sandey/blender
1
0
Fork 0
forked from bartvdbraak/blender
Code Pull Requests Activity
6b51c27414
blender/intern/cycles/util/util_cuda.cpp

496 lines
15 KiB
C++
Raw Normal View History

Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
/*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* Copyright 2011-2013 Blender Foundation
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* http://www.apache.org/licenses/LICENSE-2.0
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
*/
Cycles: prepare to make CUDA 5.0 the official version we use * Add CUDA compiler version detection to cmake/scons/runtime * Remove noinline in kernel_shader.h and reenable --use_fast_math if CUDA 5.x is used, these were workarounds for CUDA 4.2 bugs * Change max number of registers to 32 for sm 2.x (based on performance tests from Martijn Berger and confirmed here), and also for NVidia OpenCL. Overall it seems that with these changes and the latest CUDA 5.0 download, that performance is as good as or better than the 2.67b release with the scenes and graphics cards I tested.
2013-06-19 17:54:23 +00:00
#include <iostream>
Cycles: * OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU device is used, easy to enable in the code for debugging. * OpenCL kernel binaries are now cached for faster startup after the first time compiling. * CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit is installed. This means that even if the build does not have CUDA enabled, it's still possible to use it as long as you install the toolkit.
2011-09-09 12:04:39 +00:00
#include <stdlib.h>
Cycles: suppress path to nvcc appearing in the console in cases cuda toolkit is installed to different place than /usr/local/cuda (i.e. happens when using cuda toolkit from repository)
2012-10-18 13:48:02 +00:00
#include <stdio.h>
Cycles: * OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU device is used, easy to enable in the code for debugging. * OpenCL kernel binaries are now cached for faster startup after the first time compiling. * CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit is installed. This means that even if the build does not have CUDA enabled, it's still possible to use it as long as you install the toolkit.
2011-09-09 12:04:39 +00:00
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
#include "util_cuda.h"
#include "util_debug.h"
#include "util_dynlib.h"
Cycles: some warning fixes, cpu device task tweaks, avoid unnecessary tonemap in non-viewport render, and some utility functions.
2011-09-08 18:58:07 +00:00
#include "util_path.h"
#include "util_string.h"
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
Cycles: prepare to make CUDA 5.0 the official version we use * Add CUDA compiler version detection to cmake/scons/runtime * Remove noinline in kernel_shader.h and reenable --use_fast_math if CUDA 5.x is used, these were workarounds for CUDA 4.2 bugs * Change max number of registers to 32 for sm 2.x (based on performance tests from Martijn Berger and confirmed here), and also for NVidia OpenCL. Overall it seems that with these changes and the latest CUDA 5.0 download, that performance is as good as or better than the 2.67b release with the scenes and graphics cards I tested.
2013-06-19 17:54:23 +00:00
#ifdef _WIN32
#define popen _popen
#define pclose _pclose
#endif
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
/* function defininitions */
tcuInit *cuInit;
tcuDriverGetVersion *cuDriverGetVersion;
tcuDeviceGet *cuDeviceGet;
tcuDeviceGetCount *cuDeviceGetCount;
tcuDeviceGetName *cuDeviceGetName;
tcuDeviceComputeCapability *cuDeviceComputeCapability;
tcuDeviceTotalMem *cuDeviceTotalMem;
tcuDeviceGetProperties *cuDeviceGetProperties;
tcuDeviceGetAttribute *cuDeviceGetAttribute;
tcuCtxCreate *cuCtxCreate;
tcuCtxDestroy *cuCtxDestroy;
tcuCtxAttach *cuCtxAttach;
tcuCtxDetach *cuCtxDetach;
tcuCtxPushCurrent *cuCtxPushCurrent;
tcuCtxPopCurrent *cuCtxPopCurrent;
tcuCtxGetDevice *cuCtxGetDevice;
tcuCtxSynchronize *cuCtxSynchronize;
tcuModuleLoad *cuModuleLoad;
tcuModuleLoadData *cuModuleLoadData;
tcuModuleLoadDataEx *cuModuleLoadDataEx;
tcuModuleLoadFatBinary *cuModuleLoadFatBinary;
tcuModuleUnload *cuModuleUnload;
tcuModuleGetFunction *cuModuleGetFunction;
tcuModuleGetGlobal *cuModuleGetGlobal;
tcuModuleGetTexRef *cuModuleGetTexRef;
tcuModuleGetSurfRef *cuModuleGetSurfRef;
tcuMemGetInfo *cuMemGetInfo;
tcuMemAlloc *cuMemAlloc;
tcuMemAllocPitch *cuMemAllocPitch;
tcuMemFree *cuMemFree;
tcuMemGetAddressRange *cuMemGetAddressRange;
tcuMemAllocHost *cuMemAllocHost;
tcuMemFreeHost *cuMemFreeHost;
tcuMemHostAlloc *cuMemHostAlloc;
tcuMemHostGetDevicePointer *cuMemHostGetDevicePointer;
tcuMemHostGetFlags *cuMemHostGetFlags;
tcuMemcpyHtoD *cuMemcpyHtoD;
tcuMemcpyDtoH *cuMemcpyDtoH;
tcuMemcpyDtoD *cuMemcpyDtoD;
tcuMemcpyDtoA *cuMemcpyDtoA;
tcuMemcpyAtoD *cuMemcpyAtoD;
tcuMemcpyHtoA *cuMemcpyHtoA;
tcuMemcpyAtoH *cuMemcpyAtoH;
tcuMemcpyAtoA *cuMemcpyAtoA;
tcuMemcpy2D *cuMemcpy2D;
tcuMemcpy2DUnaligned *cuMemcpy2DUnaligned;
tcuMemcpy3D *cuMemcpy3D;
tcuMemcpyHtoDAsync *cuMemcpyHtoDAsync;
tcuMemcpyDtoHAsync *cuMemcpyDtoHAsync;
tcuMemcpyDtoDAsync *cuMemcpyDtoDAsync;
tcuMemcpyHtoAAsync *cuMemcpyHtoAAsync;
tcuMemcpyAtoHAsync *cuMemcpyAtoHAsync;
tcuMemcpy2DAsync *cuMemcpy2DAsync;
tcuMemcpy3DAsync *cuMemcpy3DAsync;
tcuMemsetD8 *cuMemsetD8;
tcuMemsetD16 *cuMemsetD16;
tcuMemsetD32 *cuMemsetD32;
tcuMemsetD2D8 *cuMemsetD2D8;
tcuMemsetD2D16 *cuMemsetD2D16;
tcuMemsetD2D32 *cuMemsetD2D32;
tcuFuncSetBlockShape *cuFuncSetBlockShape;
tcuFuncSetSharedSize *cuFuncSetSharedSize;
tcuFuncGetAttribute *cuFuncGetAttribute;
tcuFuncSetCacheConfig *cuFuncSetCacheConfig;
tcuArrayCreate *cuArrayCreate;
tcuArrayGetDescriptor *cuArrayGetDescriptor;
tcuArrayDestroy *cuArrayDestroy;
tcuArray3DCreate *cuArray3DCreate;
tcuArray3DGetDescriptor *cuArray3DGetDescriptor;
tcuTexRefCreate *cuTexRefCreate;
tcuTexRefDestroy *cuTexRefDestroy;
tcuTexRefSetArray *cuTexRefSetArray;
tcuTexRefSetAddress *cuTexRefSetAddress;
tcuTexRefSetAddress2D *cuTexRefSetAddress2D;
tcuTexRefSetFormat *cuTexRefSetFormat;
tcuTexRefSetAddressMode *cuTexRefSetAddressMode;
tcuTexRefSetFilterMode *cuTexRefSetFilterMode;
tcuTexRefSetFlags *cuTexRefSetFlags;
tcuTexRefGetAddress *cuTexRefGetAddress;
tcuTexRefGetArray *cuTexRefGetArray;
tcuTexRefGetAddressMode *cuTexRefGetAddressMode;
tcuTexRefGetFilterMode *cuTexRefGetFilterMode;
tcuTexRefGetFormat *cuTexRefGetFormat;
tcuTexRefGetFlags *cuTexRefGetFlags;
tcuSurfRefSetArray *cuSurfRefSetArray;
tcuSurfRefGetArray *cuSurfRefGetArray;
tcuParamSetSize *cuParamSetSize;
tcuParamSeti *cuParamSeti;
tcuParamSetf *cuParamSetf;
tcuParamSetv *cuParamSetv;
tcuParamSetTexRef *cuParamSetTexRef;
tcuLaunch *cuLaunch;
tcuLaunchGrid *cuLaunchGrid;
tcuLaunchGridAsync *cuLaunchGridAsync;
tcuEventCreate *cuEventCreate;
tcuEventRecord *cuEventRecord;
tcuEventQuery *cuEventQuery;
tcuEventSynchronize *cuEventSynchronize;
tcuEventDestroy *cuEventDestroy;
tcuEventElapsedTime *cuEventElapsedTime;
tcuStreamCreate *cuStreamCreate;
tcuStreamQuery *cuStreamQuery;
tcuStreamSynchronize *cuStreamSynchronize;
tcuStreamDestroy *cuStreamDestroy;
tcuGraphicsUnregisterResource *cuGraphicsUnregisterResource;
tcuGraphicsSubResourceGetMappedArray *cuGraphicsSubResourceGetMappedArray;
tcuGraphicsResourceGetMappedPointer *cuGraphicsResourceGetMappedPointer;
tcuGraphicsResourceSetMapFlags *cuGraphicsResourceSetMapFlags;
tcuGraphicsMapResources *cuGraphicsMapResources;
tcuGraphicsUnmapResources *cuGraphicsUnmapResources;
tcuGetExportTable *cuGetExportTable;
tcuCtxSetLimit *cuCtxSetLimit;
tcuCtxGetLimit *cuCtxGetLimit;
tcuGLCtxCreate *cuGLCtxCreate;
tcuGraphicsGLRegisterBuffer *cuGraphicsGLRegisterBuffer;
tcuGraphicsGLRegisterImage *cuGraphicsGLRegisterImage;
tcuCtxSetCurrent *cuCtxSetCurrent;
CCL_NAMESPACE_BEGIN
/* utility macros */
Hack-fix for crash-by-assertion on mingw builds on startup if the CUDA driver on the computer is too old (and probably can't be upgraded to a version that does) to support the cuDriverGetVersion method.
2011-11-12 03:59:45 +00:00
#define CUDA_LIBRARY_FIND_CHECKED(name) \
name = (t##name*)dynamic_library_find(lib, #name);
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
#define CUDA_LIBRARY_FIND(name) \
name = (t##name*)dynamic_library_find(lib, #name); \
assert(name);
#define CUDA_LIBRARY_FIND_V2(name) \
name = (t##name*)dynamic_library_find(lib, #name "_v2"); \
assert(name);
/* initialization function */
bool cuLibraryInit()
{
static bool initialized = false;
static bool result = false;
if(initialized)
return result;
initialized = true;
/* library paths */
#ifdef _WIN32
/* expected in c:/windows/system or similar, no path needed */
const char *path = "nvcuda.dll";
#elif defined(__APPLE__)
/* default installation path */
const char *path = "/usr/local/cuda/lib/libcuda.dylib";
#else
const char *path = "libcuda.so";
#endif
/* load library */
DynamicLibrary *lib = dynamic_library_open(path);
if(lib == NULL)
return false;
/* detect driver version */
int driver_version = 1000;
Hack-fix for crash-by-assertion on mingw builds on startup if the CUDA driver on the computer is too old (and probably can't be upgraded to a version that does) to support the cuDriverGetVersion method.
2011-11-12 03:59:45 +00:00
CUDA_LIBRARY_FIND_CHECKED(cuDriverGetVersion);
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
if(cuDriverGetVersion)
cuDriverGetVersion(&driver_version);
/* we require version 4.0 */
if(driver_version < 4000)
return false;
/* fetch all function pointers */
CUDA_LIBRARY_FIND(cuInit);
CUDA_LIBRARY_FIND(cuDeviceGet);
CUDA_LIBRARY_FIND(cuDeviceGetCount);
CUDA_LIBRARY_FIND(cuDeviceGetName);
CUDA_LIBRARY_FIND(cuDeviceComputeCapability);
CUDA_LIBRARY_FIND(cuDeviceTotalMem);
CUDA_LIBRARY_FIND(cuDeviceGetProperties);
CUDA_LIBRARY_FIND(cuDeviceGetAttribute);
CUDA_LIBRARY_FIND(cuCtxCreate);
CUDA_LIBRARY_FIND(cuCtxDestroy);
CUDA_LIBRARY_FIND(cuCtxAttach);
CUDA_LIBRARY_FIND(cuCtxDetach);
CUDA_LIBRARY_FIND(cuCtxPushCurrent);
CUDA_LIBRARY_FIND(cuCtxPopCurrent);
CUDA_LIBRARY_FIND(cuCtxGetDevice);
CUDA_LIBRARY_FIND(cuCtxSynchronize);
CUDA_LIBRARY_FIND(cuModuleLoad);
CUDA_LIBRARY_FIND(cuModuleLoadData);
CUDA_LIBRARY_FIND(cuModuleUnload);
CUDA_LIBRARY_FIND(cuModuleGetFunction);
CUDA_LIBRARY_FIND(cuModuleGetGlobal);
CUDA_LIBRARY_FIND(cuModuleGetTexRef);
CUDA_LIBRARY_FIND(cuMemGetInfo);
CUDA_LIBRARY_FIND(cuMemAlloc);
CUDA_LIBRARY_FIND(cuMemAllocPitch);
CUDA_LIBRARY_FIND(cuMemFree);
CUDA_LIBRARY_FIND(cuMemGetAddressRange);
CUDA_LIBRARY_FIND(cuMemAllocHost);
CUDA_LIBRARY_FIND(cuMemFreeHost);
CUDA_LIBRARY_FIND(cuMemHostAlloc);
CUDA_LIBRARY_FIND(cuMemHostGetDevicePointer);
CUDA_LIBRARY_FIND(cuMemcpyHtoD);
CUDA_LIBRARY_FIND(cuMemcpyDtoH);
CUDA_LIBRARY_FIND(cuMemcpyDtoD);
CUDA_LIBRARY_FIND(cuMemcpyDtoA);
CUDA_LIBRARY_FIND(cuMemcpyAtoD);
CUDA_LIBRARY_FIND(cuMemcpyHtoA);
CUDA_LIBRARY_FIND(cuMemcpyAtoH);
CUDA_LIBRARY_FIND(cuMemcpyAtoA);
CUDA_LIBRARY_FIND(cuMemcpy2D);
CUDA_LIBRARY_FIND(cuMemcpy2DUnaligned);
CUDA_LIBRARY_FIND(cuMemcpy3D);
CUDA_LIBRARY_FIND(cuMemcpyHtoDAsync);
CUDA_LIBRARY_FIND(cuMemcpyDtoHAsync);
CUDA_LIBRARY_FIND(cuMemcpyHtoAAsync);
CUDA_LIBRARY_FIND(cuMemcpyAtoHAsync);
CUDA_LIBRARY_FIND(cuMemcpy2DAsync);
CUDA_LIBRARY_FIND(cuMemcpy3DAsync);
CUDA_LIBRARY_FIND(cuMemsetD8);
CUDA_LIBRARY_FIND(cuMemsetD16);
CUDA_LIBRARY_FIND(cuMemsetD32);
CUDA_LIBRARY_FIND(cuMemsetD2D8);
CUDA_LIBRARY_FIND(cuMemsetD2D16);
CUDA_LIBRARY_FIND(cuMemsetD2D32);
CUDA_LIBRARY_FIND(cuFuncSetBlockShape);
CUDA_LIBRARY_FIND(cuFuncSetSharedSize);
CUDA_LIBRARY_FIND(cuFuncGetAttribute);
CUDA_LIBRARY_FIND(cuArrayCreate);
CUDA_LIBRARY_FIND(cuArrayGetDescriptor);
CUDA_LIBRARY_FIND(cuArrayDestroy);
CUDA_LIBRARY_FIND(cuArray3DCreate);
CUDA_LIBRARY_FIND(cuArray3DGetDescriptor);
CUDA_LIBRARY_FIND(cuTexRefCreate);
CUDA_LIBRARY_FIND(cuTexRefDestroy);
CUDA_LIBRARY_FIND(cuTexRefSetArray);
CUDA_LIBRARY_FIND(cuTexRefSetAddress);
CUDA_LIBRARY_FIND(cuTexRefSetAddress2D);
CUDA_LIBRARY_FIND(cuTexRefSetFormat);
CUDA_LIBRARY_FIND(cuTexRefSetAddressMode);
CUDA_LIBRARY_FIND(cuTexRefSetFilterMode);
CUDA_LIBRARY_FIND(cuTexRefSetFlags);
CUDA_LIBRARY_FIND(cuTexRefGetAddress);
CUDA_LIBRARY_FIND(cuTexRefGetArray);
CUDA_LIBRARY_FIND(cuTexRefGetAddressMode);
CUDA_LIBRARY_FIND(cuTexRefGetFilterMode);
CUDA_LIBRARY_FIND(cuTexRefGetFormat);
CUDA_LIBRARY_FIND(cuTexRefGetFlags);
CUDA_LIBRARY_FIND(cuParamSetSize);
CUDA_LIBRARY_FIND(cuParamSeti);
CUDA_LIBRARY_FIND(cuParamSetf);
CUDA_LIBRARY_FIND(cuParamSetv);
CUDA_LIBRARY_FIND(cuParamSetTexRef);
CUDA_LIBRARY_FIND(cuLaunch);
CUDA_LIBRARY_FIND(cuLaunchGrid);
CUDA_LIBRARY_FIND(cuLaunchGridAsync);
CUDA_LIBRARY_FIND(cuEventCreate);
CUDA_LIBRARY_FIND(cuEventRecord);
CUDA_LIBRARY_FIND(cuEventQuery);
CUDA_LIBRARY_FIND(cuEventSynchronize);
CUDA_LIBRARY_FIND(cuEventDestroy);
CUDA_LIBRARY_FIND(cuEventElapsedTime);
CUDA_LIBRARY_FIND(cuStreamCreate);
CUDA_LIBRARY_FIND(cuStreamQuery);
CUDA_LIBRARY_FIND(cuStreamSynchronize);
CUDA_LIBRARY_FIND(cuStreamDestroy);
/* cuda 2.1 */
CUDA_LIBRARY_FIND(cuModuleLoadDataEx);
CUDA_LIBRARY_FIND(cuModuleLoadFatBinary);
CUDA_LIBRARY_FIND(cuGLCtxCreate);
CUDA_LIBRARY_FIND(cuGraphicsGLRegisterBuffer);
CUDA_LIBRARY_FIND(cuGraphicsGLRegisterImage);
/* cuda 2.3 */
CUDA_LIBRARY_FIND(cuMemHostGetFlags);
CUDA_LIBRARY_FIND(cuGraphicsGLRegisterBuffer);
CUDA_LIBRARY_FIND(cuGraphicsGLRegisterImage);
/* cuda 3.0 */
CUDA_LIBRARY_FIND(cuMemcpyDtoDAsync);
CUDA_LIBRARY_FIND(cuFuncSetCacheConfig);
CUDA_LIBRARY_FIND(cuGraphicsUnregisterResource);
CUDA_LIBRARY_FIND(cuGraphicsSubResourceGetMappedArray);
CUDA_LIBRARY_FIND(cuGraphicsResourceGetMappedPointer);
CUDA_LIBRARY_FIND(cuGraphicsResourceSetMapFlags);
CUDA_LIBRARY_FIND(cuGraphicsMapResources);
CUDA_LIBRARY_FIND(cuGraphicsUnmapResources);
CUDA_LIBRARY_FIND(cuGetExportTable);
/* cuda 3.1 */
CUDA_LIBRARY_FIND(cuModuleGetSurfRef);
CUDA_LIBRARY_FIND(cuSurfRefSetArray);
CUDA_LIBRARY_FIND(cuSurfRefGetArray);
CUDA_LIBRARY_FIND(cuCtxSetLimit);
CUDA_LIBRARY_FIND(cuCtxGetLimit);
/* functions which changed 3.1 -> 3.2 for 64 bit stuff, the cuda library
style cleanup: block comments
2012-06-09 17:22:52 +00:00
* has both the old ones for compatibility and new ones with _v2 postfix,
* we load the _v2 ones here. */
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
CUDA_LIBRARY_FIND_V2(cuDeviceTotalMem);
CUDA_LIBRARY_FIND_V2(cuCtxCreate);
CUDA_LIBRARY_FIND_V2(cuModuleGetGlobal);
CUDA_LIBRARY_FIND_V2(cuMemGetInfo);
CUDA_LIBRARY_FIND_V2(cuMemAlloc);
CUDA_LIBRARY_FIND_V2(cuMemAllocPitch);
CUDA_LIBRARY_FIND_V2(cuMemFree);
CUDA_LIBRARY_FIND_V2(cuMemGetAddressRange);
CUDA_LIBRARY_FIND_V2(cuMemAllocHost);
CUDA_LIBRARY_FIND_V2(cuMemHostGetDevicePointer);
CUDA_LIBRARY_FIND_V2(cuMemcpyHtoD);
CUDA_LIBRARY_FIND_V2(cuMemcpyDtoH);
CUDA_LIBRARY_FIND_V2(cuMemcpyDtoD);
CUDA_LIBRARY_FIND_V2(cuMemcpyDtoA);
CUDA_LIBRARY_FIND_V2(cuMemcpyAtoD);
CUDA_LIBRARY_FIND_V2(cuMemcpyHtoA);
CUDA_LIBRARY_FIND_V2(cuMemcpyAtoH);
CUDA_LIBRARY_FIND_V2(cuMemcpyAtoA);
CUDA_LIBRARY_FIND_V2(cuMemcpyHtoAAsync);
CUDA_LIBRARY_FIND_V2(cuMemcpyAtoHAsync);
CUDA_LIBRARY_FIND_V2(cuMemcpy2D);
CUDA_LIBRARY_FIND_V2(cuMemcpy2DUnaligned);
CUDA_LIBRARY_FIND_V2(cuMemcpy3D);
CUDA_LIBRARY_FIND_V2(cuMemcpyHtoDAsync);
CUDA_LIBRARY_FIND_V2(cuMemcpyDtoHAsync);
CUDA_LIBRARY_FIND_V2(cuMemcpyDtoDAsync);
CUDA_LIBRARY_FIND_V2(cuMemcpy2DAsync);
CUDA_LIBRARY_FIND_V2(cuMemcpy3DAsync);
CUDA_LIBRARY_FIND_V2(cuMemsetD8);
CUDA_LIBRARY_FIND_V2(cuMemsetD16);
CUDA_LIBRARY_FIND_V2(cuMemsetD32);
CUDA_LIBRARY_FIND_V2(cuMemsetD2D8);
CUDA_LIBRARY_FIND_V2(cuMemsetD2D16);
CUDA_LIBRARY_FIND_V2(cuMemsetD2D32);
CUDA_LIBRARY_FIND_V2(cuArrayCreate);
CUDA_LIBRARY_FIND_V2(cuArrayGetDescriptor);
CUDA_LIBRARY_FIND_V2(cuArray3DCreate);
CUDA_LIBRARY_FIND_V2(cuArray3DGetDescriptor);
CUDA_LIBRARY_FIND_V2(cuTexRefSetAddress);
CUDA_LIBRARY_FIND_V2(cuTexRefSetAddress2D);
CUDA_LIBRARY_FIND_V2(cuTexRefGetAddress);
CUDA_LIBRARY_FIND_V2(cuGraphicsResourceGetMappedPointer);
CUDA_LIBRARY_FIND_V2(cuGLCtxCreate);
/* cuda 4.0 */
CUDA_LIBRARY_FIND(cuCtxSetCurrent);
Remove usage WITH_CYCLES_CUDA_BINARIES in code, use check for precompiled cubins instead, Logic here is following now: - If there're precompiled cubins, assume CUDA compute is available, otherwise - If cuda toolkit found, assume CUDA compute is available - In all other cases CUDA compute is not available For windows there're still check for only precompiled binaries, no runtime compilation is allowed. Ended up with such decision after discussion with Brecht. The thing is, if we'll support runtime compilation on windows we'll end up having lots of reports about different aspects of something doesn't work (you need particular toolkit version, msvc installed, environment variables set properly and so) and giving feedback on such reports will waste time.
2013-01-14 17:30:33 +00:00
if(cuHavePrecompiledKernels())
result = true;
For non-windows systems, check for CUDA compiler during runtime
2013-01-14 19:33:16 +00:00
#ifndef _WIN32
Remove usage WITH_CYCLES_CUDA_BINARIES in code, use check for precompiled cubins instead, Logic here is following now: - If there're precompiled cubins, assume CUDA compute is available, otherwise - If cuda toolkit found, assume CUDA compute is available - In all other cases CUDA compute is not available For windows there're still check for only precompiled binaries, no runtime compilation is allowed. Ended up with such decision after discussion with Brecht. The thing is, if we'll support runtime compilation on windows we'll end up having lots of reports about different aspects of something doesn't work (you need particular toolkit version, msvc installed, environment variables set properly and so) and giving feedback on such reports will waste time.
2013-01-14 17:30:33 +00:00
else if(cuCompilerPath() != "")
result = true;
Cycles: * Add back option to bundle CUDA kernel binaries with builds. * Disable runtime CUDA kernel compilation on Windows, couldn't get this working, since it seems to depend on visual studio being installed, even though for this particular case it shouldn't be needed. CMake only at the moment. * Runtime compilation on linux/mac should now work if nvcc is not installed in the default location, but available in PATH.
2011-11-10 12:52:17 +00:00
#endif
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
return result;
}
Remove usage WITH_CYCLES_CUDA_BINARIES in code, use check for precompiled cubins instead, Logic here is following now: - If there're precompiled cubins, assume CUDA compute is available, otherwise - If cuda toolkit found, assume CUDA compute is available - In all other cases CUDA compute is not available For windows there're still check for only precompiled binaries, no runtime compilation is allowed. Ended up with such decision after discussion with Brecht. The thing is, if we'll support runtime compilation on windows we'll end up having lots of reports about different aspects of something doesn't work (you need particular toolkit version, msvc installed, environment variables set properly and so) and giving feedback on such reports will waste time.
2013-01-14 17:30:33 +00:00
bool cuHavePrecompiledKernels()
{
string cubins_path = path_get("lib");
return path_exists(cubins_path);
}
Cycles: some warning fixes, cpu device task tweaks, avoid unnecessary tonemap in non-viewport render, and some utility functions.
2011-09-08 18:58:07 +00:00
string cuCompilerPath()
{
#ifdef _WIN32
Cycles: add some extra CUDA nvcc paths for runtime compile, might help for #33622.
2012-12-21 10:27:35 +00:00
const char *defaultpaths[] = {"C:/CUDA/bin", NULL};
Cycles: * OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU device is used, easy to enable in the code for debugging. * OpenCL kernel binaries are now cached for faster startup after the first time compiling. * CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit is installed. This means that even if the build does not have CUDA enabled, it's still possible to use it as long as you install the toolkit.
2011-09-09 12:04:39 +00:00
const char *executable = "nvcc.exe";
Cycles: some warning fixes, cpu device task tweaks, avoid unnecessary tonemap in non-viewport render, and some utility functions.
2011-09-08 18:58:07 +00:00
#else
Cycles: prepare to make CUDA 5.0 the official version we use * Add CUDA compiler version detection to cmake/scons/runtime * Remove noinline in kernel_shader.h and reenable --use_fast_math if CUDA 5.x is used, these were workarounds for CUDA 4.2 bugs * Change max number of registers to 32 for sm 2.x (based on performance tests from Martijn Berger and confirmed here), and also for NVidia OpenCL. Overall it seems that with these changes and the latest CUDA 5.0 download, that performance is as good as or better than the 2.67b release with the scenes and graphics cards I tested.
2013-06-19 17:54:23 +00:00
const char *defaultpaths[] = {
"/Developer/NVIDIA/CUDA-5.0/bin",
"/usr/local/cuda-5.0/bin",
"/usr/local/cuda/bin",
"/Developer/NVIDIA/CUDA-4.2/bin",
"/usr/local/cuda-4.2/bin",
"/Developer/NVIDIA/CUDA-5.5/bin",
"/usr/local/cuda-5.5/bin",
NULL};
Cycles: * OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU device is used, easy to enable in the code for debugging. * OpenCL kernel binaries are now cached for faster startup after the first time compiling. * CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit is installed. This means that even if the build does not have CUDA enabled, it's still possible to use it as long as you install the toolkit.
2011-09-09 12:04:39 +00:00
const char *executable = "nvcc";
Cycles: some warning fixes, cpu device task tweaks, avoid unnecessary tonemap in non-viewport render, and some utility functions.
2011-09-08 18:58:07 +00:00
#endif
Cycles: * OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU device is used, easy to enable in the code for debugging. * OpenCL kernel binaries are now cached for faster startup after the first time compiling. * CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit is installed. This means that even if the build does not have CUDA enabled, it's still possible to use it as long as you install the toolkit.
2011-09-09 12:04:39 +00:00
const char *binpath = getenv("CUDA_BIN_PATH");
string nvcc;
Cycles: add some extra CUDA nvcc paths for runtime compile, might help for #33622.
2012-12-21 10:27:35 +00:00
if(binpath) {
Cycles: * OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU device is used, easy to enable in the code for debugging. * OpenCL kernel binaries are now cached for faster startup after the first time compiling. * CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit is installed. This means that even if the build does not have CUDA enabled, it's still possible to use it as long as you install the toolkit.
2011-09-09 12:04:39 +00:00
nvcc = path_join(binpath, executable);
Cycles: add some extra CUDA nvcc paths for runtime compile, might help for #33622.
2012-12-21 10:27:35 +00:00
if(path_exists(nvcc))
return nvcc;
}
Cycles: * OpenCL now only uses GPU/Accelerator devices, it's only confusing if CPU device is used, easy to enable in the code for debugging. * OpenCL kernel binaries are now cached for faster startup after the first time compiling. * CUDA kernels can now be compiled and cached at runtime if the CUDA toolkit is installed. This means that even if the build does not have CUDA enabled, it's still possible to use it as long as you install the toolkit.
2011-09-09 12:04:39 +00:00
Cycles: add some extra CUDA nvcc paths for runtime compile, might help for #33622.
2012-12-21 10:27:35 +00:00
for(int i = 0; defaultpaths[i]; i++) {
nvcc = path_join(defaultpaths[i], executable);
if(path_exists(nvcc))
return nvcc;
}
Cycles: * Add back option to bundle CUDA kernel binaries with builds. * Disable runtime CUDA kernel compilation on Windows, couldn't get this working, since it seems to depend on visual studio being installed, even though for this particular case it shouldn't be needed. CMake only at the moment. * Runtime compilation on linux/mac should now work if nvcc is not installed in the default location, but available in PATH.
2011-11-10 12:52:17 +00:00
#ifndef _WIN32
Cycles: suppress path to nvcc appearing in the console in cases cuda toolkit is installed to different place than /usr/local/cuda (i.e. happens when using cuda toolkit from repository)
2012-10-18 13:48:02 +00:00
{
FILE *handle = popen("which nvcc", "r");
if(handle) {
char buffer[4096] = {0};
int len = fread(buffer, 1, sizeof(buffer) - 1, handle);
buffer[len] = '\0';
pclose(handle);
if(buffer[0])
return "nvcc";
}
}
Cycles: * Add back option to bundle CUDA kernel binaries with builds. * Disable runtime CUDA kernel compilation on Windows, couldn't get this working, since it seems to depend on visual studio being installed, even though for this particular case it shouldn't be needed. CMake only at the moment. * Runtime compilation on linux/mac should now work if nvcc is not installed in the default location, but available in PATH.
2011-11-10 12:52:17 +00:00
#endif
return "";
Cycles: some warning fixes, cpu device task tweaks, avoid unnecessary tonemap in non-viewport render, and some utility functions.
2011-09-08 18:58:07 +00:00
}
Cycles: prepare to make CUDA 5.0 the official version we use * Add CUDA compiler version detection to cmake/scons/runtime * Remove noinline in kernel_shader.h and reenable --use_fast_math if CUDA 5.x is used, these were workarounds for CUDA 4.2 bugs * Change max number of registers to 32 for sm 2.x (based on performance tests from Martijn Berger and confirmed here), and also for NVidia OpenCL. Overall it seems that with these changes and the latest CUDA 5.0 download, that performance is as good as or better than the 2.67b release with the scenes and graphics cards I tested.
2013-06-19 17:54:23 +00:00
int cuCompilerVersion()
{
string path = cuCompilerPath();
if(path == "")
return 0;
/* get --version output */
FILE *pipe = popen((path + " --version").c_str(), "r");
if(!pipe) {
fprintf(stderr, "CUDA: failed to run compiler to retrieve version");
return 0;
}
char buf[128];
string output = "";
while(!feof(pipe))
if(fgets(buf, 128, pipe) != NULL)
output += buf;
pclose(pipe);
/* parse version number */
string marker = "Cuda compilation tools, release ";
size_t offset = output.find(marker);
if(offset == string::npos) {
fprintf(stderr, "CUDA: failed to find version number in:\n\n%s\n", output.c_str());
return 0;
}
string versionstr = output.substr(offset + marker.size(), string::npos);
int major, minor;
if(sscanf(versionstr.c_str(), "%d.%d", &major, &minor) < 2) {
fprintf(stderr, "CUDA: failed to parse version number from:\n\n%s\n", output.c_str());
return 0;
}
return 10*major + minor;
}
Cycles render engine, initial commit. This is the engine itself, blender modifications and build instructions will follow later. Cycles uses code from some great open source projects, many thanks them: * BVH building and traversal code from NVidia's "Understanding the Efficiency of Ray Traversal on GPUs": http://code.google.com/p/understanding-the-efficiency-of-ray-traversal-on-gpus/ * Open Shading Language for a large part of the shading system: http://code.google.com/p/openshadinglanguage/ * Blender for procedural textures and a few other nodes. * Approximate Catmull Clark subdivision from NVidia Mesh tools: http://code.google.com/p/nvidia-mesh-tools/ * Sobol direction vectors from: http://web.maths.unsw.edu.au/~fkuo/sobol/ * Film response functions from: http://www.cs.columbia.edu/CAVE/software/softlib/dorf.php
2011-04-27 11:58:34 +00:00
CCL_NAMESPACE_END
Copy Permalink