FFmpeg
hwcontext_cuda.c
Go to the documentation of this file.
1 /*
2  * This file is part of FFmpeg.
3  *
4  * FFmpeg is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU Lesser General Public
6  * License as published by the Free Software Foundation; either
7  * version 2.1 of the License, or (at your option) any later version.
8  *
9  * FFmpeg is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12  * Lesser General Public License for more details.
13  *
14  * You should have received a copy of the GNU Lesser General Public
15  * License along with FFmpeg; if not, write to the Free Software
16  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17  */
18 
19 #include "buffer.h"
20 #include "common.h"
21 #include "hwcontext.h"
22 #include "hwcontext_internal.h"
24 #if CONFIG_VULKAN
25 #include "hwcontext_vulkan.h"
26 #endif
27 #include "cuda_check.h"
28 #include "mem.h"
29 #include "pixdesc.h"
30 #include "pixfmt.h"
31 #include "imgutils.h"
32 
33 typedef struct CUDAFramesContext {
37 
38 static const enum AVPixelFormat supported_formats[] = {
48 #if CONFIG_VULKAN
50 #endif
51 };
52 
53 #define CHECK_CU(x) FF_CUDA_CHECK_DL(device_ctx, cu, x)
54 
56  const void *hwconfig,
57  AVHWFramesConstraints *constraints)
58 {
59  int i;
60 
62  sizeof(*constraints->valid_sw_formats));
63  if (!constraints->valid_sw_formats)
64  return AVERROR(ENOMEM);
65 
66  for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++)
67  constraints->valid_sw_formats[i] = supported_formats[i];
69 
70  constraints->valid_hw_formats = av_malloc_array(2, sizeof(*constraints->valid_hw_formats));
71  if (!constraints->valid_hw_formats)
72  return AVERROR(ENOMEM);
73 
74  constraints->valid_hw_formats[0] = AV_PIX_FMT_CUDA;
75  constraints->valid_hw_formats[1] = AV_PIX_FMT_NONE;
76 
77  return 0;
78 }
79 
80 static void cuda_buffer_free(void *opaque, uint8_t *data)
81 {
82  AVHWFramesContext *ctx = opaque;
83  AVHWDeviceContext *device_ctx = ctx->device_ctx;
84  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
85  CudaFunctions *cu = hwctx->internal->cuda_dl;
86 
87  CUcontext dummy;
88 
89  CHECK_CU(cu->cuCtxPushCurrent(hwctx->cuda_ctx));
90 
91  CHECK_CU(cu->cuMemFree((CUdeviceptr)data));
92 
93  CHECK_CU(cu->cuCtxPopCurrent(&dummy));
94 }
95 
96 static AVBufferRef *cuda_pool_alloc(void *opaque, size_t size)
97 {
98  AVHWFramesContext *ctx = opaque;
99  AVHWDeviceContext *device_ctx = ctx->device_ctx;
100  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
101  CudaFunctions *cu = hwctx->internal->cuda_dl;
102 
103  AVBufferRef *ret = NULL;
104  CUcontext dummy = NULL;
105  CUdeviceptr data;
106  int err;
107 
108  err = CHECK_CU(cu->cuCtxPushCurrent(hwctx->cuda_ctx));
109  if (err < 0)
110  return NULL;
111 
112  err = CHECK_CU(cu->cuMemAlloc(&data, size));
113  if (err < 0)
114  goto fail;
115 
116  ret = av_buffer_create((uint8_t*)data, size, cuda_buffer_free, ctx, 0);
117  if (!ret) {
118  CHECK_CU(cu->cuMemFree(data));
119  goto fail;
120  }
121 
122 fail:
123  CHECK_CU(cu->cuCtxPopCurrent(&dummy));
124  return ret;
125 }
126 
128 {
129  AVHWDeviceContext *device_ctx = ctx->device_ctx;
130  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
131  CUDAFramesContext *priv = ctx->internal->priv;
132  CudaFunctions *cu = hwctx->internal->cuda_dl;
133  int err, i;
134 
135  for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) {
136  if (ctx->sw_format == supported_formats[i])
137  break;
138  }
140  av_log(ctx, AV_LOG_ERROR, "Pixel format '%s' is not supported\n",
141  av_get_pix_fmt_name(ctx->sw_format));
142  return AVERROR(ENOSYS);
143  }
144 
145  err = CHECK_CU(cu->cuDeviceGetAttribute(&priv->tex_alignment,
146  14 /* CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT */,
147  hwctx->internal->cuda_device));
148  if (err < 0)
149  return err;
150 
151  av_log(ctx, AV_LOG_DEBUG, "CUDA texture alignment: %d\n", priv->tex_alignment);
152 
153  // YUV420P is a special case.
154  // Since nvenc expects the U/V planes to have half the linesize of the Y plane
155  // alignment has to be doubled to ensure the U/V planes still end up aligned.
156  if (ctx->sw_format == AV_PIX_FMT_YUV420P)
157  priv->tex_alignment *= 2;
158 
159  av_pix_fmt_get_chroma_sub_sample(ctx->sw_format, &priv->shift_width, &priv->shift_height);
160 
161  if (!ctx->pool) {
162  int size = av_image_get_buffer_size(ctx->sw_format, ctx->width, ctx->height, priv->tex_alignment);
163  if (size < 0)
164  return size;
165 
166  ctx->internal->pool_internal = av_buffer_pool_init2(size, ctx, cuda_pool_alloc, NULL);
167  if (!ctx->internal->pool_internal)
168  return AVERROR(ENOMEM);
169  }
170 
171  return 0;
172 }
173 
175 {
176  CUDAFramesContext *priv = ctx->internal->priv;
177  int res;
178 
179  frame->buf[0] = av_buffer_pool_get(ctx->pool);
180  if (!frame->buf[0])
181  return AVERROR(ENOMEM);
182 
183  res = av_image_fill_arrays(frame->data, frame->linesize, frame->buf[0]->data,
184  ctx->sw_format, ctx->width, ctx->height, priv->tex_alignment);
185  if (res < 0)
186  return res;
187 
188  // YUV420P is a special case.
189  // Nvenc expects the U/V planes in swapped order from how ffmpeg expects them, also chroma is half-aligned
190  if (ctx->sw_format == AV_PIX_FMT_YUV420P) {
191  frame->linesize[1] = frame->linesize[2] = frame->linesize[0] / 2;
192  frame->data[2] = frame->data[1];
193  frame->data[1] = frame->data[2] + frame->linesize[2] * (ctx->height / 2);
194  }
195 
196  frame->format = AV_PIX_FMT_CUDA;
197  frame->width = ctx->width;
198  frame->height = ctx->height;
199 
200  return 0;
201 }
202 
205  enum AVPixelFormat **formats)
206 {
207  enum AVPixelFormat *fmts;
208 
209  fmts = av_malloc_array(2, sizeof(*fmts));
210  if (!fmts)
211  return AVERROR(ENOMEM);
212 
213  fmts[0] = ctx->sw_format;
214  fmts[1] = AV_PIX_FMT_NONE;
215 
216  *formats = fmts;
217 
218  return 0;
219 }
220 
222  const AVFrame *src)
223 {
224  CUDAFramesContext *priv = ctx->internal->priv;
225  AVHWDeviceContext *device_ctx = ctx->device_ctx;
226  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
227  CudaFunctions *cu = hwctx->internal->cuda_dl;
228 
229  CUcontext dummy;
230  int i, ret;
231 
232  if ((src->hw_frames_ctx && ((AVHWFramesContext*)src->hw_frames_ctx->data)->format != AV_PIX_FMT_CUDA) ||
233  (dst->hw_frames_ctx && ((AVHWFramesContext*)dst->hw_frames_ctx->data)->format != AV_PIX_FMT_CUDA))
234  return AVERROR(ENOSYS);
235 
236  ret = CHECK_CU(cu->cuCtxPushCurrent(hwctx->cuda_ctx));
237  if (ret < 0)
238  return ret;
239 
240  for (i = 0; i < FF_ARRAY_ELEMS(src->data) && src->data[i]; i++) {
241  CUDA_MEMCPY2D cpy = {
242  .srcPitch = src->linesize[i],
243  .dstPitch = dst->linesize[i],
244  .WidthInBytes = FFMIN(src->linesize[i], dst->linesize[i]),
245  .Height = src->height >> ((i == 0 || i == 3) ? 0 : priv->shift_height),
246  };
247 
248  if (src->hw_frames_ctx) {
249  cpy.srcMemoryType = CU_MEMORYTYPE_DEVICE;
250  cpy.srcDevice = (CUdeviceptr)src->data[i];
251  } else {
252  cpy.srcMemoryType = CU_MEMORYTYPE_HOST;
253  cpy.srcHost = src->data[i];
254  }
255 
256  if (dst->hw_frames_ctx) {
257  cpy.dstMemoryType = CU_MEMORYTYPE_DEVICE;
258  cpy.dstDevice = (CUdeviceptr)dst->data[i];
259  } else {
260  cpy.dstMemoryType = CU_MEMORYTYPE_HOST;
261  cpy.dstHost = dst->data[i];
262  }
263 
264  ret = CHECK_CU(cu->cuMemcpy2DAsync(&cpy, hwctx->stream));
265  if (ret < 0)
266  goto exit;
267  }
268 
269  if (!dst->hw_frames_ctx) {
270  ret = CHECK_CU(cu->cuStreamSynchronize(hwctx->stream));
271  if (ret < 0)
272  goto exit;
273  }
274 
275 exit:
276  CHECK_CU(cu->cuCtxPopCurrent(&dummy));
277 
278  return 0;
279 }
280 
281 static void cuda_device_uninit(AVHWDeviceContext *device_ctx)
282 {
283  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
284 
285  if (hwctx->internal) {
286  CudaFunctions *cu = hwctx->internal->cuda_dl;
287 
288  if (hwctx->internal->is_allocated && hwctx->cuda_ctx) {
290  CHECK_CU(cu->cuDevicePrimaryCtxRelease(hwctx->internal->cuda_device));
291  else
292  CHECK_CU(cu->cuCtxDestroy(hwctx->cuda_ctx));
293 
294  hwctx->cuda_ctx = NULL;
295  }
296 
297  cuda_free_functions(&hwctx->internal->cuda_dl);
298  }
299 
300  av_freep(&hwctx->internal);
301 }
302 
304 {
305  AVCUDADeviceContext *hwctx = ctx->hwctx;
306  int ret;
307 
308  if (!hwctx->internal) {
309  hwctx->internal = av_mallocz(sizeof(*hwctx->internal));
310  if (!hwctx->internal)
311  return AVERROR(ENOMEM);
312  }
313 
314  if (!hwctx->internal->cuda_dl) {
315  ret = cuda_load_functions(&hwctx->internal->cuda_dl, ctx);
316  if (ret < 0) {
317  av_log(ctx, AV_LOG_ERROR, "Could not dynamically load CUDA\n");
318  goto error;
319  }
320  }
321 
322  return 0;
323 
324 error:
326  return ret;
327 }
328 
329 static int cuda_context_init(AVHWDeviceContext *device_ctx, int flags) {
330  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
331  CudaFunctions *cu;
332  CUcontext dummy;
333  int ret, dev_active = 0;
334  unsigned int dev_flags = 0;
335 
336  const unsigned int desired_flags = CU_CTX_SCHED_BLOCKING_SYNC;
337 
338  cu = hwctx->internal->cuda_dl;
339 
340  hwctx->internal->flags = flags;
341 
343  ret = CHECK_CU(cu->cuDevicePrimaryCtxGetState(hwctx->internal->cuda_device,
344  &dev_flags, &dev_active));
345  if (ret < 0)
346  return ret;
347 
348  if (dev_active && dev_flags != desired_flags) {
349  av_log(device_ctx, AV_LOG_ERROR, "Primary context already active with incompatible flags.\n");
350  return AVERROR(ENOTSUP);
351  } else if (dev_flags != desired_flags) {
352  ret = CHECK_CU(cu->cuDevicePrimaryCtxSetFlags(hwctx->internal->cuda_device,
353  desired_flags));
354  if (ret < 0)
355  return ret;
356  }
357 
358  ret = CHECK_CU(cu->cuDevicePrimaryCtxRetain(&hwctx->cuda_ctx,
359  hwctx->internal->cuda_device));
360  if (ret < 0)
361  return ret;
362  } else {
363  ret = CHECK_CU(cu->cuCtxCreate(&hwctx->cuda_ctx, desired_flags,
364  hwctx->internal->cuda_device));
365  if (ret < 0)
366  return ret;
367 
368  CHECK_CU(cu->cuCtxPopCurrent(&dummy));
369  }
370 
371  hwctx->internal->is_allocated = 1;
372 
373  // Setting stream to NULL will make functions automatically use the default CUstream
374  hwctx->stream = NULL;
375 
376  return 0;
377 }
378 
380  AVDictionary *opts, int *flags)
381 {
382  AVDictionaryEntry *primary_ctx_opt = av_dict_get(opts, "primary_ctx", NULL, 0);
383 
384  if (primary_ctx_opt && strtol(primary_ctx_opt->value, NULL, 10)) {
385  av_log(device_ctx, AV_LOG_VERBOSE, "Using CUDA primary device context\n");
387  } else if (primary_ctx_opt) {
388  av_log(device_ctx, AV_LOG_VERBOSE, "Disabling use of CUDA primary device context\n");
390  }
391 
392  return 0;
393 }
394 
395 static int cuda_device_create(AVHWDeviceContext *device_ctx,
396  const char *device,
397  AVDictionary *opts, int flags)
398 {
399  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
400  CudaFunctions *cu;
401  int ret, device_idx = 0;
402 
403  ret = cuda_flags_from_opts(device_ctx, opts, &flags);
404  if (ret < 0)
405  goto error;
406 
407  if (device)
408  device_idx = strtol(device, NULL, 0);
409 
410  ret = cuda_device_init(device_ctx);
411  if (ret < 0)
412  goto error;
413 
414  cu = hwctx->internal->cuda_dl;
415 
416  ret = CHECK_CU(cu->cuInit(0));
417  if (ret < 0)
418  goto error;
419 
420  ret = CHECK_CU(cu->cuDeviceGet(&hwctx->internal->cuda_device, device_idx));
421  if (ret < 0)
422  goto error;
423 
424  ret = cuda_context_init(device_ctx, flags);
425  if (ret < 0)
426  goto error;
427 
428  return 0;
429 
430 error:
431  cuda_device_uninit(device_ctx);
432  return ret;
433 }
434 
435 static int cuda_device_derive(AVHWDeviceContext *device_ctx,
437  int flags) {
438  AVCUDADeviceContext *hwctx = device_ctx->hwctx;
439  CudaFunctions *cu;
440  const char *src_uuid = NULL;
441  int ret, i, device_count;
442 
443  ret = cuda_flags_from_opts(device_ctx, opts, &flags);
444  if (ret < 0)
445  goto error;
446 
447 #if CONFIG_VULKAN
448  VkPhysicalDeviceIDProperties vk_idp = {
449  .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES,
450  };
451 #endif
452 
453  switch (src_ctx->type) {
454 #if CONFIG_VULKAN
455 #define TYPE PFN_vkGetPhysicalDeviceProperties2
457  AVVulkanDeviceContext *vkctx = src_ctx->hwctx;
458  TYPE prop_fn = (TYPE)vkctx->get_proc_addr(vkctx->inst, "vkGetPhysicalDeviceProperties2");
459  VkPhysicalDeviceProperties2 vk_dev_props = {
460  .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2,
461  .pNext = &vk_idp,
462  };
463  prop_fn(vkctx->phys_dev, &vk_dev_props);
464  src_uuid = vk_idp.deviceUUID;
465  break;
466  }
467 #undef TYPE
468 #endif
469  default:
470  ret = AVERROR(ENOSYS);
471  goto error;
472  }
473 
474  if (!src_uuid) {
475  av_log(device_ctx, AV_LOG_ERROR,
476  "Failed to get UUID of source device.\n");
477  ret = AVERROR(EINVAL);
478  goto error;
479  }
480 
481  ret = cuda_device_init(device_ctx);
482  if (ret < 0)
483  goto error;
484 
485  cu = hwctx->internal->cuda_dl;
486 
487  ret = CHECK_CU(cu->cuInit(0));
488  if (ret < 0)
489  goto error;
490 
491  ret = CHECK_CU(cu->cuDeviceGetCount(&device_count));
492  if (ret < 0)
493  goto error;
494 
495  hwctx->internal->cuda_device = -1;
496  for (i = 0; i < device_count; i++) {
497  CUdevice dev;
498  CUuuid uuid;
499 
500  ret = CHECK_CU(cu->cuDeviceGet(&dev, i));
501  if (ret < 0)
502  goto error;
503 
504  ret = CHECK_CU(cu->cuDeviceGetUuid(&uuid, dev));
505  if (ret < 0)
506  goto error;
507 
508  if (memcmp(src_uuid, uuid.bytes, sizeof (uuid.bytes)) == 0) {
509  hwctx->internal->cuda_device = dev;
510  break;
511  }
512  }
513 
514  if (hwctx->internal->cuda_device == -1) {
515  av_log(device_ctx, AV_LOG_ERROR, "Could not derive CUDA device.\n");
516  goto error;
517  }
518 
519  ret = cuda_context_init(device_ctx, flags);
520  if (ret < 0)
521  goto error;
522 
523  return 0;
524 
525 error:
526  cuda_device_uninit(device_ctx);
527  return ret;
528 }
529 
532  .name = "CUDA",
533 
534  .device_hwctx_size = sizeof(AVCUDADeviceContext),
535  .frames_priv_size = sizeof(CUDAFramesContext),
536 
537  .device_create = cuda_device_create,
538  .device_derive = cuda_device_derive,
539  .device_init = cuda_device_init,
540  .device_uninit = cuda_device_uninit,
541  .frames_get_constraints = cuda_frames_get_constraints,
542  .frames_init = cuda_frames_init,
543  .frames_get_buffer = cuda_get_buffer,
544  .transfer_get_formats = cuda_transfer_get_formats,
545  .transfer_data_to = cuda_transfer_data,
546  .transfer_data_from = cuda_transfer_data,
547 
548  .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_CUDA, AV_PIX_FMT_NONE },
549 };
error
static void error(const char *err)
Definition: target_bsf_fuzzer.c:31
formats
formats
Definition: signature.h:48
AVHWDeviceContext::hwctx
void * hwctx
The format-specific data, allocated and freed by libavutil along with this context.
Definition: hwcontext.h:92
AVVulkanDeviceContext::phys_dev
VkPhysicalDevice phys_dev
Physical device.
Definition: hwcontext_vulkan.h:63
AV_PIX_FMT_CUDA
@ AV_PIX_FMT_CUDA
HW acceleration through CUDA.
Definition: pixfmt.h:225
AVPixelFormat
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
cuda_context_init
static int cuda_context_init(AVHWDeviceContext *device_ctx, int flags)
Definition: hwcontext_cuda.c:329
AVERROR
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
cuda_device_derive
static int cuda_device_derive(AVHWDeviceContext *device_ctx, AVHWDeviceContext *src_ctx, AVDictionary *opts, int flags)
Definition: hwcontext_cuda.c:435
hwcontext_cuda_internal.h
cuda_transfer_get_formats
static int cuda_transfer_get_formats(AVHWFramesContext *ctx, enum AVHWFrameTransferDirection dir, enum AVPixelFormat **formats)
Definition: hwcontext_cuda.c:203
cuda_flags_from_opts
static int cuda_flags_from_opts(AVHWDeviceContext *device_ctx, AVDictionary *opts, int *flags)
Definition: hwcontext_cuda.c:379
AVBufferRef::data
uint8_t * data
The data buffer.
Definition: buffer.h:90
CUDAFramesContext
Definition: hwcontext_cuda.c:33
CHECK_CU
#define CHECK_CU(x)
Definition: hwcontext_cuda.c:53
AVCUDADeviceContextInternal::is_allocated
int is_allocated
Definition: hwcontext_cuda_internal.h:33
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:325
pixdesc.h
AVVulkanDeviceContext::get_proc_addr
PFN_vkGetInstanceProcAddr get_proc_addr
Pointer to the instance-provided vkGetInstanceProcAddr loading function.
Definition: hwcontext_vulkan.h:53
data
const char data[16]
Definition: mxf.c:143
AVVulkanDeviceContext::inst
VkInstance inst
Vulkan instance.
Definition: hwcontext_vulkan.h:58
AV_LOG_VERBOSE
#define AV_LOG_VERBOSE
Detailed information.
Definition: log.h:196
AVDictionary
Definition: dict.c:30
AVHWFramesConstraints::valid_hw_formats
enum AVPixelFormat * valid_hw_formats
A list of possible values for format in the hw_frames_ctx, terminated by AV_PIX_FMT_NONE.
Definition: hwcontext.h:458
AV_PIX_FMT_VULKAN
@ AV_PIX_FMT_VULKAN
Vulkan hardware images.
Definition: pixfmt.h:348
AVFrame::data
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:346
AV_HWDEVICE_TYPE_VULKAN
@ AV_HWDEVICE_TYPE_VULKAN
Definition: hwcontext.h:39
AVHWFramesConstraints
This struct describes the constraints on hardware frames attached to a given device with a hardware-s...
Definition: hwcontext.h:453
AV_HWDEVICE_TYPE_CUDA
@ AV_HWDEVICE_TYPE_CUDA
Definition: hwcontext.h:30
fail
#define fail()
Definition: checkasm.h:131
dummy
int dummy
Definition: motion.c:65
av_buffer_pool_init2
AVBufferPool * av_buffer_pool_init2(size_t size, void *opaque, AVBufferRef *(*alloc)(void *opaque, size_t size), void(*pool_free)(void *opaque))
Allocate and initialize a buffer pool with a more complex allocator.
Definition: buffer.c:259
AVCUDADeviceContextInternal::cuda_device
CUdevice cuda_device
Definition: hwcontext_cuda_internal.h:34
av_pix_fmt_get_chroma_sub_sample
int av_pix_fmt_get_chroma_sub_sample(enum AVPixelFormat pix_fmt, int *h_shift, int *v_shift)
Utility function to access log2_chroma_w log2_chroma_h from the pixel format AVPixFmtDescriptor.
Definition: pixdesc.c:2703
AVCUDADeviceContext::cuda_ctx
CUcontext cuda_ctx
Definition: hwcontext_cuda.h:43
AVHWDeviceContext
This struct aggregates all the (hardware/vendor-specific) "high-level" state, i.e.
Definition: hwcontext.h:61
HWContextType::type
enum AVHWDeviceType type
Definition: hwcontext_internal.h:30
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:180
FF_ARRAY_ELEMS
#define FF_ARRAY_ELEMS(a)
Definition: sinewin_tablegen.c:29
AVHWFramesConstraints::valid_sw_formats
enum AVPixelFormat * valid_sw_formats
A list of possible values for sw_format in the hw_frames_ctx, terminated by AV_PIX_FMT_NONE.
Definition: hwcontext.h:465
av_dict_get
AVDictionaryEntry * av_dict_get(const AVDictionary *m, const char *key, const AVDictionaryEntry *prev, int flags)
Get a dictionary entry with matching key.
Definition: dict.c:40
av_buffer_pool_get
AVBufferRef * av_buffer_pool_get(AVBufferPool *pool)
Allocate a new AVBuffer, reusing an old buffer from the pool when available.
Definition: buffer.c:387
AV_PIX_FMT_YUVA420P
@ AV_PIX_FMT_YUVA420P
planar YUV 4:2:0, 20bpp, (1 Cr & Cb sample per 2x2 Y & A samples)
Definition: pixfmt.h:101
AV_PIX_FMT_YUV444P16
#define AV_PIX_FMT_YUV444P16
Definition: pixfmt.h:421
AV_PIX_FMT_0BGR32
#define AV_PIX_FMT_0BGR32
Definition: pixfmt.h:386
AV_LOG_DEBUG
#define AV_LOG_DEBUG
Stuff which is only useful for libav* developers.
Definition: log.h:201
ctx
AVFormatContext * ctx
Definition: movenc.c:48
cuda_device_init
static int cuda_device_init(AVHWDeviceContext *ctx)
Definition: hwcontext_cuda.c:303
AV_PIX_FMT_YUV420P
@ AV_PIX_FMT_YUV420P
planar YUV 4:2:0, 12bpp, (1 Cr & Cb sample per 2x2 Y samples)
Definition: pixfmt.h:66
AVVulkanDeviceContext
Main Vulkan context, allocated as AVHWDeviceContext.hwctx.
Definition: hwcontext_vulkan.h:42
opts
AVDictionary * opts
Definition: movenc.c:50
NULL
#define NULL
Definition: coverity.c:32
AVCUDADeviceContextInternal::flags
int flags
Definition: hwcontext_cuda_internal.h:35
TYPE
#define TYPE
Definition: ffv1.h:190
hwcontext_vulkan.h
CUDAFramesContext::shift_width
int shift_width
Definition: hwcontext_cuda.c:34
cuda_transfer_data
static int cuda_transfer_data(AVHWFramesContext *ctx, AVFrame *dst, const AVFrame *src)
Definition: hwcontext_cuda.c:221
av_buffer_create
AVBufferRef * av_buffer_create(uint8_t *data, size_t size, void(*free)(void *opaque, uint8_t *data), void *opaque, int flags)
Create an AVBuffer from an existing array.
Definition: buffer.c:55
AVCUDADeviceContext::stream
CUstream stream
Definition: hwcontext_cuda.h:44
AVCUDADeviceContext::internal
AVCUDADeviceContextInternal * internal
Definition: hwcontext_cuda.h:45
CUDAFramesContext::tex_alignment
int tex_alignment
Definition: hwcontext_cuda.c:35
av_image_fill_arrays
int av_image_fill_arrays(uint8_t *dst_data[4], int dst_linesize[4], const uint8_t *src, enum AVPixelFormat pix_fmt, int width, int height, int align)
Setup the data pointers and linesizes based on the specified image parameters and the provided array.
Definition: imgutils.c:446
CUDAFramesContext::shift_height
int shift_height
Definition: hwcontext_cuda.c:34
size
int size
Definition: twinvq_data.h:10344
ff_hwcontext_type_cuda
const HWContextType ff_hwcontext_type_cuda
Definition: hwcontext_cuda.c:530
buffer.h
av_image_get_buffer_size
int av_image_get_buffer_size(enum AVPixelFormat pix_fmt, int width, int height, int align)
Return the size in bytes of the amount of data required to store an image with the given parameters.
Definition: imgutils.c:466
AVCUDADeviceContextInternal::cuda_dl
CudaFunctions * cuda_dl
Definition: hwcontext_cuda_internal.h:32
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:269
av_malloc_array
#define av_malloc_array(a, b)
Definition: tableprint_vlc.h:31
common.h
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
av_mallocz
void * av_mallocz(size_t size)
Allocate a memory block with alignment suitable for all memory accesses (including vectors if availab...
Definition: mem.c:264
cuda_device_uninit
static void cuda_device_uninit(AVHWDeviceContext *device_ctx)
Definition: hwcontext_cuda.c:281
AV_PIX_FMT_P016
#define AV_PIX_FMT_P016
Definition: pixfmt.h:458
AVHWFrameTransferDirection
AVHWFrameTransferDirection
Definition: hwcontext.h:415
AVHWFramesContext
This struct describes a set or pool of "hardware" frames (i.e.
Definition: hwcontext.h:124
AVCUDADeviceContext
This struct is allocated as AVHWDeviceContext.hwctx.
Definition: hwcontext_cuda.h:42
ret
ret
Definition: filter_design.txt:187
AVHWDeviceContext::type
enum AVHWDeviceType type
This field identifies the underlying API used for hardware access.
Definition: hwcontext.h:79
pixfmt.h
AV_PIX_FMT_NV12
@ AV_PIX_FMT_NV12
planar YUV 4:2:0, 12bpp, 1 plane for Y and 1 plane for the UV components, which are interleaved (firs...
Definition: pixfmt.h:89
frame
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame For filters that do not use the this method is called when a frame is wanted on an output For a it should directly call filter_frame on the corresponding output For a if there are queued frames already one of these frames should be pushed If the filter should request a frame on one of its repeatedly until at least one frame has been pushed Return or at least make progress towards producing a frame
Definition: filter_design.txt:264
AV_PIX_FMT_0RGB32
#define AV_PIX_FMT_0RGB32
Definition: pixfmt.h:385
cuda_check.h
cuda_buffer_free
static void cuda_buffer_free(void *opaque, uint8_t *data)
Definition: hwcontext_cuda.c:80
AVFrame::hw_frames_ctx
AVBufferRef * hw_frames_ctx
For hwaccel-format frames, this should be a reference to the AVHWFramesContext describing the frame.
Definition: frame.h:664
AV_CUDA_USE_PRIMARY_CONTEXT
#define AV_CUDA_USE_PRIMARY_CONTEXT
Use primary device context instead of creating a new one.
Definition: hwcontext_cuda.h:63
AV_PIX_FMT_NONE
@ AV_PIX_FMT_NONE
Definition: pixfmt.h:65
cuda_device_create
static int cuda_device_create(AVHWDeviceContext *device_ctx, const char *device, AVDictionary *opts, int flags)
Definition: hwcontext_cuda.c:395
supported_formats
static enum AVPixelFormat supported_formats[]
Definition: hwcontext_cuda.c:38
AV_PIX_FMT_YUV444P
@ AV_PIX_FMT_YUV444P
planar YUV 4:4:4, 24bpp, (1 Cr & Cb sample per 1x1 Y samples)
Definition: pixfmt.h:71
cuda_get_buffer
static int cuda_get_buffer(AVHWFramesContext *ctx, AVFrame *frame)
Definition: hwcontext_cuda.c:174
AV_PIX_FMT_P010
#define AV_PIX_FMT_P010
Definition: pixfmt.h:457
mem.h
AVBufferRef
A reference to a data buffer.
Definition: buffer.h:82
cuda_pool_alloc
static AVBufferRef * cuda_pool_alloc(void *opaque, size_t size)
Definition: hwcontext_cuda.c:96
hwcontext_internal.h
AVDictionaryEntry
Definition: dict.h:79
av_freep
#define av_freep(p)
Definition: tableprint_vlc.h:34
src
INIT_CLIP pixel * src
Definition: h264pred_template.c:418
imgutils.h
flags
#define flags(name, subs,...)
Definition: cbs_av1.c:561
hwcontext.h
AVFrame::linesize
int linesize[AV_NUM_DATA_POINTERS]
For video, a positive or negative value, which is typically indicating the size in bytes of each pict...
Definition: frame.h:370
av_log
#define av_log(a,...)
Definition: tableprint_vlc.h:27
HWContextType
Definition: hwcontext_internal.h:29
cuda_frames_get_constraints
static int cuda_frames_get_constraints(AVHWDeviceContext *ctx, const void *hwconfig, AVHWFramesConstraints *constraints)
Definition: hwcontext_cuda.c:55
cuda_frames_init
static int cuda_frames_init(AVHWFramesContext *ctx)
Definition: hwcontext_cuda.c:127
AVDictionaryEntry::value
char * value
Definition: dict.h:81
av_get_pix_fmt_name
const char * av_get_pix_fmt_name(enum AVPixelFormat pix_fmt)
Return the short name for a pixel format, NULL in case pix_fmt is unknown.
Definition: pixdesc.c:2595