generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #4238 from vladimir-dudnik:d3d11-nv12-interop

Browse Source
This commit is contained in:
Alexander Alekhin
2015-07-30 10:36:25 +00:00
parent 47e7a7a7ce 6bd01a96d9
commit 2b2bc83b61
8 changed files with 673 additions and 144 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
modules/core/include/opencv2/core/directx.hpp
View File

@@ -109,19 +109,21 @@ CV_EXPORTS Context& initializeContextFromDirect3DDevice9(IDirect3DDevice9* pDire
//! @addtogroup core_directx
//! @{
//! @brief Converts InputArray to ID3D11Texture2D
//! @brief Converts InputArray to ID3D11Texture2D. If destination texture format is DXGI_FORMAT_NV12 then
//! input UMat expected to be in BGR format and data will be downsampled and color-converted to NV12.
//
//! @note Note: function does memory copy from src to
//! @note Note: Destination texture must be allocated by application. Function does memory copy from src to
//! pD3D11Texture2D
//
//! @param src - source InputArray
//! @param pD3D11Texture2D - destination D3D11 texture
CV_EXPORTS void convertToD3D11Texture2D(InputArray src, ID3D11Texture2D* pD3D11Texture2D);
//! @brief Converts ID3D11Texture2D to OutputArray
//! @brief Converts ID3D11Texture2D to OutputArray. If input texture format is DXGI_FORMAT_NV12 then
//! data will be upsampled and color-converted to BGR format.
//
//! @note Note: function does memory copy from pD3D11Texture2D
//! to dst
//! @note Note: Destination matrix will be re-allocated if it has not enough memory to match texture size.
//! function does memory copy from pD3D11Texture2D to dst
//
//! @param pD3D11Texture2D - source D3D11 texture
//! @param dst - destination OutputArray

179
modules/core/src/directx.cpp
View File

@@ -44,6 +44,7 @@
#include "opencv2/core.hpp"
#include "opencv2/core/ocl.hpp"
#include "opencv2/core/directx.hpp"
#include "opencl_kernels_core.hpp"
#ifdef HAVE_DIRECTX
#include <vector>
@@ -167,6 +168,7 @@ int getTypeFromDXGI_FORMAT(const int iDXGI_FORMAT)
//case DXGI_FORMAT_BC7_TYPELESS:
//case DXGI_FORMAT_BC7_UNORM:
//case DXGI_FORMAT_BC7_UNORM_SRGB:
case DXGI_FORMAT_NV12: return CV_8UC3;
default: break;
}
return errorType;
@@ -701,6 +703,59 @@ static void __OpenCLinitializeD3D11()
}
#endif // defined(HAVE_DIRECTX) && defined(HAVE_OPENCL)
} // namespace directx
namespace ocl {
#if defined(HAVE_DIRECTX) && defined(HAVE_OPENCL)
static
bool ocl_convert_nv12_to_bgr(
cl_mem clImageY,
cl_mem clImageUV,
cl_mem clBuffer,
int step,
int cols,
int rows)
{
ocl::Kernel k;
k.create("YUV2BGR_NV12_8u", cv::ocl::core::cvtclr_dx_oclsrc, "");
if (k.empty())
return false;
k.args(clImageY, clImageUV, clBuffer, step, cols, rows);
size_t globalsize[] = { cols, rows };
return k.run(2, globalsize, 0, false);
}
static
bool ocl_convert_bgr_to_nv12(
cl_mem clBuffer,
int step,
int cols,
int rows,
cl_mem clImageY,
cl_mem clImageUV)
{
ocl::Kernel k;
k.create("BGR2YUV_NV12_8u", cv::ocl::core::cvtclr_dx_oclsrc, "");
if (k.empty())
return false;
k.args(clBuffer, step, cols, rows, clImageY, clImageUV);
size_t globalsize[] = { cols, rows };
return k.run(2, globalsize, 0, false);
}
#endif // HAVE_DIRECTX && HAVE_OPENCL
} // namespace ocl
namespace directx {
void convertToD3D11Texture2D(InputArray src, ID3D11Texture2D* pD3D11Texture2D)
{
(void)src; (void)pD3D11Texture2D;
@@ -719,33 +774,63 @@ void convertToD3D11Texture2D(InputArray src, ID3D11Texture2D* pD3D11Texture2D)
Size srcSize = src.size();
CV_Assert(srcSize.width == (int)desc.Width && srcSize.height == (int)desc.Height);
using namespace cv::ocl;
Context& ctx = Context::getDefault();
cl_context context = (cl_context)ctx.ptr();
UMat u = src.getUMat();
// TODO Add support for roi
CV_Assert(u.offset == 0);
CV_Assert(u.isContinuous());
cl_mem clBuffer = (cl_mem)u.handle(ACCESS_READ);
using namespace cv::ocl;
Context& ctx = Context::getDefault();
cl_context context = (cl_context)ctx.ptr();
cl_int status = 0;
cl_mem clImage = clCreateFromD3D11Texture2DKHR(context, CL_MEM_WRITE_ONLY, pD3D11Texture2D, 0, &status);
cl_mem clImage = 0;
cl_mem clImageUV = 0;
clImage = clCreateFromD3D11Texture2DKHR(context, CL_MEM_WRITE_ONLY, pD3D11Texture2D, 0, &status);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clCreateFromD3D11Texture2DKHR failed");
cl_mem clBuffer = (cl_mem)u.handle(ACCESS_READ);
if(DXGI_FORMAT_NV12 == desc.Format)
{
clImageUV = clCreateFromD3D11Texture2DKHR(context, CL_MEM_WRITE_ONLY, pD3D11Texture2D, 1, &status);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clCreateFromD3D11Texture2DKHR failed");
}
cl_command_queue q = (cl_command_queue)Queue::getDefault().ptr();
status = clEnqueueAcquireD3D11ObjectsKHR(q, 1, &clImage, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed");
size_t offset = 0; // TODO
size_t dst_origin[3] = {0, 0, 0};
size_t region[3] = {u.cols, u.rows, 1};
status = clEnqueueCopyBufferToImage(q, clBuffer, clImage, offset, dst_origin, region, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueCopyBufferToImage failed");
if(DXGI_FORMAT_NV12 == desc.Format)
{
status = clEnqueueAcquireD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed");
if(!ocl::ocl_convert_bgr_to_nv12(clBuffer, (int)u.step[0], u.cols, u.rows, clImage, clImageUV))
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: ocl_convert_bgr_to_nv12 failed");
status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueReleaseD3D11ObjectsKHR failed");
}
else
{
size_t offset = 0; // TODO
size_t origin[3] = { 0, 0, 0 };
size_t region[3] = { u.cols, u.rows, 1 };
status = clEnqueueCopyBufferToImage(q, clBuffer, clImage, offset, origin, region, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueCopyBufferToImage failed");
}
status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImage, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueReleaseD3D11ObjectsKHR failed");
@@ -757,11 +842,20 @@ void convertToD3D11Texture2D(InputArray src, ID3D11Texture2D* pD3D11Texture2D)
status = clReleaseMemObject(clImage); // TODO RAII
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clReleaseMem failed");
if(DXGI_FORMAT_NV12 == desc.Format)
{
status = clReleaseMemObject(clImageUV);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clReleaseMem failed");
}
#else
// TODO memcpy
NO_OPENCL_SUPPORT_ERROR;
#endif
}
void convertFromD3D11Texture2D(ID3D11Texture2D* pD3D11Texture2D, OutputArray dst)
{
(void)pD3D11Texture2D; (void)dst;
@@ -776,10 +870,6 @@ void convertFromD3D11Texture2D(ID3D11Texture2D* pD3D11Texture2D, OutputArray dst
int textureType = getTypeFromDXGI_FORMAT(desc.Format);
CV_Assert(textureType >= 0);
using namespace cv::ocl;
Context& ctx = Context::getDefault();
cl_context context = (cl_context)ctx.ptr();
// TODO Need to specify ACCESS_WRITE here somehow to prevent useless data copying!
dst.create(Size(desc.Width, desc.Height), textureType);
UMat u = dst.getUMat();
@@ -788,23 +878,57 @@ void convertFromD3D11Texture2D(ID3D11Texture2D* pD3D11Texture2D, OutputArray dst
CV_Assert(u.offset == 0);
CV_Assert(u.isContinuous());
cl_mem clBuffer = (cl_mem)u.handle(ACCESS_READ);
using namespace cv::ocl;
Context& ctx = Context::getDefault();
cl_context context = (cl_context)ctx.ptr();
cl_int status = 0;
cl_mem clImage = clCreateFromD3D11Texture2DKHR(context, CL_MEM_READ_ONLY, pD3D11Texture2D, 0, &status);
cl_mem clImage = 0;
cl_mem clImageUV = 0;
clImage = clCreateFromD3D11Texture2DKHR(context, CL_MEM_READ_ONLY, pD3D11Texture2D, 0, &status);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clCreateFromD3D11Texture2DKHR failed");
cl_mem clBuffer = (cl_mem)u.handle(ACCESS_READ);
if(DXGI_FORMAT_NV12 == desc.Format)
{
clImageUV = clCreateFromD3D11Texture2DKHR(context, CL_MEM_READ_ONLY, pD3D11Texture2D, 1, &status);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clCreateFromD3D11Texture2DKHR failed");
}
cl_command_queue q = (cl_command_queue)Queue::getDefault().ptr();
status = clEnqueueAcquireD3D11ObjectsKHR(q, 1, &clImage, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed");
size_t offset = 0; // TODO
size_t src_origin[3] = {0, 0, 0};
size_t region[3] = {u.cols, u.rows, 1};
status = clEnqueueCopyImageToBuffer(q, clImage, clBuffer, src_origin, region, offset, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueCopyImageToBuffer failed");
if(DXGI_FORMAT_NV12 == desc.Format)
{
status = clEnqueueAcquireD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed");
if(!ocl::ocl_convert_nv12_to_bgr(clImage, clImageUV, clBuffer, (int)u.step[0], u.cols, u.rows))
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: ocl_convert_nv12_to_bgr failed");
status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueReleaseD3D11ObjectsKHR failed");
}
else
{
size_t offset = 0; // TODO
size_t origin[3] = { 0, 0, 0 };
size_t region[3] = { u.cols, u.rows, 1 };
status = clEnqueueCopyImageToBuffer(q, clImage, clBuffer, origin, region, offset, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueCopyImageToBuffer failed");
}
status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImage, 0, NULL, NULL);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueReleaseD3D11ObjectsKHR failed");
@@ -816,6 +940,13 @@ void convertFromD3D11Texture2D(ID3D11Texture2D* pD3D11Texture2D, OutputArray dst
status = clReleaseMemObject(clImage); // TODO RAII
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clReleaseMem failed");
if(DXGI_FORMAT_NV12 == desc.Format)
{
status = clReleaseMemObject(clImageUV);
if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clReleaseMem failed");
}
#else
// TODO memcpy
NO_OPENCL_SUPPORT_ERROR;

207
modules/core/src/opencl/cvtclr_dx.cl Normal file
View File

@@ -0,0 +1,207 @@
/*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) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jia Haipeng, jiahaipeng95@gmail.com
//
//
// 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 copyright holders 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*/
#ifdef DOUBLE_SUPPORT
#ifdef cl_amd_fp64
#pragma OPENCL EXTENSION cl_amd_fp64:enable
#elif defined cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64:enable
#endif
#endif
#ifdef INTEL_DEVICE
#pragma OPENCL FP_CONTRACT ON
#pragma OPENCL FP_FAST_FMAF ON
#pragma OPENCL FP_FAST_FMA ON
#endif
static
__constant
float c_YUV2RGBCoeffs_420[5] =
{
1.163999557f,
2.017999649f,
-0.390999794f,
-0.812999725f,
1.5959997177f
};
static __constant float CV_8U_MAX = 255.0f;
static __constant float CV_8U_HALF = 128.0f;
static __constant float BT601_BLACK_RANGE = 16.0f;
static __constant float CV_8U_SCALE = 1.0f / 255.0f;
static __constant float d1 = BT601_BLACK_RANGE / CV_8U_MAX;
static __constant float d2 = CV_8U_HALF / CV_8U_MAX;
#define NCHANNELS 3
__kernel
void YUV2BGR_NV12_8u(
read_only image2d_t imgY,
read_only image2d_t imgUV,
__global unsigned char* pBGR,
int bgrStep,
int cols,
int rows)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols)
{
if (y < rows)
{
__global uchar* pDstRow1 = pBGR + mad24(y, bgrStep, mad24(x, NCHANNELS, 0));
__global uchar* pDstRow2 = pDstRow1 + bgrStep;
float4 Y1 = read_imagef(imgY, (int2)(x+0, y+0));
float4 Y2 = read_imagef(imgY, (int2)(x+1, y+0));
float4 Y3 = read_imagef(imgY, (int2)(x+0, y+1));
float4 Y4 = read_imagef(imgY, (int2)(x+1, y+1));
float4 UV = read_imagef(imgUV, (int2)(x/2, y/2)) - d2;
__constant float* coeffs = c_YUV2RGBCoeffs_420;
Y1 = max(0.f, Y1 - d1) * coeffs[0];
Y2 = max(0.f, Y2 - d1) * coeffs[0];
Y3 = max(0.f, Y3 - d1) * coeffs[0];
Y4 = max(0.f, Y4 - d1) * coeffs[0];
float ruv = fma(coeffs[4], UV.y, 0.0f);
float guv = fma(coeffs[3], UV.y, fma(coeffs[2], UV.x, 0.0f));
float buv = fma(coeffs[1], UV.x, 0.0f);
float R1 = (Y1.x + ruv) * CV_8U_MAX;
float G1 = (Y1.x + guv) * CV_8U_MAX;
float B1 = (Y1.x + buv) * CV_8U_MAX;
float R2 = (Y2.x + ruv) * CV_8U_MAX;
float G2 = (Y2.x + guv) * CV_8U_MAX;
float B2 = (Y2.x + buv) * CV_8U_MAX;
float R3 = (Y3.x + ruv) * CV_8U_MAX;
float G3 = (Y3.x + guv) * CV_8U_MAX;
float B3 = (Y3.x + buv) * CV_8U_MAX;
float R4 = (Y4.x + ruv) * CV_8U_MAX;
float G4 = (Y4.x + guv) * CV_8U_MAX;
float B4 = (Y4.x + buv) * CV_8U_MAX;
pDstRow1[0*NCHANNELS + 0] = convert_uchar_sat(B1);
pDstRow1[0*NCHANNELS + 1] = convert_uchar_sat(G1);
pDstRow1[0*NCHANNELS + 2] = convert_uchar_sat(R1);
pDstRow1[1*NCHANNELS + 0] = convert_uchar_sat(B2);
pDstRow1[1*NCHANNELS + 1] = convert_uchar_sat(G2);
pDstRow1[1*NCHANNELS + 2] = convert_uchar_sat(R2);
pDstRow2[0*NCHANNELS + 0] = convert_uchar_sat(B3);
pDstRow2[0*NCHANNELS + 1] = convert_uchar_sat(G3);
pDstRow2[0*NCHANNELS + 2] = convert_uchar_sat(R3);
pDstRow2[1*NCHANNELS + 0] = convert_uchar_sat(B4);
pDstRow2[1*NCHANNELS + 1] = convert_uchar_sat(G4);
pDstRow2[1*NCHANNELS + 2] = convert_uchar_sat(R4);
}
}
}
static
__constant float c_RGB2YUVCoeffs_420[8] =
{
0.256999969f, 0.50399971f, 0.09799957f, -0.1479988098f,
-0.2909994125f, 0.438999176f, -0.3679990768f, -0.0709991455f
};
__kernel
void BGR2YUV_NV12_8u(
__global unsigned char* pBGR,
int bgrStep,
int cols,
int rows,
write_only image2d_t imgY,
write_only image2d_t imgUV)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols)
{
if (y < rows)
{
__global const uchar* pSrcRow1 = pBGR + mad24(y, bgrStep, mad24(x, NCHANNELS, 0));
__global const uchar* pSrcRow2 = pSrcRow1 + bgrStep;
float4 src_pix1 = convert_float4(vload4(0, pSrcRow1 + 0*NCHANNELS)) * CV_8U_SCALE;
float4 src_pix2 = convert_float4(vload4(0, pSrcRow1 + 1*NCHANNELS)) * CV_8U_SCALE;
float4 src_pix3 = convert_float4(vload4(0, pSrcRow2 + 0*NCHANNELS)) * CV_8U_SCALE;
float4 src_pix4 = convert_float4(vload4(0, pSrcRow2 + 1*NCHANNELS)) * CV_8U_SCALE;
__constant float* coeffs = c_RGB2YUVCoeffs_420;
float Y1 = fma(coeffs[0], src_pix1.z, fma(coeffs[1], src_pix1.y, fma(coeffs[2], src_pix1.x, d1)));
float Y2 = fma(coeffs[0], src_pix2.z, fma(coeffs[1], src_pix2.y, fma(coeffs[2], src_pix2.x, d1)));
float Y3 = fma(coeffs[0], src_pix3.z, fma(coeffs[1], src_pix3.y, fma(coeffs[2], src_pix3.x, d1)));
float Y4 = fma(coeffs[0], src_pix4.z, fma(coeffs[1], src_pix4.y, fma(coeffs[2], src_pix4.x, d1)));
float4 UV;
UV.x = fma(coeffs[3], src_pix1.z, fma(coeffs[4], src_pix1.y, fma(coeffs[5], src_pix1.x, d2)));
UV.y = fma(coeffs[5], src_pix1.z, fma(coeffs[6], src_pix1.y, fma(coeffs[7], src_pix1.x, d2)));
write_imagef(imgY, (int2)(x+0, y+0), Y1);
write_imagef(imgY, (int2)(x+1, y+0), Y2);
write_imagef(imgY, (int2)(x+0, y+1), Y3);
write_imagef(imgY, (int2)(x+1, y+1), Y4);
write_imagef(imgUV, (int2)((x/2), (y/2)), UV);
}
}
}