highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

avcodec/nvenc: use dynamically loaded CUDA

Browse Source
This commit is contained in:
Timo Rothenpieler
2016-10-10 12:55:59 +02:00
parent a0c9e76942
commit a66835bcb1
3 changed files with 26 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
configure vendored
View File

@@ -2693,6 +2693,7 @@ vaapi_encode_deps="vaapi"
hwupload_cuda_filter_deps="cuda" hwupload_cuda_filter_deps="cuda"
scale_npp_filter_deps="cuda libnpp" scale_npp_filter_deps="cuda libnpp"
nvenc_deps="cuda"
nvenc_deps_any="dlopen LoadLibrary" nvenc_deps_any="dlopen LoadLibrary"
nvenc_encoder_deps="nvenc" nvenc_encoder_deps="nvenc"
h264_cuvid_decoder_deps="cuda cuvid" h264_cuvid_decoder_deps="cuda cuvid"

125
libavcodec/nvenc.c
View File

@@ -21,56 +21,20 @@
#include "config.h" #include "config.h"
#if defined(_WIN32) || defined(__CYGWIN__) #include "nvenc.h"
# define CUDA_LIBNAME "nvcuda.dll"
# if ARCH_X86_64
# define NVENC_LIBNAME "nvEncodeAPI64.dll"
# else
# define NVENC_LIBNAME "nvEncodeAPI.dll"
# endif
#else
# define CUDA_LIBNAME "libcuda.so.1"
# define NVENC_LIBNAME "libnvidia-encode.so.1"
#endif
#if defined(_WIN32)
#include "compat/w32dlfcn.h"
#else
#include <dlfcn.h>
#endif
#include "libavutil/hwcontext_cuda.h"
#include "libavutil/hwcontext.h" #include "libavutil/hwcontext.h"
#include "libavutil/imgutils.h" #include "libavutil/imgutils.h"
#include "libavutil/avassert.h" #include "libavutil/avassert.h"
#include "libavutil/mem.h" #include "libavutil/mem.h"
#include "internal.h" #include "internal.h"
#include "nvenc.h"
#define NVENC_CAP 0x30 #define NVENC_CAP 0x30
#define IS_CBR(rc) (rc == NV_ENC_PARAMS_RC_CBR || \ #define IS_CBR(rc) (rc == NV_ENC_PARAMS_RC_CBR || \
rc == NV_ENC_PARAMS_RC_2_PASS_QUALITY || \ rc == NV_ENC_PARAMS_RC_2_PASS_QUALITY || \
rc == NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP) rc == NV_ENC_PARAMS_RC_2_PASS_FRAMESIZE_CAP)
#define LOAD_LIBRARY(l, path) \
do { \
if (!((l) = dlopen(path, RTLD_LAZY))) { \
av_log(avctx, AV_LOG_ERROR, \
"Cannot load %s\n", \
path); \
return AVERROR_UNKNOWN; \
} \
} while (0)
#define LOAD_SYMBOL(fun, lib, symbol) \
do { \
if (!((fun) = dlsym(lib, symbol))) { \
av_log(avctx, AV_LOG_ERROR, \
"Cannot load %s\n", \
symbol); \
return AVERROR_UNKNOWN; \
} \
} while (0)
const enum AVPixelFormat ff_nvenc_pix_fmts[] = { const enum AVPixelFormat ff_nvenc_pix_fmts[] = {
AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV420P,
AV_PIX_FMT_NV12, AV_PIX_FMT_NV12,
@@ -79,9 +43,7 @@ const enum AVPixelFormat ff_nvenc_pix_fmts[] = {
AV_PIX_FMT_YUV444P16, AV_PIX_FMT_YUV444P16,
AV_PIX_FMT_0RGB32, AV_PIX_FMT_0RGB32,
AV_PIX_FMT_0BGR32, AV_PIX_FMT_0BGR32,
#if CONFIG_CUDA
AV_PIX_FMT_CUDA, AV_PIX_FMT_CUDA,
#endif
AV_PIX_FMT_NONE AV_PIX_FMT_NONE
}; };
@@ -153,42 +115,19 @@ static av_cold int nvenc_load_libraries(AVCodecContext *avctx)
{ {
NvencContext *ctx = avctx->priv_data; NvencContext *ctx = avctx->priv_data;
NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs; NvencDynLoadFunctions *dl_fn = &ctx->nvenc_dload_funcs;
PNVENCODEAPIGETMAXSUPPORTEDVERSION nvenc_get_max_ver;
PNVENCODEAPICREATEINSTANCE nvenc_create_instance;
NVENCSTATUS err; NVENCSTATUS err;
uint32_t nvenc_max_ver; uint32_t nvenc_max_ver;
int ret;
#if CONFIG_CUDA ret = cuda_load_functions(&dl_fn->cuda_dl);
dl_fn->cu_init = cuInit; if (ret < 0)
dl_fn->cu_device_get_count = cuDeviceGetCount; return ret;
dl_fn->cu_device_get = cuDeviceGet;
dl_fn->cu_device_get_name = cuDeviceGetName;
dl_fn->cu_device_compute_capability = cuDeviceComputeCapability;
dl_fn->cu_ctx_create = cuCtxCreate_v2;
dl_fn->cu_ctx_pop_current = cuCtxPopCurrent_v2;
dl_fn->cu_ctx_destroy = cuCtxDestroy_v2;
#else
LOAD_LIBRARY(dl_fn->cuda, CUDA_LIBNAME);
LOAD_SYMBOL(dl_fn->cu_init, dl_fn->cuda, "cuInit"); ret = nvenc_load_functions(&dl_fn->nvenc_dl);
LOAD_SYMBOL(dl_fn->cu_device_get_count, dl_fn->cuda, "cuDeviceGetCount"); if (ret < 0)
LOAD_SYMBOL(dl_fn->cu_device_get, dl_fn->cuda, "cuDeviceGet"); return ret;
LOAD_SYMBOL(dl_fn->cu_device_get_name, dl_fn->cuda, "cuDeviceGetName");
LOAD_SYMBOL(dl_fn->cu_device_compute_capability, dl_fn->cuda,
"cuDeviceComputeCapability");
LOAD_SYMBOL(dl_fn->cu_ctx_create, dl_fn->cuda, "cuCtxCreate_v2");
LOAD_SYMBOL(dl_fn->cu_ctx_pop_current, dl_fn->cuda, "cuCtxPopCurrent_v2");
LOAD_SYMBOL(dl_fn->cu_ctx_destroy, dl_fn->cuda, "cuCtxDestroy_v2");
#endif
LOAD_LIBRARY(dl_fn->nvenc, NVENC_LIBNAME); err = dl_fn->nvenc_dl->NvEncodeAPIGetMaxSupportedVersion(&nvenc_max_ver);
LOAD_SYMBOL(nvenc_get_max_ver, dl_fn->nvenc,
"NvEncodeAPIGetMaxSupportedVersion");
LOAD_SYMBOL(nvenc_create_instance, dl_fn->nvenc,
"NvEncodeAPICreateInstance");
err = nvenc_get_max_ver(&nvenc_max_ver);
if (err != NV_ENC_SUCCESS) if (err != NV_ENC_SUCCESS)
return nvenc_print_error(avctx, err, "Failed to query nvenc max version"); return nvenc_print_error(avctx, err, "Failed to query nvenc max version");
@@ -204,7 +143,7 @@ static av_cold int nvenc_load_libraries(AVCodecContext *avctx)
dl_fn->nvenc_funcs.version = NV_ENCODE_API_FUNCTION_LIST_VER; dl_fn->nvenc_funcs.version = NV_ENCODE_API_FUNCTION_LIST_VER;
err = nvenc_create_instance(&dl_fn->nvenc_funcs); err = dl_fn->nvenc_dl->NvEncodeAPICreateInstance(&dl_fn->nvenc_funcs);
if (err != NV_ENC_SUCCESS) if (err != NV_ENC_SUCCESS)
return nvenc_print_error(avctx, err, "Failed to create nvenc instance"); return nvenc_print_error(avctx, err, "Failed to create nvenc instance");
@@ -376,7 +315,7 @@ static av_cold int nvenc_check_device(AVCodecContext *avctx, int idx)
if (ctx->device == LIST_DEVICES) if (ctx->device == LIST_DEVICES)
loglevel = AV_LOG_INFO; loglevel = AV_LOG_INFO;
cu_res = dl_fn->cu_device_get(&cu_device, idx); cu_res = dl_fn->cuda_dl->cuDeviceGet(&cu_device, idx);
if (cu_res != CUDA_SUCCESS) { if (cu_res != CUDA_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, av_log(avctx, AV_LOG_ERROR,
"Cannot access the CUDA device %d\n", "Cannot access the CUDA device %d\n",
@@ -384,11 +323,11 @@ static av_cold int nvenc_check_device(AVCodecContext *avctx, int idx)
return -1; return -1;
} }
cu_res = dl_fn->cu_device_get_name(name, sizeof(name), cu_device); cu_res = dl_fn->cuda_dl->cuDeviceGetName(name, sizeof(name), cu_device);
if (cu_res != CUDA_SUCCESS) if (cu_res != CUDA_SUCCESS)
return -1; return -1;
cu_res = dl_fn->cu_device_compute_capability(&major, &minor, cu_device); cu_res = dl_fn->cuda_dl->cuDeviceComputeCapability(&major, &minor, cu_device);
if (cu_res != CUDA_SUCCESS) if (cu_res != CUDA_SUCCESS)
return -1; return -1;
@@ -398,7 +337,7 @@ static av_cold int nvenc_check_device(AVCodecContext *avctx, int idx)
goto fail; goto fail;
} }
cu_res = dl_fn->cu_ctx_create(&ctx->cu_context_internal, 0, cu_device); cu_res = dl_fn->cuda_dl->cuCtxCreate(&ctx->cu_context_internal, 0, cu_device);
if (cu_res != CUDA_SUCCESS) { if (cu_res != CUDA_SUCCESS) {
av_log(avctx, AV_LOG_FATAL, "Failed creating CUDA context for NVENC: 0x%x\n", (int)cu_res); av_log(avctx, AV_LOG_FATAL, "Failed creating CUDA context for NVENC: 0x%x\n", (int)cu_res);
goto fail; goto fail;
@@ -406,7 +345,7 @@ static av_cold int nvenc_check_device(AVCodecContext *avctx, int idx)
ctx->cu_context = ctx->cu_context_internal; ctx->cu_context = ctx->cu_context_internal;
cu_res = dl_fn->cu_ctx_pop_current(&dummy); cu_res = dl_fn->cuda_dl->cuCtxPopCurrent(&dummy);
if (cu_res != CUDA_SUCCESS) { if (cu_res != CUDA_SUCCESS) {
av_log(avctx, AV_LOG_FATAL, "Failed popping CUDA context: 0x%x\n", (int)cu_res); av_log(avctx, AV_LOG_FATAL, "Failed popping CUDA context: 0x%x\n", (int)cu_res);
goto fail2; goto fail2;
@@ -430,7 +369,7 @@ fail3:
ctx->nvencoder = NULL; ctx->nvencoder = NULL;
fail2: fail2:
dl_fn->cu_ctx_destroy(ctx->cu_context_internal); dl_fn->cuda_dl->cuCtxDestroy(ctx->cu_context_internal);
ctx->cu_context_internal = NULL; ctx->cu_context_internal = NULL;
fail: fail:
@@ -454,7 +393,6 @@ static av_cold int nvenc_setup_device(AVCodecContext *avctx)
} }
if (avctx->pix_fmt == AV_PIX_FMT_CUDA) { if (avctx->pix_fmt == AV_PIX_FMT_CUDA) {
#if CONFIG_CUDA
AVHWFramesContext *frames_ctx; AVHWFramesContext *frames_ctx;
AVCUDADeviceContext *device_hwctx; AVCUDADeviceContext *device_hwctx;
int ret; int ret;
@@ -476,19 +414,16 @@ static av_cold int nvenc_setup_device(AVCodecContext *avctx)
av_log(avctx, AV_LOG_FATAL, "Provided device doesn't support required NVENC features\n"); av_log(avctx, AV_LOG_FATAL, "Provided device doesn't support required NVENC features\n");
return ret; return ret;
} }
#else
return AVERROR_BUG;
#endif
} else { } else {
int i, nb_devices = 0; int i, nb_devices = 0;
if ((dl_fn->cu_init(0)) != CUDA_SUCCESS) { if ((dl_fn->cuda_dl->cuInit(0)) != CUDA_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, av_log(avctx, AV_LOG_ERROR,
"Cannot init CUDA\n"); "Cannot init CUDA\n");
return AVERROR_UNKNOWN; return AVERROR_UNKNOWN;
} }
if ((dl_fn->cu_device_get_count(&nb_devices)) != CUDA_SUCCESS) { if ((dl_fn->cuda_dl->cuDeviceGetCount(&nb_devices)) != CUDA_SUCCESS) {
av_log(avctx, AV_LOG_ERROR, av_log(avctx, AV_LOG_ERROR,
"Cannot enumerate the CUDA devices\n"); "Cannot enumerate the CUDA devices\n");
return AVERROR_UNKNOWN; return AVERROR_UNKNOWN;
@@ -1265,30 +1200,14 @@ av_cold int ff_nvenc_encode_close(AVCodecContext *avctx)
ctx->nvencoder = NULL; ctx->nvencoder = NULL;
if (ctx->cu_context_internal) if (ctx->cu_context_internal)
dl_fn->cu_ctx_destroy(ctx->cu_context_internal); dl_fn->cuda_dl->cuCtxDestroy(ctx->cu_context_internal);
ctx->cu_context = ctx->cu_context_internal = NULL; ctx->cu_context = ctx->cu_context_internal = NULL;
if (dl_fn->nvenc) nvenc_free_functions(&dl_fn->nvenc_dl);
dlclose(dl_fn->nvenc); cuda_free_functions(&dl_fn->cuda_dl);
dl_fn->nvenc = NULL;
dl_fn->nvenc_device_count = 0; dl_fn->nvenc_device_count = 0;
#if !CONFIG_CUDA
if (dl_fn->cuda)
dlclose(dl_fn->cuda);
dl_fn->cuda = NULL;
#endif
dl_fn->cu_init = NULL;
dl_fn->cu_device_get_count = NULL;
dl_fn->cu_device_get = NULL;
dl_fn->cu_device_get_name = NULL;
dl_fn->cu_device_compute_capability = NULL;
dl_fn->cu_ctx_create = NULL;
dl_fn->cu_ctx_pop_current = NULL;
dl_fn->cu_ctx_destroy = NULL;
av_log(avctx, AV_LOG_VERBOSE, "Nvenc unloaded\n"); av_log(avctx, AV_LOG_VERBOSE, "Nvenc unloaded\n");
return 0; return 0;

46
libavcodec/nvenc.h
View File

@@ -23,29 +23,12 @@
#include "config.h" #include "config.h"
#include "compat/cuda/dynlink_loader.h"
#include "libavutil/fifo.h" #include "libavutil/fifo.h"
#include "libavutil/opt.h" #include "libavutil/opt.h"
#include "avcodec.h" #include "avcodec.h"
#if CONFIG_CUDA
#include "libavutil/hwcontext_cuda.h"
#else
#if defined(_WIN32)
#define CUDAAPI __stdcall
#else
#define CUDAAPI
#endif
typedef enum cudaError_enum {
CUDA_SUCCESS = 0
} CUresult;
typedef int CUdevice;
typedef void* CUcontext;
typedef void* CUdeviceptr;
#endif
#define MAX_REGISTERED_FRAMES 64 #define MAX_REGISTERED_FRAMES 64
typedef struct NvencSurface typedef struct NvencSurface
@@ -64,33 +47,10 @@ typedef struct NvencSurface
int lockCount; int lockCount;
} NvencSurface; } NvencSurface;
typedef CUresult(CUDAAPI *PCUINIT)(unsigned int Flags);
typedef CUresult(CUDAAPI *PCUDEVICEGETCOUNT)(int *count);
typedef CUresult(CUDAAPI *PCUDEVICEGET)(CUdevice *device, int ordinal);
typedef CUresult(CUDAAPI *PCUDEVICEGETNAME)(char *name, int len, CUdevice dev);
typedef CUresult(CUDAAPI *PCUDEVICECOMPUTECAPABILITY)(int *major, int *minor, CUdevice dev);
typedef CUresult(CUDAAPI *PCUCTXCREATE)(CUcontext *pctx, unsigned int flags, CUdevice dev);
typedef CUresult(CUDAAPI *PCUCTXPOPCURRENT)(CUcontext *pctx);
typedef CUresult(CUDAAPI *PCUCTXDESTROY)(CUcontext ctx);
typedef NVENCSTATUS (NVENCAPI *PNVENCODEAPIGETMAXSUPPORTEDVERSION)(uint32_t* version);
typedef NVENCSTATUS (NVENCAPI *PNVENCODEAPICREATEINSTANCE)(NV_ENCODE_API_FUNCTION_LIST *functionList);
typedef struct NvencDynLoadFunctions typedef struct NvencDynLoadFunctions
{ {
#if !CONFIG_CUDA CudaFunctions *cuda_dl;
void *cuda; NvencFunctions *nvenc_dl;
#endif
void *nvenc;
PCUINIT cu_init;
PCUDEVICEGETCOUNT cu_device_get_count;
PCUDEVICEGET cu_device_get;
PCUDEVICEGETNAME cu_device_get_name;
PCUDEVICECOMPUTECAPABILITY cu_device_compute_capability;
PCUCTXCREATE cu_ctx_create;
PCUCTXPOPCURRENT cu_ctx_pop_current;
PCUCTXDESTROY cu_ctx_destroy;
NV_ENCODE_API_FUNCTION_LIST nvenc_funcs; NV_ENCODE_API_FUNCTION_LIST nvenc_funcs;
int nvenc_device_count; int nvenc_device_count;