mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-06-07 17:44:04 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

finished with NV12 support for D3D11-interop. Now, if texture is in NV12 format then it will be converted to/from BGR UMat.

Browse Source
This commit is contained in:
Vladimir Dudnik 2015-07-29 14:31:10 +04:00
parent d4774ead43
commit 6bd01a96d9
4 changed files with 95 additions and 73 deletions
Show Stats 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 //! @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 //! pD3D11Texture2D
// //
//! @param src - source InputArray //! @param src - source InputArray
//! @param pD3D11Texture2D - destination D3D11 texture //! @param pD3D11Texture2D - destination D3D11 texture
CV_EXPORTS void convertToD3D11Texture2D(InputArray src, ID3D11Texture2D* pD3D11Texture2D); 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 //! @note Note: Destination matrix will be re-allocated if it has not enough memory to match texture size.
//! to dst //! function does memory copy from pD3D11Texture2D to dst
// //
//! @param pD3D11Texture2D - source D3D11 texture //! @param pD3D11Texture2D - source D3D11 texture
//! @param dst - destination OutputArray //! @param dst - destination OutputArray

18
modules/core/src/directx.cpp
View File

@ -168,7 +168,7 @@ int getTypeFromDXGI_FORMAT(const int iDXGI_FORMAT)
//case DXGI_FORMAT_BC7_TYPELESS: //case DXGI_FORMAT_BC7_TYPELESS:
//case DXGI_FORMAT_BC7_UNORM: //case DXGI_FORMAT_BC7_UNORM:
//case DXGI_FORMAT_BC7_UNORM_SRGB: //case DXGI_FORMAT_BC7_UNORM_SRGB:
case DXGI_FORMAT_NV12: return CV_8UC4; case DXGI_FORMAT_NV12: return CV_8UC3;
default: break; default: break;
} }
return errorType; return errorType;
@ -710,7 +710,7 @@ namespace ocl {
#if defined(HAVE_DIRECTX) && defined(HAVE_OPENCL) #if defined(HAVE_DIRECTX) && defined(HAVE_OPENCL)
static static
bool ocl_convert_nv12_to_rgba( bool ocl_convert_nv12_to_bgr(
cl_mem clImageY, cl_mem clImageY,
cl_mem clImageUV, cl_mem clImageUV,
cl_mem clBuffer, cl_mem clBuffer,
@ -719,7 +719,7 @@ bool ocl_convert_nv12_to_rgba(
int rows) int rows)
{ {
ocl::Kernel k; ocl::Kernel k;
k.create("YUV2RGBA_NV12_8u", cv::ocl::core::cvtclr_dx_oclsrc, ""); k.create("YUV2BGR_NV12_8u", cv::ocl::core::cvtclr_dx_oclsrc, "");
if (k.empty()) if (k.empty())
return false; return false;
@ -731,7 +731,7 @@ bool ocl_convert_nv12_to_rgba(
static static
bool ocl_convert_rgba_to_nv12( bool ocl_convert_bgr_to_nv12(
cl_mem clBuffer, cl_mem clBuffer,
int step, int step,
int cols, int cols,
@ -740,7 +740,7 @@ bool ocl_convert_rgba_to_nv12(
cl_mem clImageUV) cl_mem clImageUV)
{ {
ocl::Kernel k; ocl::Kernel k;
k.create("RGBA2YUV_NV12_8u", cv::ocl::core::cvtclr_dx_oclsrc, ""); k.create("BGR2YUV_NV12_8u", cv::ocl::core::cvtclr_dx_oclsrc, "");
if (k.empty()) if (k.empty())
return false; return false;
@ -813,8 +813,8 @@ void convertToD3D11Texture2D(InputArray src, ID3D11Texture2D* pD3D11Texture2D)
if (status != CL_SUCCESS) if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed"); CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed");
if(!ocl::ocl_convert_rgba_to_nv12(clBuffer, (int)u.step[0], u.cols, u.rows, clImage, clImageUV)) 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_rgba_to_nv12 failed"); CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: ocl_convert_bgr_to_nv12 failed");
status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL); status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL);
if (status != CL_SUCCESS) if (status != CL_SUCCESS)
@ -911,8 +911,8 @@ void convertFromD3D11Texture2D(ID3D11Texture2D* pD3D11Texture2D, OutputArray dst
if (status != CL_SUCCESS) if (status != CL_SUCCESS)
CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed"); CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: clEnqueueAcquireD3D11ObjectsKHR failed");
if(!ocl::ocl_convert_nv12_to_rgba(clImage, clImageUV, clBuffer, (int)u.step[0], u.cols, u.rows)) 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_rgba failed"); CV_Error(cv::Error::OpenCLApiCallError, "OpenCL: ocl_convert_nv12_to_bgr failed");
status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL); status = clEnqueueReleaseD3D11ObjectsKHR(q, 1, &clImageUV, 0, NULL, NULL);
if (status != CL_SUCCESS) if (status != CL_SUCCESS)

136
modules/core/src/opencl/cvtclr_dx.cl
View File

@ -59,6 +59,7 @@
#pragma OPENCL FP_FAST_FMA ON #pragma OPENCL FP_FAST_FMA ON
#endif #endif
static static
__constant __constant
float c_YUV2RGBCoeffs_420[5] = float c_YUV2RGBCoeffs_420[5] =
@ -70,13 +71,21 @@ float c_YUV2RGBCoeffs_420[5] =
1.5959997177f 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 __kernel
void YUV2RGBA_NV12_8u( void YUV2BGR_NV12_8u(
read_only image2d_t imgY, read_only image2d_t imgY,
read_only image2d_t imgUV, read_only image2d_t imgUV,
__global unsigned char* pRGBA, __global unsigned char* pBGR,
int rgbaStep, int bgrStep,
int cols, int cols,
int rows) int rows)
{ {
@ -87,44 +96,58 @@ void YUV2RGBA_NV12_8u(
{ {
if (y < rows) if (y < rows)
{ {
__global uchar* pDstRow1 = pRGBA + mad24(y, rgbaStep, mad24(x, 4, 0)); __global uchar* pDstRow1 = pBGR + mad24(y, bgrStep, mad24(x, NCHANNELS, 0));
__global uchar* pDstRow2 = pDstRow1 + rgbaStep; __global uchar* pDstRow2 = pDstRow1 + bgrStep;
float4 Y1 = read_imagef(imgY, (int2)(x+0, y+0)) * 255.0f; float4 Y1 = read_imagef(imgY, (int2)(x+0, y+0));
float4 Y2 = read_imagef(imgY, (int2)(x+1, y+0)) * 255.0f; float4 Y2 = read_imagef(imgY, (int2)(x+1, y+0));
float4 Y3 = read_imagef(imgY, (int2)(x+0, y+1)) * 255.0f; float4 Y3 = read_imagef(imgY, (int2)(x+0, y+1));
float4 Y4 = read_imagef(imgY, (int2)(x+1, y+1)) * 255.0f; float4 Y4 = read_imagef(imgY, (int2)(x+1, y+1));
float4 UV = read_imagef(imgUV, (int2)(x/2, y/2)) * 255.0f - 128.0f; float4 UV = read_imagef(imgUV, (int2)(x/2, y/2)) - d2;
__constant float* coeffs = c_YUV2RGBCoeffs_420; __constant float* coeffs = c_YUV2RGBCoeffs_420;
float ruv = fma(coeffs[4], UV.y, 0.5f);
float guv = fma(coeffs[3], UV.y, fma(coeffs[2], UV.x, 0.5f));
float buv = fma(coeffs[1], UV.x, 0.5f);
Y1 = max(0.f, Y1 - 16.f) * coeffs[0]; Y1 = max(0.f, Y1 - d1) * coeffs[0];
pDstRow1[0+0] = convert_uchar_sat(Y1.x + ruv); Y2 = max(0.f, Y2 - d1) * coeffs[0];
pDstRow1[1+0] = convert_uchar_sat(Y1.x + guv); Y3 = max(0.f, Y3 - d1) * coeffs[0];
pDstRow1[2+0] = convert_uchar_sat(Y1.x + buv); Y4 = max(0.f, Y4 - d1) * coeffs[0];
pDstRow1[3+0] = 255;
Y2 = max(0.f, Y2 - 16.f) * coeffs[0]; float ruv = fma(coeffs[4], UV.y, 0.0f);
pDstRow1[0+4] = convert_uchar_sat(Y2.x + ruv); float guv = fma(coeffs[3], UV.y, fma(coeffs[2], UV.x, 0.0f));
pDstRow1[1+4] = convert_uchar_sat(Y2.x + guv); float buv = fma(coeffs[1], UV.x, 0.0f);
pDstRow1[2+4] = convert_uchar_sat(Y2.x + buv);
pDstRow1[3+4] = 255;
Y3 = max(0.f, Y3 - 16.f) * coeffs[0]; float R1 = (Y1.x + ruv) * CV_8U_MAX;
pDstRow2[0+0] = convert_uchar_sat(Y3.x + ruv); float G1 = (Y1.x + guv) * CV_8U_MAX;
pDstRow2[1+0] = convert_uchar_sat(Y3.x + guv); float B1 = (Y1.x + buv) * CV_8U_MAX;
pDstRow2[2+0] = convert_uchar_sat(Y3.x + buv);
pDstRow2[3+0] = 255;
Y4 = max(0.f, Y4 - 16.f) * coeffs[0]; float R2 = (Y2.x + ruv) * CV_8U_MAX;
pDstRow2[0+4] = convert_uchar_sat(Y4.x + ruv); float G2 = (Y2.x + guv) * CV_8U_MAX;
pDstRow2[1+4] = convert_uchar_sat(Y4.x + guv); float B2 = (Y2.x + buv) * CV_8U_MAX;
pDstRow2[2+4] = convert_uchar_sat(Y4.x + buv);
pDstRow2[3+4] = 255; 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);
} }
} }
} }
@ -137,11 +160,11 @@ __constant float c_RGB2YUVCoeffs_420[8] =
-0.2909994125f, 0.438999176f, -0.3679990768f, -0.0709991455f -0.2909994125f, 0.438999176f, -0.3679990768f, -0.0709991455f
}; };
#define scn 4
__kernel __kernel
void RGBA2YUV_NV12_8u( void BGR2YUV_NV12_8u(
__global unsigned char* pRGBA, __global unsigned char* pBGR,
int rgbaStep, int bgrStep,
int cols, int cols,
int rows, int rows,
write_only image2d_t imgY, write_only image2d_t imgY,
@ -154,34 +177,31 @@ void RGBA2YUV_NV12_8u(
{ {
if (y < rows) if (y < rows)
{ {
__global const uchar* pSrcRow1 = pRGBA + mad24(y, rgbaStep, mad24(x, scn, 0)); __global const uchar* pSrcRow1 = pBGR + mad24(y, bgrStep, mad24(x, NCHANNELS, 0));
__global const uchar* pSrcRow2 = pSrcRow1 + rgbaStep; __global const uchar* pSrcRow2 = pSrcRow1 + bgrStep;
float4 src_pix1 = convert_float4(vload4(0, pSrcRow1 + 0)); float4 src_pix1 = convert_float4(vload4(0, pSrcRow1 + 0*NCHANNELS)) * CV_8U_SCALE;
float4 src_pix2 = convert_float4(vload4(0, pSrcRow1 + scn)); float4 src_pix2 = convert_float4(vload4(0, pSrcRow1 + 1*NCHANNELS)) * CV_8U_SCALE;
float4 src_pix3 = convert_float4(vload4(0, pSrcRow2 + 0)); float4 src_pix3 = convert_float4(vload4(0, pSrcRow2 + 0*NCHANNELS)) * CV_8U_SCALE;
float4 src_pix4 = convert_float4(vload4(0, pSrcRow2 + scn)); float4 src_pix4 = convert_float4(vload4(0, pSrcRow2 + 1*NCHANNELS)) * CV_8U_SCALE;
__constant float* coeffs = c_RGB2YUVCoeffs_420; __constant float* coeffs = c_RGB2YUVCoeffs_420;
uchar Y1 = convert_uchar_sat(fma(coeffs[0], src_pix1.x, fma(coeffs[1], src_pix1.y, fma(coeffs[2], src_pix1.z, 16.5f)))); float Y1 = fma(coeffs[0], src_pix1.z, fma(coeffs[1], src_pix1.y, fma(coeffs[2], src_pix1.x, d1)));
uchar Y2 = convert_uchar_sat(fma(coeffs[0], src_pix2.x, fma(coeffs[1], src_pix2.y, fma(coeffs[2], src_pix2.z, 16.5f)))); float Y2 = fma(coeffs[0], src_pix2.z, fma(coeffs[1], src_pix2.y, fma(coeffs[2], src_pix2.x, d1)));
uchar Y3 = convert_uchar_sat(fma(coeffs[0], src_pix3.x, fma(coeffs[1], src_pix3.y, fma(coeffs[2], src_pix3.z, 16.5f)))); float Y3 = fma(coeffs[0], src_pix3.z, fma(coeffs[1], src_pix3.y, fma(coeffs[2], src_pix3.x, d1)));
uchar Y4 = convert_uchar_sat(fma(coeffs[0], src_pix4.x, fma(coeffs[1], src_pix4.y, fma(coeffs[2], src_pix4.z, 16.5f)))); float Y4 = fma(coeffs[0], src_pix4.z, fma(coeffs[1], src_pix4.y, fma(coeffs[2], src_pix4.x, d1)));
write_imageui(imgY, (int2)(x+0, y+0), Y1); float4 UV;
write_imageui(imgY, (int2)(x+1, y+0), Y2); UV.x = fma(coeffs[3], src_pix1.z, fma(coeffs[4], src_pix1.y, fma(coeffs[5], src_pix1.x, d2)));
write_imageui(imgY, (int2)(x+0, y+1), Y3); UV.y = fma(coeffs[5], src_pix1.z, fma(coeffs[6], src_pix1.y, fma(coeffs[7], src_pix1.x, d2)));
write_imageui(imgY, (int2)(x+1, y+1), Y4);
float uf = fma(coeffs[3], src_pix1.x, fma(coeffs[4], src_pix1.y, fma(coeffs[5], src_pix1.z, 128.5f))); write_imagef(imgY, (int2)(x+0, y+0), Y1);
float vf = fma(coeffs[5], src_pix1.x, fma(coeffs[6], src_pix1.y, fma(coeffs[7], src_pix1.z, 128.5f))); 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);
uchar U = convert_uchar_sat(uf); write_imagef(imgUV, (int2)((x/2), (y/2)), UV);
uchar V = convert_uchar_sat(vf);
write_imageui(imgUV, (int2)((x/2)+0, (y/2)), U);
write_imageui(imgUV, (int2)((x/2)+1, (y/2)), V);
} }
} }
} }

2
samples/directx/d3d11_interop.cpp
View File

@ -162,7 +162,7 @@ public:
desc_nv12_cpu_copy.SampleDesc.Quality = 0; desc_nv12_cpu_copy.SampleDesc.Quality = 0;
desc_nv12_cpu_copy.BindFlags = 0; desc_nv12_cpu_copy.BindFlags = 0;
desc_nv12_cpu_copy.Usage = D3D11_USAGE_STAGING; desc_nv12_cpu_copy.Usage = D3D11_USAGE_STAGING;
desc_nv12_cpu_copy.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE | D3D11_CPU_ACCESS_READ; desc_nv12_cpu_copy.CPUAccessFlags = /*D3D11_CPU_ACCESS_WRITE | */D3D11_CPU_ACCESS_READ;
desc_nv12_cpu_copy.MiscFlags = 0; desc_nv12_cpu_copy.MiscFlags = 0;
r = m_pD3D11Dev->CreateTexture2D(&desc_nv12_cpu_copy, 0, &m_pSurfaceNV12_cpu_copy); r = m_pD3D11Dev->CreateTexture2D(&desc_nv12_cpu_copy, 0, &m_pSurfaceNV12_cpu_copy);