[FFmpeg-devel] [PATCH, v2] lavc/qsvdec: Add GPU-accelerated memory copy support
Li, Zhong
zhong.li at intel.com
Sun Sep 29 06:57:23 EEST 2019
> From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of Linjie Fu
> Sent: Friday, September 27, 2019 1:47 PM
> To: ffmpeg-devel at ffmpeg.org
> Cc: ChaoX A Liu <chaox.a.liu at intel.com>; Fu, Linjie <linjie.fu at intel.com>
> Subject: [FFmpeg-devel] [PATCH, v2] lavc/qsvdec: Add GPU-accelerated memory
> copy support
>
> GPU copy enables or disables GPU accelerated copying between video and
> system memory. This may lead to a notable performance improvement.
> Memory must be sequent and aligned with 128x64.
> (first introduced in FFmpeg 3.3.1)
This line should be removed. FFmpeg 3.3.1 mainline never support GPU copy.
>
> CMD:
> ffmpeg -init_hw_device qsv=hw -filter_hw_device hw -c:v h264_qsv
> -gpu_copy on -i input.h264 -f null -
> or:
> ffmpeg -c:v h264_qsv -gpu_copy on -i input.h264 -f null -
>
> Signed-off-by: Linjie Fu <linjie.fu at intel.com>
> Signed-off-by: ChaoX A Liu <chaox.a.liu at intel.com>
> ---
> Rebased and send again.
>
> libavcodec/qsv.c | 31 +++++++++++++++++-------
> libavcodec/qsv_internal.h | 7 +++---
> libavcodec/qsvdec.c | 50 ++++++++++++++++++++++++++++++++++-----
> libavcodec/qsvdec.h | 2 ++
> libavcodec/qsvdec_h2645.c | 10 ++++++++ libavcodec/qsvdec_other.c | 5 ++++
> libavcodec/qsvenc.c | 8 ++++---
> 7 files changed, 92 insertions(+), 21 deletions(-)
>
> diff --git a/libavcodec/qsv.c b/libavcodec/qsv.c index 994c9ebcb0..9e66fbc9da
> 100644
> --- a/libavcodec/qsv.c
> +++ b/libavcodec/qsv.c
> @@ -412,15 +412,19 @@ static int ff_qsv_set_display_handle(AVCodecContext
> *avctx, QSVSession *qs) #endif //AVCODEC_QSV_LINUX_SESSION_HANDLE
>
> int ff_qsv_init_internal_session(AVCodecContext *avctx, QSVSession *qs,
> - const char *load_plugins)
> + const char *load_plugins, int
> + gpu_copy)
> {
> - mfxIMPL impl = MFX_IMPL_AUTO_ANY;
> - mfxVersion ver = { { QSV_VERSION_MINOR, QSV_VERSION_MAJOR } };
> + mfxIMPL impl = MFX_IMPL_AUTO_ANY;
> + mfxVersion ver = { { QSV_VERSION_MINOR, QSV_VERSION_MAJOR } };
> + mfxInitParam init_par = { MFX_IMPL_AUTO_ANY };
>
> const char *desc;
> int ret;
>
> - ret = MFXInit(impl, &ver, &qs->session);
> + init_par.GPUCopy = gpu_copy;
GPUCopy field is introduced from API 1.16, would better to check it to avoid compile issue with old API.
> + init_par.Implementation = impl;
> + init_par.Version = ver;
> + ret = MFXInitEx(init_par, &qs->session);
> if (ret < 0)
> return ff_qsv_print_error(avctx, ret,
> "Error initializing an internal MFX session"); @@ -712,7
> +716,8 @@ static mfxStatus qsv_frame_get_hdl(mfxHDL pthis, mfxMemId mid,
> mfxHDL *hdl) }
>
> int ff_qsv_init_session_device(AVCodecContext *avctx, mfxSession *psession,
> - AVBufferRef *device_ref, const char *load_plugins)
> + AVBufferRef *device_ref, const char *load_plugins,
> + int gpu_copy)
> {
> static const mfxHandleType handle_types[] = {
> MFX_HANDLE_VA_DISPLAY,
> @@ -722,11 +727,12 @@ int ff_qsv_init_session_device(AVCodecContext
> *avctx, mfxSession *psession,
> AVHWDeviceContext *device_ctx = (AVHWDeviceContext*)device_ref-
> >data;
> AVQSVDeviceContext *device_hwctx = device_ctx->hwctx;
> mfxSession parent_session = device_hwctx->session;
> + mfxInitParam init_par = { MFX_IMPL_AUTO_ANY };
> + mfxHDL handle = NULL;
>
> mfxSession session;
> mfxVersion ver;
> mfxIMPL impl;
> - mfxHDL handle = NULL;
> mfxHandleType handle_type;
> mfxStatus err;
>
> @@ -752,7 +758,10 @@ int ff_qsv_init_session_device(AVCodecContext *avctx,
> mfxSession *psession,
> "from the session\n");
> }
>
> - err = MFXInit(impl, &ver, &session);
> + init_par.GPUCopy = gpu_copy;
> + init_par.Implementation = impl;
> + init_par.Version = ver;
> + err = MFXInitEx(init_par, &session);
> if (err != MFX_ERR_NONE)
> return ff_qsv_print_error(avctx, err,
> "Error initializing a child MFX session"); @@ -783,7 +792,7
> @@ int ff_qsv_init_session_device(AVCodecContext *avctx, mfxSession
> *psession,
>
> int ff_qsv_init_session_frames(AVCodecContext *avctx, mfxSession *psession,
> QSVFramesContext *qsv_frames_ctx,
> - const char *load_plugins, int opaque)
> + const char *load_plugins, int opaque,
> + int gpu_copy)
> {
> mfxFrameAllocator frame_allocator = {
> .pthis = qsv_frames_ctx,
> @@ -802,8 +811,12 @@ int ff_qsv_init_session_frames(AVCodecContext *avctx,
> mfxSession *psession,
>
> int ret;
>
> + if (gpu_copy == MFX_GPUCOPY_ON)
> + av_log(avctx, AV_LOG_WARNING, "GPU-accelerated memory copy "
> + "only works in
> + MFX_IOPATTERN_OUT_SYSTEM_MEMORY.\n");
This looks weird：
1. the waring log will always turn on if gpu_copy is true no matter what iopattern.
So would be better:
if (gpu_copy == MFX_GPUCOPY_ON && iopattern != system memory)
print a warning.
2. It is only added for ff_qsv_init_session_frames(), but looks like should be apply for qsv_init_session()
> ret = ff_qsv_init_session_device(avctx, &session,
> - frames_ctx->device_ref, load_plugins);
> + frames_ctx->device_ref,
> + load_plugins, gpu_copy);
> if (ret < 0)
> return ret;
>
> diff --git a/libavcodec/qsv_internal.h b/libavcodec/qsv_internal.h index
> 8b44a9b6f4..37559270e5 100644
> --- a/libavcodec/qsv_internal.h
> +++ b/libavcodec/qsv_internal.h
> @@ -127,16 +127,17 @@ enum AVPictureType ff_qsv_map_pictype(int
> mfx_pic_type); enum AVFieldOrder ff_qsv_map_picstruct(int mfx_pic_struct);
>
> int ff_qsv_init_internal_session(AVCodecContext *avctx, QSVSession *qs,
> - const char *load_plugins);
> + const char *load_plugins, int
> + gpu_copy);
>
> int ff_qsv_close_internal_session(QSVSession *qs);
>
> int ff_qsv_init_session_device(AVCodecContext *avctx, mfxSession *psession,
> - AVBufferRef *device_ref, const char *load_plugins);
> + AVBufferRef *device_ref, const char *load_plugins,
> + int gpu_copy);
>
> int ff_qsv_init_session_frames(AVCodecContext *avctx, mfxSession *session,
> QSVFramesContext *qsv_frames_ctx,
> - const char *load_plugins, int opaque);
> + const char *load_plugins, int opaque,
> + int gpu_copy);
>
> int ff_qsv_find_surface_idx(QSVFramesContext *ctx, QSVFrame *frame);
>
> diff --git a/libavcodec/qsvdec.c b/libavcodec/qsvdec.c index
> 9299596e33..a947583702 100644
> --- a/libavcodec/qsvdec.c
> +++ b/libavcodec/qsvdec.c
> @@ -34,9 +34,11 @@
> #include "libavutil/pixdesc.h"
> #include "libavutil/pixfmt.h"
> #include "libavutil/time.h"
> +#include "libavutil/imgutils.h"
>
> #include "avcodec.h"
> #include "internal.h"
> +#include "decode.h"
> #include "qsv.h"
> #include "qsv_internal.h"
> #include "qsvdec.h"
> @@ -54,6 +56,31 @@ const AVCodecHWConfigInternal *ff_qsv_hw_configs[] = {
> NULL
> };
>
> +static int ff_qsv_get_continuous_buffer(AVCodecContext *avctx, AVFrame
> +*frame, AVBufferPool *pool) {
> + int ret = 0;
> +
> + ff_decode_frame_props(avctx, frame);
> +
> + frame->width = avctx->width;
> + frame->height = avctx->height;
> + frame->linesize[0] = FFALIGN(avctx->width, 128);
> + frame->linesize[1] = frame->linesize[0];
> + frame->buf[0] = av_buffer_pool_get(pool);
> + if (!frame->buf[0])
> + return AVERROR(ENOMEM);
> +
> + frame->data[0] = frame->buf[0]->data;
> + frame->data[1] = frame->data[0] +
> + frame->linesize[0] * FFALIGN(avctx->height,
> + 64);
> +
> + ret = ff_attach_decode_data(frame);
Could you please explain why need this function? I don't see private_ref is needed from qsv decoding.
> + if (ret < 0)
> + return ret;
> +
> + return 0;
> +}
> +
> static int qsv_init_session(AVCodecContext *avctx, QSVContext *q, mfxSession
> session,
> AVBufferRef *hw_frames_ref, AVBufferRef *hw_device_ref)
> { @@ -74,7 +101,8 @@ static int qsv_init_session(AVCodecContext *avctx,
> QSVContext *q, mfxSession ses
>
> ret = ff_qsv_init_session_frames(avctx, &q->internal_qs.session,
> &q->frames_ctx, q->load_plugins,
> - q->iopattern ==
> MFX_IOPATTERN_OUT_OPAQUE_MEMORY);
> + q->iopattern ==
> MFX_IOPATTERN_OUT_OPAQUE_MEMORY,
> + q->gpu_copy);
> if (ret < 0) {
> av_buffer_unref(&q->frames_ctx.hw_frames_ctx);
> return ret;
> @@ -88,7 +116,7 @@ static int qsv_init_session(AVCodecContext *avctx,
> QSVContext *q, mfxSession ses
> }
>
> ret = ff_qsv_init_session_device(avctx, &q->internal_qs.session,
> - hw_device_ref, q->load_plugins);
> + hw_device_ref,
> + q->load_plugins, q->gpu_copy);
> if (ret < 0)
> return ret;
>
> @@ -96,7 +124,7 @@ static int qsv_init_session(AVCodecContext *avctx,
> QSVContext *q, mfxSession ses
> } else {
> if (!q->internal_qs.session) {
> ret = ff_qsv_init_internal_session(avctx, &q->internal_qs,
> - q->load_plugins);
> + q->load_plugins,
> + q->gpu_copy);
> if (ret < 0)
> return ret;
> }
> @@ -229,6 +257,9 @@ static int qsv_decode_init(AVCodecContext *avctx,
> QSVContext *q, mfxVideoParam *
>
> q->frame_info = param->mfx.FrameInfo;
>
> + if (!avctx->hw_frames_ctx)
> + q->pool = av_buffer_pool_init(av_image_get_buffer_size(avctx->pix_fmt,
> + FFALIGN(avctx->width, 128), FFALIGN(avctx->height,
> + 64), 1), av_buffer_allocz);
> return 0;
> }
>
> @@ -275,9 +306,15 @@ static int alloc_frame(AVCodecContext *avctx,
> QSVContext *q, QSVFrame *frame) {
> int ret;
>
> - ret = ff_get_buffer(avctx, frame->frame, AV_GET_BUFFER_FLAG_REF);
> - if (ret < 0)
> - return ret;
> + if (!q->pool) {
> + ret = ff_get_buffer(avctx, frame->frame, AV_GET_BUFFER_FLAG_REF);
> + if (ret < 0)
> + return ret;
> + } else {
> + ret = ff_qsv_get_continuous_buffer(avctx, frame->frame, q->pool);
> + if (ret < 0)
> + return ret;
> + }
>
> if (frame->frame->format == AV_PIX_FMT_QSV) {
> frame->surface = *(mfxFrameSurface1*)frame->frame->data[3];
> @@ -535,6 +572,7 @@ int ff_qsv_decode_close(QSVContext *q)
>
> av_buffer_unref(&q->frames_ctx.hw_frames_ctx);
> av_buffer_unref(&q->frames_ctx.mids_buf);
> + av_buffer_pool_uninit(&q->pool);
>
> return 0;
> }
> diff --git a/libavcodec/qsvdec.h b/libavcodec/qsvdec.h index
> 64dc8d2f47..dec1f61ceb 100644
> --- a/libavcodec/qsvdec.h
> +++ b/libavcodec/qsvdec.h
> @@ -59,12 +59,14 @@ typedef struct QSVContext {
> enum AVPixelFormat orig_pix_fmt;
> uint32_t fourcc;
> mfxFrameInfo frame_info;
> + AVBufferPool *pool;
>
> int initialized;
>
> // options set by the caller
> int async_depth;
> int iopattern;
> + int gpu_copy;
>
> char *load_plugins;
>
> diff --git a/libavcodec/qsvdec_h2645.c b/libavcodec/qsvdec_h2645.c index
> eb1dc336a4..d7ac00689a 100644
> --- a/libavcodec/qsvdec_h2645.c
> +++ b/libavcodec/qsvdec_h2645.c
> @@ -193,6 +193,11 @@ static const AVOption hevc_options[] = {
>
> { "load_plugins", "A :-separate list of hexadecimal plugin UIDs to load in an
> internal session",
> OFFSET(qsv.load_plugins), AV_OPT_TYPE_STRING, { .str = "" }, 0, 0, VD },
> +
> + { "gpu_copy", "A GPU-accelerated memory copy between video and system
> memory", OFFSET(qsv.gpu_copy), AV_OPT_TYPE_INT, { .i64 =
> MFX_GPUCOPY_DEFAULT }, MFX_GPUCOPY_DEFAULT, MFX_GPUCOPY_OFF, VD,
> "gpu_copy"},
> + { "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 =
> MFX_GPUCOPY_DEFAULT }, 0, 0, VD, "gpu_copy"},
> + { "on", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_GPUCOPY_ON },
> 0, 0, VD, "gpu_copy"},
> + { "off", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_GPUCOPY_OFF },
> 0, 0, VD, "gpu_copy"},
> { NULL },
> };
>
> @@ -228,6 +233,11 @@ AVCodec ff_hevc_qsv_decoder = { #if
> CONFIG_H264_QSV_DECODER static const AVOption options[] = {
> { "async_depth", "Internal parallelization depth, the higher the value the higher
> the latency.", OFFSET(qsv.async_depth), AV_OPT_TYPE_INT, { .i64 =
> ASYNC_DEPTH_DEFAULT }, 1, INT_MAX, VD },
> +
> + { "gpu_copy", "A GPU-accelerated copy between video and system memory",
> OFFSET(qsv.gpu_copy), AV_OPT_TYPE_INT, { .i64 = MFX_GPUCOPY_DEFAULT },
> MFX_GPUCOPY_DEFAULT, MFX_GPUCOPY_OFF, VD, "gpu_copy"},
> + { "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 =
> MFX_GPUCOPY_DEFAULT }, 0, 0, VD, "gpu_copy"},
> + { "on", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_GPUCOPY_ON }, 0,
> 0, VD, "gpu_copy"},
> + { "off", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_GPUCOPY_OFF }, 0,
> 0, VD, "gpu_copy"},
> { NULL },
> };
>
> diff --git a/libavcodec/qsvdec_other.c b/libavcodec/qsvdec_other.c index
> b044c11540..13b1a99113 100644
> --- a/libavcodec/qsvdec_other.c
> +++ b/libavcodec/qsvdec_other.c
> @@ -181,6 +181,11 @@ static void qsv_decode_flush(AVCodecContext *avctx)
> #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
> static const AVOption options[] = {
> { "async_depth", "Internal parallelization depth, the higher the value the higher
> the latency.", OFFSET(qsv.async_depth), AV_OPT_TYPE_INT, { .i64 =
> ASYNC_DEPTH_DEFAULT }, 1, INT_MAX, VD },
> +
> + { "gpu_copy", "A GPU-accelerated memory copy between video and system
> memory", OFFSET(qsv.gpu_copy), AV_OPT_TYPE_INT, { .i64 =
> MFX_GPUCOPY_DEFAULT }, MFX_GPUCOPY_DEFAULT, MFX_GPUCOPY_OFF, VD,
> "gpu_copy"},
> + { "default", NULL, 0, AV_OPT_TYPE_CONST, { .i64 =
> MFX_GPUCOPY_DEFAULT }, 0, 0, VD, "gpu_copy"},
> + { "on", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_GPUCOPY_ON }, 0,
> 0, VD, "gpu_copy"},
> + { "off", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = MFX_GPUCOPY_OFF }, 0,
> 0, VD, "gpu_copy"},
> { NULL },
> };
>
> diff --git a/libavcodec/qsvenc.c b/libavcodec/qsvenc.c index
> 207cdc1d61..ba85d645ca 100644
> --- a/libavcodec/qsvenc.c
> +++ b/libavcodec/qsvenc.c
> @@ -956,7 +956,8 @@ static int qsvenc_init_session(AVCodecContext *avctx,
> QSVEncContext *q)
>
> ret = ff_qsv_init_session_frames(avctx, &q->internal_qs.session,
> &q->frames_ctx, q->load_plugins,
> - q->param.IOPattern ==
> MFX_IOPATTERN_IN_OPAQUE_MEMORY);
> + q->param.IOPattern ==
> MFX_IOPATTERN_IN_OPAQUE_MEMORY,
> + MFX_GPUCOPY_OFF);
> if (ret < 0) {
> av_buffer_unref(&q->frames_ctx.hw_frames_ctx);
> return ret;
> @@ -965,14 +966,15 @@ static int qsvenc_init_session(AVCodecContext *avctx,
> QSVEncContext *q)
> q->session = q->internal_qs.session;
> } else if (avctx->hw_device_ctx) {
> ret = ff_qsv_init_session_device(avctx, &q->internal_qs.session,
> - avctx->hw_device_ctx, q->load_plugins);
> + avctx->hw_device_ctx, q->load_plugins,
> + MFX_GPUCOPY_OFF);
> if (ret < 0)
> return ret;
>
> q->session = q->internal_qs.session;
> } else {
> ret = ff_qsv_init_internal_session(avctx, &q->internal_qs,
> - q->load_plugins);
> + q->load_plugins,
> + MFX_GPUCOPY_OFF);
> if (ret < 0)
> return ret;
>
> --
> 2.17.1
More information about the ffmpeg-devel
mailing list