfaceInterop_CUDA_GL();
 
     bool registerTextureResource(GLuint tex, int plane);
     virtual void* map(SurfaceType type, const VideoFormat& fmt, void* handle, int plane);
     virtual void unmap(void *handle);
 
 private:
     typedef struct {
         GLuint tex;
         CUgraphicsResource res;
     } TextureRes;
     TextureRes tex_res[2];
     // CUDA_MEMCPY2D cp2d;
     CUdeviceptr src_dev;
 };
 typedef QSharedPointer<SurfaceInterop_CUDA_GL> SurfaceInterop_CUDA_GL_Ptr;
 
 } //namespace cuda
 } //namespace QtAV
 #endif // QTAV_SURFACEINTEROP_CUDA_GL_H
diff --git a/src/cuda/cuda_api.h b/src/cuda/cuda_api.h
@@ -28,6  28,7 @@
 
 // high version will define cuXXX macro, so functions here will be not they look like
 #if !NV_CONFIG(DLLAPI_CUDA) && !defined(CUDA_LINK)
 // TODO: use driver version, here use internal version
 #define CUDA_FORCE_API_VERSION 3010
 #endif
 #include "dllapi/nv_inc.h"

diff --git a/src/cuda/dllapi/nv_inc.h b/src/cuda/dllapi/nv_inc.h
@@ -25,6  25,15 @@
 #undef NV_CONFIG
 #define NV_CONFIG(FEATURE) (defined QTAV_HAVE_##FEATURE && QTAV_HAVE_##FEATURE)
 
 // use qt opengl to support dynamicgl
 // TODO: only define needed types not include?
 #include <QtGlobal>
 #if QT_VERSION >= QT_VERSION_CHECK(5, 0, 0)
 #include <QtGui/qopengl.h>
 #else
 #include <QtOpenGL/qgl.h>
 #endif
 
 #if NV_CONFIG(DLLAPI_CUDA)
 namespace dllapi {
 namespace cuda {
@@ -34,6  43,7 @@ namespace cuda {
 #endif /* __cplusplus */
 
 #include "cuda.h"
 #include "cudaGL.h"
 #include "nvcuvid.h"
 
 #if defined(__cplusplus)

diff --git a/src/libQtAV.pro b/src/libQtAV.pro
@@ -187,8  187,11 @@ CONFIG  = config_cuda #config_dllapi config_dllapi_cuda
 #CONFIG  = config_cuda_link
 config_cuda {
     DEFINES  = QTAV_HAVE_CUDA=1
-    HEADERS  = cuda/dllapi/nv_inc.h cuda/helper_cuda.h
-    SOURCES  = codec/video/VideoDecoderCUDA.cpp
     HEADERS  = cuda/dllapi/nv_inc.h cuda/helper_cuda.h \
             cuda/SurfaceInterop_CUDA_GL.h
     SOURCES  = \
             codec/video/VideoDecoderCUDA.cpp \
             cuda/SurfaceInterop_CUDA_GL.cpp
     INCLUDEPATH  = $$PWD/cuda cuda/dllapi
     config_dllapi:config_dllapi_cuda {
         DEFINES  = QTAV_HAVE_DLLAPI_CUDA=1
@@ -401,7  404,6 @@ HEADERS *= \
     QtAV/ColorTransform.h \
     QtAV/VideoDecoderFFmpegHW.h
 
-
 # from mkspecs/features/qt_module.prf
 # OS X and iOS frameworks
 mac_framework { # from common.pri