ffmpeg/libavfilter/qsvvpp.c

887 lines
31 KiB
C
Raw Normal View History

/*
* 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
*/
/**
* @file
* Intel Quick Sync Video VPP base function
*/
#include "libavutil/common.h"
#include "libavutil/mathematics.h"
#include "libavutil/hwcontext.h"
#include "libavutil/hwcontext_qsv.h"
#include "libavutil/time.h"
#include "libavutil/pixdesc.h"
#include "internal.h"
#include "qsvvpp.h"
#include "video.h"
#define IS_VIDEO_MEMORY(mode) (mode & (MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET | \
MFX_MEMTYPE_VIDEO_MEMORY_PROCESSOR_TARGET))
#define IS_OPAQUE_MEMORY(mode) (mode & MFX_MEMTYPE_OPAQUE_FRAME)
#define IS_SYSTEM_MEMORY(mode) (mode & MFX_MEMTYPE_SYSTEM_MEMORY)
#define MFX_IMPL_VIA_MASK(impl) (0x0f00 & (impl))
static const AVRational default_tb = { 1, 90000 };
static const struct {
int mfx_iopattern;
const char *desc;
} qsv_iopatterns[] = {
{MFX_IOPATTERN_IN_VIDEO_MEMORY, "input is video memory surface" },
{MFX_IOPATTERN_IN_SYSTEM_MEMORY, "input is system memory surface" },
{MFX_IOPATTERN_IN_OPAQUE_MEMORY, "input is opaque memory surface" },
{MFX_IOPATTERN_OUT_VIDEO_MEMORY, "output is video memory surface" },
{MFX_IOPATTERN_OUT_SYSTEM_MEMORY, "output is system memory surface" },
{MFX_IOPATTERN_OUT_OPAQUE_MEMORY, "output is opaque memory surface" },
};
int ff_qsvvpp_print_iopattern(void *log_ctx, int mfx_iopattern,
const char *extra_string)
{
const char *desc = NULL;
for (int i = 0; i < FF_ARRAY_ELEMS(qsv_iopatterns); i++) {
if (qsv_iopatterns[i].mfx_iopattern == mfx_iopattern) {
desc = qsv_iopatterns[i].desc;
}
}
if (!desc)
desc = "unknown iopattern";
av_log(log_ctx, AV_LOG_VERBOSE, "%s: %s\n", extra_string, desc);
return 0;
}
static const struct {
mfxStatus mfxerr;
int averr;
const char *desc;
} qsv_errors[] = {
{ MFX_ERR_NONE, 0, "success" },
{ MFX_ERR_UNKNOWN, AVERROR_UNKNOWN, "unknown error" },
{ MFX_ERR_NULL_PTR, AVERROR(EINVAL), "NULL pointer" },
{ MFX_ERR_UNSUPPORTED, AVERROR(ENOSYS), "unsupported" },
{ MFX_ERR_MEMORY_ALLOC, AVERROR(ENOMEM), "failed to allocate memory" },
{ MFX_ERR_NOT_ENOUGH_BUFFER, AVERROR(ENOMEM), "insufficient input/output buffer" },
{ MFX_ERR_INVALID_HANDLE, AVERROR(EINVAL), "invalid handle" },
{ MFX_ERR_LOCK_MEMORY, AVERROR(EIO), "failed to lock the memory block" },
{ MFX_ERR_NOT_INITIALIZED, AVERROR_BUG, "not initialized" },
{ MFX_ERR_NOT_FOUND, AVERROR(ENOSYS), "specified object was not found" },
/* the following 3 errors should always be handled explicitly, so those "mappings"
* are for completeness only */
{ MFX_ERR_MORE_DATA, AVERROR_UNKNOWN, "expect more data at input" },
{ MFX_ERR_MORE_SURFACE, AVERROR_UNKNOWN, "expect more surface at output" },
{ MFX_ERR_MORE_BITSTREAM, AVERROR_UNKNOWN, "expect more bitstream at output" },
{ MFX_ERR_ABORTED, AVERROR_UNKNOWN, "operation aborted" },
{ MFX_ERR_DEVICE_LOST, AVERROR(EIO), "device lost" },
{ MFX_ERR_INCOMPATIBLE_VIDEO_PARAM, AVERROR(EINVAL), "incompatible video parameters" },
{ MFX_ERR_INVALID_VIDEO_PARAM, AVERROR(EINVAL), "invalid video parameters" },
{ MFX_ERR_UNDEFINED_BEHAVIOR, AVERROR_BUG, "undefined behavior" },
{ MFX_ERR_DEVICE_FAILED, AVERROR(EIO), "device failed" },
{ MFX_ERR_INCOMPATIBLE_AUDIO_PARAM, AVERROR(EINVAL), "incompatible audio parameters" },
{ MFX_ERR_INVALID_AUDIO_PARAM, AVERROR(EINVAL), "invalid audio parameters" },
{ MFX_WRN_IN_EXECUTION, 0, "operation in execution" },
{ MFX_WRN_DEVICE_BUSY, 0, "device busy" },
{ MFX_WRN_VIDEO_PARAM_CHANGED, 0, "video parameters changed" },
{ MFX_WRN_PARTIAL_ACCELERATION, 0, "partial acceleration" },
{ MFX_WRN_INCOMPATIBLE_VIDEO_PARAM, 0, "incompatible video parameters" },
{ MFX_WRN_VALUE_NOT_CHANGED, 0, "value is saturated" },
{ MFX_WRN_OUT_OF_RANGE, 0, "value out of range" },
{ MFX_WRN_FILTER_SKIPPED, 0, "filter skipped" },
{ MFX_WRN_INCOMPATIBLE_AUDIO_PARAM, 0, "incompatible audio parameters" },
};
static int qsv_map_error(mfxStatus mfx_err, const char **desc)
{
int i;
for (i = 0; i < FF_ARRAY_ELEMS(qsv_errors); i++) {
if (qsv_errors[i].mfxerr == mfx_err) {
if (desc)
*desc = qsv_errors[i].desc;
return qsv_errors[i].averr;
}
}
if (desc)
*desc = "unknown error";
return AVERROR_UNKNOWN;
}
int ff_qsvvpp_print_error(void *log_ctx, mfxStatus err,
const char *error_string)
{
const char *desc;
int ret;
ret = qsv_map_error(err, &desc);
av_log(log_ctx, AV_LOG_ERROR, "%s: %s (%d)\n", error_string, desc, err);
return ret;
}
int ff_qsvvpp_print_warning(void *log_ctx, mfxStatus err,
const char *warning_string)
{
const char *desc;
int ret;
ret = qsv_map_error(err, &desc);
av_log(log_ctx, AV_LOG_WARNING, "%s: %s (%d)\n", warning_string, desc, err);
return ret;
}
/* functions for frameAlloc */
static mfxStatus frame_alloc(mfxHDL pthis, mfxFrameAllocRequest *req,
mfxFrameAllocResponse *resp)
{
QSVVPPContext *s = pthis;
int i;
if (!(req->Type & MFX_MEMTYPE_VIDEO_MEMORY_PROCESSOR_TARGET) ||
!(req->Type & (MFX_MEMTYPE_FROM_VPPIN | MFX_MEMTYPE_FROM_VPPOUT)) ||
!(req->Type & MFX_MEMTYPE_EXTERNAL_FRAME))
return MFX_ERR_UNSUPPORTED;
if (req->Type & MFX_MEMTYPE_FROM_VPPIN) {
resp->mids = av_mallocz(s->nb_surface_ptrs_in * sizeof(*resp->mids));
if (!resp->mids)
return AVERROR(ENOMEM);
for (i = 0; i < s->nb_surface_ptrs_in; i++)
resp->mids[i] = s->surface_ptrs_in[i]->Data.MemId;
resp->NumFrameActual = s->nb_surface_ptrs_in;
} else {
resp->mids = av_mallocz(s->nb_surface_ptrs_out * sizeof(*resp->mids));
if (!resp->mids)
return AVERROR(ENOMEM);
for (i = 0; i < s->nb_surface_ptrs_out; i++)
resp->mids[i] = s->surface_ptrs_out[i]->Data.MemId;
resp->NumFrameActual = s->nb_surface_ptrs_out;
}
return MFX_ERR_NONE;
}
static mfxStatus frame_free(mfxHDL pthis, mfxFrameAllocResponse *resp)
{
av_freep(&resp->mids);
return MFX_ERR_NONE;
}
static mfxStatus frame_lock(mfxHDL pthis, mfxMemId mid, mfxFrameData *ptr)
{
return MFX_ERR_UNSUPPORTED;
}
static mfxStatus frame_unlock(mfxHDL pthis, mfxMemId mid, mfxFrameData *ptr)
{
return MFX_ERR_UNSUPPORTED;
}
static mfxStatus frame_get_hdl(mfxHDL pthis, mfxMemId mid, mfxHDL *hdl)
{
mfxHDLPair *pair_dst = (mfxHDLPair*)hdl;
mfxHDLPair *pair_src = (mfxHDLPair*)mid;
pair_dst->first = pair_src->first;
if (pair_src->second != (mfxMemId)MFX_INFINITE)
pair_dst->second = pair_src->second;
return MFX_ERR_NONE;
}
static int pix_fmt_to_mfx_fourcc(int format)
{
switch (format) {
case AV_PIX_FMT_YUV420P:
return MFX_FOURCC_YV12;
case AV_PIX_FMT_NV12:
return MFX_FOURCC_NV12;
case AV_PIX_FMT_YUYV422:
return MFX_FOURCC_YUY2;
case AV_PIX_FMT_BGRA:
return MFX_FOURCC_RGB4;
}
return MFX_FOURCC_NV12;
}
static int map_frame_to_surface(AVFrame *frame, mfxFrameSurface1 *surface)
{
switch (frame->format) {
case AV_PIX_FMT_NV12:
case AV_PIX_FMT_P010:
surface->Data.Y = frame->data[0];
surface->Data.UV = frame->data[1];
break;
case AV_PIX_FMT_YUV420P:
surface->Data.Y = frame->data[0];
surface->Data.U = frame->data[1];
surface->Data.V = frame->data[2];
break;
case AV_PIX_FMT_YUYV422:
surface->Data.Y = frame->data[0];
surface->Data.U = frame->data[0] + 1;
surface->Data.V = frame->data[0] + 3;
break;
case AV_PIX_FMT_RGB32:
surface->Data.B = frame->data[0];
surface->Data.G = frame->data[0] + 1;
surface->Data.R = frame->data[0] + 2;
surface->Data.A = frame->data[0] + 3;
break;
default:
return MFX_ERR_UNSUPPORTED;
}
surface->Data.Pitch = frame->linesize[0];
return 0;
}
/* fill the surface info */
static int fill_frameinfo_by_link(mfxFrameInfo *frameinfo, AVFilterLink *link)
{
enum AVPixelFormat pix_fmt;
AVHWFramesContext *frames_ctx;
AVQSVFramesContext *frames_hwctx;
const AVPixFmtDescriptor *desc;
if (link->format == AV_PIX_FMT_QSV) {
if (!link->hw_frames_ctx)
return AVERROR(EINVAL);
frames_ctx = (AVHWFramesContext *)link->hw_frames_ctx->data;
frames_hwctx = frames_ctx->hwctx;
*frameinfo = frames_hwctx->surfaces[0].Info;
} else {
pix_fmt = link->format;
desc = av_pix_fmt_desc_get(pix_fmt);
if (!desc)
return AVERROR_BUG;
frameinfo->CropX = 0;
frameinfo->CropY = 0;
frameinfo->Width = FFALIGN(link->w, 32);
frameinfo->Height = FFALIGN(link->h, 32);
frameinfo->PicStruct = MFX_PICSTRUCT_PROGRESSIVE;
frameinfo->FourCC = pix_fmt_to_mfx_fourcc(pix_fmt);
frameinfo->BitDepthLuma = desc->comp[0].depth;
frameinfo->BitDepthChroma = desc->comp[0].depth;
frameinfo->Shift = desc->comp[0].depth > 8;
if (desc->log2_chroma_w && desc->log2_chroma_h)
frameinfo->ChromaFormat = MFX_CHROMAFORMAT_YUV420;
else if (desc->log2_chroma_w)
frameinfo->ChromaFormat = MFX_CHROMAFORMAT_YUV422;
else
frameinfo->ChromaFormat = MFX_CHROMAFORMAT_YUV444;
}
frameinfo->CropW = link->w;
frameinfo->CropH = link->h;
frameinfo->FrameRateExtN = link->frame_rate.num;
frameinfo->FrameRateExtD = link->frame_rate.den;
frameinfo->AspectRatioW = link->sample_aspect_ratio.num ? link->sample_aspect_ratio.num : 1;
frameinfo->AspectRatioH = link->sample_aspect_ratio.den ? link->sample_aspect_ratio.den : 1;
return 0;
}
static void clear_unused_frames(QSVFrame *list)
{
while (list) {
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
/* list->queued==1 means the frame is not cached in VPP
* process any more, it can be released to pool. */
if ((list->queued == 1) && !list->surface.Data.Locked) {
av_frame_free(&list->frame);
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
list->queued = 0;
}
list = list->next;
}
}
static void clear_frame_list(QSVFrame **list)
{
while (*list) {
QSVFrame *frame;
frame = *list;
*list = (*list)->next;
av_frame_free(&frame->frame);
av_freep(&frame);
}
}
static QSVFrame *get_free_frame(QSVFrame **list)
{
QSVFrame *out = *list;
for (; out; out = out->next) {
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
if (!out->queued) {
out->queued = 1;
break;
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
}
}
if (!out) {
out = av_mallocz(sizeof(*out));
if (!out) {
av_log(NULL, AV_LOG_ERROR, "Can't alloc new output frame.\n");
return NULL;
}
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
out->queued = 1;
out->next = *list;
*list = out;
}
return out;
}
/* get the input surface */
static QSVFrame *submit_frame(QSVVPPContext *s, AVFilterLink *inlink, AVFrame *picref)
{
QSVFrame *qsv_frame;
AVFilterContext *ctx = inlink->dst;
clear_unused_frames(s->in_frame_list);
qsv_frame = get_free_frame(&s->in_frame_list);
if (!qsv_frame)
return NULL;
/* Turn AVFrame into mfxFrameSurface1.
* For video/opaque memory mode, pix_fmt is AV_PIX_FMT_QSV, and
* mfxFrameSurface1 is stored in AVFrame->data[3];
* for system memory mode, raw video data is stored in
* AVFrame, we should map it into mfxFrameSurface1.
*/
if (!IS_SYSTEM_MEMORY(s->in_mem_mode)) {
if (picref->format != AV_PIX_FMT_QSV) {
av_log(ctx, AV_LOG_ERROR, "QSVVPP gets a wrong frame.\n");
return NULL;
}
qsv_frame->frame = av_frame_clone(picref);
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
qsv_frame->surface = *(mfxFrameSurface1 *)qsv_frame->frame->data[3];
} else {
/* make a copy if the input is not padded as libmfx requires */
if (picref->height & 31 || picref->linesize[0] & 31) {
qsv_frame->frame = ff_get_video_buffer(inlink,
FFALIGN(inlink->w, 32),
FFALIGN(inlink->h, 32));
if (!qsv_frame->frame)
return NULL;
qsv_frame->frame->width = picref->width;
qsv_frame->frame->height = picref->height;
if (av_frame_copy(qsv_frame->frame, picref) < 0) {
av_frame_free(&qsv_frame->frame);
return NULL;
}
av_frame_copy_props(qsv_frame->frame, picref);
} else
qsv_frame->frame = av_frame_clone(picref);
if (map_frame_to_surface(qsv_frame->frame,
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
&qsv_frame->surface) < 0) {
av_log(ctx, AV_LOG_ERROR, "Unsupported frame.\n");
return NULL;
}
}
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
qsv_frame->surface.Info = s->frame_infos[FF_INLINK_IDX(inlink)];
qsv_frame->surface.Data.TimeStamp = av_rescale_q(qsv_frame->frame->pts,
inlink->time_base, default_tb);
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
qsv_frame->surface.Info.PicStruct =
!qsv_frame->frame->interlaced_frame ? MFX_PICSTRUCT_PROGRESSIVE :
(qsv_frame->frame->top_field_first ? MFX_PICSTRUCT_FIELD_TFF :
MFX_PICSTRUCT_FIELD_BFF);
if (qsv_frame->frame->repeat_pict == 1)
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
qsv_frame->surface.Info.PicStruct |= MFX_PICSTRUCT_FIELD_REPEATED;
else if (qsv_frame->frame->repeat_pict == 2)
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
qsv_frame->surface.Info.PicStruct |= MFX_PICSTRUCT_FRAME_DOUBLING;
else if (qsv_frame->frame->repeat_pict == 4)
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
qsv_frame->surface.Info.PicStruct |= MFX_PICSTRUCT_FRAME_TRIPLING;
return qsv_frame;
}
/* get the output surface */
static QSVFrame *query_frame(QSVVPPContext *s, AVFilterLink *outlink)
{
AVFilterContext *ctx = outlink->src;
QSVFrame *out_frame;
int ret;
clear_unused_frames(s->out_frame_list);
out_frame = get_free_frame(&s->out_frame_list);
if (!out_frame)
return NULL;
/* For video memory, get a hw frame;
* For system memory, get a sw frame and map it into a mfx_surface. */
if (!IS_SYSTEM_MEMORY(s->out_mem_mode)) {
out_frame->frame = av_frame_alloc();
if (!out_frame->frame)
return NULL;
ret = av_hwframe_get_buffer(outlink->hw_frames_ctx, out_frame->frame, 0);
if (ret < 0) {
av_log(ctx, AV_LOG_ERROR, "Can't allocate a surface.\n");
return NULL;
}
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
out_frame->surface = *(mfxFrameSurface1 *)out_frame->frame->data[3];
} else {
/* Get a frame with aligned dimensions.
* Libmfx need system memory being 128x64 aligned */
out_frame->frame = ff_get_video_buffer(outlink,
FFALIGN(outlink->w, 128),
FFALIGN(outlink->h, 64));
if (!out_frame->frame)
return NULL;
out_frame->frame->width = outlink->w;
out_frame->frame->height = outlink->h;
ret = map_frame_to_surface(out_frame->frame,
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
&out_frame->surface);
if (ret < 0)
return NULL;
}
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
out_frame->surface.Info = s->vpp_param.vpp.Out;
return out_frame;
}
/* create the QSV session */
static int init_vpp_session(AVFilterContext *avctx, QSVVPPContext *s)
{
AVFilterLink *inlink = avctx->inputs[0];
AVFilterLink *outlink = avctx->outputs[0];
AVQSVFramesContext *in_frames_hwctx = NULL;
AVQSVFramesContext *out_frames_hwctx = NULL;
AVBufferRef *device_ref;
AVHWDeviceContext *device_ctx;
AVQSVDeviceContext *device_hwctx;
mfxHDL handle;
mfxHandleType handle_type;
mfxVersion ver;
mfxIMPL impl;
int ret, i;
if (inlink->hw_frames_ctx) {
AVHWFramesContext *frames_ctx = (AVHWFramesContext *)inlink->hw_frames_ctx->data;
device_ref = frames_ctx->device_ref;
in_frames_hwctx = frames_ctx->hwctx;
s->in_mem_mode = in_frames_hwctx->frame_type;
s->surface_ptrs_in = av_calloc(in_frames_hwctx->nb_surfaces,
sizeof(*s->surface_ptrs_in));
if (!s->surface_ptrs_in)
return AVERROR(ENOMEM);
for (i = 0; i < in_frames_hwctx->nb_surfaces; i++)
s->surface_ptrs_in[i] = in_frames_hwctx->surfaces + i;
s->nb_surface_ptrs_in = in_frames_hwctx->nb_surfaces;
} else if (avctx->hw_device_ctx) {
device_ref = avctx->hw_device_ctx;
s->in_mem_mode = MFX_MEMTYPE_SYSTEM_MEMORY;
} else {
av_log(avctx, AV_LOG_ERROR, "No hw context provided.\n");
return AVERROR(EINVAL);
}
device_ctx = (AVHWDeviceContext *)device_ref->data;
device_hwctx = device_ctx->hwctx;
if (outlink->format == AV_PIX_FMT_QSV) {
AVHWFramesContext *out_frames_ctx;
AVBufferRef *out_frames_ref = av_hwframe_ctx_alloc(device_ref);
if (!out_frames_ref)
return AVERROR(ENOMEM);
s->out_mem_mode = IS_OPAQUE_MEMORY(s->in_mem_mode) ?
MFX_MEMTYPE_OPAQUE_FRAME :
MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET | MFX_MEMTYPE_FROM_VPPOUT;
out_frames_ctx = (AVHWFramesContext *)out_frames_ref->data;
out_frames_hwctx = out_frames_ctx->hwctx;
out_frames_ctx->format = AV_PIX_FMT_QSV;
out_frames_ctx->width = FFALIGN(outlink->w, 32);
out_frames_ctx->height = FFALIGN(outlink->h, 32);
out_frames_ctx->sw_format = s->out_sw_format;
out_frames_ctx->initial_pool_size = 64;
if (avctx->extra_hw_frames > 0)
out_frames_ctx->initial_pool_size += avctx->extra_hw_frames;
out_frames_hwctx->frame_type = s->out_mem_mode;
ret = av_hwframe_ctx_init(out_frames_ref);
if (ret < 0) {
av_buffer_unref(&out_frames_ref);
av_log(avctx, AV_LOG_ERROR, "Error creating frames_ctx for output pad.\n");
return ret;
}
s->surface_ptrs_out = av_calloc(out_frames_hwctx->nb_surfaces,
sizeof(*s->surface_ptrs_out));
if (!s->surface_ptrs_out) {
av_buffer_unref(&out_frames_ref);
return AVERROR(ENOMEM);
}
for (i = 0; i < out_frames_hwctx->nb_surfaces; i++)
s->surface_ptrs_out[i] = out_frames_hwctx->surfaces + i;
s->nb_surface_ptrs_out = out_frames_hwctx->nb_surfaces;
av_buffer_unref(&outlink->hw_frames_ctx);
outlink->hw_frames_ctx = out_frames_ref;
} else
s->out_mem_mode = MFX_MEMTYPE_SYSTEM_MEMORY;
/* extract the properties of the "master" session given to us */
ret = MFXQueryIMPL(device_hwctx->session, &impl);
if (ret == MFX_ERR_NONE)
ret = MFXQueryVersion(device_hwctx->session, &ver);
if (ret != MFX_ERR_NONE) {
av_log(avctx, AV_LOG_ERROR, "Error querying the session attributes\n");
return AVERROR_UNKNOWN;
}
if (MFX_IMPL_VIA_VAAPI == MFX_IMPL_VIA_MASK(impl)) {
handle_type = MFX_HANDLE_VA_DISPLAY;
} else if (MFX_IMPL_VIA_D3D11 == MFX_IMPL_VIA_MASK(impl)) {
handle_type = MFX_HANDLE_D3D11_DEVICE;
} else if (MFX_IMPL_VIA_D3D9 == MFX_IMPL_VIA_MASK(impl)) {
handle_type = MFX_HANDLE_D3D9_DEVICE_MANAGER;
} else {
av_log(avctx, AV_LOG_ERROR, "Error unsupported handle type\n");
return AVERROR_UNKNOWN;
}
ret = MFXVideoCORE_GetHandle(device_hwctx->session, handle_type, &handle);
if (ret < 0)
return ff_qsvvpp_print_error(avctx, ret, "Error getting the session handle");
else if (ret > 0) {
ff_qsvvpp_print_warning(avctx, ret, "Warning in getting the session handle");
return AVERROR_UNKNOWN;
}
/* create a "slave" session with those same properties, to be used for vpp */
ret = MFXInit(impl, &ver, &s->session);
if (ret < 0)
return ff_qsvvpp_print_error(avctx, ret, "Error initializing a session");
else if (ret > 0) {
ff_qsvvpp_print_warning(avctx, ret, "Warning in session initialization");
return AVERROR_UNKNOWN;
}
if (handle) {
ret = MFXVideoCORE_SetHandle(s->session, handle_type, handle);
if (ret != MFX_ERR_NONE)
return AVERROR_UNKNOWN;
}
if (QSV_RUNTIME_VERSION_ATLEAST(ver, 1, 25)) {
ret = MFXJoinSession(device_hwctx->session, s->session);
if (ret != MFX_ERR_NONE)
return AVERROR_UNKNOWN;
}
if (IS_OPAQUE_MEMORY(s->in_mem_mode) || IS_OPAQUE_MEMORY(s->out_mem_mode)) {
s->opaque_alloc.In.Surfaces = s->surface_ptrs_in;
s->opaque_alloc.In.NumSurface = s->nb_surface_ptrs_in;
s->opaque_alloc.In.Type = s->in_mem_mode;
s->opaque_alloc.Out.Surfaces = s->surface_ptrs_out;
s->opaque_alloc.Out.NumSurface = s->nb_surface_ptrs_out;
s->opaque_alloc.Out.Type = s->out_mem_mode;
s->opaque_alloc.Header.BufferId = MFX_EXTBUFF_OPAQUE_SURFACE_ALLOCATION;
s->opaque_alloc.Header.BufferSz = sizeof(s->opaque_alloc);
} else if (IS_VIDEO_MEMORY(s->in_mem_mode) || IS_VIDEO_MEMORY(s->out_mem_mode)) {
mfxFrameAllocator frame_allocator = {
.pthis = s,
.Alloc = frame_alloc,
.Lock = frame_lock,
.Unlock = frame_unlock,
.GetHDL = frame_get_hdl,
.Free = frame_free,
};
ret = MFXVideoCORE_SetFrameAllocator(s->session, &frame_allocator);
if (ret != MFX_ERR_NONE)
return AVERROR_UNKNOWN;
}
return 0;
}
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
static unsigned int qsv_fifo_item_size(void)
{
return sizeof(mfxSyncPoint) + sizeof(QSVFrame*);
}
static unsigned int qsv_fifo_size(const AVFifoBuffer* fifo)
{
return av_fifo_size(fifo)/qsv_fifo_item_size();
}
int ff_qsvvpp_create(AVFilterContext *avctx, QSVVPPContext **vpp, QSVVPPParam *param)
{
int i;
int ret;
QSVVPPContext *s;
s = av_mallocz(sizeof(*s));
if (!s)
return AVERROR(ENOMEM);
s->filter_frame = param->filter_frame;
if (!s->filter_frame)
s->filter_frame = ff_filter_frame;
s->out_sw_format = param->out_sw_format;
/* create the vpp session */
ret = init_vpp_session(avctx, s);
if (ret < 0)
goto failed;
s->frame_infos = av_calloc(avctx->nb_inputs, sizeof(*s->frame_infos));
if (!s->frame_infos) {
ret = AVERROR(ENOMEM);
goto failed;
}
/* Init each input's information */
for (i = 0; i < avctx->nb_inputs; i++) {
ret = fill_frameinfo_by_link(&s->frame_infos[i], avctx->inputs[i]);
if (ret < 0)
goto failed;
}
/* Update input's frame info according to crop */
for (i = 0; i < param->num_crop; i++) {
QSVVPPCrop *crop = param->crop + i;
if (crop->in_idx > avctx->nb_inputs) {
ret = AVERROR(EINVAL);
goto failed;
}
s->frame_infos[crop->in_idx].CropX = crop->x;
s->frame_infos[crop->in_idx].CropY = crop->y;
s->frame_infos[crop->in_idx].CropW = crop->w;
s->frame_infos[crop->in_idx].CropH = crop->h;
}
s->vpp_param.vpp.In = s->frame_infos[0];
ret = fill_frameinfo_by_link(&s->vpp_param.vpp.Out, avctx->outputs[0]);
if (ret < 0) {
av_log(avctx, AV_LOG_ERROR, "Fail to get frame info from link.\n");
goto failed;
}
if (IS_OPAQUE_MEMORY(s->in_mem_mode) || IS_OPAQUE_MEMORY(s->out_mem_mode)) {
s->nb_ext_buffers = param->num_ext_buf + 1;
s->ext_buffers = av_calloc(s->nb_ext_buffers, sizeof(*s->ext_buffers));
if (!s->ext_buffers) {
ret = AVERROR(ENOMEM);
goto failed;
}
s->ext_buffers[0] = (mfxExtBuffer *)&s->opaque_alloc;
for (i = 1; i < param->num_ext_buf; i++)
s->ext_buffers[i] = param->ext_buf[i - 1];
s->vpp_param.ExtParam = s->ext_buffers;
s->vpp_param.NumExtParam = s->nb_ext_buffers;
} else {
s->vpp_param.NumExtParam = param->num_ext_buf;
s->vpp_param.ExtParam = param->ext_buf;
}
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
s->got_frame = 0;
/** keep fifo size at least 1. Even when async_depth is 0, fifo is used. */
s->async_fifo = av_fifo_alloc((param->async_depth + 1) * qsv_fifo_item_size());
s->async_depth = param->async_depth;
if (!s->async_fifo) {
ret = AVERROR(ENOMEM);
goto failed;
}
s->vpp_param.AsyncDepth = param->async_depth;
if (IS_SYSTEM_MEMORY(s->in_mem_mode))
s->vpp_param.IOPattern |= MFX_IOPATTERN_IN_SYSTEM_MEMORY;
else if (IS_VIDEO_MEMORY(s->in_mem_mode))
s->vpp_param.IOPattern |= MFX_IOPATTERN_IN_VIDEO_MEMORY;
else if (IS_OPAQUE_MEMORY(s->in_mem_mode))
s->vpp_param.IOPattern |= MFX_IOPATTERN_IN_OPAQUE_MEMORY;
if (IS_SYSTEM_MEMORY(s->out_mem_mode))
s->vpp_param.IOPattern |= MFX_IOPATTERN_OUT_SYSTEM_MEMORY;
else if (IS_VIDEO_MEMORY(s->out_mem_mode))
s->vpp_param.IOPattern |= MFX_IOPATTERN_OUT_VIDEO_MEMORY;
else if (IS_OPAQUE_MEMORY(s->out_mem_mode))
s->vpp_param.IOPattern |= MFX_IOPATTERN_OUT_OPAQUE_MEMORY;
/* Print input memory mode */
ff_qsvvpp_print_iopattern(avctx, s->vpp_param.IOPattern & 0x0F, "VPP");
/* Print output memory mode */
ff_qsvvpp_print_iopattern(avctx, s->vpp_param.IOPattern & 0xF0, "VPP");
ret = MFXVideoVPP_Init(s->session, &s->vpp_param);
if (ret < 0) {
ret = ff_qsvvpp_print_error(avctx, ret, "Failed to create a qsvvpp");
goto failed;
} else if (ret > 0)
ff_qsvvpp_print_warning(avctx, ret, "Warning When creating qsvvpp");
*vpp = s;
return 0;
failed:
ff_qsvvpp_free(&s);
return ret;
}
int ff_qsvvpp_free(QSVVPPContext **vpp)
{
QSVVPPContext *s = *vpp;
if (!s)
return 0;
if (s->session) {
MFXVideoVPP_Close(s->session);
MFXClose(s->session);
}
/* release all the resources */
clear_frame_list(&s->in_frame_list);
clear_frame_list(&s->out_frame_list);
av_freep(&s->surface_ptrs_in);
av_freep(&s->surface_ptrs_out);
av_freep(&s->ext_buffers);
av_freep(&s->frame_infos);
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
av_fifo_free(s->async_fifo);
av_freep(vpp);
return 0;
}
int ff_qsvvpp_filter_frame(QSVVPPContext *s, AVFilterLink *inlink, AVFrame *picref)
{
AVFilterContext *ctx = inlink->dst;
AVFilterLink *outlink = ctx->outputs[0];
mfxSyncPoint sync;
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
QSVFrame *in_frame, *out_frame, *tmp;
int ret, filter_ret;
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
while (s->eof && qsv_fifo_size(s->async_fifo)) {
av_fifo_generic_read(s->async_fifo, &tmp, sizeof(tmp), NULL);
av_fifo_generic_read(s->async_fifo, &sync, sizeof(sync), NULL);
if (MFXVideoCORE_SyncOperation(s->session, sync, 1000) < 0)
av_log(ctx, AV_LOG_WARNING, "Sync failed.\n");
filter_ret = s->filter_frame(outlink, tmp->frame);
if (filter_ret < 0) {
av_frame_free(&tmp->frame);
return filter_ret;
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
}
tmp->queued--;
s->got_frame = 1;
tmp->frame = NULL;
};
if (!picref)
return 0;
in_frame = submit_frame(s, inlink, picref);
if (!in_frame) {
av_log(ctx, AV_LOG_ERROR, "Failed to submit frame on input[%d]\n",
FF_INLINK_IDX(inlink));
return AVERROR(ENOMEM);
}
do {
out_frame = query_frame(s, outlink);
if (!out_frame) {
av_log(ctx, AV_LOG_ERROR, "Failed to query an output frame.\n");
return AVERROR(ENOMEM);
}
do {
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
ret = MFXVideoVPP_RunFrameVPPAsync(s->session, &in_frame->surface,
&out_frame->surface, NULL, &sync);
if (ret == MFX_WRN_DEVICE_BUSY)
av_usleep(500);
} while (ret == MFX_WRN_DEVICE_BUSY);
if (ret < 0 && ret != MFX_ERR_MORE_SURFACE) {
/* Ignore more_data error */
if (ret == MFX_ERR_MORE_DATA)
return AVERROR(EAGAIN);
break;
}
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
out_frame->frame->pts = av_rescale_q(out_frame->surface.Data.TimeStamp,
default_tb, outlink->time_base);
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
out_frame->queued++;
av_fifo_generic_write(s->async_fifo, &out_frame, sizeof(out_frame), NULL);
av_fifo_generic_write(s->async_fifo, &sync, sizeof(sync), NULL);
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
if (qsv_fifo_size(s->async_fifo) > s->async_depth) {
av_fifo_generic_read(s->async_fifo, &tmp, sizeof(tmp), NULL);
av_fifo_generic_read(s->async_fifo, &sync, sizeof(sync), NULL);
do {
ret = MFXVideoCORE_SyncOperation(s->session, sync, 1000);
} while (ret == MFX_WRN_IN_EXECUTION);
filter_ret = s->filter_frame(outlink, tmp->frame);
if (filter_ret < 0) {
av_frame_free(&tmp->frame);
return filter_ret;
lavfi/qsvvpp: support async depth Async depth will allow qsv filter cache few frames, and avoid force switch and end filter task frame by frame. This change will improve performance for some multi-task case, for example 1:N transcode( decode + vpp + encode) with all QSV plugins. Performance data test on my Coffee Lake Desktop(i7-8700K) by using the following 1:8 transcode test case improvement: 1. Fps improved from 55 to 130. 2. Render/Video usage improved from ~61%/~38% to ~100%/~70%.(Data get from intel_gpu_top) test CMD: ffmpeg -v verbose -init_hw_device qsv=hw:/dev/dri/renderD128 -filter_hw_device \ hw -hwaccel qsv -hwaccel_output_format qsv -c:v h264_qsv -i 1920x1080.264 \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - \ -vf 'vpp_qsv=w=1280:h=720:async_depth=4' -c:v h264_qsv -r:v 30 -preset 7 -g 33 -refs 2 -bf 3 -q 24 -f null - Signed-off-by: Fei Wang <fei.w.wang@intel.com> Reviewed-by: Linjie Fu <linjie.justin.fu@gmail.com> Signed-off-by: Zhong Li <zhongli_dev@126.com>
2021-03-31 04:07:44 +02:00
}
tmp->queued--;
s->got_frame = 1;
tmp->frame = NULL;
}
} while(ret == MFX_ERR_MORE_SURFACE);
if (ret < 0)
return ff_qsvvpp_print_error(ctx, ret, "Error running VPP");
else if (ret > 0)
ff_qsvvpp_print_warning(ctx, ret, "Warning in running VPP");
return 0;
}