Skip to content

Commit

Permalink
avcodec/cuviddec: Add support for decoding HEVC 4:4:4 content
Browse files Browse the repository at this point in the history 
This is the equivalent change for cuviddec after the previous change
for nvdec. I made similar changes to the copying routines to handle
pixel formats in a more generic way.

Note that unlike with nvdec, there is no confusion about the ability
of a codec to output 444 formats. This is because the cuvid parser is
used, meaning that 444 JPEG content is still indicated as using a 420
output format.
  • Loading branch information
@philipl
philipl committed Feb 16, 2019
1 parent 83c7ac2 commit 317b7b0
Showing 1 changed file with 46 additions and 20 deletions.
66 changes: 46 additions & 20 deletions libavcodec/cuviddec.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -34,8 +34,14 @@
#include "avcodec.h"
#include "decode.h"
#include "hwaccel.h"
#include "nvdec.h"
#include "internal.h"

#if !NVDECAPI_CHECK_VERSION(9, 0)
#define cudaVideoSurfaceFormat_YUV444 2
#define cudaVideoSurfaceFormat_YUV444_16Bit 3
#endif

typedef struct CuvidContext
{
AVClass *avclass;
Expand Down Expand Up @@ -106,6 +112,7 @@ static int CUDAAPI cuvid_handle_video_sequence(void *opaque, CUVIDEOFORMAT* form
CUVIDDECODECAPS *caps = NULL;
CUVIDDECODECREATEINFO cuinfo;
int surface_fmt;
int chroma_444;

int old_width = avctx->width;
int old_height = avctx->height;
Expand Down Expand Up @@ -148,17 +155,19 @@ static int CUDAAPI cuvid_handle_video_sequence(void *opaque, CUVIDEOFORMAT* form
cuinfo.target_rect.right = cuinfo.ulTargetWidth;
cuinfo.target_rect.bottom = cuinfo.ulTargetHeight;

chroma_444 = format->chroma_format == cudaVideoChromaFormat_444;

switch (format->bit_depth_luma_minus8) {
case 0: // 8-bit
pix_fmts[1] = AV_PIX_FMT_NV12;
pix_fmts[1] = chroma_444 ? AV_PIX_FMT_YUV444P : AV_PIX_FMT_NV12;
caps = &ctx->caps8;
break;
case 2: // 10-bit
pix_fmts[1] = AV_PIX_FMT_P010;
pix_fmts[1] = chroma_444 ? AV_PIX_FMT_YUV444P16 : AV_PIX_FMT_P010;
caps = &ctx->caps10;
break;
case 4: // 12-bit
pix_fmts[1] = AV_PIX_FMT_P016;
pix_fmts[1] = chroma_444 ? AV_PIX_FMT_YUV444P16 : AV_PIX_FMT_P016;
caps = &ctx->caps12;
break;
default:
Expand Down Expand Up @@ -261,12 +270,6 @@ static int CUDAAPI cuvid_handle_video_sequence(void *opaque, CUVIDEOFORMAT* form
return 0;
}

if (format->chroma_format != cudaVideoChromaFormat_420) {
av_log(avctx, AV_LOG_ERROR, "Chroma formats other than 420 are not supported\n");
ctx->internal_error = AVERROR(EINVAL);
return 0;
}

ctx->chroma_format = format->chroma_format;

cuinfo.CodecType = ctx->codec_type = format->codec;
Expand All @@ -280,8 +283,15 @@ static int CUDAAPI cuvid_handle_video_sequence(void *opaque, CUVIDEOFORMAT* form
case AV_PIX_FMT_P016:
cuinfo.OutputFormat = cudaVideoSurfaceFormat_P016;
break;
case AV_PIX_FMT_YUV444P:
cuinfo.OutputFormat = cudaVideoSurfaceFormat_YUV444;
break;
case AV_PIX_FMT_YUV444P16:
cuinfo.OutputFormat = cudaVideoSurfaceFormat_YUV444_16Bit;
break;
default:
av_log(avctx, AV_LOG_ERROR, "Output formats other than NV12, P010 or P016 are not supported\n");
av_log(avctx, AV_LOG_ERROR, "Unsupported output format: %s\n",
av_get_pix_fmt_name(avctx->sw_pix_fmt));
ctx->internal_error = AVERROR(EINVAL);
return 0;
}
Expand Down Expand Up @@ -490,6 +500,7 @@ static int cuvid_output_frame(AVCodecContext *avctx, AVFrame *frame)
return ret;

if (av_fifo_size(ctx->frame_queue)) {
const AVPixFmtDescriptor *pixdesc;
CuvidParsedFrame parsed_frame;
CUVIDPROCPARAMS params;
unsigned int pitch = 0;
Expand Down Expand Up @@ -520,7 +531,10 @@ static int cuvid_output_frame(AVCodecContext *avctx, AVFrame *frame)
goto error;
}

for (i = 0; i < 2; i++) {
pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);

for (i = 0; i < pixdesc->nb_components; i++) {
int height = avctx->height >> (i ? pixdesc->log2_chroma_h : 0);
CUDA_MEMCPY2D cpy = {
.srcMemoryType = CU_MEMORYTYPE_DEVICE,
.dstMemoryType = CU_MEMORYTYPE_DEVICE,
Expand All @@ -530,38 +544,50 @@ static int cuvid_output_frame(AVCodecContext *avctx, AVFrame *frame)
.dstPitch = frame->linesize[i],
.srcY = offset,
.WidthInBytes = FFMIN(pitch, frame->linesize[i]),
.Height = avctx->height >> (i ? 1 : 0),
.Height = height,
};

ret = CHECK_CU(ctx->cudl->cuMemcpy2DAsync(&cpy, device_hwctx->stream));
if (ret < 0)
goto error;

offset += avctx->height;
offset += height;
}

ret = CHECK_CU(ctx->cudl->cuStreamSynchronize(device_hwctx->stream));
if (ret < 0)
goto error;
} else if (avctx->pix_fmt == AV_PIX_FMT_NV12 ||
avctx->pix_fmt == AV_PIX_FMT_P010 ||
avctx->pix_fmt == AV_PIX_FMT_P016) {
} else if (avctx->pix_fmt == AV_PIX_FMT_NV12 ||
avctx->pix_fmt == AV_PIX_FMT_P010 ||
avctx->pix_fmt == AV_PIX_FMT_P016 ||
avctx->pix_fmt == AV_PIX_FMT_YUV444P ||
avctx->pix_fmt == AV_PIX_FMT_YUV444P16) {
unsigned int offset = 0;
AVFrame *tmp_frame = av_frame_alloc();
if (!tmp_frame) {
av_log(avctx, AV_LOG_ERROR, "av_frame_alloc failed\n");
ret = AVERROR(ENOMEM);
goto error;
}

pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);

tmp_frame->format = AV_PIX_FMT_CUDA;
tmp_frame->hw_frames_ctx = av_buffer_ref(ctx->hwframe);
tmp_frame->data[0] = (uint8_t*)mapped_frame;
tmp_frame->linesize[0] = pitch;
tmp_frame->data[1] = (uint8_t*)(mapped_frame + avctx->height * pitch);
tmp_frame->linesize[1] = pitch;
tmp_frame->width = avctx->width;
tmp_frame->height = avctx->height;

/*
* Note that the following logic would not work for three plane
* YUV420 because the pitch value is different for the chroma
* planes.
*/
for (i = 0; i < pixdesc->nb_components; i++) {
tmp_frame->data[i] = (uint8_t*)mapped_frame + offset;
tmp_frame->linesize[i] = pitch;
offset += pitch * (avctx->height >> (i ? pixdesc->log2_chroma_h : 0));
}

ret = ff_get_buffer(avctx, frame, 0);
if (ret < 0) {
av_log(avctx, AV_LOG_ERROR, "ff_get_buffer failed\n");
Expand Down

0 comments on commit 317b7b0

Please sign in to comment.