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/colorspace.h"
20 #include "libavutil/eval.h"
21 #include "libavutil/opt.h"
22 #include "libavutil/imgutils.h"
23 #include "avfilter.h"
24 #include "drawutils.h"
25 #include "formats.h"
26 #include "internal.h"
27 #include "opencl.h"
28 #include "opencl_source.h"
29 #include "video.h"
30
31 static const char *const var_names[] = {
32 "in_w", "iw",
33 "in_h", "ih",
34 "out_w", "ow",
35 "out_h", "oh",
36 "x",
37 "y",
38 "a",
39 "sar",
40 "dar",
41 NULL
42 };
43
44 enum var_name {
45 VAR_IN_W, VAR_IW,
46 VAR_IN_H, VAR_IH,
47 VAR_OUT_W, VAR_OW,
48 VAR_OUT_H, VAR_OH,
49 VAR_X,
50 VAR_Y,
51 VAR_A,
52 VAR_SAR,
53 VAR_DAR,
54 VARS_NB
55 };
56
57 typedef struct PadOpenCLContext {
58 OpenCLFilterContext ocf;
59 int initialized;
60 int is_rgb;
61 int is_packed;
62 int hsub, vsub;
63
64 char *w_expr;
65 char *h_expr;
66 char *x_expr;
67 char *y_expr;
68 AVRational aspect;
69
70 cl_command_queue command_queue;
71 cl_kernel kernel_pad;
72
73 int w, h;
74 int x, y;
75 uint8_t pad_rgba[4];
76 uint8_t pad_color[4];
77 cl_float4 pad_color_float;
78 cl_int2 pad_pos;
79 } PadOpenCLContext;
80
pad_opencl_init(AVFilterContext * avctx,AVFrame * input_frame)81 static int pad_opencl_init(AVFilterContext *avctx, AVFrame *input_frame)
82 {
83 PadOpenCLContext *ctx = avctx->priv;
84 AVHWFramesContext *input_frames_ctx = (AVHWFramesContext *)input_frame->hw_frames_ctx->data;
85 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(input_frames_ctx->sw_format);
86 uint8_t rgba_map[4];
87 cl_int cle;
88 int err;
89
90 ff_fill_rgba_map(rgba_map, input_frames_ctx->sw_format);
91 ctx->is_rgb = !!(desc->flags & AV_PIX_FMT_FLAG_RGB);
92 ctx->is_packed = !(desc->flags & AV_PIX_FMT_FLAG_PLANAR);
93 ctx->hsub = desc->log2_chroma_w;
94 ctx->vsub = desc->log2_chroma_h;
95
96 err = ff_opencl_filter_load_program(avctx, &ff_opencl_source_pad, 1);
97 if (err < 0)
98 goto fail;
99
100 ctx->command_queue = clCreateCommandQueue(
101 ctx->ocf.hwctx->context,
102 ctx->ocf.hwctx->device_id,
103 0,
104 &cle
105 );
106
107 if (ctx->is_rgb) {
108 ctx->pad_color[rgba_map[0]] = ctx->pad_rgba[0];
109 ctx->pad_color[rgba_map[1]] = ctx->pad_rgba[1];
110 ctx->pad_color[rgba_map[2]] = ctx->pad_rgba[2];
111 ctx->pad_color[rgba_map[3]] = ctx->pad_rgba[3];
112 } else {
113 ctx->pad_color[0] = RGB_TO_Y_BT709(ctx->pad_rgba[0], ctx->pad_rgba[1], ctx->pad_rgba[2]);
114 ctx->pad_color[1] = RGB_TO_U_BT709(ctx->pad_rgba[0], ctx->pad_rgba[1], ctx->pad_rgba[2], 0);
115 ctx->pad_color[2] = RGB_TO_V_BT709(ctx->pad_rgba[0], ctx->pad_rgba[1], ctx->pad_rgba[2], 0);
116 ctx->pad_color[3] = ctx->pad_rgba[3];
117 }
118
119 CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create OpenCL command queue %d.\n", cle);
120
121 ctx->kernel_pad = clCreateKernel(ctx->ocf.program, "pad", &cle);
122 CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to create pad kernel: %d.\n", cle);
123
124 for (int i = 0; i < 4; ++i) {
125 ctx->pad_color_float.s[i] = (float)ctx->pad_color[i] / 255.0;
126 }
127
128 ctx->pad_pos.s[0] = ctx->x;
129 ctx->pad_pos.s[1] = ctx->y;
130
131 ctx->initialized = 1;
132 return 0;
133
134 fail:
135 if (ctx->command_queue)
136 clReleaseCommandQueue(ctx->command_queue);
137 if (ctx->kernel_pad)
138 clReleaseKernel(ctx->kernel_pad);
139 return err;
140 }
141
filter_frame(AVFilterLink * link,AVFrame * input_frame)142 static int filter_frame(AVFilterLink *link, AVFrame *input_frame)
143 {
144 AVFilterContext *avctx = link->dst;
145 AVFilterLink *outlink = avctx->outputs[0];
146 PadOpenCLContext *pad_ctx = avctx->priv;
147 AVFrame *output_frame = NULL;
148 int err;
149 cl_int cle;
150 size_t global_work[2];
151 cl_mem src, dst;
152
153 if (!input_frame->hw_frames_ctx)
154 return AVERROR(EINVAL);
155
156 if (!pad_ctx->initialized) {
157 err = pad_opencl_init(avctx, input_frame);
158 if (err < 0)
159 goto fail;
160 }
161
162 output_frame = ff_get_video_buffer(outlink, outlink->w, outlink->h);
163 if (!output_frame) {
164 err = AVERROR(ENOMEM);
165 goto fail;
166 }
167
168 for (int p = 0; p < FF_ARRAY_ELEMS(output_frame->data); p++) {
169 cl_float4 pad_color_float;
170 cl_int2 pad_pos;
171
172 if (pad_ctx->is_packed) {
173 pad_color_float = pad_ctx->pad_color_float;
174 } else {
175 pad_color_float.s[0] = pad_ctx->pad_color_float.s[p];
176 pad_color_float.s[1] = pad_ctx->pad_color_float.s[2];
177 }
178
179 if (p > 0 && p < 3) {
180 pad_pos.s[0] = pad_ctx->pad_pos.s[0] >> pad_ctx->hsub;
181 pad_pos.s[1] = pad_ctx->pad_pos.s[1] >> pad_ctx->vsub;
182 } else {
183 pad_pos.s[0] = pad_ctx->pad_pos.s[0];
184 pad_pos.s[1] = pad_ctx->pad_pos.s[1];
185 }
186
187 src = (cl_mem)input_frame->data[p];
188 dst = (cl_mem)output_frame->data[p];
189
190 if (!dst)
191 break;
192
193 CL_SET_KERNEL_ARG(pad_ctx->kernel_pad, 0, cl_mem, &src);
194 CL_SET_KERNEL_ARG(pad_ctx->kernel_pad, 1, cl_mem, &dst);
195 CL_SET_KERNEL_ARG(pad_ctx->kernel_pad, 2, cl_float4, &pad_color_float);
196 CL_SET_KERNEL_ARG(pad_ctx->kernel_pad, 3, cl_int2, &pad_pos);
197
198 err = ff_opencl_filter_work_size_from_image(avctx, global_work, output_frame, p, 16);
199 if (err < 0)
200 goto fail;
201
202 cle = clEnqueueNDRangeKernel(pad_ctx->command_queue, pad_ctx->kernel_pad, 2, NULL,
203 global_work, NULL, 0, NULL, NULL);
204
205 CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to enqueue pad kernel: %d.\n", cle);
206 }
207
208 // Run queued kernel
209 cle = clFinish(pad_ctx->command_queue);
210 CL_FAIL_ON_ERROR(AVERROR(EIO), "Failed to finish command queue: %d.\n", cle);
211
212 err = av_frame_copy_props(output_frame, input_frame);
213 if (err < 0)
214 goto fail;
215
216 av_frame_free(&input_frame);
217
218 return ff_filter_frame(outlink, output_frame);
219
220 fail:
221 clFinish(pad_ctx->command_queue);
222 av_frame_free(&input_frame);
223 av_frame_free(&output_frame);
224 return err;
225 }
226
pad_opencl_uninit(AVFilterContext * avctx)227 static av_cold void pad_opencl_uninit(AVFilterContext *avctx)
228 {
229 PadOpenCLContext *ctx = avctx->priv;
230 cl_int cle;
231
232 if (ctx->kernel_pad) {
233 cle = clReleaseKernel(ctx->kernel_pad);
234 if (cle != CL_SUCCESS)
235 av_log(avctx, AV_LOG_ERROR, "Failed to release "
236 "kernel: %d.\n", cle);
237 }
238
239 if (ctx->command_queue) {
240 cle = clReleaseCommandQueue(ctx->command_queue);
241 if (cle != CL_SUCCESS)
242 av_log(avctx, AV_LOG_ERROR, "Failed to release "
243 "command queue: %d.\n", cle);
244 }
245
246 ff_opencl_filter_uninit(avctx);
247 }
248
pad_opencl_config_output(AVFilterLink * outlink)249 static int pad_opencl_config_output(AVFilterLink *outlink)
250 {
251 AVFilterContext *avctx = outlink->src;
252 AVFilterLink *inlink = avctx->inputs[0];
253 PadOpenCLContext *ctx = avctx->priv;
254 AVRational adjusted_aspect = ctx->aspect;
255 double var_values[VARS_NB], res;
256 int err, ret;
257 char *expr;
258
259 var_values[VAR_IN_W] = var_values[VAR_IW] = inlink->w;
260 var_values[VAR_IN_H] = var_values[VAR_IH] = inlink->h;
261 var_values[VAR_OUT_W] = var_values[VAR_OW] = NAN;
262 var_values[VAR_OUT_H] = var_values[VAR_OH] = NAN;
263 var_values[VAR_A] = (double) inlink->w / inlink->h;
264 var_values[VAR_SAR] = inlink->sample_aspect_ratio.num ?
265 (double) inlink->sample_aspect_ratio.num / inlink->sample_aspect_ratio.den : 1;
266 var_values[VAR_DAR] = var_values[VAR_A] * var_values[VAR_SAR];
267
268 av_expr_parse_and_eval(&res, (expr = ctx->w_expr),
269 var_names, var_values,
270 NULL, NULL, NULL, NULL, NULL, 0, ctx);
271 ctx->w = var_values[VAR_OUT_W] = var_values[VAR_OW] = res;
272 if ((ret = av_expr_parse_and_eval(&res, (expr = ctx->h_expr),
273 var_names, var_values,
274 NULL, NULL, NULL, NULL, NULL, 0, ctx)) < 0)
275 return ret;
276 ctx->h = var_values[VAR_OUT_H] = var_values[VAR_OH] = res;
277 if (!ctx->h)
278 var_values[VAR_OUT_H] = var_values[VAR_OH] = ctx->h = inlink->h;
279
280 /* evaluate the width again, as it may depend on the evaluated output height */
281 if ((ret = av_expr_parse_and_eval(&res, (expr = ctx->w_expr),
282 var_names, var_values,
283 NULL, NULL, NULL, NULL, NULL, 0, ctx)) < 0)
284 return ret;
285 ctx->w = var_values[VAR_OUT_W] = var_values[VAR_OW] = res;
286 if (!ctx->w)
287 var_values[VAR_OUT_W] = var_values[VAR_OW] = ctx->w = inlink->w;
288
289 if (adjusted_aspect.num && adjusted_aspect.den) {
290 adjusted_aspect = av_div_q(adjusted_aspect, inlink->sample_aspect_ratio);
291 if (ctx->h < av_rescale(ctx->w, adjusted_aspect.den, adjusted_aspect.num)) {
292 ctx->h = var_values[VAR_OUT_H] = var_values[VAR_OH] = av_rescale(ctx->w, adjusted_aspect.den, adjusted_aspect.num);
293 } else {
294 ctx->w = var_values[VAR_OUT_W] = var_values[VAR_OW] = av_rescale(ctx->h, adjusted_aspect.num, adjusted_aspect.den);
295 }
296 }
297
298 /* evaluate x and y */
299 av_expr_parse_and_eval(&res, (expr = ctx->x_expr),
300 var_names, var_values,
301 NULL, NULL, NULL, NULL, NULL, 0, ctx);
302 ctx->x = var_values[VAR_X] = res;
303 if ((ret = av_expr_parse_and_eval(&res, (expr = ctx->y_expr),
304 var_names, var_values,
305 NULL, NULL, NULL, NULL, NULL, 0, ctx)) < 0)
306 return ret;
307 ctx->y = var_values[VAR_Y] = res;
308 /* evaluate x again, as it may depend on the evaluated y value */
309 if ((ret = av_expr_parse_and_eval(&res, (expr = ctx->x_expr),
310 var_names, var_values,
311 NULL, NULL, NULL, NULL, NULL, 0, ctx)) < 0)
312 return ret;
313 ctx->x = var_values[VAR_X] = res;
314
315 if (ctx->x < 0 || ctx->x + inlink->w > ctx->w)
316 ctx->x = var_values[VAR_X] = (ctx->w - inlink->w) / 2;
317 if (ctx->y < 0 || ctx->y + inlink->h > ctx->h)
318 ctx->y = var_values[VAR_Y] = (ctx->h - inlink->h) / 2;
319
320 /* sanity check params */
321 if (ctx->w < inlink->w || ctx->h < inlink->h) {
322 av_log(ctx, AV_LOG_ERROR, "Padded dimensions cannot be smaller than input dimensions.\n");
323 return AVERROR(EINVAL);
324 }
325
326 if (ctx->w > avctx->inputs[0]->w) {
327 ctx->ocf.output_width = ctx->w;
328 } else {
329 ctx->ocf.output_width = avctx->inputs[0]->w;
330 }
331
332 if (ctx->h > avctx->inputs[0]->h) {
333 ctx->ocf.output_height = ctx->h;
334 } else {
335 ctx->ocf.output_height = avctx->inputs[0]->h;
336 }
337
338 if (ctx->x + avctx->inputs[0]->w > ctx->ocf.output_width ||
339 ctx->y + avctx->inputs[0]->h > ctx->ocf.output_height) {
340 return AVERROR(EINVAL);
341 }
342
343 err = ff_opencl_filter_config_output(outlink);
344 if (err < 0)
345 return err;
346
347 return 0;
348 }
349
350 static const AVFilterPad pad_opencl_inputs[] = {
351 {
352 .name = "default",
353 .type = AVMEDIA_TYPE_VIDEO,
354 .filter_frame = filter_frame,
355 .config_props = &ff_opencl_filter_config_input,
356 },
357 { NULL }
358 };
359
360 static const AVFilterPad pad_opencl_outputs[] = {
361 {
362 .name = "default",
363 .type = AVMEDIA_TYPE_VIDEO,
364 .config_props = &pad_opencl_config_output,
365 },
366 { NULL }
367 };
368
369 #define OFFSET(x) offsetof(PadOpenCLContext, x)
370 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
371
372 static const AVOption pad_opencl_options[] = {
373 { "width", "set the pad area width", OFFSET(w_expr), AV_OPT_TYPE_STRING, {.str = "iw"}, 0, 0, FLAGS },
374 { "w", "set the pad area width", OFFSET(w_expr), AV_OPT_TYPE_STRING, {.str = "iw"}, 0, 0, FLAGS },
375 { "height", "set the pad area height", OFFSET(h_expr), AV_OPT_TYPE_STRING, {.str = "ih"}, 0, 0, FLAGS },
376 { "h", "set the pad area height", OFFSET(h_expr), AV_OPT_TYPE_STRING, {.str = "ih"}, 0, 0, FLAGS },
377 { "x", "set the x offset for the input image position", OFFSET(x_expr), AV_OPT_TYPE_STRING, {.str = "0"}, 0, INT16_MAX, FLAGS },
378 { "y", "set the y offset for the input image position", OFFSET(y_expr), AV_OPT_TYPE_STRING, {.str = "0"}, 0, INT16_MAX, FLAGS },
379 { "color", "set the color of the padded area border", OFFSET(pad_rgba), AV_OPT_TYPE_COLOR, { .str = "black" }, 0, 0, FLAGS },
380 { "aspect", "pad to fit an aspect instead of a resolution", OFFSET(aspect), AV_OPT_TYPE_RATIONAL, {.dbl = 0}, 0, INT16_MAX, FLAGS },
381 { NULL }
382 };
383
384 AVFILTER_DEFINE_CLASS(pad_opencl);
385
386 AVFilter ff_vf_pad_opencl = {
387 .name = "pad_opencl",
388 .description = NULL_IF_CONFIG_SMALL("Pad the input video."),
389 .priv_size = sizeof(PadOpenCLContext),
390 .priv_class = &pad_opencl_class,
391 .init = &ff_opencl_filter_init,
392 .uninit = &pad_opencl_uninit,
393 .query_formats = &ff_opencl_filter_query_formats,
394 .inputs = pad_opencl_inputs,
395 .outputs = pad_opencl_outputs,
396 .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE
397 };
398