path: root/contrib/ffmpeg/A00-qsv.patch

diff -urN libav-11.3.org/configure libav-11.3/configure
--- libav-11.3.org/configure	2016-07-20 13:11:06.000000000 +0200
+++ libav-11.3/configure	2016-07-20 13:12:57.000000000 +0200
@@ -132,6 +132,7 @@
 
 Hardware accelerators:
   --enable-dxva2           enable DXVA2 code
+  --enable-qsv             enable QSV code
   --enable-vaapi           enable VAAPI code
   --enable-vda             enable VDA code
   --enable-vdpau           enable VDPAU code
@@ -1188,6 +1189,7 @@
 
 HWACCEL_LIST="
     dxva2
+    qsv
     vaapi
     vda
     vdpau
@@ -1919,6 +1921,7 @@
 
 # hardware accelerators
 dxva2_deps="dxva2api_h"
+qsv_deps="mfx_mfxvideo_h"
 vaapi_deps="va_va_h"
 vda_deps="VideoDecodeAcceleration_VDADecoder_h pthreads"
 vda_extralibs="-framework CoreFoundation -framework VideoDecodeAcceleration -framework QuartzCore"
@@ -1930,6 +1933,8 @@
 h263_vdpau_hwaccel_select="h263_decoder"
 h264_dxva2_hwaccel_deps="dxva2"
 h264_dxva2_hwaccel_select="h264_decoder"
+h264_qsv_decoder_deps="qsv"
+h264_qsv_decoder_select="h264_decoder"
 h264_vaapi_hwaccel_deps="vaapi"
 h264_vaapi_hwaccel_select="h264_decoder"
 h264_vda_hwaccel_deps="vda"
@@ -4080,6 +4085,7 @@
 check_header io.h
 check_header mach/mach_time.h
 check_header malloc.h
+check_header mfx/mfxvideo.h
 check_header poll.h
 check_header sys/mman.h
 check_header sys/param.h
diff -urN libav-11.3.org/libavcodec/Makefile libav-11.3/libavcodec/Makefile
--- libav-11.3.org/libavcodec/Makefile	2016-07-20 13:11:06.000000000 +0200
+++ libav-11.3/libavcodec/Makefile	2016-07-20 13:12:57.000000000 +0200
@@ -4,6 +4,7 @@
           avfft.h                                                       \
           dv_profile.h                                                  \
           dxva2.h                                                       \
+          qsv.h                                                         \
           vaapi.h                                                       \
           vda.h                                                         \
           vdpau.h                                                       \
@@ -73,6 +74,7 @@
 OBJS-$(CONFIG_MPEGVIDEOENC)            += mpegvideo_enc.o mpeg12data.o  \
                                           motion_est.o ratecontrol.o    \
                                           mpegvideoencdsp.o
+OBJS-$(CONFIG_QSV)                     += qsv.o
 OBJS-$(CONFIG_PIXBLOCKDSP)             += pixblockdsp.o
 OBJS-$(CONFIG_QPELDSP)                 += qpeldsp.o
 OBJS-$(CONFIG_RANGECODER)              += rangecoder.o
@@ -216,6 +218,7 @@
                                           h264_direct.o h264_loopfilter.o  \
                                           h264_mb.o h264_picture.o h264_ps.o \
                                           h264_refs.o h264_sei.o h264_slice.o
+OBJS-$(CONFIG_H264_QSV_DECODER)        += qsv_h264.o
 OBJS-$(CONFIG_HEVC_DECODER)            += hevc.o hevc_mvs.o hevc_ps.o hevc_sei.o \
                                           hevc_cabac.o hevc_refs.o hevcpred.o    \
                                           hevcdsp.o hevc_filter.o
@@ -709,6 +712,7 @@
 SKIPHEADERS-$(CONFIG_DXVA2)            += dxva2.h dxva2_internal.h
 SKIPHEADERS-$(CONFIG_LIBSCHROEDINGER)  += libschroedinger.h
 SKIPHEADERS-$(CONFIG_MPEG_XVMC_DECODER) += xvmc.h
+SKIPHEADERS-$(CONFIG_QSV)              += qsv.h
 SKIPHEADERS-$(CONFIG_VAAPI)            += vaapi_internal.h
 SKIPHEADERS-$(CONFIG_VDA)              += vda.h vda_internal.h
 SKIPHEADERS-$(CONFIG_VDPAU)            += vdpau.h vdpau_internal.h
diff -urN libav-11.3.org/libavcodec/allcodecs.c libav-11.3/libavcodec/allcodecs.c
--- libav-11.3.org/libavcodec/allcodecs.c	2016-07-20 13:11:06.000000000 +0200
+++ libav-11.3/libavcodec/allcodecs.c	2016-07-20 13:12:57.000000000 +0200
@@ -159,6 +159,7 @@
     REGISTER_DECODER(H263I,             h263i);
     REGISTER_ENCODER(H263P,             h263p);
     REGISTER_DECODER(H264,              h264);
+    REGISTER_DECODER(H264_QSV,          h264_qsv);
     REGISTER_DECODER(HEVC,              hevc);
     REGISTER_DECODER(HNM4_VIDEO,        hnm4_video);
     REGISTER_ENCDEC (HUFFYUV,           huffyuv);
diff -urN libav-11.3.org/libavcodec/qsv.c libav-11.3/libavcodec/qsv.c
--- libav-11.3.org/libavcodec/qsv.c	1970-01-01 01:00:00.000000000 +0100
+++ libav-11.3/libavcodec/qsv.c	2016-07-20 13:14:16.000000000 +0200
@@ -0,0 +1,650 @@
+/* ********************************************************************* *\
+
+Copyright (C) 2013 Intel Corporation.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+- Redistributions of source code must retain the above copyright notice,
+this list of conditions and the following disclaimer.
+- Redistributions in binary form must reproduce the above copyright notice,
+this list of conditions and the following disclaimer in the documentation
+and/or other materials provided with the distribution.
+- Neither the name of Intel Corporation nor the names of its contributors
+may be used to endorse or promote products derived from this software
+without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY INTEL CORPORATION "AS IS" AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+IN NO EVENT SHALL INTEL CORPORATION BE LIABLE FOR ANY DIRECT, INDIRECT,
+INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+\* ********************************************************************* */
+
+#include "qsv.h"
+
+#include "avcodec.h"
+#include "internal.h"
+
+int av_qsv_get_free_encode_task(av_qsv_list * tasks)
+{
+    int ret = MFX_ERR_NOT_FOUND;
+    int i = 0;
+    if (tasks)
+        for (i = 0; i < av_qsv_list_count(tasks); i++) {
+            av_qsv_task *task = av_qsv_list_item(tasks, i);
+            if (task->stage && task->stage->out.sync)
+                if (!(*task->stage->out.sync->p_sync)) {
+                    ret = i;
+                    break;
+                }
+        }
+    return ret;
+}
+
+int av_qsv_get_free_sync(av_qsv_space * space, av_qsv_context * qsv)
+{
+    int ret = -1;
+    int counter = 0;
+
+    while (1) {
+        for (int i = 0; i < space->sync_num; i++) {
+            if (!(*(space->p_syncp[i]->p_sync)) &&
+                0 == space->p_syncp[i]->in_use ) {
+                if (i > space->sync_num_max_used)
+                    space->sync_num_max_used = i;
+                ff_qsv_atomic_inc(&space->p_syncp[i]->in_use);
+                return i;
+            }
+        }
+#if HAVE_THREADS
+        if (++counter >= AV_QSV_REPEAT_NUM_DEFAULT) {
+#endif
+            av_log(NULL, AV_LOG_FATAL, "not enough to have %d sync point(s) allocated\n",
+                   space->sync_num);
+            break;
+#if HAVE_THREADS
+        }
+        av_qsv_sleep(5);
+#endif
+    }
+    return ret;
+}
+
+int av_qsv_get_free_surface(av_qsv_space * space, av_qsv_context * qsv,
+                     mfxFrameInfo * info, av_qsv_split part)
+{
+    int ret = -1;
+    int from = 0;
+    int up = space->surface_num;
+    int counter = 0;
+
+    while (1) {
+        from = 0;
+        up = space->surface_num;
+        if (part == QSV_PART_LOWER)
+            up /= 2;
+        if (part == QSV_PART_UPPER)
+            from = up / 2;
+
+        for (int i = from; i < up; i++) {
+            if (0 == space->p_surfaces[i]->Data.Locked) {
+                memcpy(&(space->p_surfaces[i]->Info), info,
+                       sizeof(mfxFrameInfo));
+                if (i > space->surface_num_max_used)
+                    space->surface_num_max_used = i;
+                return i;
+            }
+        }
+#if HAVE_THREADS
+        if (++counter >= AV_QSV_REPEAT_NUM_DEFAULT) {
+#endif
+            av_log(NULL, AV_LOG_FATAL,
+                   "not enough to have %d surface(s) allocated\n", up);
+            break;
+#if HAVE_THREADS
+        }
+        av_qsv_sleep(5);
+#endif
+    }
+    return ret;
+}
+
+int ff_qsv_is_surface_in_pipe(mfxFrameSurface1 * p_surface, av_qsv_context * qsv)
+{
+    int ret = 0;
+    int a, b,i;
+    av_qsv_list *list = 0;
+    av_qsv_stage *stage = 0;
+
+    if (!p_surface)
+        return ret;
+    if (!qsv->pipes)
+        return ret;
+
+    for (a = 0; a < av_qsv_list_count(qsv->pipes); a++) {
+        list = av_qsv_list_item(qsv->pipes, a);
+        for (b = 0; b < av_qsv_list_count(list); b++) {
+            stage = av_qsv_list_item(list, b);
+            if (p_surface == stage->out.p_surface)
+                return (stage->type << 16) | 2;
+            if (p_surface == stage->in.p_surface)
+                return (stage->type << 16) | 1;
+        }
+    }
+    return ret;
+}
+
+int ff_qsv_is_sync_in_pipe(mfxSyncPoint * sync, av_qsv_context * qsv)
+{
+    int ret = 0;
+    int a, b;
+    av_qsv_list *list = 0;
+    av_qsv_stage *stage = 0;
+
+    if (!sync)
+        return ret;
+    if (!qsv->pipes)
+        return ret;
+
+    for (a = 0; a < av_qsv_list_count(qsv->pipes); a++) {
+        list = av_qsv_list_item(qsv->pipes, a);
+        for (b = 0; b < av_qsv_list_count(list); b++) {
+            stage = av_qsv_list_item(list, b);
+            if (sync == stage->out.sync->p_sync) {
+                return 1;
+            }
+        }
+    }
+    return ret;
+}
+
+av_qsv_stage *av_qsv_stage_init(void)
+{
+    av_qsv_stage *stage = av_mallocz(sizeof(av_qsv_stage));
+    return stage;
+}
+
+void av_qsv_stage_clean(av_qsv_stage ** stage)
+{
+    if ((*stage)->out.sync) {
+        if ((*stage)->out.sync->p_sync)
+            *(*stage)->out.sync->p_sync = 0;
+        if ((*stage)->out.sync->in_use > 0)
+            ff_qsv_atomic_dec(&(*stage)->out.sync->in_use);
+        (*stage)->out.sync = 0;
+    }
+    if ((*stage)->out.p_surface) {
+        (*stage)->out.p_surface = 0;
+
+    }
+    if ((*stage)->in.p_surface) {
+        (*stage)->in.p_surface = 0;
+    }
+
+    av_freep(stage);
+}
+
+void av_qsv_add_context_usage(av_qsv_context * qsv, int is_threaded)
+{
+    int is_active = 0;
+#if HAVE_THREADS
+    int mut_ret = 0;
+#endif
+
+    is_active = ff_qsv_atomic_inc(&qsv->is_context_active);
+    if (is_active == 1) {
+        memset(&qsv->mfx_session, 0, sizeof(mfxSession));
+        av_qsv_pipe_list_create(&qsv->pipes, is_threaded);
+
+        qsv->dts_seq = av_qsv_list_init(is_threaded);
+
+#if HAVE_THREADS
+        if (is_threaded) {
+            qsv->qts_seq_mutex = av_mallocz(sizeof(pthread_mutex_t));
+            if (qsv->qts_seq_mutex){
+                mut_ret = pthread_mutex_init(qsv->qts_seq_mutex, NULL);
+                if(mut_ret)
+                    av_log(NULL, AV_LOG_ERROR, "pthread_mutex_init issue[%d] at %s\n",mut_ret,__FUNCTION__);
+            }
+
+        } else
+#endif
+            qsv->qts_seq_mutex = 0;
+    }
+}
+
+int av_qsv_context_clean(av_qsv_context * qsv)
+{
+    int is_active = 0;
+    mfxStatus sts = MFX_ERR_NONE;
+#if HAVE_THREADS
+    int mut_ret = 0;
+#endif
+
+    is_active = ff_qsv_atomic_dec(&qsv->is_context_active);
+
+    // spaces would have to be cleaned on the own,
+    // here we care about the rest, common stuff
+    if (is_active == 0) {
+
+        if (qsv->dts_seq) {
+            while (av_qsv_list_count(qsv->dts_seq))
+                av_qsv_dts_pop(qsv);
+
+            av_qsv_list_close(&qsv->dts_seq);
+        }
+#if HAVE_THREADS
+        if (qsv->qts_seq_mutex) {
+            mut_ret = pthread_mutex_destroy(qsv->qts_seq_mutex);
+            if(mut_ret)
+                av_log(NULL, AV_LOG_ERROR, "pthread_mutex_destroy issue[%d] at %s\n", mut_ret,__FUNCTION__);
+#endif
+            qsv->qts_seq_mutex = 0;
+#if HAVE_THREADS
+        }
+#endif
+
+        if (qsv->pipes)
+            av_qsv_pipe_list_clean(&qsv->pipes);
+
+        if (qsv->mfx_session) {
+            sts = MFXClose(qsv->mfx_session);
+            AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+            qsv->mfx_session = 0;
+        }
+    }
+    return 0;
+}
+
+void av_qsv_pipe_list_create(av_qsv_list ** list, int is_threaded)
+{
+    if (!*list)
+        *list = av_qsv_list_init(is_threaded);
+}
+
+void av_qsv_pipe_list_clean(av_qsv_list ** list)
+{
+    av_qsv_list *stage;
+    int i = 0;
+    if (*list) {
+        for (i = av_qsv_list_count(*list); i > 0; i--) {
+            stage = av_qsv_list_item(*list, i - 1);
+            av_qsv_flush_stages(*list, &stage);
+        }
+        av_qsv_list_close(list);
+    }
+}
+
+void av_qsv_add_stagee(av_qsv_list ** list, av_qsv_stage * stage, int is_threaded)
+{
+    if (!*list)
+        *list = av_qsv_list_init(is_threaded);
+    av_qsv_list_add(*list, stage);
+}
+
+av_qsv_stage *av_qsv_get_last_stage(av_qsv_list * list)
+{
+    av_qsv_stage *stage = 0;
+    int size = 0;
+
+    av_qsv_list_lock(list);
+    size = av_qsv_list_count(list);
+    if (size > 0)
+        stage = av_qsv_list_item(list, size - 1);
+    av_qsv_list_unlock(list);
+
+    return stage;
+}
+
+void av_qsv_flush_stages(av_qsv_list * list, av_qsv_list ** item)
+{
+    int i = 0;
+    int x = 0;
+    av_qsv_stage *stage = 0;
+    av_qsv_list *to_remove_list = 0;
+    av_qsv_list *to_remove_atom_list = 0;
+    av_qsv_list *to_remove_atom = 0;
+
+    for (i = 0; i < av_qsv_list_count(*item); i++) {
+        stage = av_qsv_list_item(*item, i);
+        if(stage->pending){
+            if(!to_remove_list)
+                to_remove_list = av_qsv_list_init(0);
+            av_qsv_list_add(to_remove_list, stage->pending);
+        }
+        av_qsv_stage_clean(&stage);
+        // should actually remove from the list but ok...
+    }
+    av_qsv_list_rem(list, *item);
+    av_qsv_list_close(item);
+
+    if(to_remove_list){
+        for (i = av_qsv_list_count(to_remove_list); i > 0; i--){
+            to_remove_atom_list = av_qsv_list_item(to_remove_list, i-1);
+            for (x = av_qsv_list_count(to_remove_atom_list); x > 0; x--){
+                to_remove_atom = av_qsv_list_item(to_remove_atom_list, x-1);
+                av_qsv_flush_stages(list,&to_remove_atom);
+            }
+        }
+        av_qsv_list_close(&to_remove_list);
+    }
+}
+
+av_qsv_list *av_qsv_pipe_by_stage(av_qsv_list * list, av_qsv_stage * stage)
+{
+    av_qsv_list *item = 0;
+    av_qsv_stage *cur_stage = 0;
+    int i = 0;
+    int a = 0;
+    for (i = 0; i < av_qsv_list_count(list); i++) {
+        item = av_qsv_list_item(list, i);
+        for (a = 0; a < av_qsv_list_count(item); a++) {
+            cur_stage = av_qsv_list_item(item, a);
+            if (cur_stage == stage)
+                return item;
+        }
+    }
+    return 0;
+}
+
+// no duplicate of the same value, if end == 0 : working over full length
+void av_qsv_dts_ordered_insert(av_qsv_context * qsv, int start, int end,
+                            int64_t dts, int iter)
+{
+    av_qsv_dts *cur_dts = 0;
+    av_qsv_dts *new_dts = 0;
+    int i = 0;
+#if HAVE_THREADS
+    int mut_ret = 0;
+#endif
+
+
+#if HAVE_THREADS
+    if (iter == 0 && qsv->qts_seq_mutex){
+        mut_ret = pthread_mutex_lock(qsv->qts_seq_mutex);
+        if(mut_ret)
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_lock issue[%d] at %s\n",mut_ret, __FUNCTION__);
+    }
+#endif
+
+    if (end == 0)
+        end = av_qsv_list_count(qsv->dts_seq);
+
+    if (end <= start) {
+        new_dts = av_mallocz(sizeof(av_qsv_dts));
+        if( new_dts ) {
+            new_dts->dts = dts;
+            av_qsv_list_add(qsv->dts_seq, new_dts);
+        }
+    } else
+        for (i = end; i > start; i--) {
+            cur_dts = av_qsv_list_item(qsv->dts_seq, i - 1);
+            if (cur_dts->dts < dts) {
+                new_dts = av_mallocz(sizeof(av_qsv_dts));
+                if( new_dts ) {
+                    new_dts->dts = dts;
+                    av_qsv_list_insert(qsv->dts_seq, i, new_dts);
+                }
+                break;
+            } else if (cur_dts->dts == dts)
+                break;
+        }
+#if HAVE_THREADS
+    if (iter == 0 && qsv->qts_seq_mutex){
+        mut_ret = pthread_mutex_unlock(qsv->qts_seq_mutex);
+        if(mut_ret)
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_unlock issue[%d] at %s\n",mut_ret, __FUNCTION__);
+    }
+#endif
+}
+
+void av_qsv_dts_pop(av_qsv_context * qsv)
+{
+    av_qsv_dts *item = 0;
+#if HAVE_THREADS
+    int mut_ret = 0;
+#endif
+
+#if HAVE_THREADS
+    if (qsv && qsv->qts_seq_mutex){
+        mut_ret = pthread_mutex_lock(qsv->qts_seq_mutex);
+        if(mut_ret)
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_lock issue[%d] at %s\n",mut_ret, __FUNCTION__);
+    }
+#endif
+
+    if (av_qsv_list_count(qsv->dts_seq)) {
+        item = av_qsv_list_item(qsv->dts_seq, 0);
+        av_qsv_list_rem(qsv->dts_seq, item);
+        av_free(item);
+    }
+#if HAVE_THREADS
+    if (qsv && qsv->qts_seq_mutex){
+        mut_ret = pthread_mutex_unlock(qsv->qts_seq_mutex);
+        if(mut_ret)
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_lock issue[%d] at %s\n",mut_ret, __FUNCTION__);
+        }
+#endif
+}
+
+
+av_qsv_list *av_qsv_list_init(int is_threaded)
+{
+    av_qsv_list *l;
+#if HAVE_THREADS
+    int mut_ret;
+#endif
+
+    l = av_mallocz(sizeof(av_qsv_list));
+    if (!l)
+        return 0;
+    l->items = av_mallocz(AV_QSV_JOB_SIZE_DEFAULT * sizeof(void *));
+    if (!l->items)
+        return 0;
+    l->items_alloc = AV_QSV_JOB_SIZE_DEFAULT;
+
+#if HAVE_THREADS
+    if (is_threaded) {
+        l->mutex = av_mallocz(sizeof(pthread_mutex_t));
+        if (l->mutex){
+            mut_ret = pthread_mutexattr_init(&l->mta);
+            if( mut_ret )
+                av_log(NULL, AV_LOG_ERROR, "pthread_mutexattr_init issue[%d] at %s\n",mut_ret, __FUNCTION__);
+            mut_ret = pthread_mutexattr_settype(&l->mta, PTHREAD_MUTEX_RECURSIVE /*PTHREAD_MUTEX_ERRORCHECK*/);
+            if( mut_ret )
+                av_log(NULL, AV_LOG_ERROR, "pthread_mutexattr_settype issue[%d] at %s\n",mut_ret, __FUNCTION__);
+            mut_ret = pthread_mutex_init(l->mutex, &l->mta);
+            if( mut_ret )
+                av_log(NULL, AV_LOG_ERROR, "pthread_mutex_init issue[%d] at %s\n",mut_ret, __FUNCTION__);
+        }
+    } else
+#endif
+        l->mutex = 0;
+    return l;
+}
+
+int av_qsv_list_count(av_qsv_list * l)
+{
+    int count;
+
+    av_qsv_list_lock(l);
+    count = l->items_count;
+    av_qsv_list_unlock(l);
+    return count;
+}
+
+int av_qsv_list_add(av_qsv_list * l, void *p)
+{
+    int pos = -1;
+
+    if (!p) {
+        return pos;
+    }
+
+    av_qsv_list_lock(l);
+
+    if (l->items_count == l->items_alloc) {
+        /* We need a bigger boat */
+        l->items_alloc += AV_QSV_JOB_SIZE_DEFAULT;
+        l->items = av_realloc(l->items, l->items_alloc * sizeof(void *));
+    }
+
+    l->items[l->items_count] = p;
+    pos = (l->items_count);
+    l->items_count++;
+
+    av_qsv_list_unlock(l);
+
+    return pos;
+}
+
+void av_qsv_list_rem(av_qsv_list * l, void *p)
+{
+    int i;
+
+    av_qsv_list_lock(l);
+
+    /* Find the item in the list */
+    for (i = 0; i < l->items_count; i++) {
+        if (l->items[i] == p) {
+            /* Shift all items after it sizeof( void * ) bytes earlier */
+            memmove(&l->items[i], &l->items[i + 1],
+                    (l->items_count - i - 1) * sizeof(void *));
+
+            l->items_count--;
+            break;
+        }
+    }
+
+    av_qsv_list_unlock(l);
+}
+
+void *av_qsv_list_item(av_qsv_list * l, int i)
+{
+    void *ret = NULL;
+
+    if (i < 0)
+        return NULL;
+
+    av_qsv_list_lock(l);
+    if( i < l->items_count)
+        ret = l->items[i];
+    av_qsv_list_unlock(l);
+    return ret;
+}
+
+void av_qsv_list_insert(av_qsv_list * l, int pos, void *p)
+{
+
+    if (!p)
+        return;
+
+    av_qsv_list_lock(l);
+
+    if (l->items_count == l->items_alloc) {
+        l->items_alloc += AV_QSV_JOB_SIZE_DEFAULT;
+        l->items = av_realloc(l->items, l->items_alloc * sizeof(void *));
+    }
+
+    if (l->items_count != pos) {
+        memmove(&l->items[pos + 1], &l->items[pos],
+                (l->items_count - pos) * sizeof(void *));
+    }
+
+    l->items[pos] = p;
+    l->items_count--;
+
+    av_qsv_list_unlock(l);
+}
+
+void av_qsv_list_close(av_qsv_list ** _l)
+{
+    av_qsv_list *l = *_l;
+#if HAVE_THREADS
+    int mut_ret;
+#endif
+
+    av_qsv_list_lock(l);
+
+    av_free(l->items);
+
+#if HAVE_THREADS
+    if (l->mutex){
+        mut_ret = pthread_mutex_unlock(l->mutex);
+        if( mut_ret )
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_unlock issue[%d] at %s\n",mut_ret, __FUNCTION__);
+        mut_ret = pthread_mutex_destroy(l->mutex);
+        if (mut_ret)
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_destroy issue[%d] at %s\n", mut_ret,__FUNCTION__);
+        mut_ret = pthread_mutexattr_destroy(&l->mta);
+        if (mut_ret)
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutexattr_destroy issue[%d] at %s\n", mut_ret,__FUNCTION__);
+    }
+#endif
+    av_freep(_l);
+}
+
+int av_qsv_list_lock(av_qsv_list *l){
+    int ret = 0;
+#if HAVE_THREADS
+    if (l->mutex){
+        ret = pthread_mutex_lock(l->mutex);
+        if( ret )
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_lock issue[%d] at %s\n",ret, __FUNCTION__);
+    }
+#endif
+    return ret;
+}
+
+int av_qsv_list_unlock(av_qsv_list *l){
+    int ret = 0;
+#if HAVE_THREADS
+    if (l->mutex){
+        ret = pthread_mutex_unlock(l->mutex);
+        if( ret )
+            av_log(NULL, AV_LOG_ERROR, "pthread_mutex_unlock issue[%d] at %s\n",ret, __FUNCTION__);
+    }
+#endif
+    return ret;
+}
+
+int av_is_qsv_available(mfxIMPL impl, mfxVersion * ver)
+{
+    mfxStatus sts = MFX_ERR_NONE;
+    mfxSession mfx_session;
+
+    memset(&mfx_session, 0, sizeof(mfxSession));
+    sts = MFXInit(impl, ver, &mfx_session);
+    if (sts >= 0)
+        MFXClose(mfx_session);
+    return sts;
+}
+
+void av_qsv_wait_on_sync(av_qsv_context *qsv, av_qsv_stage *stage)
+{
+    int iter = 0;
+    mfxStatus sts = MFX_ERR_NONE;
+    if( stage )
+        if(*stage->out.sync->p_sync){
+            while(1){
+                iter++;
+                sts = MFXVideoCORE_SyncOperation(qsv->mfx_session,*stage->out.sync->p_sync, AV_QSV_SYNC_TIME_DEFAULT);
+                if(MFX_WRN_IN_EXECUTION == sts){
+
+                    if(iter>20)
+                        AV_QSV_DEBUG_ASSERT(1, "Sync failed");
+
+                    av_qsv_sleep(10);
+                    continue;
+                }
+                AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+                break;
+            }
+        }
+}
diff -urN libav-11.3.org/libavcodec/qsv.h libav-11.3/libavcodec/qsv.h
--- libav-11.3.org/libavcodec/qsv.h	1970-01-01 01:00:00.000000000 +0100
+++ libav-11.3/libavcodec/qsv.h	2016-07-20 13:12:57.000000000 +0200
@@ -0,0 +1,494 @@
+/* ********************************************************************* *\
+
+Copyright (C) 2013 Intel Corporation.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+- Redistributions of source code must retain the above copyright notice,
+this list of conditions and the following disclaimer.
+- Redistributions in binary form must reproduce the above copyright notice,
+this list of conditions and the following disclaimer in the documentation
+and/or other materials provided with the distribution.
+- Neither the name of Intel Corporation nor the names of its contributors
+may be used to endorse or promote products derived from this software
+without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY INTEL CORPORATION "AS IS" AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+IN NO EVENT SHALL INTEL CORPORATION BE LIABLE FOR ANY DIRECT, INDIRECT,
+INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+\* ********************************************************************* */
+
+#ifndef AVCODEC_QSV_H
+#define AVCODEC_QSV_H
+
+/**
+ * @file
+ * @ingroup lavc_codec_hwaccel_qsv
+ * Common header for QSV/MediaSDK acceleration
+ */
+
+/**
+ * @defgroup lavc_codec_hwaccel_qsv QSV/MediaSDK based Decode/Encode and VPP
+ * @ingroup lavc_codec_hwaccel
+ *
+ *  As Intel Quick Sync Video (QSV) can decode/preprocess/encode with HW
+ *  acceleration.
+ *
+ *  Supported features:
+ *    - access:
+ *      - format AV_PIX_FMT_QSV_H264, AVCodec decoder based implementation
+ *      - name "h264_qsv", avcodec_find_decoder_by_name( "h264_qsv")
+ *    - IO Pattern:
+ *      - Opaque memory: MFX_IOPATTERN_OUT_OPAQUE_MEMORY // Video memory is
+ *                       MFX_IMPL_HARDWARE or MFX_IMPL_AUTO and runtime support,
+ *                       otherwise: System Memory
+ *      - System memory: MFX_IOPATTERN_OUT_SYSTEM_MEMORY
+ *    - Allocators:
+ *      - default allocator for System memory: MFX_MEMTYPE_SYSTEM_MEMORY
+ *    - details:
+ *      implementation as "per frame"
+ *
+ *  TODO list:
+ *    - access:
+ *      - format AV_PIX_FMT_QSV_MPEG2
+ *      - format AV_PIX_FMT_QSV_VC1
+ *      - format AV_PIX_FMT_QSV, see "details" below
+ *    - IO Pattern:
+ *      - VIDEO_MEMORY  // MFX_IOPATTERN_OUT_VIDEO_MEMORY
+ *    - Allocators:
+ *      - Video memory: MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET /
+ *                      MFX_MEMTYPE_VIDEO_MEMORY_PROCESSOR_TARGET
+ *    - details:
+ *      "per slice" support: AV_PIX_FMT_QSV with AVHWAccel based implementation
+ *
+ *  Note av_qsv_config struct required to fill in via
+ *  AVCodecContext.hwaccel_context
+ *
+ *  As per frame, note AVFrame.data[2] (qsv_atom) used for frame atom id,
+ *  data/linesize should be used together with SYSTEM_MEMORY and tested
+ *
+ *  Note: Compilation would require:
+ *   - Intel MediaSDK headers, Full SDK is avaialble from the original web site:
+ *                     http://software.intel.com/en-us/vcsource/tools/media-SDK
+ *     Will be referenced as mfx/*.h (mfxdefs.h, mfxstructures.h, ... )
+ *  and
+ *  - Final application has to link against Intel MediaSDK dispatcher, available
+ *     at MediaSDK as well
+ *
+ *  Target OS: as per available dispatcher and driver support
+ *
+ *  Implementation details:
+ *   Provided struct av_qsv_context contain several struct av_qsv_space(s) for decode,
+ *   VPP and encode.
+ *   av_qsv_space just contain needed environment for the appropriate action.
+ *   Based on this - pipeline (see pipes) will be build to pass details such as
+ *   mfxFrameSurface1* and mfxSyncPoint* from one action to the next.
+ *
+ *  Resources re-usage (av_qsv_flush_stages):
+ *     av_qsv_context *qsv = (av_qsv_context *)video_codec_ctx->priv_data;
+ *     av_qsv_list *pipe = (av_qsv_list *)video_frame->data[2];
+ *     av_qsv_flush_stages( qsv->pipes, &pipe );
+ *
+ *  DTS re-usage:
+ *     av_qsv_dts_pop(qsv);
+ *
+ *   for video,DX9/11 memory it has to be Unlock'ed as well
+ *
+ *  Implementation is thread aware and uses synchronization point(s) from MediaSDK
+ *  as per configuration.
+ *
+ *  For the details of MediaSDK usage and options available - please refer to the
+ *  available documentation at MediaSDK.
+ *
+ *  Feature set used from MSDK is defined by AV_QSV_MSDK_VERSION_MAJOR and
+ *  AV_QSV_MSDK_VERSION_MINOR
+ *
+ * @{
+ */
+
+#include <stdint.h>
+#include <string.h>
+#include "mfx/mfxvideo.h"
+#include "libavutil/mem.h"
+#include "libavutil/time.h"
+
+#ifdef HAVE_AV_CONFIG_H
+#include "config.h"
+#endif
+
+#if HAVE_THREADS
+#if defined (__GNUC__)
+#include <pthread.h>
+#define ff_qsv_atomic_inc(ptr) __sync_add_and_fetch(ptr,1)
+#define ff_qsv_atomic_dec(ptr) __sync_sub_and_fetch (ptr,1)
+#elif HAVE_WINDOWS_H            // MSVC case
+#include <windows.h>
+#if HAVE_PTHREADS
+#include <pthread.h>
+#elif HAVE_W32THREADS
+#include "w32pthreads.h"
+#endif
+#define ff_qsv_atomic_inc(ptr) InterlockedIncrement(ptr)
+#define ff_qsv_atomic_dec(ptr) InterlockedDecrement (ptr)
+#else
+// targeting only for MinGW or MSVC
+#endif
+
+#else
+#define ff_qsv_atomic_inc(ptr) ((*ptr)++)
+#define ff_qsv_atomic_dec(ptr) ((*ptr)--)
+#endif
+
+
+// sleep is defined in milliseconds
+#define av_qsv_sleep(x) av_usleep((x)*1000)
+
+#define AV_QSV_ZERO_MEMORY(VAR)                    {memset(&VAR, 0, sizeof(VAR));}
+#define AV_QSV_ALIGN32(X)                      (((mfxU32)((X)+31)) & (~ (mfxU32)31))
+#define AV_QSV_ALIGN16(value)                  (((value + 15) >> 4) << 4)
+#ifndef AV_QSV_PRINT_RET_MSG
+#define AV_QSV_PRINT_RET_MSG(ERR)              { av_log(NULL, AV_LOG_FATAL,"Error code %d,\t%s\t%d\n", ERR, __FUNCTION__, __LINE__); }
+#endif
+
+#ifndef AV_QSV_DEBUG_ASSERT
+#define AV_QSV_DEBUG_ASSERT(x,y)               {if ((x)) {av_log(NULL, AV_LOG_FATAL,"\nASSERT: %s\n",y);};}
+#endif
+
+#define AV_QSV_CHECK_RESULT(P, X, ERR)             {if ((X) > (P)) {AV_QSV_PRINT_RET_MSG(ERR); return ERR;}}
+#define AV_QSV_CHECK_POINTER(P, ERR)               {if (!(P)) {AV_QSV_PRINT_RET_MSG(ERR); return ERR;}}
+#define AV_QSV_IGNORE_MFX_STS(P, X)                {if ((X) == (P)) {P = MFX_ERR_NONE;}}
+
+#define AV_QSV_ID_BUFFER MFX_MAKEFOURCC('B','U','F','F')
+#define AV_QSV_ID_FRAME  MFX_MAKEFOURCC('F','R','M','E')
+
+#define AV_QSV_SURFACE_NUM              80
+#define AV_QSV_SYNC_NUM                 AV_QSV_SURFACE_NUM*3/4
+#define AV_QSV_BUF_SIZE_DEFAULT         4096*2160*10
+#define AV_QSV_JOB_SIZE_DEFAULT         10
+#define AV_QSV_SYNC_TIME_DEFAULT        10000
+// see av_qsv_get_free_sync, av_qsv_get_free_surface , 100 if usleep(10*1000)(10ms) == 1 sec
+#define AV_QSV_REPEAT_NUM_DEFAULT      100
+#define AV_QSV_ASYNC_DEPTH_DEFAULT     4
+
+// version of MSDK/QSV API currently used
+#define AV_QSV_MSDK_VERSION_MAJOR  1
+#define AV_QSV_MSDK_VERSION_MINOR  3
+
+typedef enum AV_QSV_STAGE_TYPE {
+
+#define AV_QSV_DECODE_MASK   0x001
+    AV_QSV_DECODE   = 0x001,
+
+#define AV_QSV_VPP_MASK      0x0F0
+    // "Mandatory VPP filter" , might be with "Hint-based VPP filters"
+    AV_QSV_VPP_DEFAULT = 0x010,
+    // "User Modules" etc
+    AV_QSV_VPP_USER = 0x020,
+
+#define av_QSV_ENCODE_MASK   0x100
+    AV_QSV_ENCODE   = 0x100
+#define AV_QSV_ANY_MASK      0xFFF
+} AV_QSV_STAGE_TYPE;
+
+
+typedef struct av_qsv_list {
+    // practically pthread_mutex_t
+    void *mutex;
+#if HAVE_THREADS
+    pthread_mutexattr_t   mta;
+#endif
+
+    void **items;
+    int items_alloc;
+
+    int items_count;
+} av_qsv_list;
+
+typedef struct av_qsv_sync {
+    mfxSyncPoint*   p_sync;
+    int             in_use;
+} av_qsv_sync;
+
+typedef struct av_qsv_stage {
+    AV_QSV_STAGE_TYPE type;
+    struct {
+        mfxBitstream *p_bs;
+        mfxFrameSurface1 *p_surface;
+    } in;
+    struct {
+        mfxBitstream *p_bs;
+        mfxFrameSurface1 *p_surface;
+        av_qsv_sync *sync;
+    } out;
+    av_qsv_list *pending;
+} av_qsv_stage;
+
+typedef struct av_qsv_task {
+    mfxBitstream *bs;
+    av_qsv_stage *stage;
+} av_qsv_task;
+
+
+typedef struct av_qsv_space {
+
+    uint8_t is_init_done;
+
+    AV_QSV_STAGE_TYPE type;
+
+    mfxVideoParam m_mfxVideoParam;
+
+    mfxFrameAllocResponse response;
+    mfxFrameAllocRequest request[2];    // [0] - in, [1] - out, if needed
+
+    mfxExtOpaqueSurfaceAlloc ext_opaque_alloc;
+    mfxExtBuffer **p_ext_params;
+    uint16_t p_ext_param_num;
+
+    uint16_t surface_num_max_used;
+    uint16_t surface_num;
+    mfxFrameSurface1 *p_surfaces[AV_QSV_SURFACE_NUM];
+
+    uint16_t sync_num_max_used;
+    uint16_t sync_num;
+    av_qsv_sync *p_syncp[AV_QSV_SYNC_NUM];
+
+    mfxBitstream bs;
+    uint8_t *p_buf;
+    size_t p_buf_max_size;
+
+    // only for encode and tasks
+    av_qsv_list *tasks;
+
+    av_qsv_list *pending;
+
+    // storage for allocations/mfxMemId*
+    mfxMemId *mids;
+} av_qsv_space;
+
+typedef struct av_qsv_context {
+    volatile int is_context_active;
+
+    mfxIMPL impl;
+    mfxSession mfx_session;
+    mfxVersion ver;
+
+    // decode
+    av_qsv_space *dec_space;
+    // encode
+    av_qsv_space *enc_space;
+    // vpp
+    av_qsv_list *vpp_space;
+
+    av_qsv_list *pipes;
+
+    // MediaSDK starting from API version 1.6 includes DecodeTimeStamp
+    // in addition to TimeStamp
+    // see also AV_QSV_MSDK_VERSION_MINOR , AV_QSV_MSDK_VERSION_MAJOR
+    av_qsv_list *dts_seq;
+
+    // practically pthread_mutex_t
+    void *qts_seq_mutex;
+
+    int is_anex;
+
+    void *qsv_config;
+
+} av_qsv_context;
+
+typedef enum {
+    QSV_PART_ANY = 0,
+    QSV_PART_LOWER,
+    QSV_PART_UPPER
+} av_qsv_split;
+
+typedef struct {
+    int64_t dts;
+} av_qsv_dts;
+
+typedef struct av_qsv_alloc_frame {
+    mfxU32 id;
+    mfxFrameInfo info;
+} av_qsv_alloc_frame;
+
+typedef struct av_qsv_alloc_buffer {
+    mfxU32 id;
+    mfxU32 nbytes;
+    mfxU16 type;
+} av_qsv_alloc_buffer;
+
+typedef struct av_qsv_allocators_space {
+    av_qsv_space *space;
+    mfxFrameAllocator frame_alloc;
+    mfxBufferAllocator buffer_alloc;
+} av_qsv_allocators_space;
+
+typedef struct av_qsv_config {
+    /**
+     * Set asynch depth of processing with QSV
+     * Format: 0 and more
+     *
+     * - encoding: Set by user.
+     * - decoding: Set by user.
+     */
+    int async_depth;
+
+    /**
+     * Range of numbers that indicate trade-offs between quality and speed.
+     * Format: from 1/MFX_TARGETUSAGE_BEST_QUALITY to 7/MFX_TARGETUSAGE_BEST_SPEED inclusive
+     *
+     * - encoding: Set by user.
+     * - decoding: unused
+     */
+    int target_usage;
+
+    /**
+     * Number of reference frames; if NumRefFrame = 0, this parameter is not specified.
+     * Format: 0 and more
+     *
+     * - encoding: Set by user.
+     * - decoding: unused
+     */
+    int num_ref_frame;
+
+    /**
+     * Distance between I- or P- key frames; if it is zero, the GOP structure is unspecified.
+     * Note: If GopRefDist = 1, there are no B-frames used.
+     *
+     * - encoding: Set by user.
+     * - decoding: unused
+     */
+     int gop_ref_dist;
+
+    /**
+     * Number of pictures within the current GOP (Group of Pictures); if GopPicSize=0,
+     * then the GOP size is unspecified. If GopPicSize=1, only I-frames are used.
+     *
+     * - encoding: Set by user.
+     * - decoding: unused
+     */
+     int gop_pic_size;
+
+    /**
+     * Set type of surfaces used with QSV
+     * Format: "IOPattern enum" of Media SDK
+     *
+     * - encoding: Set by user.
+     * - decoding: Set by user.
+     */
+    int io_pattern;
+
+    /**
+     * Set amount of additional surfaces might be needed
+     * Format: ammount of additional buffers(surfaces+syncs)
+     * to allocate in advance
+     *
+     * - encoding: Set by user.
+     * - decoding: Set by user.
+     */
+    int additional_buffers;
+
+    /**
+     * If pipeline should be sync.
+     * Format: wait time in milliseconds,
+     *         AV_QSV_SYNC_TIME_DEFAULT/10000 might be a good value
+     *
+     * - encoding: Set by user.
+     * - decoding: Set by user.
+     */
+    int sync_need;
+
+    /**
+     * Type of implementation needed
+     *
+     * - encoding: Set by user.
+     * - decoding: Set by user.
+     */
+    int impl_requested;
+
+    /**
+     * if QSV usage is multithreaded.
+     * Format: Yes/No, 1/0
+     *
+     * - encoding: Set by user.
+     * - decoding: Set by user.
+     */
+    int usage_threaded;
+
+    /**
+     * if QSV use an external allocation (valid per session/mfxSession)
+     * Format: pointer to allocators, if default: 0
+     *
+     * note that:
+     * System Memory:   can be used without provided and external allocator,
+     *  meaning MediaSDK will use an internal one
+     * Video Memory:    in this case - we must provide an external allocator
+     * Also, Media SDK session doesn't require external allocator if the application
+     *  uses opaque memory
+     *
+     * Calls SetFrameAllocator/SetBufferAllocator
+     * (MFXVideoCORE_SetFrameAllocator/MFXVideoCORE_SetBufferAllocator)
+     * are to pass allocators to Media SDK
+     *
+     * - encoding: Set by user.
+     * - decoding: Set by user.
+     */
+    av_qsv_allocators_space *allocators;
+
+} av_qsv_config;
+
+#define ANEX_UNKNOWN    0
+#define ANEX_PREFIX     1
+#define ANEX_NO_PREFIX  2
+
+static const uint8_t ff_prefix_code[] = { 0x00, 0x00, 0x00, 0x01 };
+
+int av_qsv_get_free_sync(av_qsv_space *, av_qsv_context *);
+int av_qsv_get_free_surface(av_qsv_space *, av_qsv_context *, mfxFrameInfo *,
+                     av_qsv_split);
+int av_qsv_get_free_encode_task(av_qsv_list *);
+
+int av_is_qsv_available(mfxIMPL, mfxVersion *);
+void av_qsv_wait_on_sync(av_qsv_context *, av_qsv_stage *);
+
+void av_qsv_add_context_usage(av_qsv_context *, int);
+
+void av_qsv_pipe_list_create(av_qsv_list **, int);
+void av_qsv_pipe_list_clean(av_qsv_list **);
+
+void av_qsv_add_stagee(av_qsv_list **, av_qsv_stage *, int);
+av_qsv_stage *av_qsv_get_last_stage(av_qsv_list *);
+av_qsv_list *av_qsv_pipe_by_stage(av_qsv_list *, av_qsv_stage *);
+void av_qsv_flush_stages(av_qsv_list *, av_qsv_list **);
+
+void av_qsv_dts_ordered_insert(av_qsv_context *, int, int, int64_t, int);
+void av_qsv_dts_pop(av_qsv_context *);
+
+av_qsv_stage *av_qsv_stage_init(void);
+void av_qsv_stage_clean(av_qsv_stage **);
+int av_qsv_context_clean(av_qsv_context *);
+
+int ff_qsv_is_sync_in_pipe(mfxSyncPoint *, av_qsv_context *);
+int ff_qsv_is_surface_in_pipe(mfxFrameSurface1 *, av_qsv_context *);
+
+av_qsv_list *av_qsv_list_init(int);
+int av_qsv_list_lock(av_qsv_list *);
+int av_qsv_list_unlock(av_qsv_list *);
+int av_qsv_list_add(av_qsv_list *, void *);
+void av_qsv_list_rem(av_qsv_list *, void *);
+void av_qsv_list_insert(av_qsv_list *, int, void *);
+void av_qsv_list_close(av_qsv_list **);
+
+int av_qsv_list_count(av_qsv_list *);
+void *av_qsv_list_item(av_qsv_list *, int);
+
+/* @} */
+
+#endif                          //AVCODEC_QSV_H
diff -urN libav-11.3.org/libavcodec/qsv_h264.c libav-11.3/libavcodec/qsv_h264.c
--- libav-11.3.org/libavcodec/qsv_h264.c	1970-01-01 01:00:00.000000000 +0100
+++ libav-11.3/libavcodec/qsv_h264.c	2016-07-20 13:12:57.000000000 +0200
@@ -0,0 +1,983 @@
+/* ********************************************************************* *\
+
+Copyright (C) 2013 Intel Corporation.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+- Redistributions of source code must retain the above copyright notice,
+this list of conditions and the following disclaimer.
+- Redistributions in binary form must reproduce the above copyright notice,
+this list of conditions and the following disclaimer in the documentation
+and/or other materials provided with the distribution.
+- Neither the name of Intel Corporation nor the names of its contributors
+may be used to endorse or promote products derived from this software
+without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY INTEL CORPORATION "AS IS" AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+IN NO EVENT SHALL INTEL CORPORATION BE LIABLE FOR ANY DIRECT, INDIRECT,
+INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+\* ********************************************************************* */
+
+#include "h264.h"
+#include "h264data.h"
+#include "qsv_h264.h"
+
+static av_qsv_config av_qsv_default_config = {
+    .async_depth        = AV_QSV_ASYNC_DEPTH_DEFAULT,
+    .target_usage       = MFX_TARGETUSAGE_BALANCED,
+    .num_ref_frame      = 0,
+    .gop_ref_dist       = 0,
+    .gop_pic_size       = 0,
+    .io_pattern         = MFX_IOPATTERN_OUT_OPAQUE_MEMORY,
+    .additional_buffers = 0,
+    .sync_need          = 0,
+    .impl_requested     = MFX_IMPL_HARDWARE,
+    .usage_threaded     = 0,
+    .allocators         = 0,
+};
+
+static av_qsv_allocators_space av_qsv_default_system_allocators = {
+    // fill to access mids
+    .space = 0,
+
+    .frame_alloc = {
+                    .pthis      = &av_qsv_default_system_allocators,
+                    .Alloc      = ff_qsv_mem_frame_alloc,
+                    .Lock       = ff_qsv_mem_frame_lock,
+                    .Unlock     = ff_qsv_mem_frame_unlock,
+                    .GetHDL     = ff_qsv_mem_frame_getHDL,
+                    .Free       = ff_qsv_mem_frame_free,
+                    },
+    .buffer_alloc = {
+                     .pthis     = &av_qsv_default_system_allocators,
+                     .Alloc     = ff_qsv_mem_buffer_alloc,
+                     .Lock      = ff_qsv_mem_buffer_lock,
+                     .Unlock    = ff_qsv_mem_buffer_unlock,
+                     .Free      = ff_qsv_mem_buffer_free,
+                     },
+};
+
+static const uint8_t ff_slice_code[] = { 0x00, 0x00, 0x01, 0x65 };
+
+int ff_qsv_nal_find_start_code(uint8_t * pb, size_t size)
+{
+    if ((int) size < 4)
+        return 0;
+
+    while ((4 <= size) && ((0 != pb[0]) || (0 != pb[1]) || (0 != pb[2]) || (1 != pb[3]))) {
+        pb += 1;
+        size -= 1;
+    }
+
+    if (4 <= size)
+        return 1;
+
+    return 0;
+}
+
+int ff_qsv_dec_init_clean(AVCodecContext *avctx)
+{
+    av_qsv_context *qsv = avctx->priv_data;
+    av_qsv_context_clean(qsv);
+    av_freep(&avctx->priv_data);
+    return 0;
+}
+
+int ff_qsv_dec_init(AVCodecContext * avctx)
+{
+    int ret = 0;
+    mfxStatus sts = MFX_ERR_NONE;
+    size_t current_offset = 6;
+    int header_size = 0;
+    unsigned char *current_position;
+    size_t current_size;
+
+    av_qsv_context *qsv = avctx->priv_data;
+    av_qsv_space *qsv_decode = qsv->dec_space;
+    av_qsv_config *qsv_config_context = avctx->hwaccel_context;
+
+    qsv->impl = qsv_config_context->impl_requested;
+
+    memset(&qsv->mfx_session, 0, sizeof(mfxSession));
+    qsv->ver.Major = AV_QSV_MSDK_VERSION_MAJOR;
+    qsv->ver.Minor = AV_QSV_MSDK_VERSION_MINOR;
+
+    sts = MFXInit(qsv->impl, &qsv->ver, &qsv->mfx_session);
+    AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+
+    AV_QSV_ZERO_MEMORY(qsv_decode->m_mfxVideoParam);
+    AV_QSV_ZERO_MEMORY(qsv_decode->m_mfxVideoParam.mfx);
+    qsv_decode->m_mfxVideoParam.mfx.CodecId = MFX_CODEC_AVC;
+    qsv_decode->m_mfxVideoParam.IOPattern =
+        qsv_config_context->io_pattern;
+
+    qsv_decode->m_mfxVideoParam.AsyncDepth =
+        qsv_config_context->async_depth;
+
+    AV_QSV_ZERO_MEMORY(qsv_decode->bs);
+    {
+        current_position    = avctx->extradata;
+        current_size        = avctx->extradata_size;
+
+        if (!ff_qsv_nal_find_start_code(current_position, current_size)) {
+
+            while (current_offset <= current_size) {
+                int current_nal_size =
+                    (unsigned char) current_position[current_offset] << 8 |
+                    (unsigned char) current_position[current_offset + 1];
+                unsigned char nal_type =
+                    (unsigned char) current_position[current_offset + 2] & 0x1F;
+
+                if (nal_type == NAL_SPS || nal_type == NAL_PPS) {
+                    memcpy(&qsv_decode->p_buf[header_size], ff_prefix_code,
+                           sizeof(ff_prefix_code));
+                    header_size += sizeof(ff_prefix_code);
+                    memcpy(&qsv_decode->p_buf[header_size],
+                           &current_position[current_offset + 2],
+                           current_nal_size);
+
+                    // fix for PPS as it comes after SPS, so - last
+                    if (nal_type == NAL_PPS) {
+                        // fix of MFXVideoDECODE_DecodeHeader: needs one SLICE to find, any SLICE
+                        memcpy(&qsv_decode->p_buf
+                               [header_size + current_nal_size],
+                               ff_slice_code, current_nal_size);
+                        header_size += sizeof(ff_slice_code);
+                    }
+                }
+
+                header_size += current_nal_size;
+                current_offset += current_nal_size + 3;
+            }
+        } else {
+            memcpy(&qsv_decode->p_buf[0], avctx->extradata,
+                   avctx->extradata_size);
+            header_size = avctx->extradata_size;
+            memcpy(&qsv_decode->p_buf
+                   [header_size], ff_slice_code, sizeof(ff_slice_code));
+            header_size += sizeof(ff_slice_code);
+        }
+    }
+
+    qsv_decode->bs.Data         = qsv_decode->p_buf;
+    qsv_decode->bs.DataLength   = header_size;
+    qsv_decode->bs.MaxLength    = qsv_decode->p_buf_max_size;
+
+    if (qsv_decode->bs.DataLength > qsv_decode->bs.MaxLength) {
+        av_log(avctx, AV_LOG_FATAL, "DataLength > MaxLength\n");
+        return -1;
+    }
+
+    sts = MFXVideoDECODE_DecodeHeader(qsv->mfx_session, &qsv_decode->bs,
+                                    &qsv_decode->m_mfxVideoParam);
+    if (sts < MFX_ERR_NONE && avctx->height && avctx->width)
+    {
+         av_log(avctx, AV_LOG_QUIET,"DecodeHeader failed with result:%d\n",sts);
+         sts = MFX_ERR_NONE;
+
+         // to cover absents of SPS details
+         qsv_decode->m_mfxVideoParam.mfx.FrameInfo.FourCC = MFX_FOURCC_NV12;
+         qsv_decode->m_mfxVideoParam.mfx.FrameInfo.Width  = AV_QSV_ALIGN16(avctx->width);
+         qsv_decode->m_mfxVideoParam.mfx.FrameInfo.Height = avctx->field_order > AV_FIELD_PROGRESSIVE ? AV_QSV_ALIGN32(avctx->height) : AV_QSV_ALIGN16(avctx->height);
+         qsv_decode->m_mfxVideoParam.mfx.FrameInfo.ChromaFormat = MFX_CHROMAFORMAT_YUV420;
+    }
+    AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+
+    qsv_decode->bs.DataLength   -= sizeof(ff_slice_code);
+
+    memset(&qsv_decode->request, 0, sizeof(mfxFrameAllocRequest) * 2);
+    sts = MFXVideoDECODE_QueryIOSurf(qsv->mfx_session,
+                                   &qsv_decode->m_mfxVideoParam,
+                                   &qsv_decode->request);
+
+    AV_QSV_IGNORE_MFX_STS(sts, MFX_WRN_PARTIAL_ACCELERATION);
+    AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+
+    qsv_decode->surface_num =
+        FFMIN(qsv_decode->request[0].NumFrameSuggested +
+            qsv_config_context->async_depth +
+            qsv_config_context->additional_buffers, AV_QSV_SURFACE_NUM);
+
+    if (qsv_decode->surface_num <= 0)
+        qsv_decode->surface_num = AV_QSV_SURFACE_NUM;
+
+    if (qsv_decode->m_mfxVideoParam.IOPattern ==
+        MFX_IOPATTERN_OUT_SYSTEM_MEMORY) {
+
+        // as per non-opaque memory:
+        if (!qsv_config_context->allocators) {
+            av_log(avctx, AV_LOG_INFO,
+                   "Using default allocators for QSV decode\n");
+            ((av_qsv_config *) avctx->hwaccel_context)->allocators =
+                &av_qsv_default_system_allocators;
+        }
+
+        qsv_config_context->allocators->space = qsv_decode;
+
+        qsv_decode->request[0].NumFrameMin       = qsv_decode->surface_num;
+        qsv_decode->request[0].NumFrameSuggested = qsv_decode->surface_num;
+
+        qsv_decode->request[0].Type = MFX_MEMTYPE_EXTERNAL_FRAME | MFX_MEMTYPE_FROM_DECODE;
+        // qsv_decode->request[0].Type |= m_bd3dAlloc ? MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET : MFX_MEMTYPE_SYSTEM_MEMORY;
+        qsv_decode->request[0].Type |= MFX_MEMTYPE_SYSTEM_MEMORY;
+
+        qsv_config_context->allocators->
+            frame_alloc.Alloc(qsv_config_context->allocators,
+                              &qsv_decode->request[0],
+                              &qsv_decode->response);
+    }
+
+    for (int i = 0; i < qsv_decode->surface_num; i++) {
+        qsv_decode->p_surfaces[i] = av_mallocz(sizeof(mfxFrameSurface1));
+        AV_QSV_CHECK_POINTER(qsv_decode->p_surfaces[i],
+                           AVERROR(ENOMEM));
+        memcpy(&(qsv_decode->p_surfaces[i]->Info),
+               &(qsv_decode->request[0].Info), sizeof(mfxFrameInfo));
+
+        // for an external(like DX9/11) based allocation:
+        // we bind:
+        //    m_pmfxSurfaces[i].Data.MemId = m_mfxResponse.mids[i];
+        // else, System memory:
+        if (qsv_decode->m_mfxVideoParam.IOPattern ==
+            MFX_IOPATTERN_OUT_SYSTEM_MEMORY) {
+            sts =
+                qsv_config_context->allocators->
+                frame_alloc.Lock(qsv_config_context->allocators,
+                                 qsv_decode->response.mids[i],
+                                 &(qsv_decode->p_surfaces[i]->Data));
+            AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+        }
+    }
+
+    qsv_decode->sync_num = FFMIN(qsv_decode->surface_num, AV_QSV_SYNC_NUM);
+    for (int i = 0; i < qsv_decode->sync_num; i++) {
+        qsv_decode->p_syncp[i] = av_mallocz(sizeof(av_qsv_sync));
+        AV_QSV_CHECK_POINTER(qsv_decode->p_syncp[i], AVERROR(ENOMEM));
+        qsv_decode->p_syncp[i]->p_sync = av_mallocz(sizeof(mfxSyncPoint));
+        AV_QSV_CHECK_POINTER(qsv_decode->p_syncp[i]->p_sync, AVERROR(ENOMEM));
+    }
+
+    memset(&qsv_decode->ext_opaque_alloc, 0,
+           sizeof(mfxExtOpaqueSurfaceAlloc));
+
+    if (qsv_decode->m_mfxVideoParam.IOPattern ==
+        MFX_IOPATTERN_OUT_OPAQUE_MEMORY) {
+        qsv_decode->m_mfxVideoParam.NumExtParam     = qsv_decode->p_ext_param_num = 1;
+
+        qsv_decode->p_ext_params = av_mallocz(sizeof(mfxExtBuffer *)*qsv_decode->p_ext_param_num);
+        AV_QSV_CHECK_POINTER(qsv_decode->p_ext_params, AVERROR(ENOMEM));
+
+        qsv_decode->m_mfxVideoParam.ExtParam        = qsv_decode->p_ext_params;
+
+        qsv_decode->ext_opaque_alloc.Out.Surfaces   = qsv_decode->p_surfaces;
+        qsv_decode->ext_opaque_alloc.Out.NumSurface = qsv_decode->surface_num;
+        qsv_decode->ext_opaque_alloc.Out.Type       = qsv_decode->request[0].Type;
+
+        qsv_decode->ext_opaque_alloc.Header.BufferId    = MFX_EXTBUFF_OPAQUE_SURFACE_ALLOCATION;
+        qsv_decode->ext_opaque_alloc.Header.BufferSz    = sizeof(mfxExtOpaqueSurfaceAlloc);
+        qsv_decode->p_ext_params[0]                     = (mfxExtBuffer *) &qsv_decode->ext_opaque_alloc;
+    }
+
+    sts =
+        MFXVideoDECODE_Init(qsv->mfx_session,
+                            &qsv_decode->m_mfxVideoParam);
+
+    AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+
+    qsv_decode->is_init_done = 1;
+    return ret;
+}
+
+av_cold int ff_qsv_decode_init(AVCodecContext * avctx)
+{
+    av_qsv_context *qsv;
+    av_qsv_space *qsv_decode;
+    av_qsv_config **qsv_config_context =
+        (av_qsv_config **) & avctx->hwaccel_context;
+
+    qsv = avctx->priv_data;
+
+    if (qsv && qsv->dec_space && qsv->dec_space->is_init_done || !avctx->extradata_size)
+        return 0;
+
+    if(!qsv)
+    qsv = av_mallocz(sizeof(av_qsv_context));
+    if (!qsv)
+        return AVERROR(ENOMEM);
+
+    qsv_decode = qsv->dec_space;
+    if(!qsv_decode)
+    qsv_decode = av_mallocz(sizeof(av_qsv_space));
+    if (!qsv_decode){
+        free(qsv);
+        return AVERROR(ENOMEM);
+    }
+    avctx->priv_data = qsv;
+    qsv->dec_space = qsv_decode;
+
+    qsv_decode->p_buf_max_size = AV_QSV_BUF_SIZE_DEFAULT;
+    if(!qsv_decode->p_buf)
+    qsv_decode->p_buf = av_malloc(qsv_decode->p_buf_max_size * sizeof(uint8_t));
+    if (!qsv_decode->p_buf)
+        return AVERROR(ENOMEM);
+
+    if (!(*qsv_config_context)) {
+        av_log(avctx, AV_LOG_INFO, "Using default config for QSV decode\n");
+        avctx->hwaccel_context = &av_qsv_default_config;
+    } else if ((*qsv_config_context)->io_pattern != MFX_IOPATTERN_OUT_OPAQUE_MEMORY &&
+               (*qsv_config_context)->io_pattern != MFX_IOPATTERN_OUT_SYSTEM_MEMORY) {
+        avpriv_report_missing_feature(avctx,
+                                      "Only MFX_IOPATTERN_OUT_OPAQUE_MEMORY"
+                                      " and MFX_IOPATTERN_OUT_SYSTEM_MEMORY"
+                                      " are currently supported\n");
+        return AVERROR_PATCHWELCOME;
+    }
+
+    qsv->qsv_config = avctx->hwaccel_context;
+
+    av_qsv_add_context_usage(qsv,
+                          HAVE_THREADS
+                          ? (*qsv_config_context)->usage_threaded :
+                          HAVE_THREADS);
+
+    // allocation of p_syncp and p_surfaces inside of ff_qsv_dec_init
+    return ff_qsv_dec_init(avctx);
+}
+
+static av_cold int qsv_decode_end(AVCodecContext * avctx)
+{
+    mfxStatus sts       = MFX_ERR_NONE;
+    av_qsv_context *qsv    = avctx->priv_data;
+    av_qsv_config *qsv_config_context = avctx->hwaccel_context;
+
+    if (qsv) {
+        av_qsv_space *qsv_decode = qsv->dec_space;
+        if (qsv_decode && qsv_decode->is_init_done) {
+            // todo: change to AV_LOG_INFO
+            av_log(avctx, AV_LOG_QUIET,
+                   "qsv_decode report done, max_surfaces: %u/%u , max_syncs: %u/%u\n",
+                   qsv_decode->surface_num_max_used,
+                   qsv_decode->surface_num, qsv_decode->sync_num_max_used,
+                   qsv_decode->sync_num);
+        }
+
+        if (qsv_config_context
+            && qsv_config_context->io_pattern ==
+            MFX_IOPATTERN_OUT_SYSTEM_MEMORY) {
+            if (qsv_config_context->allocators) {
+                sts =
+                    qsv_config_context->allocators->
+                    frame_alloc.Free(qsv_config_context->allocators,
+                                     &qsv_decode->response);
+                AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+            } else {
+                av_log(avctx, AV_LOG_FATAL,
+                       "No QSV allocators found for clean up\n");
+            }
+        }
+        // closing the own resources
+        av_freep(&qsv_decode->p_buf);
+
+        for (int i = 0; i < qsv_decode->surface_num; i++) {
+            av_freep(&qsv_decode->p_surfaces[i]);
+        }
+        qsv_decode->surface_num = 0;
+
+        if( qsv_decode->p_ext_param_num || qsv_decode->p_ext_params )
+            av_freep(&qsv_decode->p_ext_params);
+        qsv_decode->p_ext_param_num = 0;
+
+        for (int i = 0; i < qsv_decode->sync_num; i++) {
+            av_freep(&qsv_decode->p_syncp[i]->p_sync);
+            av_freep(&qsv_decode->p_syncp[i]);
+        }
+        qsv_decode->sync_num = 0;
+        qsv_decode->is_init_done = 0;
+
+        av_freep(&qsv->dec_space);
+
+        // closing commong stuff
+        av_qsv_context_clean(qsv);
+    }
+
+    return 0;
+}
+
+static int qsv_decode_frame(AVCodecContext * avctx, void *data,
+                            int *data_size, AVPacket * avpkt)
+{
+    mfxStatus sts = MFX_ERR_NONE;
+    av_qsv_context *qsv = avctx->priv_data;
+    av_qsv_space *qsv_decode;
+    av_qsv_config *qsv_config_context = avctx->hwaccel_context;
+    int *got_picture_ptr = data_size;
+    int ret_value = 1;
+    uint8_t *current_position = avpkt->data;
+    int current_size = avpkt->size;
+    int frame_processed = 0;
+    size_t frame_length = 0;
+    int surface_idx = 0;
+    int extra_data_workaround = 0;
+
+    int sync_idx = 0;
+    int current_nal_size;
+    unsigned char nal_type;
+    av_qsv_stage *new_stage = 0;
+    mfxBitstream *input_bs = NULL;
+    size_t current_offset = 2;
+    av_qsv_list *qsv_atom = 0;
+    av_qsv_list *pipe = 0;
+
+    AVFrame *picture = (AVFrame *) data;
+
+    *got_picture_ptr = 0;
+
+    qsv = avctx->priv_data;
+    if(!qsv){
+        extra_data_workaround = !avctx->extradata_size;
+        if(extra_data_workaround){
+            avctx->extradata = avpkt->data;
+            avctx->extradata_size = avpkt->size;
+        }
+        sts = ff_qsv_decode_init(avctx);
+        qsv = avctx->priv_data;
+        if(extra_data_workaround){
+            avctx->extradata = 0;
+            avctx->extradata_size = 0;
+        }
+        if(sts<0){
+             ff_qsv_dec_init_clean(avctx);
+             *got_picture_ptr = 0;
+             return sts;
+   	    }
+    }
+    qsv_decode = qsv->dec_space;
+
+    if (qsv_decode->bs.DataOffset + qsv_decode->bs.DataLength +
+        current_size > qsv_decode->bs.MaxLength) {
+        memmove(&qsv_decode->bs.Data[0],
+                qsv_decode->bs.Data + qsv_decode->bs.DataOffset,
+                qsv_decode->bs.DataLength);
+        qsv_decode->bs.DataOffset = 0;
+    }
+
+    if (current_size) {
+        if(qsv->is_anex == ANEX_UNKNOWN){
+            if (ff_qsv_nal_find_start_code(current_position, current_size) && current_position == avpkt->data)
+               qsv->is_anex = ANEX_PREFIX;
+            else
+               qsv->is_anex = ANEX_NO_PREFIX;
+        }
+        if (qsv->is_anex == ANEX_PREFIX){
+            memcpy(&qsv_decode->bs.Data[0] +
+                   qsv_decode->bs.DataLength +
+                   qsv_decode->bs.DataOffset,
+                   avpkt->data,
+                   avpkt->size);
+            qsv_decode->bs.DataLength += avpkt->size;
+            frame_length += avpkt->size;
+        }
+        else
+            while (current_offset <= current_size) {
+                current_nal_size =
+                    ((unsigned char) current_position[current_offset - 2] << 24 |
+                     (unsigned char) current_position[current_offset -  1] << 16 |
+                     (unsigned char) current_position[current_offset] << 8 |
+                     (unsigned char) current_position[current_offset + 1]) - 1;
+                nal_type =
+                    (unsigned char) current_position[current_offset + 2] & 0x1F;
+                {
+                    frame_length += current_nal_size;
+                    memcpy(&qsv_decode->bs.Data[0] +
+                           qsv_decode->bs.DataLength +
+                           qsv_decode->bs.DataOffset, ff_prefix_code,
+                           sizeof(ff_prefix_code));
+                    qsv_decode->bs.DataLength += sizeof(ff_prefix_code);
+                    memcpy(&qsv_decode->bs.Data[0] +
+                           qsv_decode->bs.DataLength +
+                           qsv_decode->bs.DataOffset,
+                           &current_position[current_offset + 2],
+                           current_nal_size + 1);
+                    qsv_decode->bs.DataLength += current_nal_size + 1;
+                }
+                current_offset += current_nal_size + 5;
+            }
+
+        if (qsv_decode->bs.DataLength > qsv_decode->bs.MaxLength) {
+            av_log(avctx, AV_LOG_FATAL, "DataLength > MaxLength\n");
+            return -1;
+        }
+    }
+
+    if (frame_length || current_size == 0) {
+
+        qsv_decode->bs.TimeStamp = avpkt->pts;
+
+        //not a drain
+        if ((current_size || qsv_decode->bs.DataLength))
+            av_qsv_dts_ordered_insert(qsv, 0, 0, qsv_decode->bs.TimeStamp, 0);
+
+        sts = MFX_ERR_NONE;
+        // ignore warnings, where warnings >0 , and not error codes <0
+        while (MFX_ERR_NONE <= sts || MFX_ERR_MORE_SURFACE == sts
+               || MFX_WRN_DEVICE_BUSY == sts) {
+
+            if (MFX_ERR_MORE_SURFACE == sts || MFX_ERR_NONE == sts) {
+                surface_idx =
+                    av_qsv_get_free_surface(qsv_decode, qsv,
+                                     &qsv_decode->request[0].Info,
+                                     QSV_PART_ANY);
+
+                if (surface_idx == -1) {
+                    *got_picture_ptr = 0;
+                    return 0;
+                }
+            }
+
+            if (MFX_WRN_DEVICE_BUSY == sts)
+                av_qsv_sleep(10);
+
+            sync_idx = av_qsv_get_free_sync(qsv_decode, qsv);
+
+            if (sync_idx == -1) {
+                *got_picture_ptr = 0;
+                return 0;
+            }
+            new_stage = av_qsv_stage_init();
+            input_bs = NULL;
+            // if to drain last ones
+            if (current_size || qsv_decode->bs.DataLength)
+                input_bs = &qsv_decode->bs;
+            // Decode a frame asynchronously (returns immediately)
+            // very first IDR / SLICE should be with SPS/PPS
+            sts = MFXVideoDECODE_DecodeFrameAsync(qsv->mfx_session, input_bs,
+                                                qsv_decode->p_surfaces
+                                                [surface_idx],
+                                                &new_stage->out.p_surface,
+                                                qsv_decode->p_syncp[sync_idx]->p_sync);
+
+            new_stage->out.sync = qsv_decode->p_syncp[sync_idx];
+            // have some results
+            if (MFX_ERR_NONE <= sts && MFX_WRN_DEVICE_BUSY != sts &&
+                MFX_WRN_VIDEO_PARAM_CHANGED != sts) {
+
+                ff_qsv_atomic_inc(&(new_stage->out.p_surface->Data.Locked));
+
+                new_stage->type         = AV_QSV_DECODE;
+                new_stage->in.p_bs      = input_bs;
+                new_stage->in.p_surface = qsv_decode->p_surfaces[surface_idx];
+
+                pipe = av_qsv_list_init(HAVE_THREADS ? qsv_config_context->usage_threaded : HAVE_THREADS);
+                av_qsv_add_stagee(&pipe, new_stage,
+                              HAVE_THREADS ?
+                              qsv_config_context->usage_threaded :
+                              HAVE_THREADS);
+
+                av_qsv_list_add(qsv->pipes, pipe);
+                qsv_atom = pipe;
+
+                // usage for forced decode sync and results, can be avoided if sync done by next stage
+                // also note wait time for Sync and possible usage with MFX_WRN_IN_EXECUTION check
+                if (qsv_config_context->sync_need) {
+                    sts =
+                        MFXVideoCORE_SyncOperation(qsv->mfx_session,
+                                                   qsv_decode->p_syncp[sync_idx]->p_sync,
+                                                   qsv_config_context->sync_need);
+                    AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+
+                    // no need to wait more -> force off
+                    ff_qsv_atomic_dec(&qsv_decode->p_syncp[sync_idx]->in_use);
+                    new_stage->out.sync   = 0;
+                }
+
+                sts = MFX_ERR_NONE;
+                break;
+            }
+            av_qsv_stage_clean(&new_stage);
+
+            /*
+               Can be because of:
+               - runtime situation:
+               - drain procedure:
+               At the end of the bitstream, the application continuously calls the MFXVideoDECODE_DecodeFrameAsync function with a
+               NULL bitstream pointer to drain any remaining frames cached within the Intel
+               Media SDK decoder, until the function returns MFX_ERR_MORE_DATA.
+             */
+            if (MFX_ERR_MORE_DATA == sts) {
+                // not a drain
+                if (current_size) {
+                    *got_picture_ptr = 0;
+                    return avpkt->size;
+                }
+                // drain
+                break;
+            }
+            if (MFX_ERR_MORE_SURFACE == sts  ){
+                continue;
+            }
+
+            AV_QSV_CHECK_RESULT(sts, MFX_ERR_NONE, sts);
+        }
+
+        frame_processed = 1;
+    }
+
+    if (frame_processed) {
+
+        if (current_size) {
+            *got_picture_ptr    = 1;
+            ret_value           = avpkt->size;
+        } else {
+            if (MFX_ERR_MORE_DATA != sts) {
+                *got_picture_ptr    = 1;
+                ret_value           = avpkt->size;
+            } else {
+                *got_picture_ptr = 0;
+                return 0;
+            }
+        }
+
+        picture->pkt_pts            = new_stage->out.p_surface->Data.TimeStamp;
+        picture->pts                = new_stage->out.p_surface->Data.TimeStamp;
+
+        picture->repeat_pict        = (qsv_decode->m_mfxVideoParam.mfx.FrameInfo.PicStruct & MFX_PICSTRUCT_FIELD_REPEATED);
+        picture->interlaced_frame   = !(qsv_decode->m_mfxVideoParam.mfx.FrameInfo.PicStruct & MFX_PICSTRUCT_PROGRESSIVE);
+        picture->top_field_first    = (qsv_decode->m_mfxVideoParam.mfx.FrameInfo.PicStruct & MFX_PICSTRUCT_FIELD_TFF);
+
+        // since we do not know it yet from MSDK, let's do just a simple way for now
+        picture->key_frame          = (avctx->frame_number == 0) ? 1 : 0;
+
+        if (qsv_decode->m_mfxVideoParam.IOPattern == MFX_IOPATTERN_OUT_SYSTEM_MEMORY) {
+            picture->data[0]        = new_stage->out.p_surface->Data.Y;
+            picture->data[1]        = new_stage->out.p_surface->Data.VU;
+            picture->linesize[0]    = new_stage->out.p_surface->Info.Width;
+            picture->linesize[1]    = new_stage->out.p_surface->Info.Width;
+        } else {
+            picture->data[0]        = 0;
+            picture->data[1]        = 0;
+            picture->linesize[0]    = 0;
+            picture->linesize[1]    = 0;
+        }
+
+        picture->data[2]            = qsv_atom;
+        picture->linesize[2]        = 0;
+    }
+
+    return ret_value;
+}
+
+// Will be called when seeking
+static void qsv_flush_dpb(AVCodecContext * avctx)
+{
+    av_qsv_context *qsv = avctx->priv_data;
+    av_qsv_space *qsv_decode = qsv->dec_space;
+
+    qsv_decode->bs.DataOffset = 0;
+    qsv_decode->bs.DataLength = 0;
+    qsv_decode->bs.MaxLength = qsv_decode->p_buf_max_size;
+}
+
+
+mfxStatus ff_qsv_mem_frame_alloc(mfxHDL pthis,
+                                 mfxFrameAllocRequest * request,
+                                 mfxFrameAllocResponse * response)
+{
+    mfxStatus sts = MFX_ERR_NONE;
+
+    mfxU32 numAllocated = 0;
+
+    mfxU32 width = AV_QSV_ALIGN32(request->Info.Width);
+    mfxU32 height = AV_QSV_ALIGN32(request->Info.Height);
+    mfxU32 nbytes;
+
+    av_qsv_allocators_space *this_alloc = (av_qsv_allocators_space *) pthis;
+    av_qsv_alloc_frame *fs;
+
+    if (!this_alloc->space)
+        return MFX_ERR_NOT_INITIALIZED;
+
+    switch (request->Info.FourCC) {
+    case MFX_FOURCC_YV12:
+    case MFX_FOURCC_NV12:
+        nbytes =
+            width * height + (width >> 1) * (height >> 1) +
+            (width >> 1) * (height >> 1);
+        break;
+    case MFX_FOURCC_RGB3:
+        nbytes = width * height + width * height + width * height;
+        break;
+    case MFX_FOURCC_RGB4:
+        nbytes =
+            width * height + width * height + width * height +
+            width * height;
+        break;
+    case MFX_FOURCC_YUY2:
+        nbytes =
+            width * height + (width >> 1) * (height) +
+            (width >> 1) * (height);
+        break;
+    default:
+        return MFX_ERR_UNSUPPORTED;
+    }
+
+    this_alloc->space->mids =
+        av_malloc(sizeof(mfxMemId) * request->NumFrameSuggested);
+    if (!this_alloc->space->mids)
+        return MFX_ERR_MEMORY_ALLOC;
+
+    // allocate frames
+    for (numAllocated = 0; numAllocated < request->NumFrameSuggested;
+         numAllocated++) {
+        sts =
+            this_alloc->buffer_alloc.Alloc(this_alloc->buffer_alloc.pthis,
+                                           nbytes +
+                                           AV_QSV_ALIGN32(sizeof
+                                                       (av_qsv_alloc_frame)),
+                                           request->Type,
+                                           &(this_alloc->
+                                             space->mids[numAllocated]));
+
+        if (MFX_ERR_NONE != sts)
+            break;
+
+        sts =
+            this_alloc->buffer_alloc.Lock(this_alloc->buffer_alloc.pthis,
+                                          this_alloc->
+                                          space->mids[numAllocated],
+                                          (mfxU8 **) & fs);
+
+        if (MFX_ERR_NONE != sts)
+            break;
+
+        fs->id = AV_QSV_ID_FRAME;
+        fs->info = request->Info;
+        this_alloc->buffer_alloc.Unlock(this_alloc->buffer_alloc.pthis,
+                                        this_alloc->
+                                        space->mids[numAllocated]);
+    }
+
+    // check the number of allocated frames
+    if (numAllocated < request->NumFrameMin)
+        return MFX_ERR_MEMORY_ALLOC;
+
+    response->NumFrameActual = (mfxU16) numAllocated;
+    response->mids = this_alloc->space->mids;
+
+    return MFX_ERR_NONE;
+}
+
+mfxStatus ff_qsv_mem_frame_lock(mfxHDL pthis, mfxMemId mid,
+                                mfxFrameData * ptr)
+{
+    mfxStatus sts = MFX_ERR_NONE;
+    av_qsv_alloc_frame *fs = 0;
+    mfxU16 width;
+    mfxU16 height;
+
+    av_qsv_allocators_space *this_alloc = (av_qsv_allocators_space *) pthis;
+
+    if (!this_alloc->space)
+        return MFX_ERR_NOT_INITIALIZED;
+    if (!ptr)
+        return MFX_ERR_NULL_PTR;
+
+
+    sts =
+        this_alloc->buffer_alloc.Lock(this_alloc->buffer_alloc.pthis, mid,
+                                      (mfxU8 **) & fs);
+
+    if (MFX_ERR_NONE != sts)
+        return sts;
+
+    if (AV_QSV_ID_FRAME != fs->id) {
+        this_alloc->buffer_alloc.Unlock(this_alloc->buffer_alloc.pthis,
+                                        mid);
+        return MFX_ERR_INVALID_HANDLE;
+    }
+
+    width = (mfxU16) AV_QSV_ALIGN32(fs->info.Width);
+    height = (mfxU16) AV_QSV_ALIGN32(fs->info.Height);
+    ptr->B = ptr->Y =
+        (mfxU8 *) fs + AV_QSV_ALIGN32(sizeof(av_qsv_allocators_space));
+
+    switch (fs->info.FourCC) {
+    case MFX_FOURCC_NV12:
+        ptr->U = ptr->Y + width * height;
+        ptr->V = ptr->U + 1;
+        ptr->Pitch = width;
+        break;
+    case MFX_FOURCC_YV12:
+        ptr->V = ptr->Y + width * height;
+        ptr->U = ptr->V + (width >> 1) * (height >> 1);
+        ptr->Pitch = width;
+        break;
+    case MFX_FOURCC_YUY2:
+        ptr->U = ptr->Y + 1;
+        ptr->V = ptr->Y + 3;
+        ptr->Pitch = 2 * width;
+        break;
+    case MFX_FOURCC_RGB3:
+        ptr->G = ptr->B + 1;
+        ptr->R = ptr->B + 2;
+        ptr->Pitch = 3 * width;
+        break;
+    case MFX_FOURCC_RGB4:
+        ptr->G = ptr->B + 1;
+        ptr->R = ptr->B + 2;
+        ptr->A = ptr->B + 3;
+        ptr->Pitch = 4 * width;
+        break;
+    default:
+        return MFX_ERR_UNSUPPORTED;
+    }
+
+    return MFX_ERR_NONE;
+}
+
+mfxStatus ff_qsv_mem_frame_unlock(mfxHDL pthis, mfxMemId mid,
+                                  mfxFrameData * ptr)
+{
+    mfxStatus sts = MFX_ERR_NONE;
+    av_qsv_allocators_space *this_alloc = (av_qsv_allocators_space *) pthis;
+
+    sts =
+        this_alloc->buffer_alloc.Unlock(this_alloc->buffer_alloc.pthis,
+                                        mid);
+
+    if (MFX_ERR_NONE != sts)
+        return sts;
+
+    if (NULL != ptr) {
+        ptr->Pitch = 0;
+        ptr->Y = 0;
+        ptr->U = 0;
+        ptr->V = 0;
+    }
+
+    return MFX_ERR_NONE;
+}
+
+mfxStatus ff_qsv_mem_frame_getHDL(mfxHDL pthis, mfxMemId mid,
+                                  mfxHDL * handle)
+{
+    return MFX_ERR_UNSUPPORTED;
+}
+
+mfxStatus ff_qsv_mem_frame_free(mfxHDL pthis,
+                                mfxFrameAllocResponse * response)
+{
+    mfxStatus sts = MFX_ERR_NONE;
+    av_qsv_allocators_space *this_alloc = (av_qsv_allocators_space *) pthis;
+    mfxU32 i;
+
+    if (!response)
+        return MFX_ERR_NULL_PTR;
+
+    if (!this_alloc->space)
+        return MFX_ERR_NOT_INITIALIZED;
+
+    if (response->mids)
+        for (i = 0; i < response->NumFrameActual; i++) {
+            if (response->mids[i]) {
+                sts =
+                    this_alloc->buffer_alloc.Free(this_alloc->
+                                                  buffer_alloc.pthis,
+                                                  response->mids[i]);
+                if (MFX_ERR_NONE != sts)
+                    return sts;
+            }
+        }
+
+    av_freep(&response->mids);
+
+    return sts;
+}
+
+
+mfxStatus ff_qsv_mem_buffer_alloc(mfxHDL pthis, mfxU32 nbytes, mfxU16 type,
+                                  mfxMemId * mid)
+{
+    av_qsv_alloc_buffer *bs;
+    mfxU32 header_size;
+    mfxU8 *buffer_ptr;
+
+    if (!mid)
+        return MFX_ERR_NULL_PTR;
+
+    if (0 == (type & MFX_MEMTYPE_SYSTEM_MEMORY))
+        return MFX_ERR_UNSUPPORTED;
+
+    header_size = AV_QSV_ALIGN32(sizeof(av_qsv_alloc_buffer));
+    buffer_ptr = (mfxU8 *) av_malloc(header_size + nbytes);
+
+    if (!buffer_ptr)
+        return MFX_ERR_MEMORY_ALLOC;
+
+    bs = (av_qsv_alloc_buffer *) buffer_ptr;
+    bs->id = AV_QSV_ID_BUFFER;
+    bs->type = type;
+    bs->nbytes = nbytes;
+    *mid = (mfxHDL) bs;
+    return MFX_ERR_NONE;
+}
+
+mfxStatus ff_qsv_mem_buffer_lock(mfxHDL pthis, mfxMemId mid, mfxU8 ** ptr)
+{
+    av_qsv_alloc_buffer *bs;
+
+    if (!ptr)
+        return MFX_ERR_NULL_PTR;
+
+    bs = (av_qsv_alloc_buffer *) mid;
+
+    if (!bs)
+        return MFX_ERR_INVALID_HANDLE;
+    if (AV_QSV_ID_BUFFER != bs->id)
+        return MFX_ERR_INVALID_HANDLE;
+
+    *ptr = (mfxU8 *) bs + AV_QSV_ALIGN32(sizeof(av_qsv_alloc_buffer));
+    return MFX_ERR_NONE;
+}
+
+mfxStatus ff_qsv_mem_buffer_unlock(mfxHDL pthis, mfxMemId mid)
+{
+    av_qsv_alloc_buffer *bs = (av_qsv_alloc_buffer *) mid;
+
+    if (!bs || AV_QSV_ID_BUFFER != bs->id)
+        return MFX_ERR_INVALID_HANDLE;
+
+    return MFX_ERR_NONE;
+}
+
+mfxStatus ff_qsv_mem_buffer_free(mfxHDL pthis, mfxMemId mid)
+{
+    av_qsv_alloc_buffer *bs = (av_qsv_alloc_buffer *) mid;
+    if (!bs || AV_QSV_ID_BUFFER != bs->id)
+        return MFX_ERR_INVALID_HANDLE;
+
+    av_freep(&bs);
+    return MFX_ERR_NONE;
+}
+
+
+AVCodec ff_h264_qsv_decoder = {
+    .name           = "h264_qsv",
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = AV_CODEC_ID_H264,
+    .init           = ff_qsv_decode_init,
+    .close          = qsv_decode_end,
+    .decode         = qsv_decode_frame,
+    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
+    .flush          = qsv_flush_dpb,
+    .long_name      = NULL_IF_CONFIG_SMALL("H.264 / AVC / Intel QSV"),
+    .pix_fmts       = (const enum PixelFormat[]) {AV_PIX_FMT_QSV_H264,
+                                                  AV_PIX_FMT_NONE},
+};
diff -urN libav-11.3.org/libavcodec/qsv_h264.h libav-11.3/libavcodec/qsv_h264.h
--- libav-11.3.org/libavcodec/qsv_h264.h	1970-01-01 01:00:00.000000000 +0100
+++ libav-11.3/libavcodec/qsv_h264.h	2016-07-20 13:12:57.000000000 +0200
@@ -0,0 +1,65 @@
+/* ********************************************************************* *\
+
+Copyright (C) 2013 Intel Corporation.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+- Redistributions of source code must retain the above copyright notice,
+this list of conditions and the following disclaimer.
+- Redistributions in binary form must reproduce the above copyright notice,
+this list of conditions and the following disclaimer in the documentation
+and/or other materials provided with the distribution.
+- Neither the name of Intel Corporation nor the names of its contributors
+may be used to endorse or promote products derived from this software
+without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY INTEL CORPORATION "AS IS" AND ANY EXPRESS OR
+IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+IN NO EVENT SHALL INTEL CORPORATION BE LIABLE FOR ANY DIRECT, INDIRECT,
+INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
+NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
+THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+\* ********************************************************************* */
+
+#ifndef AVCODEC_QSV_H264_H
+#define AVCODEC_QSV_H264_H
+
+#include "qsv.h"
+
+int ff_qsv_dec_init(AVCodecContext *);
+int ff_qsv_nal_find_start_code(uint8_t * pb, size_t size);
+
+int ff_qsv_dec_init_clean(AVCodecContext *avctx);
+av_cold int ff_qsv_decode_init(AVCodecContext * avctx);
+static av_cold int qsv_decode_end(AVCodecContext * avctx);
+static int qsv_decode_frame(AVCodecContext * avctx, void *data,
+                            int *data_size, AVPacket * avpkt);
+static void qsv_flush_dpb(AVCodecContext * avctx);
+
+
+// Default for SYSTEM MEMORY
+// as from MFXFrameAllocator
+mfxStatus ff_qsv_mem_frame_alloc(mfxHDL pthis,
+                                 mfxFrameAllocRequest * request,
+                                 mfxFrameAllocResponse * response);
+mfxStatus ff_qsv_mem_frame_lock(mfxHDL pthis, mfxMemId mid,
+                                mfxFrameData * ptr);
+mfxStatus ff_qsv_mem_frame_unlock(mfxHDL pthis, mfxMemId mid,
+                                  mfxFrameData * ptr);
+mfxStatus ff_qsv_mem_frame_getHDL(mfxHDL pthis, mfxMemId mid,
+                                  mfxHDL * handle);
+mfxStatus ff_qsv_mem_frame_free(mfxHDL pthis,
+                                mfxFrameAllocResponse * response);
+// as from mfxBufferAllocator
+mfxStatus ff_qsv_mem_buffer_alloc(mfxHDL pthis, mfxU32 nbytes, mfxU16 type,
+                                  mfxMemId * mid);
+mfxStatus ff_qsv_mem_buffer_lock(mfxHDL pthis, mfxMemId mid, mfxU8 ** ptr);
+mfxStatus ff_qsv_mem_buffer_unlock(mfxHDL pthis, mfxMemId mid);
+mfxStatus ff_qsv_mem_buffer_free(mfxHDL pthis, mfxMemId mid);
+
+#endif                          //AVCODEC_QSV_H264_H
diff -urN libav-11.3.org/libavutil/pixfmt.h libav-11.3/libavutil/pixfmt.h
--- libav-11.3.org/libavutil/pixfmt.h	2016-07-20 13:11:07.000000000 +0200
+++ libav-11.3/libavutil/pixfmt.h	2016-07-20 13:12:57.000000000 +0200
@@ -193,6 +193,7 @@
     AV_PIX_FMT_NV16,         ///< interleaved chroma YUV 4:2:2, 16bpp, (1 Cr & Cb sample per 2x1 Y samples)
     AV_PIX_FMT_NV20LE,       ///< interleaved chroma YUV 4:2:2, 20bpp, (1 Cr & Cb sample per 2x1 Y samples), little-endian
     AV_PIX_FMT_NV20BE,       ///< interleaved chroma YUV 4:2:2, 20bpp, (1 Cr & Cb sample per 2x1 Y samples), big-endian
+    AV_PIX_FMT_QSV_H264,     ///< H.264 HW decoding with QSV, data[2] contains qsv_atom information for MFX_IOPATTERN_OUT_OPAQUE_MEMORY, MFX_IOPATTERN_OUT_VIDEO_MEMORY
 
     AV_PIX_FMT_RGBA64BE,     ///< packed RGBA 16:16:16:16, 64bpp, 16R, 16G, 16B, 16A, the 2-byte value for each R/G/B/A component is stored as big-endian
     AV_PIX_FMT_RGBA64LE,     ///< packed RGBA 16:16:16:16, 64bpp, 16R, 16G, 16B, 16A, the 2-byte value for each R/G/B/A component is stored as little-endian