On 02/08/2017 09:52, Huang, Zhengxu wrote:
From fc3c6ff4c738d5cf6de84fb6d210efbe3f1a491f Mon Sep 17 00:00:00 2001
From: "Huang, Zhengxu" <[email protected]>
Date: Tue, 25 Jul 2017 21:55:50 +0800
Subject: [PATCH 1/2] libavfilter/overlay_qsv: Add QSV overlay vpp filter
Add intel QSV overlay filter. Now it supports two input. And it also supports
the second input scale(implicit) during composition compared to the sw overlay.
Code has been separated into common interface part and qsv overlay implement
part.
The common part mainly creates the qsv session and manages the surface which are
nearly the same for all qsv filters. So the qsvvpp.c/qsvvpp.h API can be used by
other QSV vpp filters to reduce code redundancy.
Usage:
-hwaccel qsv -c:v mpeg2_qsv -r 25 -i in.m2v -hwaccel qsv -c:v h264_qsv -i
in.h264 -filter_complex
"overlay_qsv=eof_action=repeat:x=(W-w)/2:y=(H-h)/2" -b 2M -maxrate 3M -c:v
h264_qsv -y out.h264
two input should have different size otherwise one will be completely covered
or you need scale the
second input as follow:
-hwaccel qsv -c:v mpeg2_qsv -r 25 -i in.m2v -hwaccel qsv -c:v h264_qsv -i
in.h264 -filter_complex
"overlay_qsv=w=720:h=576:x=(W-w)/2:y=(H-h)/2" -b 2M -maxrate 3M -c:v h264_qsv
-y out.h264
Signed-off-by: ChaoX A Liu <[email protected]>
Signed-off-by: Zhengxu Huang <[email protected]>
Signed-off-by: Andrew Zhang <[email protected]>
Change-Id: I5c381febb0af6e2f9622c54ba00490ab99d48297
---
Changelog | 1 +
configure | 3 +
libavfilter/Makefile | 1 +
libavfilter/allfilters.c | 1 +
libavfilter/qsvvpp.c | 697 +++++++++++++++++++++++++++++++++++++++++++
libavfilter/qsvvpp.h | 66 ++++
libavfilter/vf_overlay_qsv.c | 482 ++++++++++++++++++++++++++++++
7 files changed, 1251 insertions(+)
create mode 100644 libavfilter/qsvvpp.c
create mode 100644 libavfilter/qsvvpp.h
create mode 100644 libavfilter/vf_overlay_qsv.c
diff --git a/Changelog b/Changelog
index 92a7249..51c3f85 100644
--- a/Changelog
+++ b/Changelog
@@ -19,6 +19,7 @@ version <next>:
- Cinepak encoder
- Intel QSV-accelerated MJPEG encoding
- NVIDIA CUVID-accelerated H.264 and HEVC decoding
+- Intel QSV-accelerated overlay filter
version 12:
diff --git a/configure b/configure
index 4510100..a80f24f 100755
--- a/configure
+++ b/configure
@@ -1786,6 +1786,7 @@ CONFIG_EXTRA="
qsv
qsvdec
qsvenc
+ qsvvpp
rangecoder
riffdec
riffenc
@@ -2528,6 +2529,8 @@ hqdn3d_filter_deps="gpl"
interlace_filter_deps="gpl"
movie_filter_deps="avcodec avformat"
ocv_filter_deps="libopencv"
+overlay_qsv_filter_deps="libmfx qsv"
+overlay_qsv_filter_select="qsvvpp"
resample_filter_deps="avresample"
scale_filter_deps="swscale"
scale_qsv_filter_deps="libmfx"
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 348ad92..2edb6c7 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -75,6 +75,7 @@ OBJS-$(CONFIG_NOFORMAT_FILTER) += vf_format.o
OBJS-$(CONFIG_NULL_FILTER) += vf_null.o
OBJS-$(CONFIG_OCV_FILTER) += vf_libopencv.o
OBJS-$(CONFIG_OVERLAY_FILTER) += vf_overlay.o
+OBJS-$(CONFIG_OVERLAY_QSV_FILTER) += qsvvpp.o vf_overlay_qsv.o
OBJS-$(CONFIG_PAD_FILTER) += vf_pad.o
OBJS-$(CONFIG_PIXDESCTEST_FILTER) += vf_pixdesctest.o
OBJS-$(CONFIG_SCALE_FILTER) += vf_scale.o
The .h should be added to the SKIPHEADERS otherwise make check would
fail on qsv-less systems, something along:
SKIPHEADERS-$(CONFIG_LIBMFX) += qsvvpp.h
Should do.
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index 3f09f46..dc59ccd 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -98,6 +98,7 @@ void avfilter_register_all(void)
REGISTER_FILTER(NULL, null, vf);
REGISTER_FILTER(OCV, ocv, vf);
REGISTER_FILTER(OVERLAY, overlay, vf);
+ REGISTER_FILTER(OVERLAY_QSV, overlay_qsv, vf);
REGISTER_FILTER(PAD, pad, vf);
REGISTER_FILTER(PIXDESCTEST, pixdesctest, vf);
REGISTER_FILTER(SCALE, scale, vf);
diff --git a/libavfilter/qsvvpp.c b/libavfilter/qsvvpp.c
new file mode 100644
index 0000000..464a030
--- /dev/null
+++ b/libavfilter/qsvvpp.c
@@ -0,0 +1,697 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; 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
+ */
This description could be put in the .h as well.
+
+#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"
+
+typedef struct QSVFrame {
+ AVFrame *frame;
+ mfxFrameSurface1 *surface;
+ mfxFrameSurface1 surface_internal; ///< for system memory
+ struct QSVFrame *next;
+} QSVFrame;
+
+/* abstract struct for all QSV filters */
+struct FFQSVVPPContext {
+ mfxSession session;
+ int (*filter_frame) (AVFilterLink *outlink, AVFrame *frame);///< callback
function
+ enum AVPixelFormat out_sw_format; ///< Real output format
+ mfxVideoParam vpp_param;
+ mfxFrameInfo *frame_infos; ///< frame info for each input
+
+ /* members releated to the input/output surface */
+ int in_mem_mode;
+ int out_mem_mode;
+ QSVFrame *in_frame_list;
+ QSVFrame *out_frame_list;
+ int nb_surface_ptrs_in;
+ int nb_surface_ptrs_out;
+ mfxFrameSurface1 **surface_ptrs_in;
+ mfxFrameSurface1 **surface_ptrs_out;
+
+ /* MFXVPP extern parameters*/
+ mfxExtOpaqueSurfaceAlloc opaque_alloc;
+ mfxExtBuffer **ext_buffers;
+ int nb_ext_buffers;
+};
+
+static const mfxHandleType handle_types[] = {
+ MFX_HANDLE_VA_DISPLAY,
+ MFX_HANDLE_D3D9_DEVICE_MANAGER,
+ MFX_HANDLE_D3D11_DEVICE,
+};
+
+static const AVRational default_tb = {1, 90000};
+
+/* functions for frameAlloc */
+static mfxStatus frame_alloc(mfxHDL pthis, mfxFrameAllocRequest *req,
+ mfxFrameAllocResponse *resp)
+{
+ FFQSVVPPContext *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_in * 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)
+{
+ *hdl = mid;
+ 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_RGB32:
+ 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:
+ 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;
+ 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;
+ pix_fmt = frames_ctx->sw_format;
+ } 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->CropW = link->w;
+ frameinfo->CropH = link->h;
+ frameinfo->Width = FFALIGN(link->w, 32);
+ frameinfo->Height = FFALIGN(link->h, 32);
+ frameinfo->PicStruct = MFX_PICSTRUCT_PROGRESSIVE;
+ frameinfo->FrameRateExtN = link->frame_rate.num;
+ frameinfo->FrameRateExtD = link->frame_rate.den;
+ 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->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) {
+ if (list->surface && !list->surface->Data.Locked) {
+ list->surface = NULL;
+ av_frame_free(&list->frame);
+ }
+ list = list->next;
+ }
+}
+
+static void clear_frame_list(QSVFrame **list)
+{
+ QSVFrame *frame;
+
+ while (*list) {
+ 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) {
+ if (!out->surface)
+ break;
+ }
+
+ if (!out) {
+ out = av_mallocz(sizeof(*out));
+ if (!out)
+ return NULL;
+ out->next = *list;
+ *list = out;
+ }
+
+ return out;
+}
+
+/* get the input surface */
+static QSVFrame *submit_frame(FFQSVVPPContext *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) {
+ av_log(ctx, AV_LOG_ERROR, "Can't alloc new frame.\n");
+ return NULL;
+ }
+
+ qsv_frame->frame = picref;
+
+ /**
+ * 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 (s->in_mem_mode != MFX_MEMTYPE_SYSTEM_MEMORY) {
+ if (qsv_frame->frame->format != AV_PIX_FMT_QSV) {
+ av_log(ctx, AV_LOG_ERROR, "QSVVPP gets a wrong frame.\n");
+ return NULL;
+ }
+ qsv_frame->surface = (mfxFrameSurface1*)qsv_frame->frame->data[3];
+ } else {
+ if (map_frame_to_surface(qsv_frame->frame,
+ &qsv_frame->surface_internal) < 0) {
+ av_log(ctx, AV_LOG_ERROR, "Unsupported frame.\n");
+ return NULL;
+ }
+ qsv_frame->surface = &qsv_frame->surface_internal;
+ }
+
+ 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);
+
+ return qsv_frame;
+}
+
+/* get the output surface */
+static QSVFrame *query_frame(FFQSVVPPContext *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) {
+ av_log(ctx, AV_LOG_ERROR, "Can't alloc new output frame.\n");
+ return NULL;
+ }
+
+ /**
+ * For video memory, get a hw frame;
+ * for system memory, get a sw frame and map it into a mfx_surface.
+ */
+ if (s->out_mem_mode != MFX_MEMTYPE_SYSTEM_MEMORY) {
+ 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;
+ }
+ 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,
+ &out_frame->surface_internal);
+ if (ret < 0)
+ return NULL;
+
+ out_frame->surface = &out_frame->surface_internal;
+ }
+
+ out_frame->surface->Info = s->vpp_param.vpp.Out;
+
+ return out_frame;
+}
+
+/* create the QSV session */
+static int init_vpp_session(AVFilterContext *avctx, FFQSVVPPContext *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;
+
+ /**
+ * @mem-mode
+ * 1.Input has a HW context, Video or Opaque mode
+ * 2.Otherwise, System mode.
+ */
+ 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;
+
+ if (in_frames_hwctx->frame_type & MFX_MEMTYPE_OPAQUE_FRAME)
+ s->in_mem_mode = MFX_MEMTYPE_OPAQUE_FRAME;
+ else
+ s->in_mem_mode = MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET;
+
+ s->surface_ptrs_in = av_mallocz_array(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);
+
+ 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;
+ out_frames_hwctx->frame_type = in_frames_hwctx->frame_type;
+ s->out_mem_mode = s->in_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_mallocz_array(out_frames_hwctx->nb_surfaces,
+ sizeof(*s->surface_ptrs_out));
+ if (!s->surface_ptrs_out)
+ 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;
+ }
+
+ for (i = 0; i < FF_ARRAY_ELEMS(handle_types); i++) {
+ ret = MFXVideoCORE_GetHandle(device_hwctx->session, handle_types[i],
&handle);
+ if (ret == MFX_ERR_NONE) {
+ handle_type = handle_types[i];
+ break;
+ }
+ }
+
+ /* create a "slave" session with those same properties, to be used for
vpp*/
+ ret = MFXInit(impl, &ver, &s->session);
+ if (ret != MFX_ERR_NONE) {
+ av_log(avctx, AV_LOG_ERROR, "Error initializing a session for
scaling\n");
+ return AVERROR_UNKNOWN;
+ }
+
+ if (handle) {
+ ret = MFXVideoCORE_SetHandle(s->session, handle_type, handle);
+ if (ret != MFX_ERR_NONE)
+ return AVERROR_UNKNOWN;
+ }
+
+ if (s->in_mem_mode == MFX_MEMTYPE_OPAQUE_FRAME || s->out_mem_mode ==
MFX_MEMTYPE_OPAQUE_FRAME) {
+ s->opaque_alloc.In.Surfaces = s->surface_ptrs_in;
+ s->opaque_alloc.In.NumSurface = s->nb_surface_ptrs_in;
+ s->opaque_alloc.In.Type = in_frames_hwctx ?
in_frames_hwctx->frame_type :
+ MFX_MEMTYPE_SYSTEM_MEMORY;
+
+ s->opaque_alloc.Out.Surfaces = s->surface_ptrs_out;
+ s->opaque_alloc.Out.NumSurface = s->nb_surface_ptrs_out;
+ s->opaque_alloc.Out.Type = out_frames_hwctx ?
out_frames_hwctx->frame_type :
+ MFX_MEMTYPE_SYSTEM_MEMORY;
+
+ s->opaque_alloc.Header.BufferId =
MFX_EXTBUFF_OPAQUE_SURFACE_ALLOCATION;
+ s->opaque_alloc.Header.BufferSz = sizeof(s->opaque_alloc);
+ } else if (s->in_mem_mode == MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET ||
+ s->out_mem_mode == MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET) {
+ 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;
+}
+
+int ff_qsvvpp_create(AVFilterContext *avctx, FFQSVVPPContext **vpp,
FFQSVVPPParam *param)
+{
+ int ret = 0, i;
+ FFQSVVPPContext *s;
+
+ if (!avctx || !vpp)
+ return AVERROR(EINVAL);
+
+ s = av_mallocz(sizeof(FFQSVVPPContext));
+ if (!s) {
+ ret = AVERROR(ENOMEM);
+ goto failed;
+ }
+
+ 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_mallocz_array(sizeof(*s->frame_infos),
avctx->nb_inputs);
+ 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++) {
+ FFQSVVPPCrop *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 (s->in_mem_mode == MFX_MEMTYPE_OPAQUE_FRAME || s->out_mem_mode ==
MFX_MEMTYPE_OPAQUE_FRAME) {
+ s->nb_ext_buffers = param->num_ext_buf + 1;
+ s->ext_buffers = av_mallocz(sizeof(*s->ext_buffers) *
s->nb_ext_buffers);
+ if (!s->ext_buffers)
+ return AVERROR(ENOMEM);
+
+ 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;
+ }
+
+ s->vpp_param.AsyncDepth = 1;
+
+ if (s->in_mem_mode == MFX_MEMTYPE_SYSTEM_MEMORY)
+ s->vpp_param.IOPattern |= MFX_IOPATTERN_IN_SYSTEM_MEMORY;
+ else if (s->in_mem_mode == MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET)
+ s->vpp_param.IOPattern |= MFX_IOPATTERN_IN_VIDEO_MEMORY;
+ else if (s->in_mem_mode == MFX_MEMTYPE_OPAQUE_FRAME)
+ s->vpp_param.IOPattern |= MFX_IOPATTERN_IN_OPAQUE_MEMORY;
+
+
+ if (s->out_mem_mode == MFX_MEMTYPE_SYSTEM_MEMORY)
+ s->vpp_param.IOPattern |= MFX_IOPATTERN_OUT_SYSTEM_MEMORY;
+ else if (s->out_mem_mode == MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET)
+ s->vpp_param.IOPattern |= MFX_IOPATTERN_OUT_VIDEO_MEMORY;
+ else if (s->out_mem_mode == MFX_MEMTYPE_OPAQUE_FRAME)
+ s->vpp_param.IOPattern |= MFX_IOPATTERN_OUT_OPAQUE_MEMORY;
+
+ ret = MFXVideoVPP_Init(s->session, &s->vpp_param);
+ if (ret < 0) {
+ av_log(avctx, AV_LOG_ERROR, "Failed to create a qsvvpp, ret = %d.\n",
ret);
+ goto failed;
+ }
+
+ *vpp = s;
+ return 0;
+
+failed:
+ if (s)
+ ff_qsvvpp_free(&s);
+
+ return ret;
+}
+
+int ff_qsvvpp_free(FFQSVVPPContext **vpp)
+{
+ FFQSVVPPContext *s = *vpp;
+
+ if (!s)
+ return 0;
+
+ 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);
+ av_freep(vpp);
+
+ return 0;
+}
+
+int ff_qsvvpp_filter_frame(FFQSVVPPContext *s, AVFilterLink *inlink, AVFrame
*picref)
+{
+ AVFilterContext *ctx = inlink->dst;
+ AVFilterLink *outlink = ctx->outputs[0];
+ mfxSyncPoint sync;
+ QSVFrame *in_frame, *out_frame;
+ int ret, filter_ret;
+
+ in_frame = submit_frame(s, inlink, picref);
+ if (!in_frame) {
+ av_log(ctx, AV_LOG_ERROR, "Fail to submit frame on input[%d]\n",
+ FF_INLINK_IDX(inlink));
+ return AVERROR(EINVAL);
+ }
+
+ do {
+ out_frame = query_frame(s, outlink);
+ if(!out_frame) {
+ av_log(ctx, AV_LOG_ERROR, "Fail to query an output frame.\n");
+ return AVERROR(ENOMEM);
+ }
+
+ do {
+ 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)
+ ret = AVERROR(EAGAIN);
+ break;
+ }
+
+ if (MFXVideoCORE_SyncOperation(s->session, sync, 1000) < 0)
+ av_log(ctx, AV_LOG_WARNING, "Sync failed.\n");
+
+ out_frame->frame->pts =
av_rescale_q(out_frame->surface->Data.TimeStamp,
+ default_tb, outlink->time_base);
+
+ filter_ret = s->filter_frame(outlink, out_frame->frame);
+ if (filter_ret < 0) {
+ av_frame_free(&out_frame->frame);
+ ret = filter_ret;
+ break;
+ }
+ out_frame->frame = NULL;
+ } while(ret == MFX_ERR_MORE_SURFACE);
+
+ return ret;
+}
diff --git a/libavfilter/qsvvpp.h b/libavfilter/qsvvpp.h
new file mode 100644
index 0000000..b8ada20
--- /dev/null
+++ b/libavfilter/qsvvpp.h
@@ -0,0 +1,66 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVFILTER_QSVVPP_H
+#define AVFILTER_QSVVPP_H
+
+#include "avfilter.h"
+#include <mfx/mfxvideo.h>
+
+#define FF_INLINK_IDX(link) ((int)((link)->dstpad - (link)->dst->input_pads))
+#define FF_OUTLINK_IDX(link) ((int)((link)->srcpad - (link)->src->output_pads))
+
+typedef struct FFQSVVPPContext FFQSVVPPContext;
+
+typedef struct FFQSVVPPCrop {
+ int in_idx; ///< Input index
+ int x, y, w, h; ///< Crop rectangle
+} FFQSVVPPCrop;
+
+typedef struct FFQSVVPPParam {
+ /* default is ff_filter_frame */
+ int (*filter_frame) (AVFilterLink *outlink, AVFrame *frame);
+
+ /* To fill with MFX enhanced filter configurations */
+ int num_ext_buf;
+ mfxExtBuffer **ext_buf;
+
+ /* Real output format */
+ enum AVPixelFormat out_sw_format;
+
+ /* Crop information for each input, if needed */
+ int num_crop;
+ FFQSVVPPCrop *crop;
+} FFQSVVPPParam;
+
+/**
+ * Blow APIs are abstracted to be called by different QSV VPP filters,> + *
most codes of different QSV filters
(scale/deinterlace/framerate/overlay) are nearly
+ * the same so use the common APIs to reduce code redundancy
+ */
Probably the comment above could be dropped or simplified as "common
code for VPP-based filters".
+
+/* create and initialize the QSV session*/
+int ff_qsvvpp_create(AVFilterContext *avctx, FFQSVVPPContext **vpp,
FFQSVVPPParam *param);
+
+/* release the resources eg.surfaces */
+int ff_qsvvpp_free(FFQSVVPPContext **vpp);
+
+/* vpp filter frame and call the cb if needed */
+int ff_qsvvpp_filter_frame(FFQSVVPPContext *vpp, AVFilterLink *inlink, AVFrame
*frame);
+
+#endif /* QSVVPP_H */
diff --git a/libavfilter/vf_overlay_qsv.c b/libavfilter/vf_overlay_qsv.c
new file mode 100644
index 0000000..f4c78fa
--- /dev/null
+++ b/libavfilter/vf_overlay_qsv.c
@@ -0,0 +1,482 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+/**
+ * @vf_overlay_qsv
+ * A hardware accelerated overlay filter based on Intel Quick Sync Video VPP
+ */
+
+#include "internal.h"
+#include "avfilter.h"
+#include "formats.h"
+#include "libavutil/opt.h"
+#include "libavutil/common.h"
+#include "libavutil/pixdesc.h"
+#include "libavutil/eval.h"
+#include "libavutil/hwcontext.h"
+#include "libavutil/avstring.h"
+#include "libavutil/avassert.h"
+#include "libavutil/imgutils.h"
+#include "libavutil/mathematics.h"
+#include "video.h"
+
+#include "qsvvpp.h"
+
+#define MAIN 0
+#define OVERLAY 1
+
+#define OFFSET(x) offsetof(QSVOverlayContext, x)
+#define FLAGS AV_OPT_FLAG_VIDEO_PARAM
+
+enum var_name {
+ VAR_MAIN_iW, VAR_MW,
+ VAR_MAIN_iH, VAR_MH,
+ VAR_OVERLAY_iW,
+ VAR_OVERLAY_iH,
+ VAR_OVERLAY_X, VAR_OX,
+ VAR_OVERLAY_Y, VAR_OY,
+ VAR_OVERLAY_W, VAR_OW,
+ VAR_OVERLAY_H, VAR_OH,
+ VAR_VARS_NB
+};
+
+enum EOFAction {
+ EOF_ACTION_REPEAT,
+ EOF_ACTION_ENDALL
+};
While at it, a newline here.
+typedef struct QSVOverlayContext {
+ AVClass *class;
+
+ FFQSVVPPContext *qsv;
+ FFQSVVPPParam qsv_param;
+ mfxExtVPPComposite comp_conf;
+ double var_values[VAR_VARS_NB];
+
+ char *overlay_ox, *overlay_oy, *overlay_ow, *overlay_oh;
+ uint16_t overlay_alpha, overlay_pixel_alpha;
+
+ enum EOFAction eof_action; ///< action to take on EOF from source
+
+ AVFrame *main;
+ AVFrame *over_prev, *over_next;
+} QSVOverlayContext;
+
+/**
+ * qsv composite support overlay implicitly scale so the params much
+ * more complicated than sw overlay
+ */
+static const char *const var_names[] = {
+ "main_w", "W", ///< input width of the main layer
+ "main_h", "H", ///< input height of the main layer
+ "overlay_iw", ///< input width of the overlay layer
+ "overlay_ih", ///< input height of the overlay layer
+ "overlay_x", "x", ///< x position of the overlay layer inside of main
+ "overlay_y", "y", ///< y position of the overlay layer inside of main
+ "overlay_w", "w", ///< output width of overlay layer
+ "overlay_h", "h", ///< output height of overlay layer
+ NULL
+};
+
+static const AVOption options[] = {
+ { "x", "Overlay x position", OFFSET(overlay_ox), AV_OPT_TYPE_STRING,
{.str="0"}, 0, 255, .flags = FLAGS},
+ { "y", "Overlay y position", OFFSET(overlay_oy), AV_OPT_TYPE_STRING,
{.str="0"}, 0, 255, .flags = FLAGS},
+ { "w", "Overlay width", OFFSET(overlay_ow), AV_OPT_TYPE_STRING,
{.str="overlay_iw"}, 0, 255, .flags = FLAGS},
+ { "h", "Overlay height", OFFSET(overlay_oh), AV_OPT_TYPE_STRING,
{.str="overlay_ih*w/overlay_iw"}, 0, 255, .flags = FLAGS},
+ { "alpha", "Overlay global alpha", OFFSET(overlay_alpha), AV_OPT_TYPE_INT,
{.i64 = 255}, 0, 255, .flags = FLAGS},
+ { "eof_action", "Action to take when encountering EOF from secondary input
",
+ OFFSET(eof_action), AV_OPT_TYPE_INT, { .i64 = EOF_ACTION_REPEAT },
+ EOF_ACTION_REPEAT, EOF_ACTION_ENDALL, .flags = FLAGS, "eof_action" },
+ { "repeat", "Repeat the previous frame.", 0, AV_OPT_TYPE_CONST, { .i64 =
EOF_ACTION_REPEAT }, .flags = FLAGS, "eof_action" },
+ { "endall", "End both streams.", 0, AV_OPT_TYPE_CONST, { .i64 =
EOF_ACTION_ENDALL }, .flags = FLAGS, "eof_action" },
+ { NULL }
+};
+
+static int eval_expr(AVFilterContext *ctx)
+{
+ QSVOverlayContext *vpp = ctx->priv;
+ double *var_values = vpp->var_values;
+ AVExpr *ox_expr, *oy_expr, *ow_expr, *oh_expr;
+ int ret;
+
+#define PASS_EXPR(e, s) {\
+ ret = av_expr_parse(&e, s, var_names, NULL, NULL, NULL, NULL, 0, ctx); \
+ if (ret < 0) {\
+ av_log(ctx, AV_LOG_ERROR, "Error when passing '%s'.\n", s);\
+ return ret;\
+ }\
+}
+ PASS_EXPR(ox_expr, vpp->overlay_ox);
+ PASS_EXPR(oy_expr, vpp->overlay_oy);
+ PASS_EXPR(ow_expr, vpp->overlay_ow);
+ PASS_EXPR(oh_expr, vpp->overlay_oh);
+#undef PASS_EXPR
+
+ var_values[VAR_OVERLAY_W] =
+ var_values[VAR_OW] = av_expr_eval(ow_expr, var_values, NULL);
+ var_values[VAR_OVERLAY_H] =
+ var_values[VAR_OH] = av_expr_eval(oh_expr, var_values, NULL);
+
+ /* calc again in case ow is relative to oh */
+ var_values[VAR_OVERLAY_W] =
+ var_values[VAR_OW] = av_expr_eval(ow_expr, var_values, NULL);
+
+ var_values[VAR_OVERLAY_X] =
+ var_values[VAR_OX] = av_expr_eval(ox_expr, var_values, NULL);
+ var_values[VAR_OVERLAY_Y] =
+ var_values[VAR_OY] = av_expr_eval(oy_expr, var_values, NULL);
+
+ /* calc again in case ox is relative to oy */
+ var_values[VAR_OVERLAY_X] =
+ var_values[VAR_OX] = av_expr_eval(ox_expr, var_values, NULL);
+
+ /* calc overlay_w and overlay_h again incase relative to ox,oy */
+ var_values[VAR_OVERLAY_W] =
+ var_values[VAR_OW] = av_expr_eval(ow_expr, var_values, NULL);
+ var_values[VAR_OVERLAY_H] =
+ var_values[VAR_OH] = av_expr_eval(oh_expr, var_values, NULL);
+ var_values[VAR_OVERLAY_W] =
+ var_values[VAR_OW] = av_expr_eval(ow_expr, var_values, NULL);
+
+ av_expr_free(ox_expr);
+ av_expr_free(oy_expr);
+ av_expr_free(ow_expr);
+ av_expr_free(oh_expr);
+
+ return 0;
+}
+
+static int have_alpha_planar(AVFilterLink *link)
+{
+ enum AVPixelFormat pix_fmt;
+ const AVPixFmtDescriptor *desc;
+ AVHWFramesContext *fctx;
+
+ if (link->format == AV_PIX_FMT_QSV) {
+ fctx = (AVHWFramesContext *)link->hw_frames_ctx->data;
+ pix_fmt = fctx->sw_format;
+ }
+
+ desc = av_pix_fmt_desc_get(pix_fmt);
+ if (!desc)
+ return 0;
+
+ return !!(desc->flags & AV_PIX_FMT_FLAG_ALPHA);
+}
+
+static int config_main_input(AVFilterLink *inlink)
+{
+ AVFilterContext *ctx = inlink->dst;
+ QSVOverlayContext *vpp = ctx->priv;
+ mfxVPPCompInputStream *st = &vpp->comp_conf.InputStream[0];
+
+ av_log(ctx, AV_LOG_DEBUG, "Input[%d] is of %s.\n", FF_INLINK_IDX(inlink),
+ av_get_pix_fmt_name(inlink->format));
+
+ vpp->var_values[VAR_MAIN_iW] =
+ vpp->var_values[VAR_MW] = inlink->w;
+ vpp->var_values[VAR_MAIN_iH] =
+ vpp->var_values[VAR_MH] = inlink->h;
+
+ st->DstX = 0;
+ st->DstY = 0;
+ st->DstW = inlink->w;
+ st->DstH = inlink->h;
+ st->GlobalAlphaEnable = 0;
+ st->PixelAlphaEnable = 0;
+
+ return 0;
+}
+
+static int config_overlay_input(AVFilterLink *inlink)
+{
+ AVFilterContext *ctx = inlink->dst;
+ AVFilterLink *in0 = ctx->inputs[0];
+ QSVOverlayContext *vpp = ctx->priv;
+ mfxVPPCompInputStream *st = &vpp->comp_conf.InputStream[1];
+ int ret;
+
+ av_log(ctx, AV_LOG_DEBUG, "Input[%d] is of %s.\n", FF_INLINK_IDX(inlink),
+ av_get_pix_fmt_name(inlink->format));
+ if ((in0->format == AV_PIX_FMT_QSV && inlink->format != AV_PIX_FMT_QSV) ||
+ (in0->format != AV_PIX_FMT_QSV && inlink->format == AV_PIX_FMT_QSV)) {
+ av_log(ctx, AV_LOG_ERROR, "One of the inputs is of AV_PIX_FMT_QSV,"
+ "but the other is of soft pixel format.\n");
+ av_log(ctx, AV_LOG_ERROR, "HW/SW mixed format is not supported
now.\n");
+ return AVERROR(EINVAL);
+ }
+
+ vpp->var_values[VAR_OVERLAY_iW] = inlink->w;
+ vpp->var_values[VAR_OVERLAY_iH] = inlink->h;
+ ret = eval_expr(ctx);
+ if (ret < 0)
+ return ret;
+
+ st->DstX = vpp->var_values[VAR_OX];
+ st->DstY = vpp->var_values[VAR_OY];
+ st->DstW = vpp->var_values[VAR_OW];
+ st->DstH = vpp->var_values[VAR_OH];
+ st->GlobalAlpha = vpp->overlay_alpha;
+ st->GlobalAlphaEnable = (st->GlobalAlpha < 255);
+ st->PixelAlphaEnable = have_alpha_planar(inlink);
+
+ return 0;
+}
+
+static int config_output(AVFilterLink *outlink)
+{
+ AVFilterContext *ctx = outlink->src;
+ QSVOverlayContext *vpp = ctx->priv;
+ AVFilterLink *inlink = ctx->inputs[0];
+
+ av_log(ctx, AV_LOG_DEBUG, "Output is of %s.\n",
av_get_pix_fmt_name(outlink->format));
+
+ outlink->w = vpp->var_values[VAR_MW];
+ outlink->h = vpp->var_values[VAR_MH];
+ outlink->frame_rate = inlink->frame_rate;
+ outlink->time_base = av_inv_q(outlink->frame_rate);
+
+ return ff_qsvvpp_create(ctx, &vpp->qsv, &vpp->qsv_param);
+}
+
+static int blend_frame(AVFilterContext *ctx,
+ AVFrame *mpic, AVFrame *opic)
+{
+ int ret = 0;
+ QSVOverlayContext *vpp = ctx->priv;
+ AVFrame *opic_copy = NULL;
+
+ ret = ff_qsvvpp_filter_frame(vpp->qsv, ctx->inputs[0], mpic);
+ if (ret == 0 || ret == AVERROR(EAGAIN)) {
+ /**
+ * Make a copy of the overlay frame. Because:
+ * 1. ff_qsvvpp_filter_frame will take control of the given frame
+ * 2. We need to repeat the overlay frame when 2nd input goes into EOF
+ */
+ opic_copy = av_frame_clone(opic);
+ if (!opic_copy)
+ return AVERROR(ENOMEM);
+
+ ret = ff_qsvvpp_filter_frame(vpp->qsv, ctx->inputs[1], opic_copy);
+ }
+
+ return ret;
+}
+
+static int handle_overlay_eof(AVFilterContext *ctx)
+{
+ int ret = 0;
+ QSVOverlayContext *s = ctx->priv;
+ /* Repeat previous frame on secondary input */
+ if (s->over_prev && s->eof_action == EOF_ACTION_REPEAT)
+ ret = blend_frame(ctx, s->main, s->over_prev);
+ /* End both streams */
+ else if (s->eof_action == EOF_ACTION_ENDALL)
+ return AVERROR_EOF;
+
+ s->main = NULL;
+
+ return ret;
+}
+
+static int request_frame(AVFilterLink *outlink)
+{
+ AVFilterContext *ctx = outlink->src;
+ QSVOverlayContext *s = ctx->priv;
+ AVRational tb_main = ctx->inputs[MAIN]->time_base;
+ AVRational tb_over = ctx->inputs[OVERLAY]->time_base;
+ int ret = 0;
+
+ /* get a frame on the main input */
+ if (!s->main) {
+ ret = ff_request_frame(ctx->inputs[MAIN]);
+ if (ret < 0)
+ return ret;
+ }
+
+ /* get a new frame on the overlay input, on EOF check setting 'eof_action'
*/
+ if (!s->over_next) {
+ ret = ff_request_frame(ctx->inputs[OVERLAY]);
+ if (ret == AVERROR_EOF)
+ return handle_overlay_eof(ctx);
+ else if (ret < 0)
+ return ret;
+ }
+
+ while (s->main->pts != AV_NOPTS_VALUE &&
+ s->over_next->pts != AV_NOPTS_VALUE &&
+ av_compare_ts(s->over_next->pts, tb_over, s->main->pts, tb_main) <
0) {
+ av_frame_free(&s->over_prev);
+ FFSWAP(AVFrame*, s->over_prev, s->over_next);
+
+ ret = ff_request_frame(ctx->inputs[OVERLAY]);
+ if (ret == AVERROR_EOF)
+ return handle_overlay_eof(ctx);
+ else if (ret < 0)
+ return ret;
+ }
+
+ if (s->main->pts == AV_NOPTS_VALUE ||
+ s->over_next->pts == AV_NOPTS_VALUE ||
+ !av_compare_ts(s->over_next->pts, tb_over, s->main->pts, tb_main)) {
+ ret = blend_frame(ctx, s->main, s->over_next);
+ av_frame_free(&s->over_prev);
+ FFSWAP(AVFrame*, s->over_prev, s->over_next);
+ } else if (s->over_prev) {
+ ret = blend_frame(ctx, s->main, s->over_prev);
+ } else {
+ av_frame_free(&s->main);
+ ret = AVERROR(EAGAIN);
+ }
+
+ s->main = NULL;
+
+ return ret ;
+}
+
+static int filter_frame_main(AVFilterLink *inlink, AVFrame *frame)
+{
+ QSVOverlayContext *s = inlink->dst->priv;
+
+ av_assert0(!s->main);
+ s->main = frame;
+
+ return 0;
+}
+
+static int filter_frame_overlay(AVFilterLink *inlink, AVFrame *frame)
+{
+ QSVOverlayContext *s = inlink->dst->priv;
+
+ av_assert0(!s->over_next);
+ s->over_next = frame;
+
+ return 0;
+}
+
+static int overlay_qsv_init(AVFilterContext *ctx)
+{
+ QSVOverlayContext *vpp = ctx->priv;
+
+ /* fill composite config */
+ vpp->comp_conf.Header.BufferId = MFX_EXTBUFF_VPP_COMPOSITE;
+ vpp->comp_conf.Header.BufferSz = sizeof(vpp->comp_conf);
+ vpp->comp_conf.NumInputStream = ctx->nb_inputs;
+ vpp->comp_conf.InputStream =
+ av_mallocz(sizeof(*vpp->comp_conf.InputStream) * ctx->nb_inputs);
+ if (!vpp->comp_conf.InputStream)
+ return AVERROR(ENOMEM);
+
+ /* initiallize QSVVPP params */
typo here ^^^
+ vpp->qsv_param.filter_frame = NULL;
+ vpp->qsv_param.ext_buf = av_mallocz(sizeof(*vpp->qsv_param.ext_buf));
+ if (!vpp->qsv_param.ext_buf)
+ return AVERROR(ENOMEM);
+
+ vpp->qsv_param.ext_buf[0] = (mfxExtBuffer *)&vpp->comp_conf;
+ vpp->qsv_param.num_ext_buf = 1;
+ vpp->qsv_param.out_sw_format = AV_PIX_FMT_NV12;
+ vpp->qsv_param.num_crop = 0;
+
+ return 0;
+}
+
+static void overlay_qsv_uninit(AVFilterContext *ctx)
+{
+ QSVOverlayContext *vpp = ctx->priv;
+
+ av_frame_free(&vpp->main);
+ av_frame_free(&vpp->over_prev);
+ av_frame_free(&vpp->over_next);
+ ff_qsvvpp_free(&vpp->qsv);
+ av_freep(&vpp->comp_conf.InputStream);
+ av_freep(&vpp->qsv_param.ext_buf);
+}
+
+static int overlay_qsv_query_formats(AVFilterContext *ctx)
+{
+ int i;
+
+ static const enum AVPixelFormat main_in_fmts[] = {
+ AV_PIX_FMT_YUV420P,
+ AV_PIX_FMT_NV12,
+ AV_PIX_FMT_YUYV422,
+ AV_PIX_FMT_RGB32,
+ AV_PIX_FMT_QSV,
+ AV_PIX_FMT_NONE
+ };
+ static const enum AVPixelFormat out_pix_fmts[] = {
+ AV_PIX_FMT_NV12,
+ AV_PIX_FMT_QSV,
+ AV_PIX_FMT_NONE
+ };
+
+ for(i = 0; i < ctx->nb_inputs; i++)
+ ff_formats_ref(ff_make_format_list(main_in_fmts),
&ctx->inputs[i]->out_formats);
+
+ ff_formats_ref(ff_make_format_list(out_pix_fmts),
&ctx->outputs[0]->in_formats);
+
+ return 0;
+}
+
+static const AVClass overlay_qsv_class = {
+ .class_name = "overlay_qsv",
+ .item_name = av_default_item_name,
+ .option = options,
+ .version = LIBAVUTIL_VERSION_INT,
+};
+
+static const AVFilterPad overlay_qsv_inputs[] = {
+ {
+ .name = "main",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .filter_frame = filter_frame_main,
+ .config_props = config_main_input,
+ .needs_fifo = 1,
+ },
+ {
+ .name = "overlay",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .filter_frame = filter_frame_overlay,
+ .config_props = config_overlay_input,
+ .needs_fifo = 1,
+ },
+ { NULL }
+};
+
+static const AVFilterPad overlay_qsv_outputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .config_props = config_output,
+ .request_frame = request_frame,
+ },
+ { NULL }
+};
+
+AVFilter ff_vf_overlay_qsv = {
+ .name = "overlay_qsv",
+ .description = NULL_IF_CONFIG_SMALL("Quick Sync Video overlay."),
+ .priv_size = sizeof(QSVOverlayContext),
+ .query_formats = overlay_qsv_query_formats,
+ .init = overlay_qsv_init,
+ .uninit = overlay_qsv_uninit,
+ .inputs = overlay_qsv_inputs,
+ .outputs = overlay_qsv_outputs,
+ .priv_class = &overlay_qsv_class,
+ .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
+};
