FFmpeg
vf_neighbor_opencl.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018 Danil Iashchenko
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * FFmpeg is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20 
21 #include "config_components.h"
22 
23 #include "libavutil/common.h"
24 #include "libavutil/imgutils.h"
25 #include "libavutil/opt.h"
26 #include "libavutil/pixdesc.h"
27 #include "libavutil/avstring.h"
28 
29 
30 #include "avfilter.h"
31 #include "internal.h"
32 #include "opencl.h"
33 #include "opencl_source.h"
34 #include "video.h"
35 
36 typedef struct NeighborOpenCLContext {
38 
40  cl_kernel kernel;
41  cl_command_queue command_queue;
42 
43  char *matrix_str[4];
44 
45  cl_float threshold[4];
46  cl_int coordinates;
47  cl_mem coord;
48 
50 
52 {
53  NeighborOpenCLContext *ctx = avctx->priv;
54  const char *kernel_name;
55  cl_int cle;
56  int err;
57 
59  if (err < 0)
60  goto fail;
61 
62  ctx->command_queue = clCreateCommandQueue(ctx->ocf.hwctx->context,
63  ctx->ocf.hwctx->device_id,
64  0, &cle);
65  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create OpenCL "
66  "command queue %d.\n", cle);
67 
68  if (!strcmp(avctx->filter->name, "erosion_opencl")){
69  kernel_name = "erosion_global";
70  } else if (!strcmp(avctx->filter->name, "dilation_opencl")){
71  kernel_name = "dilation_global";
72  }
73  ctx->kernel = clCreateKernel(ctx->ocf.program, kernel_name, &cle);
74  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create "
75  "kernel %d.\n", cle);
76 
77  ctx->initialised = 1;
78  return 0;
79 
80 fail:
81  if (ctx->command_queue)
82  clReleaseCommandQueue(ctx->command_queue);
83  if (ctx->kernel)
84  clReleaseKernel(ctx->kernel);
85  return err;
86 }
87 
89 {
90  NeighborOpenCLContext *ctx = avctx->priv;
91  cl_int matrix[9];
92  cl_mem buffer;
93  cl_int cle;
94  int i;
95 
96  for (i = 0; i < 4; i++) {
97  ctx->threshold[i] /= 255.0;
98  }
99 
100  matrix[4] = 0;
101  for (i = 0; i < 8; i++) {
102  if (ctx->coordinates & (1 << i)) {
103  matrix[i > 3 ? i + 1: i] = 1;
104  }
105  }
106  buffer = clCreateBuffer(ctx->ocf.hwctx->context,
107  CL_MEM_READ_ONLY |
108  CL_MEM_COPY_HOST_PTR |
109  CL_MEM_HOST_NO_ACCESS,
110  9 * sizeof(cl_int), matrix, &cle);
111  if (!buffer) {
112  av_log(avctx, AV_LOG_ERROR, "Failed to create matrix buffer: "
113  "%d.\n", cle);
114  return AVERROR(EIO);
115  }
116  ctx->coord = buffer;
117 
118  return 0;
119 }
120 
121 
123 {
124  AVFilterContext *avctx = inlink->dst;
125  AVFilterLink *outlink = avctx->outputs[0];
126  NeighborOpenCLContext *ctx = avctx->priv;
127  AVFrame *output = NULL;
128  cl_int cle;
129  size_t global_work[2];
130  cl_mem src, dst;
131  int err, p;
132  size_t origin[3] = {0, 0, 0};
133  size_t region[3] = {0, 0, 1};
134 
135  av_log(ctx, AV_LOG_DEBUG, "Filter input: %s, %ux%u (%"PRId64").\n",
136  av_get_pix_fmt_name(input->format),
137  input->width, input->height, input->pts);
138 
139  if (!input->hw_frames_ctx)
140  return AVERROR(EINVAL);
141 
142  if (!ctx->initialised) {
143  err = neighbor_opencl_init(avctx);
144  if (err < 0)
145  goto fail;
146 
148  if (err < 0)
149  goto fail;
150 
151  }
152 
153  output = ff_get_video_buffer(outlink, outlink->w, outlink->h);
154  if (!output) {
155  err = AVERROR(ENOMEM);
156  goto fail;
157  }
158 
159  for (p = 0; p < FF_ARRAY_ELEMS(output->data); p++) {
160  src = (cl_mem) input->data[p];
161  dst = (cl_mem)output->data[p];
162 
163  if (!dst)
164  break;
165 
166  if (ctx->threshold[p] == 0) {
167  err = ff_opencl_filter_work_size_from_image(avctx, region, output, p, 0);
168  if (err < 0)
169  goto fail;
170 
171  cle = clEnqueueCopyImage(ctx->command_queue, src, dst,
172  origin, origin, region, 0, NULL, NULL);
173  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to copy plane %d: %d.\n",
174  p, cle);
175  } else {
176  CL_SET_KERNEL_ARG(ctx->kernel, 0, cl_mem, &dst);
177  CL_SET_KERNEL_ARG(ctx->kernel, 1, cl_mem, &src);
178  CL_SET_KERNEL_ARG(ctx->kernel, 2, cl_float, &ctx->threshold[p]);
179  CL_SET_KERNEL_ARG(ctx->kernel, 3, cl_mem, &ctx->coord);
180 
181  err = ff_opencl_filter_work_size_from_image(avctx, global_work, output, p, 0);
182  if (err < 0)
183  goto fail;
184 
185  av_log(avctx, AV_LOG_DEBUG, "Run kernel on plane %d "
186  "(%"SIZE_SPECIFIER"x%"SIZE_SPECIFIER").\n",
187  p, global_work[0], global_work[1]);
188 
189  cle = clEnqueueNDRangeKernel(ctx->command_queue, ctx->kernel, 2, NULL,
190  global_work, NULL,
191  0, NULL, NULL);
192  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to enqueue "
193  "kernel: %d.\n", cle);
194  }
195  }
196 
197  cle = clFinish(ctx->command_queue);
198  CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to finish command queue: %d.\n", cle);
199 
201  if (err < 0)
202  goto fail;
203 
205 
206  av_log(ctx, AV_LOG_DEBUG, "Filter output: %s, %ux%u (%"PRId64").\n",
207  av_get_pix_fmt_name(output->format),
208  output->width, output->height, output->pts);
209 
210  return ff_filter_frame(outlink, output);
211 
212 fail:
213  clFinish(ctx->command_queue);
216  return err;
217 }
218 
220 {
221  NeighborOpenCLContext *ctx = avctx->priv;
222  cl_int cle;
223 
224  clReleaseMemObject(ctx->coord);
225 
226  if (ctx->kernel) {
227  cle = clReleaseKernel(ctx->kernel);
228  if (cle != CL_SUCCESS)
229  av_log(avctx, AV_LOG_ERROR, "Failed to release "
230  "kernel: %d.\n", cle);
231  }
232 
233  if (ctx->command_queue) {
234  cle = clReleaseCommandQueue(ctx->command_queue);
235  if (cle != CL_SUCCESS)
236  av_log(avctx, AV_LOG_ERROR, "Failed to release "
237  "command queue: %d.\n", cle);
238  }
239 
241 }
242 
244  {
245  .name = "default",
246  .type = AVMEDIA_TYPE_VIDEO,
247  .filter_frame = &neighbor_opencl_filter_frame,
248  .config_props = &ff_opencl_filter_config_input,
249  },
250 };
251 
253  {
254  .name = "default",
255  .type = AVMEDIA_TYPE_VIDEO,
256  .config_props = &ff_opencl_filter_config_output,
257  },
258 };
259 
260 #define OFFSET(x) offsetof(NeighborOpenCLContext, x)
261 #define FLAGS (AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
262 
263 #if CONFIG_EROSION_OPENCL_FILTER
264 
265 static const AVOption erosion_opencl_options[] = {
266  { "threshold0", "set threshold for 1st plane", OFFSET(threshold[0]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
267  { "threshold1", "set threshold for 2nd plane", OFFSET(threshold[1]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
268  { "threshold2", "set threshold for 3rd plane", OFFSET(threshold[2]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
269  { "threshold3", "set threshold for 4th plane", OFFSET(threshold[3]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
270  { "coordinates", "set coordinates", OFFSET(coordinates), AV_OPT_TYPE_INT, {.i64=255}, 0, 255, FLAGS },
271  { NULL }
272 };
273 
274 AVFILTER_DEFINE_CLASS(erosion_opencl);
275 
277  .name = "erosion_opencl",
278  .description = NULL_IF_CONFIG_SMALL("Apply erosion effect"),
279  .priv_size = sizeof(NeighborOpenCLContext),
280  .priv_class = &erosion_opencl_class,
286  .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
287 };
288 
289 #endif /* CONFIG_EROSION_OPENCL_FILTER */
290 
291 #if CONFIG_DILATION_OPENCL_FILTER
292 
293 static const AVOption dilation_opencl_options[] = {
294  { "threshold0", "set threshold for 1st plane", OFFSET(threshold[0]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
295  { "threshold1", "set threshold for 2nd plane", OFFSET(threshold[1]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
296  { "threshold2", "set threshold for 3rd plane", OFFSET(threshold[2]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
297  { "threshold3", "set threshold for 4th plane", OFFSET(threshold[3]), AV_OPT_TYPE_FLOAT, {.dbl=65535.0}, 0.0, 65535, FLAGS },
298  { "coordinates", "set coordinates", OFFSET(coordinates), AV_OPT_TYPE_INT, {.i64=255}, 0, 255, FLAGS },
299  { NULL }
300 };
301 
302 AVFILTER_DEFINE_CLASS(dilation_opencl);
303 
305  .name = "dilation_opencl",
306  .description = NULL_IF_CONFIG_SMALL("Apply dilation effect"),
307  .priv_size = sizeof(NeighborOpenCLContext),
308  .priv_class = &dilation_opencl_class,
314  .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
315  .flags = AVFILTER_FLAG_HWDEVICE,
316 };
317 
318 #endif /* CONFIG_DILATION_OPENCL_FILTER */
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:112
neighbor_opencl_inputs
static const AVFilterPad neighbor_opencl_inputs[]
Definition: vf_neighbor_opencl.c:243
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
CL_SET_KERNEL_ARG
#define CL_SET_KERNEL_ARG(kernel, arg_num, type, arg)
set argument to specific Kernel.
Definition: opencl.h:61
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:351
ff_filter_frame
int ff_filter_frame(AVFilterLink *link, AVFrame *frame)
Send a frame of data to the next filter.
Definition: avfilter.c:1018
matrix
Definition: vc1dsp.c:42
neighbor_opencl_uninit
static av_cold void neighbor_opencl_uninit(AVFilterContext *avctx)
Definition: vf_neighbor_opencl.c:219
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
inlink
The exact code depends on how similar the blocks are and how related they are to the and needs to apply these operations to the correct inlink or outlink if there are several Macros are available to factor that when no extra processing is inlink
Definition: filter_design.txt:212
av_frame_free
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:88
test::height
int height
Definition: vc1dsp.c:39
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:340
pixdesc.h
opencl.h
AVOption
AVOption.
Definition: opt.h:346
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
neighbor_opencl_init
static int neighbor_opencl_init(AVFilterContext *avctx)
Definition: vf_neighbor_opencl.c:51
AVFilter::name
const char * name
Filter name.
Definition: avfilter.h:170
ff_vf_dilation_opencl
const AVFilter ff_vf_dilation_opencl
video.h
NeighborOpenCLContext::ocf
OpenCLFilterContext ocf
Definition: vf_neighbor_opencl.c:37
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:422
fail
#define fail()
Definition: checkasm.h:179
neighbor_opencl_make_filter_params
static int neighbor_opencl_make_filter_params(AVFilterContext *avctx)
Definition: vf_neighbor_opencl.c:88
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:33
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
av_cold
#define av_cold
Definition: attributes.h:90
NeighborOpenCLContext::initialised
int initialised
Definition: vf_neighbor_opencl.c:39
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
NeighborOpenCLContext::kernel
cl_kernel kernel
Definition: vf_neighbor_opencl.c:40
FILTER_INPUTS
#define FILTER_INPUTS(array)
Definition: internal.h:182
OFFSET
#define OFFSET(x)
Definition: vf_neighbor_opencl.c:260
neighbor_opencl_filter_frame
static int neighbor_opencl_filter_frame(AVFilterLink *inlink, AVFrame *input)
Definition: vf_neighbor_opencl.c:122
NULL
#define NULL
Definition: coverity.c:32
NeighborOpenCLContext::threshold
cl_float threshold[4]
Definition: vf_neighbor_opencl.c:45
NeighborOpenCLContext
Definition: vf_neighbor_opencl.c:36
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:637
AV_PIX_FMT_OPENCL
@ AV_PIX_FMT_OPENCL
Hardware surfaces for OpenCL.
Definition: pixfmt.h:358
NeighborOpenCLContext::command_queue
cl_command_queue command_queue
Definition: vf_neighbor_opencl.c:41
ff_vf_erosion_opencl
const AVFilter ff_vf_erosion_opencl
test::width
int width
Definition: vc1dsp.c:38
init
int(* init)(AVBSFContext *ctx)
Definition: dts2pts.c:365
neighbor_opencl_outputs
static const AVFilterPad neighbor_opencl_outputs[]
Definition: vf_neighbor_opencl.c:252
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:106
AVFILTER_FLAG_HWDEVICE
#define AVFILTER_FLAG_HWDEVICE
The filter can create hardware frames using AVFilterContext.hw_device_ctx.
Definition: avfilter.h:138
opencl_source.h
input
and forward the test the status of outputs and forward it to the corresponding return FFERROR_NOT_READY If the filters stores internally one or a few frame for some input
Definition: filter_design.txt:172
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
AVFILTER_DEFINE_CLASS
#define AVFILTER_DEFINE_CLASS(fname)
Definition: internal.h:323
AV_OPT_TYPE_FLOAT
@ AV_OPT_TYPE_FLOAT
Definition: opt.h:238
FILTER_SINGLE_PIXFMT
#define FILTER_SINGLE_PIXFMT(pix_fmt_)
Definition: internal.h:172
uninit
static void uninit(AVBSFContext *ctx)
Definition: pcm_rechunk.c:68
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:255
common.h
FLAGS
#define FLAGS
Definition: vf_neighbor_opencl.c:261
AVFilterPad::name
const char * name
Pad name.
Definition: internal.h:39
AVFilter
Filter definition.
Definition: avfilter.h:166
ff_opencl_filter_init
int ff_opencl_filter_init(AVFilterContext *avctx)
Initialise an OpenCL filter context.
Definition: opencl.c:132
SIZE_SPECIFIER
#define SIZE_SPECIFIER
Definition: internal.h:141
buffer
the frame and frame reference mechanism is intended to as much as expensive copies of that data while still allowing the filters to produce correct results The data is stored in buffers represented by AVFrame structures Several references can point to the same frame buffer
Definition: filter_design.txt:49
AV_OPT_TYPE_INT
@ AV_OPT_TYPE_INT
Definition: opt.h:235
avfilter.h
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
NeighborOpenCLContext::coordinates
cl_int coordinates
Definition: vf_neighbor_opencl.c:46
AVFilterContext
An instance of a filter.
Definition: avfilter.h:407
AVMEDIA_TYPE_VIDEO
@ AVMEDIA_TYPE_VIDEO
Definition: avutil.h:201
ff_source_neighbor_cl
const char * ff_source_neighbor_cl
FILTER_OUTPUTS
#define FILTER_OUTPUTS(array)
Definition: internal.h:183
src
INIT_CLIP pixel * src
Definition: h264pred_template.c:418
imgutils.h
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
NeighborOpenCLContext::coord
cl_mem coord
Definition: vf_neighbor_opencl.c:47
avstring.h
AVFilterContext::filter
const AVFilter * filter
the AVFilter of which this is an instance
Definition: avfilter.h:410
NeighborOpenCLContext::matrix_str
char * matrix_str[4]
Definition: vf_neighbor_opencl.c:43
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:2882
AVFilterContext::outputs
AVFilterLink ** outputs
array of pointers to output links
Definition: avfilter.h:419