Brecht Van Lommel
1dafe759ed
This brings separate initialization for libcuda and libnvrtc, which fixes Cycles nvrtc compilation not working on build machines without CUDA hardware available. Differential Revision: https://developer.blender.org/D3045
238 lines
7.1 KiB
C++
238 lines
7.1 KiB
C++
/*
|
|
* Adopted from OpenSubdiv with the following license:
|
|
*
|
|
* Copyright 2015 Pixar
|
|
*
|
|
* Licensed under the Apache License, Version 2.0 (the "Apache License")
|
|
* with the following modification; you may not use this file except in
|
|
* compliance with the Apache License and the following modification to it:
|
|
* Section 6. Trademarks. is deleted and replaced with:
|
|
*
|
|
* 6. Trademarks. This License does not grant permission to use the trade
|
|
* names, trademarks, service marks, or product names of the Licensor
|
|
* and its affiliates, except as required to comply with Section 4(c) of
|
|
* the License and to reproduce the content of the NOTICE file.
|
|
*
|
|
* You may obtain a copy of the Apache License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the Apache License with the above modification is
|
|
* distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
|
|
* KIND, either express or implied. See the Apache License for the specific
|
|
* language governing permissions and limitations under the Apache License.
|
|
*/
|
|
|
|
#ifdef OPENSUBDIV_HAS_CUDA
|
|
|
|
#ifdef _MSC_VER
|
|
# include "iso646.h"
|
|
#endif
|
|
|
|
#include "opensubdiv_device_context_cuda.h"
|
|
|
|
#if defined(_WIN32)
|
|
# include <windows.h>
|
|
#elif defined(__APPLE__)
|
|
# include <OpenGL/OpenGL.h>
|
|
#else
|
|
# include <X11/Xlib.h>
|
|
# include <GL/glx.h>
|
|
#endif
|
|
|
|
#include <cstdio>
|
|
#include <algorithm>
|
|
#include <cuda.h>
|
|
#include <cuda_runtime_api.h>
|
|
#include <cuda_gl_interop.h>
|
|
|
|
#define message(fmt, ...)
|
|
//#define message(fmt, ...) fprintf(stderr, fmt, __VA_ARGS__)
|
|
#define error(fmt, ...) fprintf(stderr, fmt, __VA_ARGS__)
|
|
|
|
static int _GetCudaDeviceForCurrentGLContext()
|
|
{
|
|
// Find and use the CUDA device for the current GL context
|
|
unsigned int interopDeviceCount = 0;
|
|
int interopDevices[1];
|
|
cudaError_t status = cudaGLGetDevices(&interopDeviceCount, interopDevices,
|
|
1, cudaGLDeviceListCurrentFrame);
|
|
if (status == cudaErrorNoDevice or interopDeviceCount != 1) {
|
|
message("CUDA no interop devices found.\n");
|
|
return 0;
|
|
}
|
|
int device = interopDevices[0];
|
|
|
|
#if defined(_WIN32)
|
|
return device;
|
|
|
|
#elif defined(__APPLE__)
|
|
return device;
|
|
|
|
#else // X11
|
|
Display * display = glXGetCurrentDisplay();
|
|
int screen = DefaultScreen(display);
|
|
if (device != screen) {
|
|
error("The CUDA interop device (%d) does not match "
|
|
"the screen used by the current GL context (%d), "
|
|
"which may cause slow performance on systems "
|
|
"with multiple GPU devices.", device, screen);
|
|
}
|
|
message("CUDA init using device for current GL context: %d\n", device);
|
|
return device;
|
|
#endif
|
|
}
|
|
|
|
/* From "NVIDIA GPU Computing SDK 4.2/C/common/inc/cutil_inline_runtime.h": */
|
|
|
|
/* Beginning of GPU Architecture definitions */
|
|
inline int _ConvertSMVer2Cores_local(int major, int minor)
|
|
{
|
|
/* Defines for GPU Architecture types (using the SM version to determine
|
|
* the # of cores per SM
|
|
*/
|
|
typedef struct {
|
|
int SM; /* 0xMm (hexidecimal notation),
|
|
* M = SM Major version,
|
|
* and m = SM minor version
|
|
*/
|
|
int Cores;
|
|
} sSMtoCores;
|
|
|
|
sSMtoCores nGpuArchCoresPerSM[] =
|
|
{ { 0x10, 8 }, /* Tesla Generation (SM 1.0) G80 class */
|
|
{ 0x11, 8 }, /* Tesla Generation (SM 1.1) G8x class */
|
|
{ 0x12, 8 }, /* Tesla Generation (SM 1.2) G9x class */
|
|
{ 0x13, 8 }, /* Tesla Generation (SM 1.3) GT200 class */
|
|
{ 0x20, 32 }, /* Fermi Generation (SM 2.0) GF100 class */
|
|
{ 0x21, 48 }, /* Fermi Generation (SM 2.1) GF10x class */
|
|
{ 0x30, 192}, /* Fermi Generation (SM 3.0) GK10x class */
|
|
{ -1, -1 }
|
|
};
|
|
|
|
int index = 0;
|
|
while (nGpuArchCoresPerSM[index].SM != -1) {
|
|
if (nGpuArchCoresPerSM[index].SM == ((major << 4) + minor)) {
|
|
return nGpuArchCoresPerSM[index].Cores;
|
|
}
|
|
index++;
|
|
}
|
|
printf("MapSMtoCores undefined SMversion %d.%d!\n", major, minor);
|
|
return -1;
|
|
}
|
|
/* End of GPU Architecture definitions. */
|
|
|
|
/* This function returns the best GPU (with maximum GFLOPS) */
|
|
inline int cutGetMaxGflopsDeviceId()
|
|
{
|
|
int current_device = 0, sm_per_multiproc = 0;
|
|
int max_compute_perf = 0, max_perf_device = -1;
|
|
int device_count = 0, best_SM_arch = 0;
|
|
int compat_major, compat_minor;
|
|
|
|
cuDeviceGetCount(&device_count);
|
|
/* Find the best major SM Architecture GPU device. */
|
|
while (current_device < device_count) {
|
|
cuDeviceComputeCapability(&compat_major, &compat_minor, current_device);
|
|
if (compat_major > 0 && compat_major < 9999) {
|
|
best_SM_arch = std::max(best_SM_arch, compat_major);
|
|
}
|
|
current_device++;
|
|
}
|
|
|
|
/* Find the best CUDA capable GPU device. */
|
|
current_device = 0;
|
|
while (current_device < device_count) {
|
|
cuDeviceComputeCapability(&compat_major, &compat_minor, current_device);
|
|
if (compat_major == 9999 && compat_minor == 9999) {
|
|
sm_per_multiproc = 1;
|
|
} else {
|
|
sm_per_multiproc = _ConvertSMVer2Cores_local(compat_major,
|
|
compat_minor);
|
|
}
|
|
int multi_processor_count;
|
|
cuDeviceGetAttribute(&multi_processor_count,
|
|
CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
|
|
current_device);
|
|
int clock_rate;
|
|
cuDeviceGetAttribute(&clock_rate,
|
|
CU_DEVICE_ATTRIBUTE_CLOCK_RATE,
|
|
current_device);
|
|
int compute_perf = multi_processor_count * sm_per_multiproc * clock_rate;
|
|
if (compute_perf > max_compute_perf) {
|
|
/* If we find GPU with SM major > 2, search only these */
|
|
if (best_SM_arch > 2) {
|
|
/* If our device==dest_SM_arch, choose this, or else pass. */
|
|
if (compat_major == best_SM_arch) {
|
|
max_compute_perf = compute_perf;
|
|
max_perf_device = current_device;
|
|
}
|
|
} else {
|
|
max_compute_perf = compute_perf;
|
|
max_perf_device = current_device;
|
|
}
|
|
}
|
|
++current_device;
|
|
}
|
|
return max_perf_device;
|
|
}
|
|
|
|
bool CudaDeviceContext::HAS_CUDA_VERSION_4_0()
|
|
{
|
|
#ifdef OPENSUBDIV_HAS_CUDA
|
|
static bool cudaInitialized = false;
|
|
static bool cudaLoadSuccess = true;
|
|
if (!cudaInitialized) {
|
|
cudaInitialized = true;
|
|
|
|
# ifdef OPENSUBDIV_HAS_CUEW
|
|
cudaLoadSuccess = cuewInit(CUEW_INIT_CUDA) == CUEW_SUCCESS;
|
|
if (!cudaLoadSuccess) {
|
|
fprintf(stderr, "Loading CUDA failed.\n");
|
|
}
|
|
# endif
|
|
// Need to initialize CUDA here so getting device
|
|
// with the maximum FPLOS works fine.
|
|
if (cuInit(0) == CUDA_SUCCESS) {
|
|
// This is to deal with cases like NVidia Optimus,
|
|
// when there might be CUDA library installed but
|
|
// NVidia card is not being active.
|
|
if (cutGetMaxGflopsDeviceId() < 0) {
|
|
cudaLoadSuccess = false;
|
|
}
|
|
}
|
|
else {
|
|
cudaLoadSuccess = false;
|
|
}
|
|
}
|
|
return cudaLoadSuccess;
|
|
#else
|
|
return false;
|
|
#endif
|
|
}
|
|
|
|
CudaDeviceContext::CudaDeviceContext()
|
|
: _initialized(false) {
|
|
}
|
|
|
|
CudaDeviceContext::~CudaDeviceContext() {
|
|
cudaDeviceReset();
|
|
}
|
|
|
|
bool CudaDeviceContext::Initialize()
|
|
{
|
|
/* See if any cuda device is available. */
|
|
int deviceCount = 0;
|
|
cudaGetDeviceCount(&deviceCount);
|
|
message("CUDA device count: %d\n", deviceCount);
|
|
if (deviceCount <= 0) {
|
|
return false;
|
|
}
|
|
cudaGLSetGLDevice(_GetCudaDeviceForCurrentGLContext());
|
|
_initialized = true;
|
|
return true;
|
|
}
|
|
|
|
#endif /* OPENSUBDIV_HAS_CUDA */
|