FFmpeg
vf_overlay_opencl.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 "libavutil/log.h"
20 #include "libavutil/opt.h"
21 #include "libavutil/pixdesc.h"
22 
23 #include "avfilter.h"
24 #include "framesync.h"
25 #include "internal.h"
26 #include "opencl.h"
27 #include "opencl_source.h"
28 #include "video.h"
29 
30 typedef struct OverlayOpenCLContext {
32 
34  cl_kernel kernel;
35  cl_command_queue command_queue;
36 
38 
39  int nb_planes;
43 
47 
49  enum AVPixelFormat main_format,
50  enum AVPixelFormat overlay_format)
51 {
52  OverlayOpenCLContext *ctx = avctx->priv;
53  cl_int cle;
54  const char *source = ff_opencl_source_overlay;
55  const char *kernel;
56  const AVPixFmtDescriptor *main_desc, *overlay_desc;
57  int err, i, main_planes, overlay_planes;
58 
59  main_desc = av_pix_fmt_desc_get(main_format);
60  overlay_desc = av_pix_fmt_desc_get(overlay_format);
61 
62  main_planes = overlay_planes = 0;
63  for (i = 0; i < main_desc->nb_components; i++)
64  main_planes = FFMAX(main_planes,
65  main_desc->comp[i].plane + 1);
66  for (i = 0; i < overlay_desc->nb_components; i++)
67  overlay_planes = FFMAX(overlay_planes,
68  overlay_desc->comp[i].plane + 1);
69 
70  ctx->nb_planes = main_planes;
71  ctx->x_subsample = 1 << main_desc->log2_chroma_w;
72  ctx->y_subsample = 1 << main_desc->log2_chroma_h;
73 
74  if (ctx->x_position % ctx->x_subsample ||
75  ctx->y_position % ctx->y_subsample) {
76  av_log(avctx, AV_LOG_WARNING, "Warning: overlay position (%d, %d) "
77  "does not match subsampling (%d, %d).\n",
78  ctx->x_position, ctx->y_position,
79  ctx->x_subsample, ctx->y_subsample);
80  }
81 
82  if (main_planes == overlay_planes) {
83  if (main_desc->nb_components == overlay_desc->nb_components)
84  kernel = "overlay_no_alpha";
85  else
86  kernel = "overlay_internal_alpha";
87  ctx->alpha_separate = 0;
88  } else {
89  kernel = "overlay_external_alpha";
90  ctx->alpha_separate = 1;
91  }
92 
93  av_log(avctx, AV_LOG_DEBUG, "Using kernel %s.\n", kernel);
94 
95  err = ff_opencl_filter_load_program(avctx, &source, 1);
96  if (err < 0)
97  goto fail;
98 
99  ctx->command_queue = clCreateCommandQueue(ctx->ocf.hwctx->context,
100  ctx->ocf.hwctx->device_id,
101  0, &cle);
102  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create OpenCL "
103  "command queue %d.\n", cle);
104 
105  ctx->kernel = clCreateKernel(ctx->ocf.program, kernel, &cle);
106  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create kernel %d.\n", cle);
107 
108  ctx->initialised = 1;
109  return 0;
110 
111 fail:
112  if (ctx->command_queue)
113  clReleaseCommandQueue(ctx->command_queue);
114  if (ctx->kernel)
115  clReleaseKernel(ctx->kernel);
116  return err;
117 }
118 
120 {
121  AVFilterContext *avctx = fs->parent;
122  AVFilterLink *outlink = avctx->outputs[0];
123  OverlayOpenCLContext *ctx = avctx->priv;
124  AVFrame *input_main, *input_overlay;
125  AVFrame *output;
126  cl_mem mem;
127  cl_int cle, x, y;
128  size_t global_work[2];
129  int kernel_arg = 0;
130  int err, plane;
131 
132  err = ff_framesync_get_frame(fs, 0, &input_main, 0);
133  if (err < 0)
134  return err;
135  err = ff_framesync_get_frame(fs, 1, &input_overlay, 0);
136  if (err < 0)
137  return err;
138 
139  if (!ctx->initialised) {
140  AVHWFramesContext *main_fc =
141  (AVHWFramesContext*)input_main->hw_frames_ctx->data;
142  AVHWFramesContext *overlay_fc =
143  (AVHWFramesContext*)input_overlay->hw_frames_ctx->data;
144 
145  err = overlay_opencl_load(avctx, main_fc->sw_format,
146  overlay_fc->sw_format);
147  if (err < 0)
148  return err;
149  }
150 
151  output = ff_get_video_buffer(outlink, outlink->w, outlink->h);
152  if (!output) {
153  err = AVERROR(ENOMEM);
154  goto fail;
155  }
156 
157  for (plane = 0; plane < ctx->nb_planes; plane++) {
158  kernel_arg = 0;
159 
160  mem = (cl_mem)output->data[plane];
161  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_mem, &mem);
162  kernel_arg++;
163 
164  mem = (cl_mem)input_main->data[plane];
165  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_mem, &mem);
166  kernel_arg++;
167 
168  mem = (cl_mem)input_overlay->data[plane];
169  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_mem, &mem);
170  kernel_arg++;
171 
172  if (ctx->alpha_separate) {
173  mem = (cl_mem)input_overlay->data[ctx->nb_planes];
174  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_mem, &mem);
175  kernel_arg++;
176  }
177 
178  x = ctx->x_position / (plane == 0 ? 1 : ctx->x_subsample);
179  y = ctx->y_position / (plane == 0 ? 1 : ctx->y_subsample);
180 
181  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_int, &x);
182  kernel_arg++;
183  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_int, &y);
184  kernel_arg++;
185 
186  if (ctx->alpha_separate) {
187  cl_int alpha_adj_x = plane == 0 ? 1 : ctx->x_subsample;
188  cl_int alpha_adj_y = plane == 0 ? 1 : ctx->y_subsample;
189 
190  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_int, &alpha_adj_x);
191  kernel_arg++;
192  CL_SET_KERNEL_ARG(ctx->kernel, kernel_arg, cl_int, &alpha_adj_y);
193  kernel_arg++;
194  }
195 
196  err = ff_opencl_filter_work_size_from_image(avctx, global_work,
197  output, plane, 0);
198  if (err < 0)
199  goto fail;
200 
201  cle = clEnqueueNDRangeKernel(ctx->command_queue, ctx->kernel, 2, NULL,
202  global_work, NULL, 0, NULL, NULL);
203  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to enqueue overlay kernel "
204  "for plane %d: %d.\n", plane, cle);
205  }
206 
207  cle = clFinish(ctx->command_queue);
208  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to finish command queue: %d.\n", cle);
209 
210  err = av_frame_copy_props(output, input_main);
211 
212  av_log(avctx, AV_LOG_DEBUG, "Filter output: %s, %ux%u (%"PRId64").\n",
213  av_get_pix_fmt_name(output->format),
214  output->width, output->height, output->pts);
215 
216  return ff_filter_frame(outlink, output);
217 
218 fail:
220  return err;
221 }
222 
224 {
225  AVFilterContext *avctx = outlink->src;
226  OverlayOpenCLContext *ctx = avctx->priv;
227  int err;
228 
229  err = ff_opencl_filter_config_output(outlink);
230  if (err < 0)
231  return err;
232 
233  err = ff_framesync_init_dualinput(&ctx->fs, avctx);
234  if (err < 0)
235  return err;
236 
237  return ff_framesync_configure(&ctx->fs);
238 }
239 
241 {
242  OverlayOpenCLContext *ctx = avctx->priv;
243 
244  ctx->fs.on_event = &overlay_opencl_blend;
245 
246  return ff_opencl_filter_init(avctx);
247 }
248 
250 {
251  OverlayOpenCLContext *ctx = avctx->priv;
252 
253  return ff_framesync_activate(&ctx->fs);
254 }
255 
257 {
258  OverlayOpenCLContext *ctx = avctx->priv;
259  cl_int cle;
260 
261  if (ctx->kernel) {
262  cle = clReleaseKernel(ctx->kernel);
263  if (cle != CL_SUCCESS)
264  av_log(avctx, AV_LOG_ERROR, "Failed to release "
265  "kernel: %d.\n", cle);
266  }
267 
268  if (ctx->command_queue) {
269  cle = clReleaseCommandQueue(ctx->command_queue);
270  if (cle != CL_SUCCESS)
271  av_log(avctx, AV_LOG_ERROR, "Failed to release "
272  "command queue: %d.\n", cle);
273  }
274 
276 
277  ff_framesync_uninit(&ctx->fs);
278 }
279 
280 #define OFFSET(x) offsetof(OverlayOpenCLContext, x)
281 #define FLAGS (AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
283  { "x", "Overlay x position",
284  OFFSET(x_position), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, .flags = FLAGS },
285  { "y", "Overlay y position",
286  OFFSET(y_position), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, .flags = FLAGS },
287  { NULL },
288 };
289 
290 AVFILTER_DEFINE_CLASS(overlay_opencl);
291 
293  {
294  .name = "main",
295  .type = AVMEDIA_TYPE_VIDEO,
296  .config_props = &ff_opencl_filter_config_input,
297  },
298  {
299  .name = "overlay",
300  .type = AVMEDIA_TYPE_VIDEO,
301  .config_props = &ff_opencl_filter_config_input,
302  },
303 };
304 
306  {
307  .name = "default",
308  .type = AVMEDIA_TYPE_VIDEO,
309  .config_props = &overlay_opencl_config_output,
310  },
311 };
312 
314  .name = "overlay_opencl",
315  .description = NULL_IF_CONFIG_SMALL("Overlay one video on top of another"),
316  .priv_size = sizeof(OverlayOpenCLContext),
317  .priv_class = &overlay_opencl_class,
324  .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
325 };
ff_get_video_buffer
AVFrame * ff_get_video_buffer(AVFilterLink *link, int w, int h)
Request a picture buffer with a specific set of permissions.
Definition: video.c:101
ff_framesync_configure
int ff_framesync_configure(FFFrameSync *fs)
Configure a frame sync structure.
Definition: framesync.c:119
AV_LOG_WARNING
#define AV_LOG_WARNING
Something somehow does not look correct.
Definition: log.h:186
AVPixelFormat
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
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
opt.h
OverlayOpenCLContext::nb_planes
int nb_planes
Definition: vf_overlay_opencl.c:39
CL_SET_KERNEL_ARG
#define CL_SET_KERNEL_ARG(kernel, arg_num, type, arg)
set argument to specific Kernel.
Definition: opencl.h:61
ff_framesync_uninit
void ff_framesync_uninit(FFFrameSync *fs)
Free all memory currently allocated.
Definition: framesync.c:285
AVFILTER_DEFINE_CLASS
AVFILTER_DEFINE_CLASS(overlay_opencl)
FF_FILTER_FLAG_HWFRAME_AWARE
#define FF_FILTER_FLAG_HWFRAME_AWARE
The filter is aware of hardware frames, and any hardware frame context should not be automatically pr...
Definition: internal.h:370
ff_filter_frame
int ff_filter_frame(AVFilterLink *link, AVFrame *frame)
Send a frame of data to the next filter.
Definition: avfilter.c:999
av_pix_fmt_desc_get
const AVPixFmtDescriptor * av_pix_fmt_desc_get(enum AVPixelFormat pix_fmt)
Definition: pixdesc.c:2662
AVBufferRef::data
uint8_t * data
The data buffer.
Definition: buffer.h:90
ff_framesync_get_frame
int ff_framesync_get_frame(FFFrameSync *fs, unsigned in, AVFrame **rframe, unsigned get)
Get the current frame in an input.
Definition: framesync.c:248
overlay_opencl_load
static int overlay_opencl_load(AVFilterContext *avctx, enum AVPixelFormat main_format, enum AVPixelFormat overlay_format)
Definition: vf_overlay_opencl.c:48
output
filter_frame For filters that do not use the this method is called when a frame is pushed to the filter s input It can be called at any time except in a reentrant way If the input frame is enough to produce output
Definition: filter_design.txt:225
OverlayOpenCLContext::ocf
OpenCLFilterContext ocf
Definition: vf_overlay_opencl.c:31
av_frame_free
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:111
OverlayOpenCLContext::command_queue
cl_command_queue command_queue
Definition: vf_overlay_opencl.c:35
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:325
pixdesc.h
opencl.h
AVOption
AVOption.
Definition: opt.h:251
overlay_opencl_activate
static int overlay_opencl_activate(AVFilterContext *avctx)
Definition: vf_overlay_opencl.c:249
ff_opencl_filter_load_program
int ff_opencl_filter_load_program(AVFilterContext *avctx, const char **program_source_array, int nb_strings)
Load a new OpenCL program from strings in memory.
Definition: opencl.c:156
FFMAX
#define FFMAX(a, b)
Definition: macros.h:47
ff_opencl_source_overlay
const char * ff_opencl_source_overlay
AVFilter::name
const char * name
Filter name.
Definition: avfilter.h:175
FFFrameSync
Frame sync structure.
Definition: framesync.h:146
OverlayOpenCLContext
Definition: vf_overlay_opencl.c:30
video.h
OverlayOpenCLContext::fs
FFFrameSync fs
Definition: vf_overlay_opencl.c:37
AVFrame::data
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:346
init
static int init
Definition: av_tx.c:47
ff_opencl_filter_work_size_from_image
int ff_opencl_filter_work_size_from_image(AVFilterContext *avctx, size_t *work_size, AVFrame *frame, int plane, int block_alignment)
Find the work size needed needed for a given plane of an image.
Definition: opencl.c:263
AVFilterContext::priv
void * priv
private data for use by the filter
Definition: avfilter.h:423
OverlayOpenCLContext::initialised
int initialised
Definition: vf_overlay_opencl.c:33
fail
#define fail()
Definition: checkasm.h:131
ff_opencl_filter_config_output
int ff_opencl_filter_config_output(AVFilterLink *outlink)
Create a suitable hardware frames context for the output.
Definition: opencl.c:81
AVFilterPad
A filter pad used for either input or output.
Definition: internal.h:49
OverlayOpenCLContext::y_position
int y_position
Definition: vf_overlay_opencl.c:45
OverlayOpenCLContext::alpha_separate
int alpha_separate
Definition: vf_overlay_opencl.c:42
AV_LOG_ERROR
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:180
av_cold
#define av_cold
Definition: attributes.h:90
overlay_opencl_blend
static int overlay_opencl_blend(FFFrameSync *fs)
Definition: vf_overlay_opencl.c:119
overlay_opencl_inputs
static const AVFilterPad overlay_opencl_inputs[]
Definition: vf_overlay_opencl.c:292
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
overlay_opencl_init
static av_cold int overlay_opencl_init(AVFilterContext *avctx)
Definition: vf_overlay_opencl.c:240
AVPixFmtDescriptor::log2_chroma_w
uint8_t log2_chroma_w
Amount to shift the luma width right to find the chroma width.
Definition: pixdesc.h:80
FILTER_INPUTS
#define FILTER_INPUTS(array)
Definition: internal.h:190
ff_vf_overlay_opencl
const AVFilter ff_vf_overlay_opencl
Definition: vf_overlay_opencl.c:313
NULL
#define NULL
Definition: coverity.c:32
AVHWFramesContext::sw_format
enum AVPixelFormat sw_format
The pixel format identifying the actual data layout of the hardware frames.
Definition: hwcontext.h:222
av_frame_copy_props
int av_frame_copy_props(AVFrame *dst, const AVFrame *src)
Copy only "metadata" fields from src to dst.
Definition: frame.c:596
fs
#define fs(width, name, subs,...)
Definition: cbs_vp9.c:258
AVPixFmtDescriptor::nb_components
uint8_t nb_components
The number of components each pixel has, (1-4)
Definition: pixdesc.h:71
activate
filter_frame For filters that do not use the activate() callback
OFFSET
#define OFFSET(x)
Definition: vf_overlay_opencl.c:280
AVComponentDescriptor::plane
int plane
Which of the 4 planes contains the component.
Definition: pixdesc.h:34
AV_PIX_FMT_OPENCL
@ AV_PIX_FMT_OPENCL
Hardware surfaces for OpenCL.
Definition: pixfmt.h:327
OverlayOpenCLContext::x_position
int x_position
Definition: vf_overlay_opencl.c:44
source
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 source
Definition: filter_design.txt:255
overlay_opencl_config_output
static int overlay_opencl_config_output(AVFilterLink *outlink)
Definition: vf_overlay_opencl.c:223
NULL_IF_CONFIG_SMALL
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification.
Definition: internal.h:117
ff_framesync_init_dualinput
int ff_framesync_init_dualinput(FFFrameSync *fs, AVFilterContext *parent)
Initialize a frame sync structure for dualinput.
Definition: framesync.c:353
OverlayOpenCLContext::y_subsample
int y_subsample
Definition: vf_overlay_opencl.c:41
opencl_source.h
ff_opencl_filter_config_input
int ff_opencl_filter_config_input(AVFilterLink *inlink)
Check that the input link contains a suitable hardware frames context and extract the device from it.
Definition: opencl.c:45
internal.h
FILTER_SINGLE_PIXFMT
#define FILTER_SINGLE_PIXFMT(pix_fmt_)
Definition: internal.h:180
log.h
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:269
overlay_opencl_options
static const AVOption overlay_opencl_options[]
Definition: vf_overlay_opencl.c:282
AVFilterPad::name
const char * name
Pad name.
Definition: internal.h:55
FLAGS
#define FLAGS
Definition: vf_overlay_opencl.c:281
OverlayOpenCLContext::kernel
cl_kernel kernel
Definition: vf_overlay_opencl.c:34
AVFilter
Filter definition.
Definition: avfilter.h:171
AVHWFramesContext
This struct describes a set or pool of "hardware" frames (i.e.
Definition: hwcontext.h:124
ff_opencl_filter_init
int ff_opencl_filter_init(AVFilterContext *avctx)
Initialise an OpenCL filter context.
Definition: opencl.c:132
OverlayOpenCLContext::x_subsample
int x_subsample
Definition: vf_overlay_opencl.c:40
overlay_opencl_uninit
static av_cold void overlay_opencl_uninit(AVFilterContext *avctx)
Definition: vf_overlay_opencl.c:256
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:659
framesync.h
AV_OPT_TYPE_INT
@ AV_OPT_TYPE_INT
Definition: opt.h:225
avfilter.h
AVPixFmtDescriptor::comp
AVComponentDescriptor comp[4]
Parameters that describe how pixels are packed.
Definition: pixdesc.h:105
OpenCLFilterContext
Definition: opencl.h:36
ff_opencl_filter_uninit
void ff_opencl_filter_uninit(AVFilterContext *avctx)
Uninitialise an OpenCL filter context.
Definition: opencl.c:141
AVFilterContext
An instance of a filter.
Definition: avfilter.h:408
AVMEDIA_TYPE_VIDEO
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
AVPixFmtDescriptor
Descriptor that unambiguously describes how the bits of a pixel are stored in the up to 4 data planes...
Definition: pixdesc.h:69
FILTER_OUTPUTS
#define FILTER_OUTPUTS(array)
Definition: internal.h:191
av_log
#define av_log(a,...)
Definition: tableprint_vlc.h:27
CL_FAIL_ON_ERROR
#define CL_FAIL_ON_ERROR(errcode,...)
A helper macro to handle OpenCL errors.
Definition: opencl.h:74
uninit
static av_cold int uninit(AVCodecContext *avctx)
Definition: crystalhd.c:285
ff_framesync_activate
int ff_framesync_activate(FFFrameSync *fs)
Examine the frames in the filter's input and try to produce output.
Definition: framesync.c:336
overlay_opencl_outputs
static const AVFilterPad overlay_opencl_outputs[]
Definition: vf_overlay_opencl.c:305
AVPixFmtDescriptor::log2_chroma_h
uint8_t log2_chroma_h
Amount to shift the luma height right to find the chroma height.
Definition: pixdesc.h:89
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:2582
AVFilterContext::outputs
AVFilterLink ** outputs
array of pointers to output links
Definition: avfilter.h:420