gstreamer/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudanvrtc.c
Seungha Yang a874c9f2d4 cuda: Fix runtime compiler library loading on Windows
The cuda is a part of GPU driver but runtime compiler is a part of
cuda toolkit, which means the version number can be different.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3545>
2022-12-09 19:24:08 +09:00

343 lines
9.1 KiB
C

/* GStreamer
* Copyright (C) 2019 Seungha Yang <seungha.yang@navercorp.com>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
* Boston, MA 02110-1301, USA.
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "cuda-gst.h"
#include "gstcudanvrtc.h"
#include "gstcudaloader.h"
#include <nvrtc.h>
#include <gmodule.h>
GST_DEBUG_CATEGORY_STATIC (gst_cuda_nvrtc_debug);
#define GST_CAT_DEFAULT gst_cuda_nvrtc_debug
#ifndef G_OS_WIN32
#define NVRTC_LIBNAME "libnvrtc.so"
#else
#define NVRTC_LIBNAME "nvrtc64_%d%d_0.dll"
#endif
#define LOAD_SYMBOL(name,func) G_STMT_START { \
if (!g_module_symbol (module, G_STRINGIFY (name), (gpointer *) &vtable->func)) { \
GST_ERROR ("Failed to load '%s', %s", G_STRINGIFY (name), g_module_error()); \
goto error; \
} \
} G_STMT_END;
/* *INDENT-OFF* */
typedef struct _GstCudaNvrtcVTable
{
gboolean loaded;
nvrtcResult (*NvrtcCompileProgram) (nvrtcProgram prog, int numOptions,
const char **options);
nvrtcResult (*NvrtcCreateProgram) (nvrtcProgram * prog, const char *src,
const char *name, int numHeaders, const char **headers,
const char **includeNames);
nvrtcResult (*NvrtcDestroyProgram) (nvrtcProgram * prog);
nvrtcResult (*NvrtcGetPTX) (nvrtcProgram prog, char *ptx);
nvrtcResult (*NvrtcGetPTXSize) (nvrtcProgram prog, size_t * ptxSizeRet);
nvrtcResult (*NvrtcGetProgramLog) (nvrtcProgram prog, char *log);
nvrtcResult (*NvrtcGetProgramLogSize) (nvrtcProgram prog,
size_t * logSizeRet);
} GstCudaNvrtcVTable;
/* *INDENT-ON* */
static GstCudaNvrtcVTable gst_cuda_nvrtc_vtable = { 0, };
#ifdef G_OS_WIN32
static GModule *
gst_cuda_nvrtc_load_library_once_win32 (void)
{
gchar *dll_name = NULL;
GModule *module = NULL;
gint cuda_version;
gint cuda_major_version;
gint cuda_minor_version;
gint major, minor;
CUresult rst;
rst = CuDriverGetVersion (&cuda_version);
if (rst != CUDA_SUCCESS) {
GST_WARNING ("Couldn't get driver version, 0x%x", (guint) rst);
return NULL;
}
cuda_major_version = cuda_version / 1000;
cuda_minor_version = (cuda_version % 1000) / 10;
GST_INFO ("CUDA version %d / %d", cuda_major_version, cuda_minor_version);
/* First path for searching nvrtc library using system CUDA version */
for (minor = cuda_minor_version; minor >= 0; minor--) {
g_clear_pointer (&dll_name, g_free);
dll_name = g_strdup_printf (NVRTC_LIBNAME, cuda_major_version, minor);
module = g_module_open (dll_name, G_MODULE_BIND_LAZY);
if (module) {
GST_INFO ("%s is available", dll_name);
g_free (dll_name);
return module;
}
GST_DEBUG ("Couldn't open library %s", dll_name);
}
/* CUDA is a part for driever installation, but nvrtc library is a part of
* CUDA-toolkit. So CUDA-toolkit version may be lower than
* CUDA version. Do search the dll again */
for (major = cuda_major_version; major >= 9; major--) {
for (minor = 5; minor >= 0; minor--) {
g_clear_pointer (&dll_name, g_free);
dll_name = g_strdup_printf (NVRTC_LIBNAME, major, minor);
module = g_module_open (dll_name, G_MODULE_BIND_LAZY);
if (module) {
GST_INFO ("%s is available", dll_name);
g_free (dll_name);
return module;
}
GST_DEBUG ("Couldn't open library %s", dll_name);
}
}
g_free (dll_name);
return NULL;
}
#endif
static gboolean
gst_cuda_nvrtc_load_library_once (void)
{
GModule *module = NULL;
const gchar *filename_env;
GstCudaNvrtcVTable *vtable;
filename_env = g_getenv ("GST_CUDA_NVRTC_LIBNAME");
if (filename_env)
module = g_module_open (filename_env, G_MODULE_BIND_LAZY);
if (!module) {
#ifndef G_OS_WIN32
module = g_module_open (NVRTC_LIBNAME, G_MODULE_BIND_LAZY);
#else
module = gst_cuda_nvrtc_load_library_once_win32 ();
#endif
}
if (module == NULL) {
GST_WARNING ("Could not open nvrtc library %s", g_module_error ());
return FALSE;
}
vtable = &gst_cuda_nvrtc_vtable;
LOAD_SYMBOL (nvrtcCompileProgram, NvrtcCompileProgram);
LOAD_SYMBOL (nvrtcCreateProgram, NvrtcCreateProgram);
LOAD_SYMBOL (nvrtcDestroyProgram, NvrtcDestroyProgram);
LOAD_SYMBOL (nvrtcGetPTX, NvrtcGetPTX);
LOAD_SYMBOL (nvrtcGetPTXSize, NvrtcGetPTXSize);
LOAD_SYMBOL (nvrtcGetProgramLog, NvrtcGetProgramLog);
LOAD_SYMBOL (nvrtcGetProgramLogSize, NvrtcGetProgramLogSize);
vtable->loaded = TRUE;
return TRUE;
error:
g_module_close (module);
return FALSE;
}
/**
* gst_cuda_nvrtc_load_library:
*
* Loads the nvrtc library.
*
* Returns: %TRUE if the library could be loaded, %FALSE otherwise
*
* Since: 1.22
*/
gboolean
gst_cuda_nvrtc_load_library (void)
{
static gsize init_once = 0;
if (g_once_init_enter (&init_once)) {
GST_DEBUG_CATEGORY_INIT (gst_cuda_nvrtc_debug, "cudanvrtc", 0,
"CUDA runtime compiler");
if (gst_cuda_load_library ())
gst_cuda_nvrtc_load_library_once ();
g_once_init_leave (&init_once, 1);
}
return gst_cuda_nvrtc_vtable.loaded;
}
/* *INDENT-OFF* */
static nvrtcResult
NvrtcCompileProgram (nvrtcProgram prog, int numOptions, const char **options)
{
g_assert (gst_cuda_nvrtc_vtable.NvrtcCompileProgram != NULL);
return gst_cuda_nvrtc_vtable.NvrtcCompileProgram (prog, numOptions, options);
}
static nvrtcResult
NvrtcCreateProgram (nvrtcProgram * prog, const char *src, const char *name,
int numHeaders, const char **headers, const char **includeNames)
{
g_assert (gst_cuda_nvrtc_vtable.NvrtcCreateProgram != NULL);
return gst_cuda_nvrtc_vtable.NvrtcCreateProgram (prog, src, name, numHeaders,
headers, includeNames);
}
static nvrtcResult
NvrtcDestroyProgram (nvrtcProgram * prog)
{
g_assert (gst_cuda_nvrtc_vtable.NvrtcDestroyProgram != NULL);
return gst_cuda_nvrtc_vtable.NvrtcDestroyProgram (prog);
}
static nvrtcResult
NvrtcGetPTX (nvrtcProgram prog, char *ptx)
{
g_assert (gst_cuda_nvrtc_vtable.NvrtcGetPTX != NULL);
return gst_cuda_nvrtc_vtable.NvrtcGetPTX (prog, ptx);
}
static nvrtcResult
NvrtcGetPTXSize (nvrtcProgram prog, size_t *ptxSizeRet)
{
g_assert (gst_cuda_nvrtc_vtable.NvrtcGetPTXSize != NULL);
return gst_cuda_nvrtc_vtable.NvrtcGetPTXSize (prog, ptxSizeRet);
}
static nvrtcResult
NvrtcGetProgramLog (nvrtcProgram prog, char *log)
{
g_assert (gst_cuda_nvrtc_vtable.NvrtcGetProgramLog != NULL);
return gst_cuda_nvrtc_vtable.NvrtcGetProgramLog (prog, log);
}
static nvrtcResult
NvrtcGetProgramLogSize (nvrtcProgram prog, size_t *logSizeRet)
{
g_assert (gst_cuda_nvrtc_vtable.NvrtcGetProgramLogSize != NULL);
return gst_cuda_nvrtc_vtable.NvrtcGetProgramLogSize (prog, logSizeRet);
}
/* *INDENT-ON* */
/**
* gst_cuda_nvrtc_compile:
* @source: Source code to compile
*
* Since: 1.22
*/
gchar *
gst_cuda_nvrtc_compile (const gchar * source)
{
nvrtcProgram prog;
nvrtcResult ret;
CUresult curet;
const gchar *opts[] = { "--gpu-architecture=compute_30" };
gsize ptx_size;
gchar *ptx = NULL;
int driverVersion;
g_return_val_if_fail (source != NULL, NULL);
if (!gst_cuda_nvrtc_load_library ()) {
return NULL;
}
GST_TRACE ("CUDA kernel source \n%s", source);
curet = CuDriverGetVersion (&driverVersion);
if (curet != CUDA_SUCCESS) {
GST_ERROR ("Failed to query CUDA Driver version, ret %d", curet);
return NULL;
}
GST_DEBUG ("CUDA Driver Version %d.%d", driverVersion / 1000,
(driverVersion % 1000) / 10);
ret = NvrtcCreateProgram (&prog, source, NULL, 0, NULL, NULL);
if (ret != NVRTC_SUCCESS) {
GST_ERROR ("couldn't create nvrtc program, ret %d", ret);
return NULL;
}
/* Starting from CUDA 11, the lowest supported architecture is 5.2 */
if (driverVersion >= 11000)
opts[0] = "--gpu-architecture=compute_52";
ret = NvrtcCompileProgram (prog, 1, opts);
if (ret != NVRTC_SUCCESS) {
gsize log_size;
GST_ERROR ("couldn't compile nvrtc program, ret %d", ret);
if (NvrtcGetProgramLogSize (prog, &log_size) == NVRTC_SUCCESS &&
log_size > 0) {
gchar *compile_log = g_alloca (log_size);
if (NvrtcGetProgramLog (prog, compile_log) == NVRTC_SUCCESS) {
GST_ERROR ("nvrtc compile log %s", compile_log);
}
}
goto error;
}
ret = NvrtcGetPTXSize (prog, &ptx_size);
if (ret != NVRTC_SUCCESS) {
GST_ERROR ("unknown ptx size, ret %d", ret);
goto error;
}
ptx = g_malloc0 (ptx_size);
ret = NvrtcGetPTX (prog, ptx);
if (ret != NVRTC_SUCCESS) {
GST_ERROR ("couldn't get ptx, ret %d", ret);
g_free (ptx);
goto error;
}
NvrtcDestroyProgram (&prog);
GST_TRACE ("compiled CUDA PTX %s\n", ptx);
return ptx;
error:
NvrtcDestroyProgram (&prog);
return NULL;
}