AuroraMiddleware/OpenSubdiv
1
0
Fork 0
mirror of https://github.com/PixarAnimationStudios/OpenSubdiv synced 2024-11-26 21:40:07 +00:00
Code Issues Releases Wiki Activity

Fix DX-OpenCL/CUDA interop.

Browse Source 
- resolves DX-CL interop functions in Osd::ClD3D11VertexBuffer.
- enable CL kernels in DX build.
- more cleanup in test harnesses, adding D3D11 initializations into DeviceContext.
- add new defines OPENSUBDIV_HAS_OPENGL and OPENSUBDIV_HAS_DX for convenience.
This commit is contained in:
Takahito Tejima 2015-04-29 11:51:12 -07:00
parent 766ed5b316
commit dcb022e1db
10 changed files with 353 additions and 143 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
CMakeLists.txt
View File

@ -345,6 +345,9 @@ if(NOT NO_PTEX)
find_package(PTex 2.0)
endif()
if (OPENGL_FOUND AND NOT IOS)
add_definitions(
-DOPENSUBDIV_HAS_OPENGL
)
if (APPLE)
find_package(GLEW)
else()
@ -353,7 +356,12 @@ if (OPENGL_FOUND AND NOT IOS)
endif()
if (WIN32 AND NOT NO_DX)
find_package(DXSDK)
find_package(DXSDK)
if(DXSDK_FOUND)
add_definitions(
-DOPENSUBDIV_HAS_DX
)
endif()
endif()
if (NOT NO_MAYA)

256
examples/common/clDeviceContext.cpp
View File

@ -36,6 +36,98 @@
#include <cstring>
#include <string>
#if defined(OPENSUBDIV_HAS_DX)
#include <D3D11.h>
#include <CL/cl_d3d11.h>
#endif
#define message(...) // fprintf(stderr, __VA_ARGS__)
#define error(...) fprintf(stderr, __VA_ARGS__)
// returns the first found platform.
//
static cl_platform_id
findPlatform() {
cl_uint numPlatforms;
cl_int ciErrNum = clGetPlatformIDs(0, NULL, &numPlatforms);
if (ciErrNum != CL_SUCCESS) {
error("Error %d in clGetPlatformIDs call.\n", ciErrNum);
return NULL;
}
if (numPlatforms == 0) {
error("No OpenCL platform found.\n");
return NULL;
}
cl_platform_id *clPlatformIDs = new cl_platform_id[numPlatforms];
ciErrNum = clGetPlatformIDs(numPlatforms, clPlatformIDs, NULL);
char chBuffer[1024];
for (cl_uint i = 0; i < numPlatforms; ++i) {
ciErrNum = clGetPlatformInfo(clPlatformIDs[i], CL_PLATFORM_NAME,
1024, chBuffer,NULL);
if (ciErrNum == CL_SUCCESS) {
cl_platform_id platformId = clPlatformIDs[i];
delete[] clPlatformIDs;
return platformId;
}
}
delete[] clPlatformIDs;
return NULL;
}
// returns the device in clDevices which supports the extension.
//
static int
findExtensionSupportedDevice(cl_device_id *clDevices,
int numDevices,
const char *extensionName) {
// find a device that supports sharing with GL/D3D11
// (SLI / X-fire configurations)
cl_int ciErrNum;
for (int i = 0; i < numDevices; ++i) {
// get extensions string size
size_t extensionSize;
ciErrNum = clGetDeviceInfo(clDevices[i],
CL_DEVICE_EXTENSIONS, 0, NULL,
&extensionSize );
if (ciErrNum != CL_SUCCESS) {
error("Error %d in clGetDeviceInfo\n", ciErrNum);
return -1;
}
if (extensionSize>0) {
// get extensions string
char *extensions = new char[extensionSize];
ciErrNum = clGetDeviceInfo(clDevices[i], CL_DEVICE_EXTENSIONS,
extensionSize, extensions,
&extensionSize);
if (ciErrNum != CL_SUCCESS) {
error("Error %d in clGetDeviceInfo\n", ciErrNum);
delete[] extensions;
continue;
}
std::string extString(extensions);
delete[] extensions;
// parse string. This is bit deficient since the extentions
// is space separated.
//
// The actual string would be "cl_khr_d3d11_sharing"
// or "cl_nv_d3d11_sharing"
if (extString.find(extensionName) != std::string::npos) {
return i;
}
}
}
return -1;
}
// --------------------------------------------------------------------------
CLDeviceContext::CLDeviceContext() :
_clContext(NULL), _clCommandQueue(NULL) {
}
@ -59,7 +151,7 @@ CLDeviceContext::HAS_CL_VERSION_1_1 () {
clewInitialized = true;
clewLoadSuccess = clewInit() == CLEW_SUCCESS;
if (not clewLoadSuccess) {
fprintf(stderr, "Loading OpenCL failed.\n");
error(stderr, "Loading OpenCL failed.\n");
}
}
return clewLoadSuccess;
@ -72,33 +164,13 @@ CLDeviceContext::Initialize() {
#ifdef OPENSUBDIV_HAS_CLEW
if (!clGetPlatformIDs) {
printf("Error clGetPlatformIDs function not bound.\n");
error("Error clGetPlatformIDs function not bound.\n");
return false;
}
#endif
cl_int ciErrNum;
cl_platform_id cpPlatform = 0;
cl_uint num_platforms;
ciErrNum = clGetPlatformIDs(0, NULL, &num_platforms);
if (ciErrNum != CL_SUCCESS) {
printf("Error %d in clGetPlatformIDs call.\n", ciErrNum);
return false;
}
if (num_platforms == 0) {
printf("No OpenCL platform found.\n");
return false;
}
cl_platform_id *clPlatformIDs = new cl_platform_id[num_platforms];
ciErrNum = clGetPlatformIDs(num_platforms, clPlatformIDs, NULL);
char chBuffer[1024];
for (cl_uint i = 0; i < num_platforms; ++i) {
ciErrNum = clGetPlatformInfo(clPlatformIDs[i], CL_PLATFORM_NAME, 1024, chBuffer,NULL);
if (ciErrNum == CL_SUCCESS) {
cpPlatform = clPlatformIDs[i];
}
}
cl_platform_id cpPlatform = findPlatform();
#if defined(_WIN32)
cl_context_properties props[] = {
@ -122,33 +194,36 @@ CLDeviceContext::Initialize() {
0
};
#endif
delete[] clPlatformIDs;
int clDeviceUsed = 0;
#if defined(__APPLE__)
_clContext = clCreateContext(props, 0, NULL, clLogMessagesToStdoutAPPLE, NULL, &ciErrNum);
_clContext = clCreateContext(props, 0, NULL, clLogMessagesToStdoutAPPLE,
NULL, &ciErrNum);
if (ciErrNum != CL_SUCCESS) {
printf("Error %d in clCreateContext\n", ciErrNum);
error("Error %d in clCreateContext\n", ciErrNum);
return false;
}
size_t devicesSize = 0;
clGetGLContextInfoAPPLE(_clContext, kCGLContext, CL_CGL_DEVICES_FOR_SUPPORTED_VIRTUAL_SCREENS_APPLE, 0, NULL, &devicesSize);
clGetGLContextInfoAPPLE(_clContext, kCGLContext,
CL_CGL_DEVICES_FOR_SUPPORTED_VIRTUAL_SCREENS_APPLE,
0, NULL, &devicesSize);
int numDevices = int(devicesSize / sizeof(cl_device_id));
if (numDevices == 0) {
printf("No sharable devices.\n");
error("No sharable devices.\n");
return false;
}
cl_device_id *clDevices = new cl_device_id[numDevices];
clGetGLContextInfoAPPLE(_clContext, kCGLContext, CL_CGL_DEVICES_FOR_SUPPORTED_VIRTUAL_SCREENS_APPLE, numDevices * sizeof(cl_device_id), clDevices, NULL);
#else
clGetGLContextInfoAPPLE(_clContext, kCGLContext,
CL_CGL_DEVICES_FOR_SUPPORTED_VIRTUAL_SCREENS_APPLE,
numDevices * sizeof(cl_device_id), clDevices, NULL);
#else // not __APPLE__
// get the number of GPU devices available to the platform
cl_uint numDevices = 0;
clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices);
if (numDevices == 0) {
printf("No CL GPU device found.\n");
error("No CL GPU device found.\n");
return false;
}
@ -156,46 +231,77 @@ CLDeviceContext::Initialize() {
cl_device_id *clDevices = new cl_device_id[numDevices];
clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, numDevices, clDevices, NULL);
#define GL_SHARING_EXTENSION "cl_khr_gl_sharing"
const char *extension = "cl_khr_gl_sharing";
int clDeviceUsed = findExtensionSupportedDevice(clDevices, numDevices,
extension);
// find a device that supports sharing with GL (SLI / X-fire configurations)
bool sharingSupported=false;
for (int i=0; i<(int)numDevices; ++i) {
size_t extensionSize;
ciErrNum = clGetDeviceInfo(clDevices[i], CL_DEVICE_EXTENSIONS, 0, NULL, &extensionSize );
if (ciErrNum != CL_SUCCESS) {
printf("Error %d in clGetDeviceInfo\n", ciErrNum);
return false;
}
if (extensionSize>0) {
char* extensions = (char*)malloc(extensionSize);
ciErrNum = clGetDeviceInfo(clDevices[i], CL_DEVICE_EXTENSIONS, extensionSize, extensions, &extensionSize);
if (ciErrNum != CL_SUCCESS) {
printf("Error %d in clGetDeviceInfo\n", ciErrNum);
return false;
}
std::string stdDevString(extensions);
free(extensions);
size_t szOldPos = 0, szSpacePos = stdDevString.find(' ', szOldPos); // extensions string is space delimited
while (szSpacePos != stdDevString.npos) {
if (strcmp(GL_SHARING_EXTENSION,
stdDevString.substr(szOldPos, szSpacePos - szOldPos).c_str())==0) {
clDeviceUsed = i;
sharingSupported = true;
break;
}
do {
szOldPos = szSpacePos + 1;
szSpacePos = stdDevString.find(' ', szOldPos);
} while (szSpacePos == szOldPos);
}
}
if (clDeviceUsed < 0) {
error("No device found that supports CL/GL context sharing\n");
delete[] clDevices;
return false;
}
if (not sharingSupported) {
printf("No device found that supports CL/GL context sharing\n");
_clContext = clCreateContext(props, 1, &clDevices[clDeviceUsed],
NULL, NULL, &ciErrNum);
#endif // not __APPLE__
if (ciErrNum != CL_SUCCESS) {
error("Error %d in clCreateContext\n", ciErrNum);
delete[] clDevices;
return false;
}
_clCommandQueue = clCreateCommandQueue(_clContext, clDevices[clDeviceUsed],
0, &ciErrNum);
delete[] clDevices;
if (ciErrNum != CL_SUCCESS) {
error("Error %d in clCreateCommandQueue\n", ciErrNum);
return false;
}
return true;
}
// ---------------------------------------------------------------------------
bool
CLD3D11DeviceContext::Initialize(ID3D11DeviceContext *d3dDeviceContext) {
#if defined(OPENSUBDIV_HAS_DX)
_d3dDeviceContext = d3dDeviceContext;
cl_int ciErrNum;
cl_platform_id cpPlatform = findPlatform();
ID3D11Device *device;
d3dDeviceContext->GetDevice(&device);
cl_context_properties props[] = {
CL_CONTEXT_D3D11_DEVICE_KHR, (cl_context_properties)device,
CL_CONTEXT_PLATFORM, (cl_context_properties)cpPlatform,
0
};
// get the number of GPU devices available to the platform
cl_uint numDevices = 0;
clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, 0, NULL, &numDevices);
if (numDevices == 0) {
error("No CL GPU device found.\n");
return false;
}
// create the device list
cl_device_id *clDevices = new cl_device_id[numDevices];
clGetDeviceIDs(cpPlatform, CL_DEVICE_TYPE_GPU, numDevices, clDevices, NULL);
// we're cheating a little bit.
// try to find both cl_khr_d3d11_sharing and cl_nv_d3d11_sharing.
const char *extension = "_d3d11_sharing";
int clDeviceUsed = findExtensionSupportedDevice(clDevices, numDevices,
extension);
if (clDeviceUsed < 0) {
error("No device found that supports CL/D3D11 context sharing\n");
delete[] clDevices;
return false;
}
@ -203,19 +309,21 @@ CLDeviceContext::Initialize() {
_clContext = clCreateContext(props, 1, &clDevices[clDeviceUsed],
NULL, NULL, &ciErrNum);
if (ciErrNum != CL_SUCCESS) {
printf("Error %d in clCreateContext\n", ciErrNum);
error("Error %d in clCreateContext\n", ciErrNum);
delete[] clDevices;
return false;
}
#endif
_clCommandQueue = clCreateCommandQueue(_clContext, clDevices[clDeviceUsed],
0, &ciErrNum);
delete[] clDevices;
if (ciErrNum != CL_SUCCESS) {
printf("Error %d in clCreateCommandQueue\n", ciErrNum);
error("Error %d in clCreateCommandQueue\n", ciErrNum);
return false;
}
return true;
#else
(void)d3dDeviceContext; // unused
return false;
#endif
}

15
examples/common/clDeviceContext.h
View File

@ -47,11 +47,24 @@ public:
return _clCommandQueue;
}
private:
protected:
cl_context _clContext;
cl_command_queue _clCommandQueue;
};
struct ID3D11DeviceContext;
class CLD3D11DeviceContext : public CLDeviceContext {
public:
bool Initialize(ID3D11DeviceContext *deviceContext);
ID3D11DeviceContext *GetDeviceContext() const {
return _d3dDeviceContext;
}
private:
ID3D11DeviceContext *_d3dDeviceContext;
};
#endif // OSD_EXAMPLES_COMMON_CL_DEVICE_CONTEXT_H

16
examples/common/cudaDeviceContext.cpp
View File

@ -38,6 +38,10 @@
#include <cuda_runtime_api.h>
#include <cuda_gl_interop.h>
#if defined(OPENSUBDIV_HAS_DX)
#include <cuda_d3d11_interop.h>
#endif
#define message(fmt, ...)
//#define message(fmt, ...) fprintf(stderr, fmt, __VA_ARGS__)
#define error(fmt, ...) fprintf(stderr, fmt, __VA_ARGS__)
@ -135,3 +139,15 @@ CudaDeviceContext::Initialize() {
_initialized = true;
return true;
}
bool
CudaDeviceContext::Initialize(ID3D11Device *device) {
#if defined(OPENSUBDIV_HAS_DX)
cudaD3D11SetDirect3DDevice(device);
return true;
#else
(void)device; // unused
return false;
#endif
}

7
examples/common/cudaDeviceContext.h
View File

@ -25,13 +25,20 @@
#ifndef OSD_EXAMPLES_COMMON_CUDA_DEVICE_CONTEXT_H
#define OSD_EXAMPLES_COMMON_CUDA_DEVICE_CONTEXT_H
struct ID3D11Device;
class CudaDeviceContext {
public:
CudaDeviceContext();
~CudaDeviceContext();
/// Initialze cuda device from the current GL context
bool Initialize();
/// Initialze cuda device from the ID3D11Device
bool Initialize(ID3D11Device *device);
/// Returns true if the cuda device has already been initialized
bool IsInitialized() const {
return _initialized;
}

1
examples/dxPtexViewer/CMakeLists.txt
View File

@ -51,7 +51,6 @@ include_directories("${CMAKE_CURRENT_BINARY_DIR}")
_add_possibly_cuda_executable(dxPtexViewer WIN32
"${SOURCE_FILES}"
"${SHADER_FILES}"
"${INC_FILES}"
$<TARGET_OBJECTS:regression_common_obj>
$<TARGET_OBJECTS:regression_vtr_utils_obj>

43
examples/dxPtexViewer/dxPtexViewer.cpp
View File

@ -41,22 +41,14 @@ OpenSubdiv::Osd::CpuComputeController * g_cpuComputeController = NULL;
OpenSubdiv::Osd::OmpComputeController * g_ompComputeController = NULL;
#endif
#undef OPENSUBDIV_HAS_OPENCL // XXX: dyu OpenCL D3D11 interop needs work...
#ifdef OPENSUBDIV_HAS_OPENCL
#include <osd/clD3D11VertexBuffer.h>
#include <osd/clComputeContext.h>
#include <osd/clComputeController.h>
#include "../common/clInit.h"
#include "../common/clDeviceContext.h"
struct CLContext {
cl_context GetContext() const { return clContext; }
cl_command_queue GetCommandQueue() const { return clQueue; }
cl_context clContext;
cl_command_queue clQueue;
ID3D11DeviceContext *pd3dDeviceContext;
};
CLContext g_clContext;
CLD3D11DeviceContext g_clDeviceContext;
OpenSubdiv::Osd::CLComputeController * g_clComputeController = NULL;
#endif
@ -65,10 +57,9 @@ OpenSubdiv::Osd::CpuComputeController * g_cpuComputeController = NULL;
#include <osd/cudaComputeContext.h>
#include <osd/cudaComputeController.h>
#include <cuda_runtime_api.h>
#include <cuda_d3d11_interop.h>
#include "../common/cudaDeviceContext.h"
bool g_cudaInitialized = false;
CudaDeviceContext g_cudaDeviceContext;
OpenSubdiv::Osd::CudaComputeController * g_cudaComputeController = NULL;
#endif
@ -695,6 +686,8 @@ createOsdMesh(int level, int kernel) {
OpenSubdiv::Osd::MeshBitset bits;
bits.set(OpenSubdiv::Osd::MeshAdaptive, doAdaptive);
bits.set(OpenSubdiv::Osd::MeshPtexData, true);
// gregory basis hasn't supported yet in D3D11Mesh
bits.set(OpenSubdiv::Osd::MeshEndCapLegacyGregory, true);
int numVertexElements = 6; //g_adaptive ? 3 : 6;
int numVaryingElements = 0;
@ -731,17 +724,18 @@ createOsdMesh(int level, int kernel) {
} else if (kernel == kCL) {
if (not g_clComputeController) {
g_clComputeController = new OpenSubdiv::Osd::CLComputeController(
g_clContext.clContext, g_clContext.clQueue);
g_clDeviceContext.GetContext(),
g_clDeviceContext.GetCommandQueue());
}
g_mesh = new OpenSubdiv::Osd::Mesh<OpenSubdiv::Osd::CLD3D11VertexBuffer,
OpenSubdiv::Osd::CLComputeController,
OpenSubdiv::Osd::D3D11DrawContext,
CLContext>(
CLD3D11DeviceContext>(
g_clComputeController,
refiner,
numVertexElements,
numVaryingElements,
level, bits, &g_clContext);
level, bits, &g_clDeviceContext);
#endif
#ifdef OPENSUBDIV_HAS_CUDA
} else if (kernel == kCUDA) {
@ -1137,12 +1131,10 @@ quit() {
#ifdef OPENSUBDIV_HAS_OPENCL
delete g_clComputeController;
uninitCL(g_clContext.clContext, g_clContext.clQueue);
#endif
#ifdef OPENSUBDIV_HAS_CUDA
delete g_cudaComputeController;
cudaDeviceReset();
#endif
delete g_d3d11ComputeController;
@ -1179,18 +1171,19 @@ callbackKernel(int k) {
g_kernel = k;
#ifdef OPENSUBDIV_HAS_OPENCL
if (g_kernel == kCL and g_clContext.clContext == NULL) {
if (initCL(&g_clContext.clContext, &g_clContext.clQueue) == false) {
if (g_kernel == kCL and (not g_clDeviceContext.IsInitialized())) {
if (g_clDeviceContext.Initialize(g_pd3dDeviceContext) == false) {
printf("Error in initializing OpenCL\n");
exit(1);
}
g_clContext.pd3dDeviceContext = g_pd3dDeviceContext;
}
#endif
#ifdef OPENSUBDIV_HAS_CUDA
if (g_kernel == kCUDA and g_cudaInitialized == false) {
g_cudaInitialized = true;
cudaD3D11SetDirect3DDevice( g_pd3dDevice );
if (g_kernel == kCUDA and (not g_cudaDeviceContext.IsInitialized())) {
if (g_cudaDeviceContext.Initialize(g_pd3dDevice) == false) {
printf("Error in initializing Cuda\n");
exit(1);
}
}
#endif
@ -1288,7 +1281,7 @@ initHUD() {
g_hud->AddRadioButton(0, "CUDA", false, 10, 50, callbackKernel, kCUDA, 'K');
#endif
#ifdef OPENSUBDIV_HAS_OPENCL
if (HAS_CL_VERSION_1_1()) {
if (CLDeviceContext::HAS_CL_VERSION_1_1()) {
g_hud->AddRadioButton(0, "OPENCL", false, 10, 70, callbackKernel, kCL, 'K');
}
#endif

44
examples/dxViewer/dxviewer.cpp Executable file → Normal file
View File

@ -45,25 +45,14 @@ OpenSubdiv::Osd::CpuComputeController * g_cpuComputeController = NULL;
OpenSubdiv::Osd::TbbComputeController *g_tbbComputeController = NULL;
#endif
#undef OPENSUBDIV_HAS_OPENCL // XXX: dyu OpenCL D3D11 interop needs work...
#ifdef OPENSUBDIV_HAS_OPENCL
#include <osd/clD3D11VertexBuffer.h>
#include <osd/clComputeContext.h>
#include <osd/clComputeController.h>
#include "../common/clInit.h"
struct CLContext {
cl_context GetContext() const { return clContext; }
cl_command_queue GetCommandQueue() const { return clQueue; }
ID3D11DeviceContext *GetDeviceContext() const { return pd3dDeviceContext; }
cl_context clContext;
cl_command_queue clQueue;
ID3D11DeviceContext *pd3dDeviceContext;
};
CLContext g_clContext;
#include "../common/clDeviceContext.h"
CLD3D11DeviceContext g_clDeviceContext;
OpenSubdiv::Osd::CLComputeController * g_clComputeController = NULL;
#endif
@ -72,10 +61,9 @@ OpenSubdiv::Osd::CpuComputeController * g_cpuComputeController = NULL;
#include <osd/cudaComputeContext.h>
#include <osd/cudaComputeController.h>
#include <cuda_runtime_api.h>
#include <cuda_d3d11_interop.h>
#include "../common/cudaDeviceContext.h"
bool g_cudaInitialized = false;
CudaDeviceContext g_cudaDeviceContext;
OpenSubdiv::Osd::CudaComputeController * g_cudaComputeController = NULL;
#endif
@ -385,18 +373,19 @@ createOsdMesh(ShapeDesc const & shapeDesc, int level, int kernel, Scheme scheme=
} else if(kernel == kCL) {
if (not g_clComputeController) {
g_clComputeController = new OpenSubdiv::Osd::CLComputeController(
g_clContext.clContext, g_clContext.clQueue);
g_clDeviceContext.GetContext(),
g_clDeviceContext.GetCommandQueue());
}
g_mesh = new OpenSubdiv::Osd::Mesh<OpenSubdiv::Osd::CLD3D11VertexBuffer,
OpenSubdiv::Osd::CLComputeController,
OpenSubdiv::Osd::D3D11DrawContext,
CLContext>(
CLD3D11DeviceContext>(
g_clComputeController,
refiner,
numVertexElements,
numVaryingElements,
level, bits,
&g_clContext);
&g_clDeviceContext);
#endif
#ifdef OPENSUBDIV_HAS_CUDA
} else if (g_kernel == kCUDA) {
@ -1019,12 +1008,10 @@ quit() {
#ifdef OPENSUBDIV_HAS_OPENCL
delete g_clComputeController;
uninitCL(g_clContext.clContext, g_clContext.clQueue);
#endif
#ifdef OPENSUBDIV_HAS_CUDA
delete g_cudaComputeController;
cudaDeviceReset();
#endif
delete g_d3d11ComputeController;
@ -1061,18 +1048,19 @@ callbackKernel(int k) {
g_kernel = k;
#ifdef OPENSUBDIV_HAS_OPENCL
if (g_kernel == kCL and g_clContext.clContext == NULL) {
if (initCL(&g_clContext.clContext, &g_clContext.clQueue) == false) {
if (g_kernel == kCL and (not g_clDeviceContext.IsInitialized())) {
if (g_clDeviceContext.Initialize(g_pd3dDeviceContext) == false) {
printf("Error in initializing OpenCL\n");
exit(1);
}
g_clContext.pd3dDeviceContext = g_pd3dDeviceContext;
}
#endif
#ifdef OPENSUBDIV_HAS_CUDA
if (g_kernel == kCUDA and g_cudaInitialized == false) {
g_cudaInitialized = true;
cudaD3D11SetDirect3DDevice( g_pd3dDevice );
if (g_kernel == kCUDA and (not g_cudaDeviceContext.IsInitialized())) {
if (g_cudaDeviceContext.Initialize(g_pd3dDevice) == false) {
printf("Error in initializing Cuda\n");
exit(1);
}
}
#endif
@ -1184,7 +1172,7 @@ initHUD() {
g_hud->AddPullDownButton(compute_pulldown, "CUDA", kCUDA);
#endif
#ifdef OPENSUBDIV_HAS_OPENCL
if (HAS_CL_VERSION_1_1()) {
if (CLDeviceContext::HAS_CL_VERSION_1_1()) {
g_hud->AddPullDownButton(compute_pulldown, "OpenCL", kCL);
}
#endif

17
opensubdiv/osd/CMakeLists.txt
View File

@ -322,15 +322,14 @@ if ( OPENCL_FOUND )
list(APPEND PUBLIC_HEADER_FILES clGLVertexBuffer.h)
endif()
# OpenCL D3D11 interop needs work...
#if ( DXSDK_FOUND )
# list(APPEND GPU_SOURCE_FILES
# clD3D11VertexBuffer.cpp
# )
# list(APPEND PUBLIC_HEADER_FILES
# clD3D11VertexBuffer.h)
# )
#endif()
if ( DXSDK_FOUND )
list(APPEND GPU_SOURCE_FILES
clD3D11VertexBuffer.cpp
)
list(APPEND PUBLIC_HEADER_FILES
clD3D11VertexBuffer.h
)
endif()
endif()
list(APPEND DOXY_HEADER_FILES ${OPENCL_PUBLIC_HEADERS})

87
opensubdiv/osd/clD3D11VertexBuffer.cpp Normal file → Executable file
View File

@ -35,6 +35,51 @@ namespace OPENSUBDIV_VERSION {
namespace Osd {
static clCreateFromD3D11BufferKHR_fn clCreateFromD3D11Buffer = NULL;
static clEnqueueAcquireD3D11ObjectsKHR_fn clEnqueueAcquireD3D11Objects = NULL;
static clEnqueueReleaseD3D11ObjectsKHR_fn clEnqueueReleaseD3D11Objects = NULL;
// XXX: clGetExtensionFunctionAddress is marked as deprecated and
// clGetExtensionFunctionAddressForPlatform should be used,
// however it requires OpenCL 1.2.
// until we bump the requirement to 1.2, mute the deprecated warning.
#if defined(_MSC_VER)
#pragma warning(disable: 4996)
#endif
static void resolveInteropFunctions() {
if (not clCreateFromD3D11Buffer) {
clCreateFromD3D11Buffer =
(clCreateFromD3D11BufferKHR_fn)
clGetExtensionFunctionAddress("clCreateFromD3D11BufferKHR");
}
if (not clCreateFromD3D11Buffer) {
clCreateFromD3D11Buffer =
(clCreateFromD3D11BufferKHR_fn)
clGetExtensionFunctionAddress("clCreateFromD3D11BufferNV");
}
if (not clEnqueueAcquireD3D11Objects) {
clEnqueueAcquireD3D11Objects = (clEnqueueAcquireD3D11ObjectsKHR_fn)
clGetExtensionFunctionAddress("clEnqueueAcquireD3D11ObjectsKHR");
}
if (not clEnqueueAcquireD3D11Objects) {
clEnqueueAcquireD3D11Objects = (clEnqueueAcquireD3D11ObjectsKHR_fn)
clGetExtensionFunctionAddress("clEnqueueAcquireD3D11ObjectsNV");
}
if (not clEnqueueReleaseD3D11Objects) {
clEnqueueReleaseD3D11Objects = (clEnqueueReleaseD3D11ObjectsKHR_fn)
clGetExtensionFunctionAddress("clEnqueueReleaseD3D11ObjectsKHR");
}
if (not clEnqueueReleaseD3D11Objects) {
clEnqueueReleaseD3D11Objects = (clEnqueueReleaseD3D11ObjectsKHR_fn)
clGetExtensionFunctionAddress("clEnqueueReleaseD3D11ObjectsNV");
}
}
CLD3D11VertexBuffer::CLD3D11VertexBuffer(int numElements, int numVertices)
: _numElements(numElements), _numVertices(numVertices),
_d3d11Buffer(NULL), _clMemory(NULL), _clQueue(NULL), _clMapped(false) {
@ -73,6 +118,18 @@ CLD3D11VertexBuffer::UpdateData(const float *src, int startVertex,
clEnqueueWriteBuffer(queue, _clMemory, true, offset, size, src, 0, NULL, NULL);
}
int
CLD3D11VertexBuffer::GetNumElements() const {
return _numElements;
}
int
CLD3D11VertexBuffer::GetNumVertices() const {
return _numVertices;
}
cl_mem
CLD3D11VertexBuffer::BindCLBuffer(cl_command_queue queue) {
@ -105,9 +162,17 @@ CLD3D11VertexBuffer::allocate(cl_context clContext, ID3D11Device *device) {
return false;
}
if (not clCreateFromD3D11Buffer) {
resolveInteropFunctions();
if (not clCreateFromD3D11Buffer) {
return false;
}
}
// register d3d11buffer as cl memory
cl_int err;
_clMemory = clCreateFromD3D11BufferKHR(clContext, CL_MEM_READ_WRITE, _d3d11Buffer, &err);
_clMemory = clCreateFromD3D11Buffer(clContext, CL_MEM_READ_WRITE, _d3d11Buffer, &err);
if (err != CL_SUCCESS) return false;
return true;
@ -118,15 +183,29 @@ CLD3D11VertexBuffer::map(cl_command_queue queue) {
if (_clMapped) return;
_clQueue = queue;
clEnqueueAcquireD3D11ObjectsKHR(queue, 1, &_clMemory, 0, 0, 0);
if (not clEnqueueAcquireD3D11Objects) {
resolveInteropFunctions();
if (not clEnqueueAcquireD3D11Objects) {
return;
}
}
clEnqueueAcquireD3D11Objects(queue, 1, &_clMemory, 0, 0, 0);
_clMapped = true;
}
void
CLD3D11VertexBuffer::unmap() {
if (not _clMapped) return;
clEnqueueReleaseD3D11ObjectsKHR(_clQueue, 1, &_clMemory, 0, 0, 0);
if (not clEnqueueReleaseD3D11Objects) {
resolveInteropFunctions();
if (not clEnqueueReleaseD3D11Objects) {
return;
}
}
clEnqueueReleaseD3D11Objects(_clQueue, 1, &_clMemory, 0, 0, 0);
_clMapped = false;
}