aboutsummaryrefslogtreecommitdiffstats
path: root/libavfilter/vf_convolution_opencl.c
blob: 3871ddcff4dc4c4c4956f302e43305e385f66dc6 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
/*
 * Copyright (c) 2018 Danil Iashchenko
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/common.h"
#include "libavutil/imgutils.h"
#include "libavutil/mem.h"
#include "libavutil/opt.h"
#include "libavutil/pixdesc.h"
#include "libavutil/avstring.h"


#include "avfilter.h"
#include "internal.h"
#include "opencl.h"
#include "opencl_source.h"
#include "video.h"

typedef struct ConvolutionOpenCLContext {
    OpenCLFilterContext ocf;

    int              initialised;
    cl_kernel        kernel;
    cl_command_queue command_queue;

    char *matrix_str[4];

    cl_mem matrix[4];
    cl_int matrix_sizes[4];
    cl_int dims[4];
    cl_float rdivs[4];
    cl_float biases[4];

    cl_int planes;
    cl_float scale;
    cl_float delta;

} ConvolutionOpenCLContext;

static int convolution_opencl_init(AVFilterContext *avctx)
{
    ConvolutionOpenCLContext *ctx = avctx->priv;
    const char *kernel_name;
    cl_int cle;
    int err;

    err = ff_opencl_filter_load_program(avctx, &ff_opencl_source_convolution, 1);
    if (err < 0)
        goto fail;

    ctx->command_queue = clCreateCommandQueue(ctx->ocf.hwctx->context,
                                              ctx->ocf.hwctx->device_id,
                                              0, &cle);
    CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create OpenCL "
                     "command queue %d.\n", cle);

    if (!strcmp(avctx->filter->name, "convolution_opencl")) {
        kernel_name = "convolution_global";
    } else if (!strcmp(avctx->filter->name, "sobel_opencl")) {
        kernel_name = "sobel_global";
    } else if (!strcmp(avctx->filter->name, "prewitt_opencl")){
        kernel_name = "prewitt_global";
    } else if (!strcmp(avctx->filter->name, "roberts_opencl")){
        kernel_name = "roberts_global";
    }
    ctx->kernel = clCreateKernel(ctx->ocf.program, kernel_name, &cle);
    CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create "
                     "kernel %d.\n", cle);

    ctx->initialised = 1;
    return 0;

fail:
    if (ctx->command_queue)
        clReleaseCommandQueue(ctx->command_queue);
    if (ctx->kernel)
        clReleaseKernel(ctx->kernel);
    return err;
}



static int convolution_opencl_make_filter_params(AVFilterContext *avctx)
{
    ConvolutionOpenCLContext *ctx = avctx->priv;
    float *matrix = NULL;
    size_t matrix_bytes;
    cl_mem buffer;
    cl_int cle;
    int i, j;
    int sscanf_err;
    char *p, *arg, *saveptr = NULL;
    float input_matrix[4][49];

    for (i = 0; i < 4; i++) {
        ctx->biases[i] = ctx->biases[i] / 255.0;
    }

    for (i = 0; i < 4; i++) {
        p = ctx->matrix_str[i];
        while (ctx->matrix_sizes[i] < 49) {
            arg = av_strtok(p, " ", &saveptr);
            if (!arg) {
                break;
            }
            p = NULL;
            sscanf_err = sscanf(arg, "%f", &input_matrix[i][ctx->matrix_sizes[i]]);
            if (sscanf_err != 1) {
                av_log(ctx, AV_LOG_ERROR, "Matrix is sequence of 9, 25 or 49 signed numbers\n");
                return AVERROR(EINVAL);
            }
            ctx->matrix_sizes[i]++;
        }
        if (ctx->matrix_sizes[i] == 9) {
            ctx->dims[i] = 3;
        } else if (ctx->matrix_sizes[i] == 25) {
            ctx->dims[i] = 5;
        } else if (ctx->matrix_sizes[i] == 49) {
            ctx->dims[i] = 7;
        } else {
            av_log(ctx, AV_LOG_ERROR, "Invalid matrix size:%d\n", ctx->matrix_sizes[i]);
            return AVERROR(EINVAL);
        }

    }

    for (j = 0; j < 4; j++) {
        matrix_bytes = sizeof(float)*ctx->matrix_sizes[j];
        matrix = av_malloc(matrix_bytes);
        if (!matrix) {
            av_freep(&matrix);
            return AVERROR(ENOMEM);
        }

        for (i = 0; i < ctx->matrix_sizes[j]; i++)
            matrix[i] = input_matrix[j][i];

        buffer = clCreateBuffer(ctx->ocf.hwctx->context,
                                CL_MEM_READ_ONLY |
                                CL_MEM_COPY_HOST_PTR |
                                CL_MEM_HOST_NO_ACCESS,
                                matrix_bytes, matrix, &cle);
        if (!buffer) {
            av_log(avctx, AV_LOG_ERROR, "Failed to create matrix buffer: "
                   "%d.\n", cle);
            av_freep(&matrix);
            return AVERROR(EIO);
        }
        ctx->matrix[j] = buffer;
        av_freep(&matrix);
    }

    return 0;
}

static int convolution_opencl_filter_frame(AVFilterLink *inlink, AVFrame *input)
{
    AVFilterContext *avctx = inlink->dst;
    AVFilterLink *outlink = avctx->outputs[0];
    ConvolutionOpenCLContext *ctx = avctx->priv;
    AVFrame *output = NULL;
    cl_int cle;
    size_t global_work[2];
    cl_mem src, dst;
    int err, p;
    size_t origin[3] = {0, 0, 0};
    size_t region[3] = {0, 0, 1};

    av_log(ctx, AV_LOG_DEBUG, "Filter input: %s, %ux%u (%"PRId64").\n",
           av_get_pix_fmt_name(input->format),
           input->width, input->height, input->pts);

    if (!input->hw_frames_ctx)
        return AVERROR(EINVAL);

    if (!ctx->initialised) {
        err = convolution_opencl_init(avctx);
        if (err < 0)
            goto fail;

        if (!strcmp(avctx->filter->name, "convolution_opencl")) {
            err = convolution_opencl_make_filter_params(avctx);
            if (err < 0)
                goto fail;
        } else {
            ctx->delta /= 255.0;
        }

    }

    output = ff_get_video_buffer(outlink, outlink->w, outlink->h);
    if (!output) {
        err = AVERROR(ENOMEM);
        goto fail;
    }

    for (p = 0; p < FF_ARRAY_ELEMS(output->data); p++) {
        src = (cl_mem) input->data[p];
        dst = (cl_mem)output->data[p];

        if (!dst)
            break;

        if (!strcmp(avctx->filter->name, "convolution_opencl")) {
            CL_SET_KERNEL_ARG(ctx->kernel, 0, cl_mem,   &dst);
            CL_SET_KERNEL_ARG(ctx->kernel, 1, cl_mem,   &src);
            CL_SET_KERNEL_ARG(ctx->kernel, 2, cl_int,   &ctx->dims[p]);
            CL_SET_KERNEL_ARG(ctx->kernel, 3, cl_mem,   &ctx->matrix[p]);
            CL_SET_KERNEL_ARG(ctx->kernel, 4, cl_float, &ctx->rdivs[p]);
            CL_SET_KERNEL_ARG(ctx->kernel, 5, cl_float, &ctx->biases[p]);

            err = ff_opencl_filter_work_size_from_image(avctx, global_work, output, p, 0);
            if (err < 0)
                goto fail;

            av_log(avctx, AV_LOG_DEBUG, "Run kernel on plane %d "
                   "(%"SIZE_SPECIFIER"x%"SIZE_SPECIFIER").\n",
                   p, global_work[0], global_work[1]);

            cle = clEnqueueNDRangeKernel(ctx->command_queue, ctx->kernel, 2, NULL,
                                         global_work, NULL,
                                         0, NULL, NULL);
            CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to enqueue "
                             "kernel: %d.\n", cle);
        } else {
            if (!(ctx->planes & (1 << p))) {
                err = ff_opencl_filter_work_size_from_image(avctx, region, output, p, 0);
                if (err < 0)
                    goto fail;

                cle = clEnqueueCopyImage(ctx->command_queue, src, dst,
                                         origin, origin, region, 0, NULL, NULL);
                CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to copy plane %d: %d.\n",
                                 p, cle);
            } else {
                CL_SET_KERNEL_ARG(ctx->kernel, 0, cl_mem,   &dst);
                CL_SET_KERNEL_ARG(ctx->kernel, 1, cl_mem,   &src);
                CL_SET_KERNEL_ARG(ctx->kernel, 2, cl_float, &ctx->scale);
                CL_SET_KERNEL_ARG(ctx->kernel, 3, cl_float, &ctx->delta);

                err = ff_opencl_filter_work_size_from_image(avctx, global_work, output, p, 0);
                if (err < 0)
                    goto fail;

                av_log(avctx, AV_LOG_DEBUG, "Run kernel on plane %d "
                       "(%"SIZE_SPECIFIER"x%"SIZE_SPECIFIER").\n",
                       p, global_work[0], global_work[1]);

                cle = clEnqueueNDRangeKernel(ctx->command_queue, ctx->kernel, 2, NULL,
                                                         global_work, NULL,
                                                         0, NULL, NULL);
                CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to enqueue "
                                 "kernel: %d.\n", cle);
            }
        }
    }

    cle = clFinish(ctx->command_queue);
    CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to finish command queue: %d.\n", cle);

    err = av_frame_copy_props(output, input);
    if (err < 0)
        goto fail;

    av_frame_free(&input);

    av_log(ctx, AV_LOG_DEBUG, "Filter output: %s, %ux%u (%"PRId64").\n",
           av_get_pix_fmt_name(output->format),
           output->width, output->height, output->pts);

    return ff_filter_frame(outlink, output);

fail:
    clFinish(ctx->command_queue);
    av_frame_free(&input);
    av_frame_free(&output);
    return err;
}

static av_cold void convolution_opencl_uninit(AVFilterContext *avctx)
{
    ConvolutionOpenCLContext *ctx = avctx->priv;
    cl_int cle;
    int i;

    for (i = 0; i < 4; i++) {
        clReleaseMemObject(ctx->matrix[i]);
    }

    if (ctx->kernel) {
        cle = clReleaseKernel(ctx->kernel);
        if (cle != CL_SUCCESS)
            av_log(avctx, AV_LOG_ERROR, "Failed to release "
                   "kernel: %d.\n", cle);
    }

    if (ctx->command_queue) {
        cle = clReleaseCommandQueue(ctx->command_queue);
        if (cle != CL_SUCCESS)
            av_log(avctx, AV_LOG_ERROR, "Failed to release "
                   "command queue: %d.\n", cle);
    }

    ff_opencl_filter_uninit(avctx);
}

static const AVFilterPad convolution_opencl_inputs[] = {
    {
        .name         = "default",
        .type         = AVMEDIA_TYPE_VIDEO,
        .filter_frame = &convolution_opencl_filter_frame,
        .config_props = &ff_opencl_filter_config_input,
    },
};

static const AVFilterPad convolution_opencl_outputs[] = {
    {
        .name         = "default",
        .type         = AVMEDIA_TYPE_VIDEO,
        .config_props = &ff_opencl_filter_config_output,
    },
};

#define OFFSET(x) offsetof(ConvolutionOpenCLContext, x)
#define FLAGS (AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)

#if CONFIG_CONVOLUTION_OPENCL_FILTER

static const AVOption convolution_opencl_options[] = {
    { "0m", "set matrix for 2nd plane", OFFSET(matrix_str[0]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
    { "1m", "set matrix for 2nd plane", OFFSET(matrix_str[1]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
    { "2m", "set matrix for 3rd plane", OFFSET(matrix_str[2]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
    { "3m", "set matrix for 4th plane", OFFSET(matrix_str[3]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
    { "0rdiv", "set rdiv for 1nd plane", OFFSET(rdivs[0]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
    { "1rdiv", "set rdiv for 2nd plane", OFFSET(rdivs[1]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
    { "2rdiv", "set rdiv for 3rd plane", OFFSET(rdivs[2]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
    { "3rdiv", "set rdiv for 4th plane", OFFSET(rdivs[3]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
    { "0bias", "set bias for 1st plane", OFFSET(biases[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
    { "1bias", "set bias for 2nd plane", OFFSET(biases[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
    { "2bias", "set bias for 3rd plane", OFFSET(biases[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
    { "3bias", "set bias for 4th plane", OFFSET(biases[3]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
    { NULL }
};

AVFILTER_DEFINE_CLASS(convolution_opencl);

const AVFilter ff_vf_convolution_opencl = {
    .name           = "convolution_opencl",
    .description    = NULL_IF_CONFIG_SMALL("Apply convolution mask to input video"),
    .priv_size      = sizeof(ConvolutionOpenCLContext),
    .priv_class     = &convolution_opencl_class,
    .init           = &ff_opencl_filter_init,
    .uninit         = &convolution_opencl_uninit,
    FILTER_INPUTS(convolution_opencl_inputs),
    FILTER_OUTPUTS(convolution_opencl_outputs),
    FILTER_QUERY_FUNC(&ff_opencl_filter_query_formats),
    .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
};

#endif /* CONFIG_CONVOLUTION_OPENCL_FILTER */

#if CONFIG_SOBEL_OPENCL_FILTER

static const AVOption sobel_opencl_options[] = {
    { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT,  {.i64=15}, 0, 15, FLAGS},
    { "scale",  "set scale",            OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0,  65535, FLAGS},
    { "delta",  "set delta",            OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
    { NULL }
};

AVFILTER_DEFINE_CLASS(sobel_opencl);

const AVFilter ff_vf_sobel_opencl = {
    .name           = "sobel_opencl",
    .description    = NULL_IF_CONFIG_SMALL("Apply sobel operator"),
    .priv_size      = sizeof(ConvolutionOpenCLContext),
    .priv_class     = &sobel_opencl_class,
    .init           = &ff_opencl_filter_init,
    .uninit         = &convolution_opencl_uninit,
    FILTER_INPUTS(convolution_opencl_inputs),
    FILTER_OUTPUTS(convolution_opencl_outputs),
    FILTER_QUERY_FUNC(&ff_opencl_filter_query_formats),
    .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
};

#endif /* CONFIG_SOBEL_OPENCL_FILTER */

#if CONFIG_PREWITT_OPENCL_FILTER

static const AVOption prewitt_opencl_options[] = {
    { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT,  {.i64=15}, 0, 15, FLAGS},
    { "scale",  "set scale",            OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0,  65535, FLAGS},
    { "delta",  "set delta",            OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
    { NULL }
};

AVFILTER_DEFINE_CLASS(prewitt_opencl);

const AVFilter ff_vf_prewitt_opencl = {
    .name           = "prewitt_opencl",
    .description    = NULL_IF_CONFIG_SMALL("Apply prewitt operator"),
    .priv_size      = sizeof(ConvolutionOpenCLContext),
    .priv_class     = &prewitt_opencl_class,
    .init           = &ff_opencl_filter_init,
    .uninit         = &convolution_opencl_uninit,
    FILTER_INPUTS(convolution_opencl_inputs),
    FILTER_OUTPUTS(convolution_opencl_outputs),
    FILTER_QUERY_FUNC(&ff_opencl_filter_query_formats),
    .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
};

#endif /* CONFIG_PREWITT_OPENCL_FILTER */

#if CONFIG_ROBERTS_OPENCL_FILTER

static const AVOption roberts_opencl_options[] = {
    { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT,  {.i64=15}, 0, 15, FLAGS},
    { "scale",  "set scale",            OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0,  65535, FLAGS},
    { "delta",  "set delta",            OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
    { NULL }
};

AVFILTER_DEFINE_CLASS(roberts_opencl);

const AVFilter ff_vf_roberts_opencl = {
    .name           = "roberts_opencl",
    .description    = NULL_IF_CONFIG_SMALL("Apply roberts operator"),
    .priv_size      = sizeof(ConvolutionOpenCLContext),
    .priv_class     = &roberts_opencl_class,
    .init           = &ff_opencl_filter_init,
    .uninit         = &convolution_opencl_uninit,
    FILTER_INPUTS(convolution_opencl_inputs),
    FILTER_OUTPUTS(convolution_opencl_outputs),
    FILTER_QUERY_FUNC(&ff_opencl_filter_query_formats),
    .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
};

#endif /* CONFIG_ROBERTS_OPENCL_FILTER */