aboutsummaryrefslogtreecommitdiffstats
path: root/libavfilter/vf_yadif_cuda.c
blob: d777757e6566744e5f36e6f97b38675ec201bf50 (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
/*
 * Copyright (C) 2018 Philip Langdale <philipl@overt.org>
 *
 * 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/avassert.h"
#include "libavutil/hwcontext.h"
#include "libavutil/hwcontext_cuda_internal.h"
#include "libavutil/cuda_check.h"
#include "internal.h"
#include "yadif.h"

#include "cuda/load_helper.h"

extern const unsigned char ff_vf_yadif_cuda_ptx_data[];
extern const unsigned int ff_vf_yadif_cuda_ptx_len;

typedef struct DeintCUDAContext {
    YADIFContext yadif;

    AVCUDADeviceContext *hwctx;
    AVBufferRef         *device_ref;
    AVBufferRef         *input_frames_ref;
    AVHWFramesContext   *input_frames;

    CUmodule    cu_module;
    CUfunction  cu_func_uchar;
    CUfunction  cu_func_uchar2;
    CUfunction  cu_func_ushort;
    CUfunction  cu_func_ushort2;
} DeintCUDAContext;

#define DIV_UP(a, b) ( ((a) + (b) - 1) / (b) )
#define ALIGN_UP(a, b) (((a) + (b) - 1) & ~((b) - 1))
#define BLOCKX 32
#define BLOCKY 16

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

static CUresult call_kernel(AVFilterContext *ctx, CUfunction func,
                            CUdeviceptr prev, CUdeviceptr cur, CUdeviceptr next,
                            CUarray_format format, int channels,
                            int src_width,  // Width is pixels per channel
                            int src_height, // Height is pixels per channel
                            int src_pitch,  // Pitch is bytes
                            CUdeviceptr dst,
                            int dst_width,  // Width is pixels per channel
                            int dst_height, // Height is pixels per channel
                            int dst_pitch,  // Pitch is pixels per channel
                            int parity, int tff)
{
    DeintCUDAContext *s = ctx->priv;
    CudaFunctions *cu = s->hwctx->internal->cuda_dl;
    CUtexObject tex_prev = 0, tex_cur = 0, tex_next = 0;
    int ret;
    int skip_spatial_check = s->yadif.mode&2;

    void *args[] = { &dst, &tex_prev, &tex_cur, &tex_next,
                     &dst_width, &dst_height, &dst_pitch,
                     &src_width, &src_height, &parity, &tff,
                     &skip_spatial_check };

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

    CUDA_RESOURCE_DESC res_desc = {
        .resType = CU_RESOURCE_TYPE_PITCH2D,
        .res.pitch2D.format = format,
        .res.pitch2D.numChannels = channels,
        .res.pitch2D.width = src_width,
        .res.pitch2D.height = src_height,
        .res.pitch2D.pitchInBytes = src_pitch,
    };

    res_desc.res.pitch2D.devPtr = (CUdeviceptr)prev;
    ret = CHECK_CU(cu->cuTexObjectCreate(&tex_prev, &res_desc, &tex_desc, NULL));
    if (ret < 0)
        goto exit;

    res_desc.res.pitch2D.devPtr = (CUdeviceptr)cur;
    ret = CHECK_CU(cu->cuTexObjectCreate(&tex_cur, &res_desc, &tex_desc, NULL));
    if (ret < 0)
        goto exit;

    res_desc.res.pitch2D.devPtr = (CUdeviceptr)next;
    ret = CHECK_CU(cu->cuTexObjectCreate(&tex_next, &res_desc, &tex_desc, NULL));
    if (ret < 0)
        goto exit;

    ret = CHECK_CU(cu->cuLaunchKernel(func,
                                      DIV_UP(dst_width, BLOCKX), DIV_UP(dst_height, BLOCKY), 1,
                                      BLOCKX, BLOCKY, 1,
                                      0, s->hwctx->stream, args, NULL));

exit:
    if (tex_prev)
        CHECK_CU(cu->cuTexObjectDestroy(tex_prev));
    if (tex_cur)
        CHECK_CU(cu->cuTexObjectDestroy(tex_cur));
    if (tex_next)
        CHECK_CU(cu->cuTexObjectDestroy(tex_next));

    return ret;
}

static void filter(AVFilterContext *ctx, AVFrame *dst,
                   int parity, int tff)
{
    DeintCUDAContext *s = ctx->priv;
    YADIFContext *y = &s->yadif;
    CudaFunctions *cu = s->hwctx->internal->cuda_dl;
    CUcontext dummy;
    int i, ret;

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

    for (i = 0; i < y->csp->nb_components; i++) {
        CUfunction func;
        CUarray_format format;
        int pixel_size, channels;
        const AVComponentDescriptor *comp = &y->csp->comp[i];

        if (comp->plane < i) {
            // We process planes as a whole, so don't reprocess
            // them for additional components
            continue;
        }

        pixel_size = (comp->depth + comp->shift) / 8;
        channels = comp->step / pixel_size;
        if (pixel_size > 2 || channels > 2) {
            av_log(ctx, AV_LOG_ERROR, "Unsupported pixel format: %s\n", y->csp->name);
            goto exit;
        }
        switch (pixel_size) {
        case 1:
            func = channels == 1 ? s->cu_func_uchar : s->cu_func_uchar2;
            format = CU_AD_FORMAT_UNSIGNED_INT8;
            break;
        case 2:
            func = channels == 1 ? s->cu_func_ushort : s->cu_func_ushort2;
            format = CU_AD_FORMAT_UNSIGNED_INT16;
            break;
        default:
            av_log(ctx, AV_LOG_ERROR, "Unsupported pixel format: %s\n", y->csp->name);
            goto exit;
        }
        av_log(ctx, AV_LOG_TRACE,
               "Deinterlacing plane %d: pixel_size: %d channels: %d\n",
               comp->plane, pixel_size, channels);
        call_kernel(ctx, func,
                    (CUdeviceptr)y->prev->data[i],
                    (CUdeviceptr)y->cur->data[i],
                    (CUdeviceptr)y->next->data[i],
                    format, channels,
                    AV_CEIL_RSHIFT(y->cur->width, i ? y->csp->log2_chroma_w : 0),
                    AV_CEIL_RSHIFT(y->cur->height, i ? y->csp->log2_chroma_h : 0),
                    y->cur->linesize[i],
                    (CUdeviceptr)dst->data[i],
                    AV_CEIL_RSHIFT(dst->width, i ? y->csp->log2_chroma_w : 0),
                    AV_CEIL_RSHIFT(dst->height, i ? y->csp->log2_chroma_h : 0),
                    dst->linesize[i] / comp->step,
                    parity, tff);
    }

exit:
    CHECK_CU(cu->cuCtxPopCurrent(&dummy));
    return;
}

static av_cold void deint_cuda_uninit(AVFilterContext *ctx)
{
    CUcontext dummy;
    DeintCUDAContext *s = ctx->priv;
    YADIFContext *y = &s->yadif;

    if (s->hwctx && s->cu_module) {
        CudaFunctions *cu = s->hwctx->internal->cuda_dl;
        CHECK_CU(cu->cuCtxPushCurrent(s->hwctx->cuda_ctx));
        CHECK_CU(cu->cuModuleUnload(s->cu_module));
        CHECK_CU(cu->cuCtxPopCurrent(&dummy));
    }

    av_frame_free(&y->prev);
    av_frame_free(&y->cur);
    av_frame_free(&y->next);
    ff_ccfifo_uninit(&y->cc_fifo);

    av_buffer_unref(&s->device_ref);
    s->hwctx = NULL;
    av_buffer_unref(&s->input_frames_ref);
    s->input_frames = NULL;
}

static int config_input(AVFilterLink *inlink)
{
    AVFilterContext *ctx = inlink->dst;
    DeintCUDAContext *s  = ctx->priv;

    if (!inlink->hw_frames_ctx) {
        av_log(ctx, AV_LOG_ERROR, "A hardware frames reference is "
               "required to associate the processing device.\n");
        return AVERROR(EINVAL);
    }

    s->input_frames_ref = av_buffer_ref(inlink->hw_frames_ctx);
    if (!s->input_frames_ref) {
        av_log(ctx, AV_LOG_ERROR, "A input frames reference create "
               "failed.\n");
        return AVERROR(ENOMEM);
    }
    s->input_frames = (AVHWFramesContext*)s->input_frames_ref->data;

    return 0;
}

static int config_output(AVFilterLink *link)
{
    AVHWFramesContext *output_frames;
    AVFilterContext *ctx = link->src;
    DeintCUDAContext *s = ctx->priv;
    YADIFContext *y = &s->yadif;
    CudaFunctions *cu;
    int ret = 0;
    CUcontext dummy;

    av_assert0(s->input_frames);
    s->device_ref = av_buffer_ref(s->input_frames->device_ref);
    if (!s->device_ref) {
        av_log(ctx, AV_LOG_ERROR, "A device reference create "
               "failed.\n");
        return AVERROR(ENOMEM);
    }
    s->hwctx = ((AVHWDeviceContext*)s->device_ref->data)->hwctx;
    cu = s->hwctx->internal->cuda_dl;

    link->hw_frames_ctx = av_hwframe_ctx_alloc(s->device_ref);
    if (!link->hw_frames_ctx) {
        av_log(ctx, AV_LOG_ERROR, "Failed to create HW frame context "
               "for output.\n");
        ret = AVERROR(ENOMEM);
        goto exit;
    }

    output_frames = (AVHWFramesContext*)link->hw_frames_ctx->data;

    output_frames->format    = AV_PIX_FMT_CUDA;
    output_frames->sw_format = s->input_frames->sw_format;
    output_frames->width     = ctx->inputs[0]->w;
    output_frames->height    = ctx->inputs[0]->h;

    output_frames->initial_pool_size = 4;

    ret = ff_filter_init_hw_frames(ctx, link, 10);
    if (ret < 0)
        goto exit;

    ret = av_hwframe_ctx_init(link->hw_frames_ctx);
    if (ret < 0) {
        av_log(ctx, AV_LOG_ERROR, "Failed to initialise CUDA frame "
               "context for output: %d\n", ret);
        goto exit;
    }

    link->time_base = av_mul_q(ctx->inputs[0]->time_base, (AVRational){1, 2});
    link->w         = ctx->inputs[0]->w;
    link->h         = ctx->inputs[0]->h;

    if(y->mode & 1)
        link->frame_rate = av_mul_q(ctx->inputs[0]->frame_rate,
                                    (AVRational){2, 1});
    else
        link->frame_rate = ctx->inputs[0]->frame_rate;

    ret = ff_ccfifo_init(&y->cc_fifo, link->frame_rate, ctx);
    if (ret < 0) {
        av_log(ctx, AV_LOG_ERROR, "Failure to setup CC FIFO queue\n");
        goto exit;
    }

    if (link->w < 3 || link->h < 3) {
        av_log(ctx, AV_LOG_ERROR, "Video of less than 3 columns or lines is not supported\n");
        ret = AVERROR(EINVAL);
        goto exit;
    }

    y->csp = av_pix_fmt_desc_get(output_frames->sw_format);
    y->filter = filter;

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

    ret = ff_cuda_load_module(ctx, s->hwctx, &s->cu_module, ff_vf_yadif_cuda_ptx_data, ff_vf_yadif_cuda_ptx_len);
    if (ret < 0)
        goto exit;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uchar, s->cu_module, "yadif_uchar"));
    if (ret < 0)
        goto exit;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uchar2, s->cu_module, "yadif_uchar2"));
    if (ret < 0)
        goto exit;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_ushort, s->cu_module, "yadif_ushort"));
    if (ret < 0)
        goto exit;

    ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_ushort2, s->cu_module, "yadif_ushort2"));
    if (ret < 0)
        goto exit;

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

    return ret;
}

static const AVClass yadif_cuda_class = {
    .class_name = "yadif_cuda",
    .item_name  = av_default_item_name,
    .option     = ff_yadif_options,
    .version    = LIBAVUTIL_VERSION_INT,
    .category   = AV_CLASS_CATEGORY_FILTER,
};

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

static const AVFilterPad deint_cuda_outputs[] = {
    {
        .name          = "default",
        .type          = AVMEDIA_TYPE_VIDEO,
        .request_frame = ff_yadif_request_frame,
        .config_props  = config_output,
    },
};

const AVFilter ff_vf_yadif_cuda = {
    .name           = "yadif_cuda",
    .description    = NULL_IF_CONFIG_SMALL("Deinterlace CUDA frames"),
    .priv_size      = sizeof(DeintCUDAContext),
    .priv_class     = &yadif_cuda_class,
    .uninit         = deint_cuda_uninit,
    FILTER_SINGLE_PIXFMT(AV_PIX_FMT_CUDA),
    FILTER_INPUTS(deint_cuda_inputs),
    FILTER_OUTPUTS(deint_cuda_outputs),
    .flags          = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
    .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
};