aboutsummaryrefslogtreecommitdiffstats
path: root/libavfilter/vf_thumbnail_cuda.c
blob: a7a246c9591c942c41d03a2d2790f8d9b8b3ff41 (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
/*
* Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*/

#include "libavutil/hwcontext.h"
#include "libavutil/hwcontext_cuda_internal.h"
#include "libavutil/cuda_check.h"
#include "libavutil/opt.h"
#include "libavutil/pixdesc.h"

#include "avfilter.h"
#include "internal.h"

#include "cuda/load_helper.h"

#define CHECK_CU(x) FF_CUDA_CHECK_DL(ctx, s->hwctx->internal->cuda_dl, x)

#define HIST_SIZE (3*256)
#define DIV_UP(a, b) ( ((a) + (b) - 1) / (b) )
#define BLOCKX 32
#define BLOCKY 16

static const enum AVPixelFormat supported_formats[] = {
    AV_PIX_FMT_NV12,
    AV_PIX_FMT_YUV420P,
    AV_PIX_FMT_YUV444P,
    AV_PIX_FMT_P010,
    AV_PIX_FMT_P016,
    AV_PIX_FMT_YUV444P16,
};

struct thumb_frame {
    AVFrame *buf;               ///< cached frame
    int histogram[HIST_SIZE];   ///< RGB color distribution histogram of the frame
};

typedef struct ThumbnailCudaContext {
    const AVClass *class;
    int n;                      ///< current frame
    int n_frames;               ///< number of frames for analysis
    struct thumb_frame *frames; ///< the n_frames frames
    AVRational tb;              ///< copy of the input timebase to ease access

    AVBufferRef *hw_frames_ctx;
    AVCUDADeviceContext *hwctx;

    CUmodule    cu_module;

    CUfunction  cu_func_uchar;
    CUfunction  cu_func_uchar2;
    CUfunction  cu_func_ushort;
    CUfunction  cu_func_ushort2;
    CUstream    cu_stream;

    CUdeviceptr data;

} ThumbnailCudaContext;

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

static const AVOption thumbnail_cuda_options[] = {
    { "n", "set the frames batch size", OFFSET(n_frames), AV_OPT_TYPE_INT, {.i64=100}, 2, INT_MAX, FLAGS },
    { NULL }
};

AVFILTER_DEFINE_CLASS(thumbnail_cuda);

static av_cold int init(AVFilterContext *ctx)
{
    ThumbnailCudaContext *s = ctx->priv;

    s->frames = av_calloc(s->n_frames, sizeof(*s->frames));
    if (!s->frames) {
        av_log(ctx, AV_LOG_ERROR,
               "Allocation failure, try to lower the number of frames\n");
        return AVERROR(ENOMEM);
    }
    av_log(ctx, AV_LOG_VERBOSE, "batch size: %d frames\n", s->n_frames);
    return 0;
}

/**
 * @brief        Compute Sum-square deviation to estimate "closeness".
 * @param hist   color distribution histogram
 * @param median average color distribution histogram
 * @return       sum of squared errors
 */
static double frame_sum_square_err(const int *hist, const double *median)
{
    int i;
    double err, sum_sq_err = 0;

    for (i = 0; i < HIST_SIZE; i++) {
        err = median[i] - (double)hist[i];
        sum_sq_err += err*err;
    }
    return sum_sq_err;
}

static AVFrame *get_best_frame(AVFilterContext *ctx)
{
    AVFrame *picref;
    ThumbnailCudaContext *s = ctx->priv;
    int i, j, best_frame_idx = 0;
    int nb_frames = s->n;
    double avg_hist[HIST_SIZE] = {0}, sq_err, min_sq_err = -1;

    // average histogram of the N frames
    for (j = 0; j < FF_ARRAY_ELEMS(avg_hist); j++) {
        for (i = 0; i < nb_frames; i++)
            avg_hist[j] += (double)s->frames[i].histogram[j];
        avg_hist[j] /= nb_frames;
    }

    // find the frame closer to the average using the sum of squared errors
    for (i = 0; i < nb_frames; i++) {
        sq_err = frame_sum_square_err(s->frames[i].histogram, avg_hist);
        if (i == 0 || sq_err < min_sq_err)
            best_frame_idx = i, min_sq_err = sq_err;
    }

    // free and reset everything (except the best frame buffer)
    for (i = 0; i < nb_frames; i++) {
        memset(s->frames[i].histogram, 0, sizeof(s->frames[i].histogram));
        if (i != best_frame_idx)
            av_frame_free(&s->frames[i].buf);
    }
    s->n = 0;

    // raise the chosen one
    picref = s->frames[best_frame_idx].buf;
    av_log(ctx, AV_LOG_INFO, "frame id #%d (pts_time=%f) selected "
           "from a set of %d images\n", best_frame_idx,
           picref->pts * av_q2d(s->tb), nb_frames);
    s->frames[best_frame_idx].buf = NULL;

    return picref;
}

static int thumbnail_kernel(AVFilterContext *ctx, CUfunction func, int channels,
    int *histogram, uint8_t *src_dptr, int src_width, int src_height, int src_pitch, int pixel_size)
{
    int ret;
    ThumbnailCudaContext *s = ctx->priv;
    CudaFunctions *cu = s->hwctx->internal->cuda_dl;
    CUtexObject tex = 0;
    void *args[] = { &tex, &histogram, &src_width, &src_height };

    CUDA_TEXTURE_DESC tex_desc = {
        .filterMode = CU_TR_FILTER_MODE_LINEAR,
        .flags = CU_TRSF_READ_AS_INTEGER,
    };

    CUDA_RESOURCE_DESC res_desc = {
        .resType = CU_RESOURCE_TYPE_PITCH2D,
        .res.pitch2D.format = pixel_size == 1 ?
                              CU_AD_FORMAT_UNSIGNED_INT8 :
                              CU_AD_FORMAT_UNSIGNED_INT16,
        .res.pitch2D.numChannels = channels,
        .res.pitch2D.width = src_width,
        .res.pitch2D.height = src_height,
        .res.pitch2D.pitchInBytes = src_pitch,
        .res.pitch2D.devPtr = (CUdeviceptr)src_dptr,
    };

    ret = CHECK_CU(cu->cuTexObjectCreate(&tex, &res_desc, &tex_desc, NULL));
    if (ret < 0)
        goto exit;

    ret = CHECK_CU(cu->cuLaunchKernel(func,
                                      DIV_UP(src_width, BLOCKX), DIV_UP(src_height, BLOCKY), 1,
                                      BLOCKX, BLOCKY, 1, 0, s->cu_stream, args, NULL));
exit:
    if (tex)
        CHECK_CU(cu->cuTexObjectDestroy(tex));

    return ret;
}

static int thumbnail(AVFilterContext *ctx, int *histogram, AVFrame *in)
{
    AVHWFramesContext *in_frames_ctx = (AVHWFramesContext*)in->hw_frames_ctx->data;
    ThumbnailCudaContext *s = ctx->priv;

    switch (in_frames_ctx->sw_format) {
    case AV_PIX_FMT_NV12:
        thumbnail_kernel(ctx, s->cu_func_uchar, 1,
            histogram, in->data[0], in->width, in->height, in->linesize[0], 1);
        thumbnail_kernel(ctx, s->cu_func_uchar2, 2,
            histogram + 256, in->data[1], in->width / 2, in->height / 2, in->linesize[1], 1);
        break;
    case AV_PIX_FMT_YUV420P:
        thumbnail_kernel(ctx, s->cu_func_uchar, 1,
            histogram, in->data[0], in->width, in->height, in->linesize[0], 1);
        thumbnail_kernel(ctx, s->cu_func_uchar, 1,
            histogram + 256, in->data[1], in->width / 2, in->height / 2, in->linesize[1], 1);
        thumbnail_kernel(ctx, s->cu_func_uchar, 1,
            histogram + 512, in->data[2], in->width / 2, in->height / 2, in->linesize[2], 1);
        break;
    case AV_PIX_FMT_YUV444P:
        thumbnail_kernel(ctx, s->cu_func_uchar, 1,
            histogram, in->data[0], in->width, in->height, in->linesize[0], 1);
        thumbnail_kernel(ctx, s->cu_func_uchar, 1,
            histogram + 256, in->data[1], in->width, in->height, in->linesize[1], 1);
        thumbnail_kernel(ctx, s->cu_func_uchar, 1,
            histogram + 512, in->data[2], in->width, in->height, in->linesize[2], 1);
        break;
    case AV_PIX_FMT_P010LE:
    case AV_PIX_FMT_P016LE:
        thumbnail_kernel(ctx, s->cu_func_ushort, 1,
            histogram, in->data[0], in->width, in->height, in->linesize[0], 2);
        thumbnail_kernel(ctx, s->cu_func_ushort2, 2,
            histogram + 256, in->data[1], in->width / 2, in->height / 2, in->linesize[1], 2);
        break;
    case AV_PIX_FMT_YUV444P16:
        thumbnail_kernel(ctx, s->cu_func_ushort2, 1,
            histogram, in->data[0], in->width, in->height, in->linesize[0], 2);
        thumbnail_kernel(ctx, s->cu_func_ushort2, 1,
            histogram + 256, in->data[1], in->width, in->height, in->linesize[1], 2);
        thumbnail_kernel(ctx, s->cu_func_ushort2, 1,
            histogram + 512, in->data[2], in->width, in->height, in->linesize[2], 2);
        break;
    default:
        return AVERROR_BUG;
    }

    return 0;
}

static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
{
    AVFilterContext *ctx  = inlink->dst;
    ThumbnailCudaContext *s   = ctx->priv;
    CudaFunctions *cu = s->hwctx->internal->cuda_dl;
    AVFilterLink *outlink = ctx->outputs[0];
    int *hist = s->frames[s->n].histogram;
    AVHWFramesContext *hw_frames_ctx = (AVHWFramesContext*)s->hw_frames_ctx->data;
    CUcontext dummy;
    CUDA_MEMCPY2D cpy = { 0 };
    int ret = 0;

    // keep a reference of each frame
    s->frames[s->n].buf = frame;

    ret = CHECK_CU(cu->cuCtxPushCurrent(s->hwctx->cuda_ctx));
    if (ret < 0)
        return ret;

    CHECK_CU(cu->cuMemsetD8Async(s->data, 0, HIST_SIZE * sizeof(int), s->cu_stream));

    thumbnail(ctx, (int*)s->data, frame);

    cpy.srcMemoryType = CU_MEMORYTYPE_DEVICE;
    cpy.dstMemoryType = CU_MEMORYTYPE_HOST;
    cpy.srcDevice = s->data;
    cpy.dstHost = hist;
    cpy.srcPitch = HIST_SIZE * sizeof(int);
    cpy.dstPitch = HIST_SIZE * sizeof(int);
    cpy.WidthInBytes = HIST_SIZE * sizeof(int);
    cpy.Height = 1;

    ret = CHECK_CU(cu->cuMemcpy2DAsync(&cpy, s->cu_stream));
    if (ret < 0)
        return ret;

    if (hw_frames_ctx->sw_format == AV_PIX_FMT_NV12 || hw_frames_ctx->sw_format == AV_PIX_FMT_YUV420P ||
        hw_frames_ctx->sw_format == AV_PIX_FMT_P010LE || hw_frames_ctx->sw_format == AV_PIX_FMT_P016LE)
    {
        int i;
        for (i = 256; i < HIST_SIZE; i++)
            hist[i] = 4 * hist[i];
    }

    CHECK_CU(cu->cuCtxPopCurrent(&dummy));
    if (ret < 0)
        return ret;

    // no selection until the buffer of N frames is filled up
    s->n++;
    if (s->n < s->n_frames)
        return 0;

    return ff_filter_frame(outlink, get_best_frame(ctx));
}

static av_cold void uninit(AVFilterContext *ctx)
{
    int i;
    ThumbnailCudaContext *s = ctx->priv;
    CudaFunctions *cu = s->hwctx->internal->cuda_dl;

    if (s->data) {
        CHECK_CU(cu->cuMemFree(s->data));
        s->data = 0;
    }

    if (s->cu_module) {
        CHECK_CU(cu->cuModuleUnload(s->cu_module));
        s->cu_module = NULL;
    }

    for (i = 0; i < s->n_frames && s->frames[i].buf; i++)
        av_frame_free(&s->frames[i].buf);
    av_freep(&s->frames);
}

static int request_frame(AVFilterLink *link)
{
    AVFilterContext *ctx = link->src;
    ThumbnailCudaContext *s = ctx->priv;
    int ret = ff_request_frame(ctx->inputs[0]);

    if (ret == AVERROR_EOF && s->n) {
        ret = ff_filter_frame(link, get_best_frame(ctx));
        if (ret < 0)
            return ret;
        ret = AVERROR_EOF;
    }
    if (ret < 0)
        return ret;
    return 0;
}

static int format_is_supported(enum AVPixelFormat fmt)
{
    int i;

    for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++)
        if (supported_formats[i] == fmt)
            return 1;
    return 0;
}

static int config_props(AVFilterLink *inlink)
{
    AVFilterContext *ctx = inlink->dst;
    ThumbnailCudaContext *s = ctx->priv;
    AVHWFramesContext     *hw_frames_ctx = (AVHWFramesContext*)inlink->hw_frames_ctx->data;
    AVCUDADeviceContext *device_hwctx = hw_frames_ctx->device_ctx->hwctx;
    CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
    CudaFunctions *cu = device_hwctx->internal->cuda_dl;
    int ret;

    extern const unsigned char ff_vf_thumbnail_cuda_ptx_data[];
    extern const unsigned int ff_vf_thumbnail_cuda_ptx_len;

    s->hwctx = device_hwctx;
    s->cu_stream = s->hwctx->stream;

    ret = CHECK_CU(cu->cuCtxPushCurrent(cuda_ctx));
    if (ret < 0)
        return ret;

    ret = ff_cuda_load_module(ctx, device_hwctx, &s->cu_module, ff_vf_thumbnail_cuda_ptx_data, ff_vf_thumbnail_cuda_ptx_len);
    if (ret < 0)
        return ret;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uchar, s->cu_module, "Thumbnail_uchar"));
    if (ret < 0)
        return ret;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uchar2, s->cu_module, "Thumbnail_uchar2"));
    if (ret < 0)
        return ret;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_ushort, s->cu_module, "Thumbnail_ushort"));
    if (ret < 0)
        return ret;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_ushort2, s->cu_module, "Thumbnail_ushort2"));
    if (ret < 0)
        return ret;

    ret = CHECK_CU(cu->cuMemAlloc(&s->data, HIST_SIZE * sizeof(int)));
    if (ret < 0)
        return ret;

    CHECK_CU(cu->cuCtxPopCurrent(&dummy));

    s->hw_frames_ctx = ctx->inputs[0]->hw_frames_ctx;

    ctx->outputs[0]->hw_frames_ctx = av_buffer_ref(s->hw_frames_ctx);
    if (!ctx->outputs[0]->hw_frames_ctx)
        return AVERROR(ENOMEM);

    s->tb = inlink->time_base;

    if (!format_is_supported(hw_frames_ctx->sw_format)) {
        av_log(ctx, AV_LOG_ERROR, "Unsupported input format: %s\n", av_get_pix_fmt_name(hw_frames_ctx->sw_format));
        return AVERROR(ENOSYS);
    }

    return 0;
}

static int query_formats(AVFilterContext *ctx)
{
    static const enum AVPixelFormat pix_fmts[] = {
        AV_PIX_FMT_CUDA,
        AV_PIX_FMT_NONE
    };
    return ff_set_common_formats_from_list(ctx, pix_fmts);
}

static const AVFilterPad thumbnail_cuda_inputs[] = {
    {
        .name         = "default",
        .type         = AVMEDIA_TYPE_VIDEO,
        .config_props = config_props,
        .filter_frame = filter_frame,
    },
};

static const AVFilterPad thumbnail_cuda_outputs[] = {
    {
        .name          = "default",
        .type          = AVMEDIA_TYPE_VIDEO,
        .request_frame = request_frame,
    },
};

const AVFilter ff_vf_thumbnail_cuda = {
    .name          = "thumbnail_cuda",
    .description   = NULL_IF_CONFIG_SMALL("Select the most representative frame in a given sequence of consecutive frames."),
    .priv_size     = sizeof(ThumbnailCudaContext),
    .init          = init,
    .uninit        = uninit,
    FILTER_INPUTS(thumbnail_cuda_inputs),
    FILTER_OUTPUTS(thumbnail_cuda_outputs),
    FILTER_QUERY_FUNC(query_formats),
    .priv_class    = &thumbnail_cuda_class,
    .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
};