2 * Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
19 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
20 * DEALINGS IN THE SOFTWARE.
26 #include "libavutil/avstring.h"
27 #include "libavutil/common.h"
28 #include "libavutil/hwcontext.h"
29 #include "libavutil/hwcontext_cuda_internal.h"
30 #include "libavutil/cuda_check.h"
31 #include "libavutil/internal.h"
32 #include "libavutil/opt.h"
33 #include "libavutil/pixdesc.h"
41 static const enum AVPixelFormat supported_formats[] = {
50 #define DIV_UP(a, b) ( ((a) + (b) - 1) / (b) )
51 #define ALIGN_UP(a, b) (((a) + (b) - 1) & ~((b) - 1))
56 #define CHECK_CU(x) FF_CUDA_CHECK_DL(ctx, s->hwctx->internal->cuda_dl, x)
58 typedef struct CUDAScaleContext {
61 AVCUDADeviceContext *hwctx;
63 enum AVPixelFormat in_fmt;
64 enum AVPixelFormat out_fmt;
69 } planes_in[3], planes_out[3];
71 AVBufferRef *frames_ctx;
78 * Output sw format. AV_PIX_FMT_NONE for no conversion.
80 enum AVPixelFormat format;
82 char *w_expr; ///< width expression string
83 char *h_expr; ///< height expression string
87 CUfunction cu_func_uchar;
88 CUfunction cu_func_uchar2;
89 CUfunction cu_func_uchar4;
90 CUfunction cu_func_ushort;
91 CUfunction cu_func_ushort2;
92 CUfunction cu_func_ushort4;
95 CUdeviceptr srcBuffer;
96 CUdeviceptr dstBuffer;
100 static av_cold int cudascale_init(AVFilterContext *ctx)
102 CUDAScaleContext *s = ctx->priv;
104 s->format = AV_PIX_FMT_NONE;
105 s->frame = av_frame_alloc();
107 return AVERROR(ENOMEM);
109 s->tmp_frame = av_frame_alloc();
111 return AVERROR(ENOMEM);
116 static av_cold void cudascale_uninit(AVFilterContext *ctx)
118 CUDAScaleContext *s = ctx->priv;
120 av_frame_free(&s->frame);
121 av_buffer_unref(&s->frames_ctx);
122 av_frame_free(&s->tmp_frame);
125 static int cudascale_query_formats(AVFilterContext *ctx)
127 static const enum AVPixelFormat pixel_formats[] = {
128 AV_PIX_FMT_CUDA, AV_PIX_FMT_NONE,
130 AVFilterFormats *pix_fmts = ff_make_format_list(pixel_formats);
132 return ff_set_common_formats(ctx, pix_fmts);
135 static av_cold int init_stage(CUDAScaleContext *s, AVBufferRef *device_ctx)
137 AVBufferRef *out_ref = NULL;
138 AVHWFramesContext *out_ctx;
139 int in_sw, in_sh, out_sw, out_sh;
142 av_pix_fmt_get_chroma_sub_sample(s->in_fmt, &in_sw, &in_sh);
143 av_pix_fmt_get_chroma_sub_sample(s->out_fmt, &out_sw, &out_sh);
144 if (!s->planes_out[0].width) {
145 s->planes_out[0].width = s->planes_in[0].width;
146 s->planes_out[0].height = s->planes_in[0].height;
149 for (i = 1; i < FF_ARRAY_ELEMS(s->planes_in); i++) {
150 s->planes_in[i].width = s->planes_in[0].width >> in_sw;
151 s->planes_in[i].height = s->planes_in[0].height >> in_sh;
152 s->planes_out[i].width = s->planes_out[0].width >> out_sw;
153 s->planes_out[i].height = s->planes_out[0].height >> out_sh;
156 out_ref = av_hwframe_ctx_alloc(device_ctx);
158 return AVERROR(ENOMEM);
159 out_ctx = (AVHWFramesContext*)out_ref->data;
161 out_ctx->format = AV_PIX_FMT_CUDA;
162 out_ctx->sw_format = s->out_fmt;
163 out_ctx->width = FFALIGN(s->planes_out[0].width, 32);
164 out_ctx->height = FFALIGN(s->planes_out[0].height, 32);
166 ret = av_hwframe_ctx_init(out_ref);
170 av_frame_unref(s->frame);
171 ret = av_hwframe_get_buffer(out_ref, s->frame, 0);
175 s->frame->width = s->planes_out[0].width;
176 s->frame->height = s->planes_out[0].height;
178 av_buffer_unref(&s->frames_ctx);
179 s->frames_ctx = out_ref;
183 av_buffer_unref(&out_ref);
187 static int format_is_supported(enum AVPixelFormat fmt)
191 for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++)
192 if (supported_formats[i] == fmt)
197 static av_cold int init_processing_chain(AVFilterContext *ctx, int in_width, int in_height,
198 int out_width, int out_height)
200 CUDAScaleContext *s = ctx->priv;
202 AVHWFramesContext *in_frames_ctx;
204 enum AVPixelFormat in_format;
205 enum AVPixelFormat out_format;
208 /* check that we have a hw context */
209 if (!ctx->inputs[0]->hw_frames_ctx) {
210 av_log(ctx, AV_LOG_ERROR, "No hw context provided on input\n");
211 return AVERROR(EINVAL);
213 in_frames_ctx = (AVHWFramesContext*)ctx->inputs[0]->hw_frames_ctx->data;
214 in_format = in_frames_ctx->sw_format;
215 out_format = (s->format == AV_PIX_FMT_NONE) ? in_format : s->format;
217 if (!format_is_supported(in_format)) {
218 av_log(ctx, AV_LOG_ERROR, "Unsupported input format: %s\n",
219 av_get_pix_fmt_name(in_format));
220 return AVERROR(ENOSYS);
222 if (!format_is_supported(out_format)) {
223 av_log(ctx, AV_LOG_ERROR, "Unsupported output format: %s\n",
224 av_get_pix_fmt_name(out_format));
225 return AVERROR(ENOSYS);
228 if (in_width == out_width && in_height == out_height)
231 s->in_fmt = in_format;
232 s->out_fmt = out_format;
234 s->planes_in[0].width = in_width;
235 s->planes_in[0].height = in_height;
236 s->planes_out[0].width = out_width;
237 s->planes_out[0].height = out_height;
239 ret = init_stage(s, in_frames_ctx->device_ref);
243 ctx->outputs[0]->hw_frames_ctx = av_buffer_ref(s->frames_ctx);
244 if (!ctx->outputs[0]->hw_frames_ctx)
245 return AVERROR(ENOMEM);
250 static av_cold int cudascale_config_props(AVFilterLink *outlink)
252 AVFilterContext *ctx = outlink->src;
253 AVFilterLink *inlink = outlink->src->inputs[0];
254 CUDAScaleContext *s = ctx->priv;
255 AVHWFramesContext *frames_ctx = (AVHWFramesContext*)inlink->hw_frames_ctx->data;
256 AVCUDADeviceContext *device_hwctx = frames_ctx->device_ctx->hwctx;
257 CUcontext dummy, cuda_ctx = device_hwctx->cuda_ctx;
258 CudaFunctions *cu = device_hwctx->internal->cuda_dl;
262 extern char vf_scale_cuda_ptx[];
264 s->hwctx = device_hwctx;
265 s->cu_stream = s->hwctx->stream;
267 ret = CHECK_CU(cu->cuCtxPushCurrent(cuda_ctx));
271 ret = CHECK_CU(cu->cuModuleLoadData(&s->cu_module, vf_scale_cuda_ptx));
275 CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uchar, s->cu_module, "Subsample_Bilinear_uchar"));
279 CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uchar2, s->cu_module, "Subsample_Bilinear_uchar2"));
283 CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uchar4, s->cu_module, "Subsample_Bilinear_uchar4"));
287 CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_ushort, s->cu_module, "Subsample_Bilinear_ushort"));
291 CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_ushort2, s->cu_module, "Subsample_Bilinear_ushort2"));
295 CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_ushort4, s->cu_module, "Subsample_Bilinear_ushort4"));
300 CHECK_CU(cu->cuCtxPopCurrent(&dummy));
302 if ((ret = ff_scale_eval_dimensions(s,
303 s->w_expr, s->h_expr,
308 if (((int64_t)h * inlink->w) > INT_MAX ||
309 ((int64_t)w * inlink->h) > INT_MAX)
310 av_log(ctx, AV_LOG_ERROR, "Rescaled value for width or height is too big.\n");
315 ret = init_processing_chain(ctx, inlink->w, inlink->h, w, h);
319 av_log(ctx, AV_LOG_VERBOSE, "w:%d h:%d -> w:%d h:%d\n",
320 inlink->w, inlink->h, outlink->w, outlink->h);
322 if (inlink->sample_aspect_ratio.num) {
323 outlink->sample_aspect_ratio = av_mul_q((AVRational){outlink->h*inlink->w,
324 outlink->w*inlink->h},
325 inlink->sample_aspect_ratio);
327 outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
336 static int call_resize_kernel(AVFilterContext *ctx, CUfunction func, int channels,
337 uint8_t *src_dptr, int src_width, int src_height, int src_pitch,
338 uint8_t *dst_dptr, int dst_width, int dst_height, int dst_pitch,
341 CUDAScaleContext *s = ctx->priv;
342 CudaFunctions *cu = s->hwctx->internal->cuda_dl;
343 CUdeviceptr dst_devptr = (CUdeviceptr)dst_dptr;
345 void *args_uchar[] = { &tex, &dst_devptr, &dst_width, &dst_height, &dst_pitch, &src_width, &src_height };
348 CUDA_TEXTURE_DESC tex_desc = {
349 .filterMode = CU_TR_FILTER_MODE_LINEAR,
350 .flags = CU_TRSF_READ_AS_INTEGER,
353 CUDA_RESOURCE_DESC res_desc = {
354 .resType = CU_RESOURCE_TYPE_PITCH2D,
355 .res.pitch2D.format = pixel_size == 1 ?
356 CU_AD_FORMAT_UNSIGNED_INT8 :
357 CU_AD_FORMAT_UNSIGNED_INT16,
358 .res.pitch2D.numChannels = channels,
359 .res.pitch2D.width = src_width,
360 .res.pitch2D.height = src_height,
361 .res.pitch2D.pitchInBytes = src_pitch * pixel_size,
362 .res.pitch2D.devPtr = (CUdeviceptr)src_dptr,
365 ret = CHECK_CU(cu->cuTexObjectCreate(&tex, &res_desc, &tex_desc, NULL));
369 ret = CHECK_CU(cu->cuLaunchKernel(func,
370 DIV_UP(dst_width, BLOCKX), DIV_UP(dst_height, BLOCKY), 1,
371 BLOCKX, BLOCKY, 1, 0, s->cu_stream, args_uchar, NULL));
375 CHECK_CU(cu->cuTexObjectDestroy(tex));
380 static int scalecuda_resize(AVFilterContext *ctx,
381 AVFrame *out, AVFrame *in)
383 AVHWFramesContext *in_frames_ctx = (AVHWFramesContext*)in->hw_frames_ctx->data;
384 CUDAScaleContext *s = ctx->priv;
386 switch (in_frames_ctx->sw_format) {
387 case AV_PIX_FMT_YUV420P:
388 call_resize_kernel(ctx, s->cu_func_uchar, 1,
389 in->data[0], in->width, in->height, in->linesize[0],
390 out->data[0], out->width, out->height, out->linesize[0],
392 call_resize_kernel(ctx, s->cu_func_uchar, 1,
393 in->data[1], in->width/2, in->height/2, in->linesize[0]/2,
394 out->data[1], out->width/2, out->height/2, out->linesize[0]/2,
396 call_resize_kernel(ctx, s->cu_func_uchar, 1,
397 in->data[2], in->width/2, in->height/2, in->linesize[0]/2,
398 out->data[2], out->width/2, out->height/2, out->linesize[0]/2,
401 case AV_PIX_FMT_YUV444P:
402 call_resize_kernel(ctx, s->cu_func_uchar, 1,
403 in->data[0], in->width, in->height, in->linesize[0],
404 out->data[0], out->width, out->height, out->linesize[0],
406 call_resize_kernel(ctx, s->cu_func_uchar, 1,
407 in->data[1], in->width, in->height, in->linesize[0],
408 out->data[1], out->width, out->height, out->linesize[0],
410 call_resize_kernel(ctx, s->cu_func_uchar, 1,
411 in->data[2], in->width, in->height, in->linesize[0],
412 out->data[2], out->width, out->height, out->linesize[0],
415 case AV_PIX_FMT_YUV444P16:
416 call_resize_kernel(ctx, s->cu_func_ushort, 1,
417 in->data[0], in->width, in->height, in->linesize[0] / 2,
418 out->data[0], out->width, out->height, out->linesize[0] / 2,
420 call_resize_kernel(ctx, s->cu_func_ushort, 1,
421 in->data[1], in->width, in->height, in->linesize[1] / 2,
422 out->data[1], out->width, out->height, out->linesize[1] / 2,
424 call_resize_kernel(ctx, s->cu_func_ushort, 1,
425 in->data[2], in->width, in->height, in->linesize[2] / 2,
426 out->data[2], out->width, out->height, out->linesize[2] / 2,
429 case AV_PIX_FMT_NV12:
430 call_resize_kernel(ctx, s->cu_func_uchar, 1,
431 in->data[0], in->width, in->height, in->linesize[0],
432 out->data[0], out->width, out->height, out->linesize[0],
434 call_resize_kernel(ctx, s->cu_func_uchar2, 2,
435 in->data[1], in->width/2, in->height/2, in->linesize[1],
436 out->data[1], out->width/2, out->height/2, out->linesize[1]/2,
439 case AV_PIX_FMT_P010LE:
440 call_resize_kernel(ctx, s->cu_func_ushort, 1,
441 in->data[0], in->width, in->height, in->linesize[0]/2,
442 out->data[0], out->width, out->height, out->linesize[0]/2,
444 call_resize_kernel(ctx, s->cu_func_ushort2, 2,
445 in->data[1], in->width / 2, in->height / 2, in->linesize[1]/2,
446 out->data[1], out->width / 2, out->height / 2, out->linesize[1] / 4,
449 case AV_PIX_FMT_P016LE:
450 call_resize_kernel(ctx, s->cu_func_ushort, 1,
451 in->data[0], in->width, in->height, in->linesize[0] / 2,
452 out->data[0], out->width, out->height, out->linesize[0] / 2,
454 call_resize_kernel(ctx, s->cu_func_ushort2, 2,
455 in->data[1], in->width / 2, in->height / 2, in->linesize[1] / 2,
456 out->data[1], out->width / 2, out->height / 2, out->linesize[1] / 4,
466 static int cudascale_scale(AVFilterContext *ctx, AVFrame *out, AVFrame *in)
468 CUDAScaleContext *s = ctx->priv;
472 ret = scalecuda_resize(ctx, s->frame, src);
477 ret = av_hwframe_get_buffer(src->hw_frames_ctx, s->tmp_frame, 0);
481 av_frame_move_ref(out, s->frame);
482 av_frame_move_ref(s->frame, s->tmp_frame);
484 ret = av_frame_copy_props(out, in);
491 static int cudascale_filter_frame(AVFilterLink *link, AVFrame *in)
493 AVFilterContext *ctx = link->dst;
494 CUDAScaleContext *s = ctx->priv;
495 AVFilterLink *outlink = ctx->outputs[0];
496 CudaFunctions *cu = s->hwctx->internal->cuda_dl;
502 out = av_frame_alloc();
504 ret = AVERROR(ENOMEM);
508 ret = CHECK_CU(cu->cuCtxPushCurrent(s->hwctx->cuda_ctx));
512 ret = cudascale_scale(ctx, out, in);
514 CHECK_CU(cu->cuCtxPopCurrent(&dummy));
518 av_reduce(&out->sample_aspect_ratio.num, &out->sample_aspect_ratio.den,
519 (int64_t)in->sample_aspect_ratio.num * outlink->h * link->w,
520 (int64_t)in->sample_aspect_ratio.den * outlink->w * link->h,
524 return ff_filter_frame(outlink, out);
531 #define OFFSET(x) offsetof(CUDAScaleContext, x)
532 #define FLAGS (AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM)
533 static const AVOption options[] = {
534 { "w", "Output video width", OFFSET(w_expr), AV_OPT_TYPE_STRING, { .str = "iw" }, .flags = FLAGS },
535 { "h", "Output video height", OFFSET(h_expr), AV_OPT_TYPE_STRING, { .str = "ih" }, .flags = FLAGS },
539 static const AVClass cudascale_class = {
540 .class_name = "cudascale",
541 .item_name = av_default_item_name,
543 .version = LIBAVUTIL_VERSION_INT,
546 static const AVFilterPad cudascale_inputs[] = {
549 .type = AVMEDIA_TYPE_VIDEO,
550 .filter_frame = cudascale_filter_frame,
555 static const AVFilterPad cudascale_outputs[] = {
558 .type = AVMEDIA_TYPE_VIDEO,
559 .config_props = cudascale_config_props,
564 AVFilter ff_vf_scale_cuda = {
565 .name = "scale_cuda",
566 .description = NULL_IF_CONFIG_SMALL("GPU accelerated video resizer"),
568 .init = cudascale_init,
569 .uninit = cudascale_uninit,
570 .query_formats = cudascale_query_formats,
572 .priv_size = sizeof(CUDAScaleContext),
573 .priv_class = &cudascale_class,
575 .inputs = cudascale_inputs,
576 .outputs = cudascale_outputs,
578 .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,