gstreamer/subprojects/gst-plugins-bad/sys/qsv/gstqsvencoder.cpp

1550 lines
46 KiB
C++
Raw Normal View History

qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
/* GStreamer
* Copyright (C) 2021 Seungha Yang <seungha@centricular.com>
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library 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
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
* Boston, MA 02110-1301, USA.
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "gstqsvencoder.h"
#include <mfxvideo++.h>
#include <string.h>
#include <string>
#ifdef G_OS_WIN32
#include "gstqsvallocator_d3d11.h"
#include <wrl.h>
/* *INDENT-OFF* */
using namespace Microsoft::WRL;
/* *INDENT-ON* */
#else
#include "gstqsvallocator_va.h"
#endif /* G_OS_WIN32 */
GST_DEBUG_CATEGORY_EXTERN (gst_qsv_encoder_debug);
#define GST_CAT_DEFAULT gst_qsv_encoder_debug
GType
gst_qsv_coding_option_get_type (void)
{
static GType coding_opt_type = 0;
static const GEnumValue coding_opts[] = {
{MFX_CODINGOPTION_UNKNOWN, "Unknown", "unknown"},
{MFX_CODINGOPTION_ON, "On", "on"},
{MFX_CODINGOPTION_OFF, "Off", "off"},
{0, nullptr, nullptr}
};
if (g_once_init_enter (&coding_opt_type)) {
GType type = g_enum_register_static ("GstQsvCodingOption",
coding_opts);
g_once_init_leave (&coding_opt_type, type);
}
return coding_opt_type;
}
enum
{
PROP_0,
PROP_TARGET_USAGE,
PROP_LOW_LATENCY,
};
#define DEFAULT_TARGET_USAGE MFX_TARGETUSAGE_BALANCED
#define DEFAULT_LOW_LATENCY FALSE
typedef struct _GstQsvEncoderSurface
{
mfxFrameSurface1 surface;
mfxEncodeCtrl encode_control;
/* array of mfxPayload (e.g., SEI data) associated with this surface */
GPtrArray *payload;
/* holds ownership */
GstQsvFrame *qsv_frame;
} GstQsvEncoderSurface;
typedef struct _GstQsvEncoderTask
{
mfxSyncPoint sync_point;
mfxBitstream bitstream;
} GstQsvEncoderTask;
struct _GstQsvEncoderPrivate
{
GstObject *device;
GstVideoCodecState *input_state;
GstQsvAllocator *allocator;
/* API specific alignment requirement (multiple of 16 or 32) */
GstVideoInfo aligned_info;
mfxSession session;
mfxVideoParam video_param;
/* List of mfxExtBuffer configured by subclass, subclass will hold
* allocated memory for each mfxExtBuffer */
GPtrArray *extra_params;
MFXVideoENCODE *encoder;
GstQsvMemoryType mem_type;
/* Internal buffer pool used to allocate fallback buffer when input buffer
* is not compatible with expected format/type/resolution etc */
GstBufferPool *internal_pool;
/* Array of GstQsvEncoderSurface, holding ownership */
GArray *surface_pool;
guint next_surface_index;
/* Array of GstQsvEncoderTask, holding ownership */
GArray *task_pool;
GQueue free_tasks;
GQueue pending_tasks;
/* Properties */
guint target_usage;
gboolean low_latency;
};
#define gst_qsv_encoder_parent_class parent_class
G_DEFINE_ABSTRACT_TYPE_WITH_PRIVATE (GstQsvEncoder, gst_qsv_encoder,
GST_TYPE_VIDEO_ENCODER);
static void gst_qsv_encoder_dispose (GObject * object);
static void gst_qsv_encoder_finalize (GObject * object);
static void gst_qsv_encoder_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec);
static void gst_qsv_encoder_get_property (GObject * object, guint prop_id,
GValue * value, GParamSpec * pspec);
static void gst_qsv_encoder_set_context (GstElement * element,
GstContext * context);
static gboolean gst_qsv_encoder_open (GstVideoEncoder * encoder);
static gboolean gst_qsv_encoder_stop (GstVideoEncoder * encoder);
static gboolean gst_qsv_encoder_close (GstVideoEncoder * encoder);
static gboolean gst_qsv_encoder_set_format (GstVideoEncoder * encoder,
GstVideoCodecState * state);
static GstFlowReturn gst_qsv_encoder_handle_frame (GstVideoEncoder * encoder,
GstVideoCodecFrame * frame);
static GstFlowReturn gst_qsv_encoder_finish (GstVideoEncoder * encoder);
static gboolean gst_qsv_encoder_flush (GstVideoEncoder * encoder);
static gboolean gst_qsv_encoder_sink_query (GstVideoEncoder * encoder,
GstQuery * query);
static gboolean gst_qsv_encoder_src_query (GstVideoEncoder * encoder,
GstQuery * query);
static gboolean gst_qsv_encoder_propose_allocation (GstVideoEncoder * encoder,
GstQuery * query);
static void gst_qsv_encoder_surface_clear (GstQsvEncoderSurface * task);
static void gst_qsv_encoder_task_clear (GstQsvEncoderTask * task);
static void
gst_qsv_encoder_class_init (GstQsvEncoderClass * klass)
{
GObjectClass *object_class = G_OBJECT_CLASS (klass);
GstElementClass *element_class = GST_ELEMENT_CLASS (klass);
GstVideoEncoderClass *videoenc_class = GST_VIDEO_ENCODER_CLASS (klass);
object_class->dispose = gst_qsv_encoder_dispose;
object_class->finalize = gst_qsv_encoder_finalize;
object_class->set_property = gst_qsv_encoder_set_property;
object_class->get_property = gst_qsv_encoder_get_property;
g_object_class_install_property (object_class, PROP_TARGET_USAGE,
g_param_spec_uint ("target-usage", "Target Usage",
"1: Best quality, 4: Balanced, 7: Best speed",
1, 7, DEFAULT_TARGET_USAGE,
(GParamFlags) (G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS)));
g_object_class_install_property (object_class, PROP_LOW_LATENCY,
g_param_spec_boolean ("low-latency", "Low Latency",
"Enables low-latency encoding", DEFAULT_LOW_LATENCY,
(GParamFlags) (G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS)));
element_class->set_context = GST_DEBUG_FUNCPTR (gst_qsv_encoder_set_context);
videoenc_class->open = GST_DEBUG_FUNCPTR (gst_qsv_encoder_open);
videoenc_class->stop = GST_DEBUG_FUNCPTR (gst_qsv_encoder_stop);
videoenc_class->close = GST_DEBUG_FUNCPTR (gst_qsv_encoder_close);
videoenc_class->set_format = GST_DEBUG_FUNCPTR (gst_qsv_encoder_set_format);
videoenc_class->handle_frame =
GST_DEBUG_FUNCPTR (gst_qsv_encoder_handle_frame);
videoenc_class->finish = GST_DEBUG_FUNCPTR (gst_qsv_encoder_finish);
videoenc_class->flush = GST_DEBUG_FUNCPTR (gst_qsv_encoder_flush);
videoenc_class->sink_query = GST_DEBUG_FUNCPTR (gst_qsv_encoder_sink_query);
videoenc_class->src_query = GST_DEBUG_FUNCPTR (gst_qsv_encoder_src_query);
videoenc_class->propose_allocation =
GST_DEBUG_FUNCPTR (gst_qsv_encoder_propose_allocation);
}
static void
gst_qsv_encoder_init (GstQsvEncoder * self)
{
GstQsvEncoderPrivate *priv;
priv = self->priv =
(GstQsvEncoderPrivate *) gst_qsv_encoder_get_instance_private (self);
priv->extra_params = g_ptr_array_sized_new (8);
priv->surface_pool = g_array_new (FALSE, TRUE, sizeof (GstQsvEncoderSurface));
g_array_set_clear_func (priv->surface_pool,
(GDestroyNotify) gst_qsv_encoder_surface_clear);
priv->task_pool = g_array_new (FALSE, TRUE, sizeof (GstQsvEncoderTask));
g_array_set_clear_func (priv->task_pool,
(GDestroyNotify) gst_qsv_encoder_task_clear);
g_queue_init (&priv->free_tasks);
g_queue_init (&priv->pending_tasks);
priv->target_usage = DEFAULT_TARGET_USAGE;
priv->low_latency = DEFAULT_LOW_LATENCY;
}
static void
gst_qsv_encoder_dispose (GObject * object)
{
GstQsvEncoder *self = GST_QSV_ENCODER (object);
GstQsvEncoderPrivate *priv = self->priv;
gst_clear_object (&priv->device);
G_OBJECT_CLASS (parent_class)->dispose (object);
}
static void
gst_qsv_encoder_finalize (GObject * object)
{
GstQsvEncoder *self = GST_QSV_ENCODER (object);
GstQsvEncoderPrivate *priv = self->priv;
g_ptr_array_unref (priv->extra_params);
g_array_unref (priv->task_pool);
g_array_unref (priv->surface_pool);
G_OBJECT_CLASS (parent_class)->finalize (object);
}
static void
gst_qsv_encoder_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec)
{
GstQsvEncoder *self = GST_QSV_ENCODER (object);
GstQsvEncoderPrivate *priv = self->priv;
switch (prop_id) {
case PROP_TARGET_USAGE:
priv->target_usage = g_value_get_uint (value);
break;
case PROP_LOW_LATENCY:
priv->low_latency = g_value_get_boolean (value);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static void
gst_qsv_encoder_get_property (GObject * object, guint prop_id, GValue * value,
GParamSpec * pspec)
{
GstQsvEncoder *self = GST_QSV_ENCODER (object);
GstQsvEncoderPrivate *priv = self->priv;
switch (prop_id) {
case PROP_TARGET_USAGE:
g_value_set_uint (value, priv->target_usage);
break;
case PROP_LOW_LATENCY:
g_value_set_boolean (value, priv->low_latency);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static void
gst_qsv_encoder_set_context (GstElement * element, GstContext * context)
{
GstQsvEncoder *self = GST_QSV_ENCODER (element);
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (element);
GstQsvEncoderPrivate *priv = self->priv;
#ifdef G_OS_WIN32
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_d3d11_handle_set_context_for_adapter_luid (element,
context, klass->adapter_luid, (GstD3D11Device **) & priv->device);
#else
gst_va_handle_set_context (element, context, klass->display_path,
(GstVaDisplay **) & priv->device);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
#endif
GST_ELEMENT_CLASS (parent_class)->set_context (element, context);
}
#ifdef G_OS_WIN32
static gboolean
gst_qsv_encoder_open_platform_device (GstQsvEncoder * self)
{
GstQsvEncoderPrivate *priv = self->priv;
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (self);
ComPtr < ID3D10Multithread > multi_thread;
HRESULT hr;
ID3D11Device *device_handle;
mfxStatus status;
GstD3D11Device *device;
if (!gst_d3d11_ensure_element_data_for_adapter_luid (GST_ELEMENT (self),
klass->adapter_luid, (GstD3D11Device **) & priv->device)) {
GST_ERROR_OBJECT (self, "d3d11 device is unavailable");
return FALSE;
}
device = GST_D3D11_DEVICE_CAST (priv->device);
priv->allocator = gst_qsv_d3d11_allocator_new (device);
/* For D3D11 device handle to be used by QSV, multithread protection layer
* must be enabled before the MFXVideoCORE_SetHandle() call.
*
* TODO: Need to check performance impact by this mutithread protection layer,
* since it may have a negative impact on overall pipeline performance.
* If so, we should create encoding session dedicated d3d11 device and
* make use of shared resource */
device_handle = gst_d3d11_device_get_device_handle (device);
hr = device_handle->QueryInterface (IID_PPV_ARGS (&multi_thread));
if (!gst_d3d11_result (hr, device)) {
GST_ERROR_OBJECT (self, "ID3D10Multithread interface is unavailable");
return FALSE;
}
multi_thread->SetMultithreadProtected (TRUE);
status = MFXVideoCORE_SetHandle (priv->session, MFX_HANDLE_D3D11_DEVICE,
device_handle);
if (status != MFX_ERR_NONE) {
GST_ERROR_OBJECT (self, "Failed to set d3d11 device handle");
return FALSE;
}
/* NOTE: We never use this mfxFrameAllocator to allocate memory from our side,
* but required for QSV because:
* 1) QSV may request memory allocation for encoder's internal usage,
* MFX_FOURCC_P8 for example
* 2) Our mfxFrameAllocator provides bridge layer for
* gst_video_frame_{map,unmap} and mfxFrameAllocator::{Lock,Unlock},
* including mfxFrameAllocator::GetHDL.
* 3) GstQsvAllocator provides GstQsvFrame pool, and therefore allocated
* GstQsvFrame struct can be re-used without per-frame malloc/free
*/
status = MFXVideoCORE_SetFrameAllocator (priv->session,
gst_qsv_allocator_get_allocator_handle (priv->allocator));
if (status != MFX_ERR_NONE) {
GST_ERROR_OBJECT (self, "Failed to set frame allocator %d", status);
return FALSE;
}
return TRUE;
}
#else
static gboolean
gst_qsv_encoder_open_platform_device (GstQsvEncoder * self)
{
GstQsvEncoderPrivate *priv = self->priv;
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (self);
mfxStatus status;
GstVaDisplay *display;
if (!gst_va_ensure_element_data (GST_ELEMENT (self), klass->display_path,
(GstVaDisplay **) & priv->device)) {
GST_ERROR_OBJECT (self, "VA display is unavailable");
return FALSE;
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
}
display = GST_VA_DISPLAY (priv->device);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
priv->allocator = gst_qsv_va_allocator_new (display);
status = MFXVideoCORE_SetHandle (priv->session, MFX_HANDLE_VA_DISPLAY,
gst_va_display_get_va_dpy (display));
if (status != MFX_ERR_NONE) {
GST_ERROR_OBJECT (self, "Failed to set VA display handle");
return FALSE;
}
status = MFXVideoCORE_SetFrameAllocator (priv->session,
gst_qsv_allocator_get_allocator_handle (priv->allocator));
if (status != MFX_ERR_NONE) {
GST_ERROR_OBJECT (self, "Failed to set frame allocator %d", status);
return FALSE;
}
return TRUE;
}
#endif
static gboolean
gst_qsv_encoder_open (GstVideoEncoder * encoder)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
GstQsvEncoderPrivate *priv = self->priv;
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (self);
mfxStatus status;
status = MFXCreateSession (gst_qsv_get_loader (), klass->impl_index,
&priv->session);
if (status != MFX_ERR_NONE) {
GST_ERROR_OBJECT (self, "Failed to create session");
return FALSE;
}
if (!gst_qsv_encoder_open_platform_device (self)) {
g_clear_pointer (&priv->session, MFXClose);
gst_clear_object (&priv->allocator);
gst_clear_object (&priv->device);
return FALSE;
}
return TRUE;
}
static gboolean
gst_qsv_encoder_reset (GstQsvEncoder * self)
{
GstQsvEncoderPrivate *priv = self->priv;
if (priv->encoder) {
delete priv->encoder;
priv->encoder = nullptr;
}
if (priv->internal_pool) {
gst_buffer_pool_set_active (priv->internal_pool, FALSE);
gst_clear_object (&priv->internal_pool);
}
g_array_set_size (priv->surface_pool, 0);
g_array_set_size (priv->task_pool, 0);
g_queue_clear (&priv->free_tasks);
g_queue_clear (&priv->pending_tasks);
return TRUE;
}
static gboolean
gst_qsv_encoder_stop (GstVideoEncoder * encoder)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
GstQsvEncoderPrivate *priv = self->priv;
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_qsv_encoder_reset (self);
g_clear_pointer (&priv->input_state, gst_video_codec_state_unref);
return TRUE;
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
}
static gboolean
gst_qsv_encoder_close (GstVideoEncoder * encoder)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
GstQsvEncoderPrivate *priv = self->priv;
g_clear_pointer (&priv->session, MFXClose);
gst_clear_object (&priv->allocator);
gst_clear_object (&priv->device);
return TRUE;
}
static void
gst_qsv_encoder_payload_clear (mfxPayload * payload)
{
if (!payload)
return;
g_free (payload->Data);
g_free (payload);
}
static void
gst_qsv_encoder_surface_reset (GstQsvEncoderSurface * surface)
{
if (!surface)
return;
gst_clear_qsv_frame (&surface->qsv_frame);
g_ptr_array_set_size (surface->payload, 0);
memset (&surface->encode_control, 0, sizeof (mfxEncodeCtrl));
}
static void
gst_qsv_encoder_surface_clear (GstQsvEncoderSurface * surface)
{
if (!surface)
return;
gst_qsv_encoder_surface_reset (surface);
g_clear_pointer (&surface->payload, g_ptr_array_unref);
memset (&surface->surface, 0, sizeof (mfxFrameSurface1));
}
static void
gst_qsv_encoder_task_reset (GstQsvEncoder * self, GstQsvEncoderTask * task)
{
GstQsvEncoderPrivate *priv = self->priv;
if (!task)
return;
task->sync_point = nullptr;
task->bitstream.DataLength = 0;
g_queue_push_head (&priv->free_tasks, task);
}
static void
gst_qsv_encoder_task_clear (GstQsvEncoderTask * task)
{
if (!task)
return;
g_clear_pointer (&task->bitstream.Data, g_free);
memset (&task->bitstream, 0, sizeof (mfxBitstream));
}
static GstQsvEncoderSurface *
gst_qsv_encoder_get_next_surface (GstQsvEncoder * self)
{
GstQsvEncoderPrivate *priv = self->priv;
GstQsvEncoderSurface *surface = nullptr;
for (guint i = priv->next_surface_index; i < priv->surface_pool->len; i++) {
GstQsvEncoderSurface *iter =
&g_array_index (priv->surface_pool, GstQsvEncoderSurface, i);
/* This means surface is still being used by QSV */
if (iter->surface.Data.Locked > 0)
continue;
surface = iter;
priv->next_surface_index = i;
goto out;
}
for (guint i = 0; i < priv->next_surface_index; i++) {
GstQsvEncoderSurface *iter =
&g_array_index (priv->surface_pool, GstQsvEncoderSurface, i);
/* This means surface is still being used by QSV */
if (iter->surface.Data.Locked > 0)
continue;
surface = iter;
priv->next_surface_index = i;
goto out;
}
/* Magic number to avoid too large pool size */
if (priv->surface_pool->len > 64) {
GST_ERROR_OBJECT (self,
"No availble surface but pool size is too large already");
return nullptr;
}
/* Something went wrong, increase surface pool size */
GST_INFO_OBJECT (self, "No useable surfaces, increasing pool size to %d",
priv->surface_pool->len + 1);
g_array_set_size (priv->surface_pool, priv->surface_pool->len + 1);
surface = &g_array_index (priv->surface_pool, GstQsvEncoderSurface,
priv->surface_pool->len - 1);
memset (surface, 0, sizeof (GstQsvEncoderSurface));
surface->surface.Info =
g_array_index (priv->surface_pool, GstQsvEncoderSurface, 0).surface.Info;
surface->payload = g_ptr_array_new_with_free_func ((GDestroyNotify)
gst_qsv_encoder_payload_clear);
out:
priv->next_surface_index++;
priv->next_surface_index %= priv->surface_pool->len;
gst_qsv_encoder_surface_reset (surface);
return surface;
}
static mfxStatus
gst_qsv_encoder_encode_frame (GstQsvEncoder * self,
GstQsvEncoderSurface * surface, GstQsvEncoderTask * task, mfxU64 timestamp)
{
mfxFrameSurface1 *s;
GstQsvEncoderPrivate *priv = self->priv;
mfxStatus status;
guint retry_count = 0;
/* magic number */
const guint retry_threshold = 100;
mfxEncodeCtrl *encode_ctrl;
if (surface) {
s = &surface->surface;
s->Data.MemId = (mfxMemId) surface->qsv_frame;
s->Data.TimeStamp = timestamp;
encode_ctrl = &surface->encode_control;
} else {
/* draining */
s = nullptr;
encode_ctrl = nullptr;
}
do {
status = priv->encoder->EncodeFrameAsync (encode_ctrl,
s, &task->bitstream, &task->sync_point);
/* XXX: probably we should try to drain pending tasks if any in this case
* as documented? */
if (status == MFX_WRN_DEVICE_BUSY && retry_count < retry_threshold) {
GST_INFO_OBJECT (self, "GPU is busy, retry count (%d/%d)",
retry_count, retry_threshold);
retry_count++;
/* Magic number 10ms */
g_usleep (10000);
continue;
}
break;
} while (TRUE);
return status;
}
static GstVideoCodecFrame *
gst_qsv_encoder_find_output_frame (GstQsvEncoder * self, GstClockTime pts)
{
GList *frames, *iter;
GstVideoCodecFrame *ret = nullptr;
GstVideoCodecFrame *closest = nullptr;
guint64 min_pts_abs_diff = 0;
/* give up, just returns the oldest frame */
if (!GST_CLOCK_TIME_IS_VALID (pts))
return gst_video_encoder_get_oldest_frame (GST_VIDEO_ENCODER (self));
frames = gst_video_encoder_get_frames (GST_VIDEO_ENCODER (self));
for (iter = frames; iter; iter = g_list_next (iter)) {
GstVideoCodecFrame *frame = (GstVideoCodecFrame *) iter->data;
guint64 abs_diff;
if (!GST_CLOCK_TIME_IS_VALID (frame->pts))
continue;
if (pts == frame->pts) {
ret = frame;
break;
}
if (pts >= frame->pts)
abs_diff = pts - frame->pts;
else
abs_diff = frame->pts - pts;
if (!closest || abs_diff < min_pts_abs_diff) {
closest = frame;
min_pts_abs_diff = abs_diff;
}
}
if (!ret && closest)
ret = closest;
if (ret) {
gst_video_codec_frame_ref (ret);
} else {
ret = gst_video_encoder_get_oldest_frame (GST_VIDEO_ENCODER (self));
}
if (frames)
g_list_free_full (frames, (GDestroyNotify) gst_video_codec_frame_unref);
return ret;
}
static GstFlowReturn
gst_qsv_encoder_finish_frame (GstQsvEncoder * self, GstQsvEncoderTask * task,
gboolean discard)
{
GstQsvEncoderPrivate *priv = self->priv;
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (self);
mfxStatus status;
mfxBitstream *bs;
GstVideoCodecFrame *frame;
GstClockTime pts = GST_CLOCK_TIME_NONE;
GstClockTime dts = GST_CLOCK_TIME_NONE;
GstBuffer *buffer;
gboolean keyframe = FALSE;
guint retry_count = 0;
/* magic number */
const guint retry_threshold = 100;
status = MFX_ERR_NONE;
do {
/* magic number 100 ms */
status = MFXVideoCORE_SyncOperation (priv->session, task->sync_point, 100);
/* Retry up to 10 sec (100 ms x 100 times), that should be enough time for
* encoding a frame using hardware */
if (status == MFX_WRN_IN_EXECUTION && retry_count < retry_threshold) {
GST_DEBUG_OBJECT (self,
"Operation is still in execution, retry count (%d/%d)",
retry_count, retry_threshold);
retry_count++;
continue;
}
break;
} while (TRUE);
if (discard) {
gst_qsv_encoder_task_reset (self, task);
return GST_FLOW_OK;
}
if (status != MFX_ERR_NONE && status != MFX_ERR_NONE_PARTIAL_OUTPUT) {
gst_qsv_encoder_task_reset (self, task);
if (status == MFX_ERR_ABORTED) {
GST_INFO_OBJECT (self, "Operation was aborted");
return GST_FLOW_FLUSHING;
}
GST_WARNING_OBJECT (self, "SyncOperation returned %d (%s)",
QSV_STATUS_ARGS (status));
return GST_FLOW_ERROR;
}
bs = &task->bitstream;
pts = gst_qsv_timestamp_to_gst (bs->TimeStamp);
dts = gst_qsv_timestamp_to_gst ((mfxU64) bs->DecodeTimeStamp);
if ((bs->FrameType & MFX_FRAMETYPE_IDR) != 0)
keyframe = TRUE;
if (klass->create_output_buffer) {
buffer = klass->create_output_buffer (self, bs);
} else {
buffer = gst_buffer_new_memdup (bs->Data + bs->DataOffset, bs->DataLength);
}
gst_qsv_encoder_task_reset (self, task);
if (!buffer) {
GST_ERROR_OBJECT (self, "No output buffer");
return GST_FLOW_ERROR;
}
frame = gst_qsv_encoder_find_output_frame (self, pts);
if (frame) {
frame->pts = pts;
frame->dts = dts;
frame->output_buffer = buffer;
if (keyframe)
GST_VIDEO_CODEC_FRAME_SET_SYNC_POINT (frame);
return gst_video_encoder_finish_frame (GST_VIDEO_ENCODER (self), frame);
}
/* Empty available frame, something went wrong but we can just push this
* buffer */
GST_WARNING_OBJECT (self, "Failed to find corresponding frame");
GST_BUFFER_PTS (buffer) = pts;
GST_BUFFER_DTS (buffer) = dts;
if (!keyframe)
GST_BUFFER_FLAG_SET (buffer, GST_BUFFER_FLAG_DELTA_UNIT);
return gst_pad_push (GST_VIDEO_ENCODER_SRC_PAD (self), buffer);
}
static GstFlowReturn
gst_qsv_encoder_drain (GstQsvEncoder * self, gboolean discard)
{
GstQsvEncoderPrivate *priv = self->priv;
mfxStatus status = MFX_ERR_NONE;
GstFlowReturn ret = GST_FLOW_OK;
GstQsvEncoderTask *task;
if (!priv->session || !priv->encoder)
return GST_FLOW_OK;
GST_DEBUG_OBJECT (self, "Drain");
/* Drain pending tasks first if any */
while (g_queue_get_length (&priv->pending_tasks) > 0) {
task = (GstQsvEncoderTask *) g_queue_pop_tail (&priv->pending_tasks);
ret = gst_qsv_encoder_finish_frame (self, task, discard);
}
while (status == MFX_ERR_NONE) {
task = (GstQsvEncoderTask *) g_queue_pop_tail (&priv->free_tasks);
status = gst_qsv_encoder_encode_frame (self,
nullptr, task, MFX_TIMESTAMP_UNKNOWN);
/* once it's fully drained, then driver will return more data */
if (status == MFX_ERR_NONE && task->sync_point) {
ret = gst_qsv_encoder_finish_frame (self, task, discard);
continue;
}
if (status != MFX_ERR_MORE_DATA)
GST_WARNING_OBJECT (self, "Unexpected status return %d (%s)",
QSV_STATUS_ARGS (status));
g_queue_push_head (&priv->free_tasks, task);
}
/* Release GstQsvFrame objects */
for (guint i = 0; i < priv->surface_pool->len; i++) {
GstQsvEncoderSurface *iter =
&g_array_index (priv->surface_pool, GstQsvEncoderSurface, i);
if (iter->surface.Data.Locked > 0) {
GST_WARNING_OBJECT (self,
"Encoder was drained but QSV is holding surface %d", i);
continue;
}
gst_qsv_encoder_surface_reset (iter);
}
return ret;
}
#ifdef G_OS_WIN32
static gboolean
gst_qsv_encoder_prepare_d3d11_pool (GstQsvEncoder * self,
GstCaps * caps, GstVideoInfo * aligned_info)
{
GstQsvEncoderPrivate *priv = self->priv;
GstStructure *config;
GstD3D11AllocationParams *params;
GstD3D11Device *device = GST_D3D11_DEVICE_CAST (priv->device);
priv->internal_pool = gst_d3d11_buffer_pool_new (device);
config = gst_buffer_pool_get_config (priv->internal_pool);
params = gst_d3d11_allocation_params_new (device, aligned_info,
(GstD3D11AllocationFlags) 0, 0);
gst_buffer_pool_config_set_d3d11_allocation_params (config, params);
gst_d3d11_allocation_params_free (params);
gst_buffer_pool_config_set_params (config, caps,
GST_VIDEO_INFO_SIZE (aligned_info), 0, 0);
gst_buffer_pool_set_config (priv->internal_pool, config);
gst_buffer_pool_set_active (priv->internal_pool, TRUE);
return TRUE;
}
#else
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
static gboolean
gst_qsv_encoder_prepare_va_pool (GstQsvEncoder * self,
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
GstCaps * caps, GstVideoInfo * aligned_info)
{
GstQsvEncoderPrivate *priv = self->priv;
GstAllocator *allocator;
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
GstStructure *config;
GArray *formats;
GstAllocationParams params;
GstVaDisplay *display = GST_VA_DISPLAY (priv->device);
formats = g_array_new (FALSE, FALSE, sizeof (GstVideoFormat));
g_array_append_val (formats, GST_VIDEO_INFO_FORMAT (aligned_info));
allocator = gst_va_allocator_new (display, formats);
if (!allocator) {
GST_ERROR_OBJECT (self, "Failed to create allocator");
return FALSE;
}
gst_allocation_params_init (&params);
priv->internal_pool = gst_va_pool_new_with_config (caps,
GST_VIDEO_INFO_SIZE (aligned_info), 0, 0,
VA_SURFACE_ATTRIB_USAGE_HINT_GENERIC, GST_VA_FEATURE_AUTO,
allocator, &params);
gst_object_unref (allocator);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
if (!priv->internal_pool) {
GST_ERROR_OBJECT (self, "Failed to create va pool");
return FALSE;
}
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
config = gst_buffer_pool_get_config (priv->internal_pool);
gst_buffer_pool_config_add_option (config, GST_BUFFER_POOL_OPTION_VIDEO_META);
gst_buffer_pool_config_set_params (config, caps,
GST_VIDEO_INFO_SIZE (aligned_info), 0, 0);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_buffer_pool_set_config (priv->internal_pool, config);
gst_buffer_pool_set_active (priv->internal_pool, TRUE);
return TRUE;
}
#endif
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
/* Prepare internal pool, which is used to allocate fallback buffer
* when upstream buffer is not directly accessible by QSV */
static gboolean
gst_qsv_encoder_prepare_pool (GstQsvEncoder * self, GstCaps * caps,
GstVideoInfo * aligned_info)
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
{
GstQsvEncoderPrivate *priv = self->priv;
gboolean ret = FALSE;
GstCaps *aligned_caps;
if (priv->internal_pool) {
gst_buffer_pool_set_active (priv->internal_pool, FALSE);
gst_clear_object (&priv->internal_pool);
}
aligned_caps = gst_video_info_to_caps (aligned_info);
#ifdef G_OS_WIN32
ret = gst_qsv_encoder_prepare_d3d11_pool (self, aligned_caps, aligned_info);
#else
ret = gst_qsv_encoder_prepare_va_pool (self, aligned_caps, aligned_info);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
#endif
gst_caps_unref (aligned_caps);
return ret;
}
static gboolean
gst_qsv_encoder_init_encode_session (GstQsvEncoder * self)
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
{
GstQsvEncoderPrivate *priv = self->priv;
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (self);
GstVideoInfo *info = &priv->input_state->info;
GstCaps *caps = priv->input_state->caps;
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
mfxVideoParam param;
mfxFrameInfo *frame_info;
mfxFrameAllocRequest alloc_request;
mfxStatus status;
MFXVideoENCODE *encoder_handle = nullptr;
guint bitstream_size;
gboolean ret;
guint64 min_delay_frames, max_delay_frames;
GstClockTime min_latency, max_latency;
gst_qsv_encoder_drain (self, FALSE);
gst_qsv_encoder_reset (self);
encoder_handle = new MFXVideoENCODE (priv->session);
memset (&param, 0, sizeof (mfxVideoParam));
g_ptr_array_set_size (priv->extra_params, 0);
g_assert (klass->set_format);
if (!klass->set_format (self, priv->input_state, &param, priv->extra_params)) {
GST_ERROR_OBJECT (self, "Subclass failed to set format");
goto error;
}
/* LowPower mode supports smaller set of features, don't enable it for now */
param.mfx.LowPower = MFX_CODINGOPTION_OFF;
if (priv->low_latency)
param.AsyncDepth = 1;
else
param.AsyncDepth = 4;
param.mfx.TargetUsage = priv->target_usage;
frame_info = &param.mfx.FrameInfo;
gst_video_info_set_interlaced_format (&priv->aligned_info,
GST_VIDEO_INFO_FORMAT (info), GST_VIDEO_INFO_INTERLACE_MODE (info),
frame_info->Width, frame_info->Height);
/* Always video memory, even when upstream is non-hardware element */
priv->mem_type = GST_QSV_VIDEO_MEMORY | GST_QSV_ENCODER_IN_MEMORY;
param.IOPattern = MFX_IOPATTERN_IN_VIDEO_MEMORY;
if (!gst_qsv_encoder_prepare_pool (self, caps, &priv->aligned_info)) {
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
GST_ERROR_OBJECT (self, "Failed to prepare pool");
goto error;
}
status = encoder_handle->Query (&param, &param);
/* If device is unhappy with LowPower = OFF, try again with unknown */
if (status < MFX_ERR_NONE) {
GST_INFO_OBJECT (self, "LowPower - OFF returned %d (%s)",
QSV_STATUS_ARGS (status));
param.mfx.LowPower = MFX_CODINGOPTION_UNKNOWN;
}
status = encoder_handle->Query (&param, &param);
QSV_CHECK_STATUS (self, status, MFXVideoENCODE::Query);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
status = encoder_handle->QueryIOSurf (&param, &alloc_request);
QSV_CHECK_STATUS (self, status, MFXVideoENCODE::QueryIOSurf);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
status = encoder_handle->Init (&param);
QSV_CHECK_STATUS (self, status, MFXVideoENCODE::Init);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
status = encoder_handle->GetVideoParam (&param);
QSV_CHECK_STATUS (self, status, MFXVideoENCODE::GetVideoParam);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
GST_DEBUG_OBJECT (self, "NumFrameSuggested: %d, AsyncDepth %d",
alloc_request.NumFrameSuggested, param.AsyncDepth);
g_assert (klass->set_output_state);
ret = klass->set_output_state (self, priv->input_state, priv->session);
if (!ret) {
GST_ERROR_OBJECT (self, "Subclass failed to set output state");
goto error;
}
/* Prepare surface pool with size NumFrameSuggested, then if it's not
* sufficient while encoding, we can increse the pool size dynamically
* if needed */
g_array_set_size (priv->surface_pool, alloc_request.NumFrameSuggested);
for (guint i = 0; i < priv->surface_pool->len; i++) {
GstQsvEncoderSurface *surface = &g_array_index (priv->surface_pool,
GstQsvEncoderSurface, i);
surface->surface.Info = param.mfx.FrameInfo;
surface->payload = g_ptr_array_new_with_free_func ((GDestroyNotify)
gst_qsv_encoder_payload_clear);
}
priv->next_surface_index = 0;
g_array_set_size (priv->task_pool, param.AsyncDepth);
if (klass->codec_id == MFX_CODEC_JPEG) {
gdouble factor = 4.0;
/* jpeg zero returns buffer size */
switch (GST_VIDEO_INFO_FORMAT (info)) {
case GST_VIDEO_FORMAT_NV12:
factor = 1.5;
break;
case GST_VIDEO_FORMAT_YUY2:
factor = 2.0;
break;
default:
break;
}
bitstream_size = (guint)
(factor * GST_VIDEO_INFO_WIDTH (info) * GST_VIDEO_INFO_HEIGHT (info));
} else {
bitstream_size =
(guint) param.mfx.BufferSizeInKB * param.mfx.BRCParamMultiplier * 1024;
}
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
for (guint i = 0; i < priv->task_pool->len; i++) {
GstQsvEncoderTask *task = &g_array_index (priv->task_pool,
GstQsvEncoderTask, i);
task->bitstream.Data = (mfxU8 *) g_malloc (bitstream_size);
task->bitstream.MaxLength = bitstream_size;
g_queue_push_head (&priv->free_tasks, task);
}
min_delay_frames = priv->task_pool->len;
max_delay_frames = priv->surface_pool->len + min_delay_frames;
min_latency = gst_util_uint64_scale (min_delay_frames * GST_SECOND,
param.mfx.FrameInfo.FrameRateExtD, param.mfx.FrameInfo.FrameRateExtN);
max_latency = gst_util_uint64_scale (max_delay_frames * GST_SECOND,
param.mfx.FrameInfo.FrameRateExtD, param.mfx.FrameInfo.FrameRateExtN);
gst_video_encoder_set_latency (GST_VIDEO_ENCODER (self),
min_latency, max_latency);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
priv->video_param = param;
priv->encoder = encoder_handle;
return TRUE;
error:
if (encoder_handle)
delete encoder_handle;
gst_qsv_encoder_reset (self);
return FALSE;
}
static gboolean
gst_qsv_encoder_reset_encode_session (GstQsvEncoder * self)
{
GstQsvEncoderPrivate *priv = self->priv;
GPtrArray *extra_params = priv->extra_params;
mfxStatus status;
mfxExtEncoderResetOption reset_opt;
if (!priv->encoder) {
GST_WARNING_OBJECT (self, "Encoder was not configured");
return gst_qsv_encoder_init_encode_session (self);
}
reset_opt.Header.BufferId = MFX_EXTBUFF_ENCODER_RESET_OPTION;
reset_opt.Header.BufferSz = sizeof (mfxExtEncoderResetOption);
reset_opt.StartNewSequence = MFX_CODINGOPTION_OFF;
gst_qsv_encoder_drain (self, FALSE);
g_ptr_array_add (extra_params, &reset_opt);
priv->video_param.ExtParam = (mfxExtBuffer **) extra_params->pdata;
priv->video_param.NumExtParam = extra_params->len;
status = priv->encoder->Reset (&priv->video_param);
g_ptr_array_remove_index (extra_params, extra_params->len - 1);
priv->video_param.NumExtParam = extra_params->len;
if (status != MFX_ERR_NONE) {
GST_WARNING_OBJECT (self, "MFXVideoENCODE_Reset returned %d (%s)",
QSV_STATUS_ARGS (status));
return gst_qsv_encoder_init_encode_session (self);
}
GST_DEBUG_OBJECT (self, "Encode session reset done");
return TRUE;
}
static gboolean
gst_qsv_encoder_set_format (GstVideoEncoder * encoder,
GstVideoCodecState * state)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
GstQsvEncoderPrivate *priv = self->priv;
g_clear_pointer (&priv->input_state, gst_video_codec_state_unref);
priv->input_state = gst_video_codec_state_ref (state);
return gst_qsv_encoder_init_encode_session (self);
}
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
static mfxU16
gst_qsv_encoder_get_pic_struct (GstQsvEncoder * self,
GstVideoCodecFrame * frame)
{
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (self);
GstQsvEncoderPrivate *priv = self->priv;
GstVideoInfo *info = &priv->input_state->info;
if (klass->codec_id != MFX_CODEC_AVC)
return MFX_PICSTRUCT_PROGRESSIVE;
if (!GST_VIDEO_INFO_IS_INTERLACED (info))
return MFX_PICSTRUCT_PROGRESSIVE;
if (GST_VIDEO_INFO_INTERLACE_MODE (info) == GST_VIDEO_INTERLACE_MODE_MIXED) {
if (!GST_BUFFER_FLAG_IS_SET (frame->input_buffer,
GST_VIDEO_BUFFER_FLAG_INTERLACED)) {
return MFX_PICSTRUCT_PROGRESSIVE;
}
if (GST_BUFFER_FLAG_IS_SET (frame->input_buffer, GST_VIDEO_BUFFER_FLAG_TFF))
return MFX_PICSTRUCT_FIELD_TFF;
return MFX_PICSTRUCT_FIELD_BFF;
}
switch (GST_VIDEO_INFO_FIELD_ORDER (info)) {
case GST_VIDEO_FIELD_ORDER_TOP_FIELD_FIRST:
return MFX_PICSTRUCT_FIELD_TFF;
break;
case GST_VIDEO_FIELD_ORDER_BOTTOM_FIELD_FIRST:
return MFX_PICSTRUCT_FIELD_BFF;
break;
default:
break;
}
if (GST_BUFFER_FLAG_IS_SET (frame->input_buffer, GST_VIDEO_BUFFER_FLAG_TFF))
return MFX_PICSTRUCT_FIELD_TFF;
return MFX_PICSTRUCT_FIELD_BFF;
}
static GstFlowReturn
gst_qsv_encoder_handle_frame (GstVideoEncoder * encoder,
GstVideoCodecFrame * frame)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
GstQsvEncoderPrivate *priv = self->priv;
GstQsvEncoderClass *klass = GST_QSV_ENCODER_GET_CLASS (self);
GstFlowReturn ret = GST_FLOW_ERROR;
GstQsvEncoderSurface *surface;
GstQsvEncoderTask *task;
mfxU64 timestamp;
mfxStatus status;
if (klass->check_reconfigure && priv->encoder) {
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
GstQsvEncoderReconfigure reconfigure;
reconfigure = klass->check_reconfigure (self, priv->session,
&priv->video_param, priv->extra_params);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
switch (reconfigure) {
case GST_QSV_ENCODER_RECONFIGURE_BITRATE:
if (!gst_qsv_encoder_reset_encode_session (self)) {
GST_ERROR_OBJECT (self, "Failed to reset session");
gst_video_encoder_finish_frame (encoder, frame);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
return GST_FLOW_ERROR;
}
break;
case GST_QSV_ENCODER_RECONFIGURE_FULL:
if (!gst_qsv_encoder_init_encode_session (self)) {
GST_ERROR_OBJECT (self, "Failed to init session");
gst_video_encoder_finish_frame (encoder, frame);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
return GST_FLOW_ERROR;
}
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
break;
default:
break;
}
}
if (!priv->encoder) {
GST_ERROR_OBJECT (self, "Encoder object was not configured");
gst_video_encoder_finish_frame (encoder, frame);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
return GST_FLOW_NOT_NEGOTIATED;
}
surface = gst_qsv_encoder_get_next_surface (self);
if (!surface) {
GST_ERROR_OBJECT (self, "No available surface");
goto out;
}
task = (GstQsvEncoderTask *) g_queue_pop_tail (&priv->free_tasks);
g_assert (task);
surface->qsv_frame =
gst_qsv_allocator_acquire_frame (priv->allocator, priv->mem_type,
&priv->input_state->info, gst_buffer_ref (frame->input_buffer),
priv->internal_pool);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
if (!surface->qsv_frame) {
GST_ERROR_OBJECT (self, "Failed to wrap buffer with qsv frame");
gst_qsv_encoder_task_reset (self, task);
goto out;
}
surface->surface.Info.PicStruct =
gst_qsv_encoder_get_pic_struct (self, frame);
if (GST_VIDEO_CODEC_FRAME_IS_FORCE_KEYFRAME (frame)) {
surface->encode_control.FrameType =
MFX_FRAMETYPE_IDR | MFX_FRAMETYPE_I | MFX_FRAMETYPE_REF;
} else {
surface->encode_control.FrameType = MFX_FRAMETYPE_UNKNOWN;
}
if (klass->attach_payload) {
klass->attach_payload (self, frame, surface->payload);
if (surface->payload->len > 0) {
surface->encode_control.NumPayload = surface->payload->len;
surface->encode_control.Payload = (mfxPayload **) surface->payload->pdata;
}
}
timestamp = gst_qsv_timestamp_from_gst (frame->pts);
status = gst_qsv_encoder_encode_frame (self, surface, task, timestamp);
if (status != MFX_ERR_NONE && status != MFX_ERR_MORE_DATA) {
GST_ERROR_OBJECT (self, "Failed to encode frame, ret %d (%s)",
QSV_STATUS_ARGS (status));
gst_qsv_encoder_task_reset (self, task);
goto out;
}
if (status == MFX_ERR_NONE && task->sync_point) {
g_queue_push_head (&priv->pending_tasks, task);
} else {
gst_qsv_encoder_task_reset (self, task);
}
ret = GST_FLOW_OK;
/* Do not sync immediately, but record tasks which have output buffer here
* to improve throughput.
* In this way, hardware may be able to run encoding job from its background
* threads (if any). We will do sync only when there's no more free task item
*/
while (g_queue_get_length (&priv->pending_tasks) >= priv->task_pool->len) {
GstQsvEncoderTask *task =
(GstQsvEncoderTask *) g_queue_pop_tail (&priv->pending_tasks);
ret = gst_qsv_encoder_finish_frame (self, task, FALSE);
}
out:
gst_video_codec_frame_unref (frame);
return ret;
}
static GstFlowReturn
gst_qsv_encoder_finish (GstVideoEncoder * encoder)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
return gst_qsv_encoder_drain (self, FALSE);
}
static gboolean
gst_qsv_encoder_flush (GstVideoEncoder * encoder)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
gst_qsv_encoder_drain (self, TRUE);
return TRUE;
}
static gboolean
gst_qsv_encoder_handle_context_query (GstQsvEncoder * self, GstQuery * query)
{
GstQsvEncoderPrivate *priv = self->priv;
#ifdef G_OS_WIN32
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
return gst_d3d11_handle_context_query (GST_ELEMENT (self), query,
(GstD3D11Device *) priv->device);
#else
return gst_va_handle_context_query (GST_ELEMENT (self), query,
(GstVaDisplay *) priv->device);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
#endif
}
static gboolean
gst_qsv_encoder_sink_query (GstVideoEncoder * encoder, GstQuery * query)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
switch (GST_QUERY_TYPE (query)) {
case GST_QUERY_CONTEXT:
if (gst_qsv_encoder_handle_context_query (self, query))
return TRUE;
break;
default:
break;
}
return GST_VIDEO_ENCODER_CLASS (parent_class)->sink_query (encoder, query);
}
static gboolean
gst_qsv_encoder_src_query (GstVideoEncoder * encoder, GstQuery * query)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
switch (GST_QUERY_TYPE (query)) {
case GST_QUERY_CONTEXT:
if (gst_qsv_encoder_handle_context_query (self, query))
return TRUE;
break;
default:
break;
}
return GST_VIDEO_ENCODER_CLASS (parent_class)->src_query (encoder, query);
}
#ifdef G_OS_WIN32
static gboolean
gst_qsv_encoder_propose_allocation (GstVideoEncoder * encoder, GstQuery * query)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
GstQsvEncoderPrivate *priv = self->priv;
GstD3D11Device *device = GST_D3D11_DEVICE (priv->device);
GstVideoInfo info;
GstBufferPool *pool;
GstCaps *caps;
guint size;
GstStructure *config;
GstCapsFeatures *features;
gboolean is_d3d11 = FALSE;
gst_query_parse_allocation (query, &caps, nullptr);
if (!caps) {
GST_WARNING_OBJECT (self, "null caps in query");
return FALSE;
}
if (!gst_video_info_from_caps (&info, caps)) {
GST_WARNING_OBJECT (self, "Failed to convert caps into info");
return FALSE;
}
features = gst_caps_get_features (caps, 0);
if (features && gst_caps_features_contains (features,
GST_CAPS_FEATURE_MEMORY_D3D11_MEMORY)) {
GST_DEBUG_OBJECT (self, "upstream support d3d11 memory");
pool = gst_d3d11_buffer_pool_new (device);
is_d3d11 = TRUE;
} else {
pool = gst_d3d11_staging_buffer_pool_new (device);
}
config = gst_buffer_pool_get_config (pool);
gst_buffer_pool_config_add_option (config, GST_BUFFER_POOL_OPTION_VIDEO_META);
if (is_d3d11) {
GstD3D11AllocationParams *d3d11_params;
GstVideoAlignment align;
/* d3d11 buffer pool doesn't support generic video alignment
* because memory layout of CPU accessible staging texture is uncontrollable.
* Do D3D11 specific handling */
gst_video_alignment_reset (&align);
align.padding_right = GST_VIDEO_INFO_WIDTH (&priv->aligned_info) -
GST_VIDEO_INFO_WIDTH (&info);
align.padding_bottom = GST_VIDEO_INFO_HEIGHT (&priv->aligned_info) -
GST_VIDEO_INFO_HEIGHT (&info);
d3d11_params = gst_d3d11_allocation_params_new (device, &info,
(GstD3D11AllocationFlags) 0, 0);
gst_d3d11_allocation_params_alignment (d3d11_params, &align);
gst_buffer_pool_config_set_d3d11_allocation_params (config, d3d11_params);
gst_d3d11_allocation_params_free (d3d11_params);
}
size = GST_VIDEO_INFO_SIZE (&info);
gst_buffer_pool_config_set_params (config,
caps, size, priv->surface_pool->len, 0);
if (!gst_buffer_pool_set_config (pool, config)) {
GST_WARNING_OBJECT (self, "Failed to set pool config");
gst_object_unref (pool);
return FALSE;
}
/* d3d11 buffer pool will update actual CPU accessible buffer size based on
* allocated staging texture per gst_buffer_pool_set_config() call,
* need query again to get the size */
config = gst_buffer_pool_get_config (pool);
gst_buffer_pool_config_get_params (config, nullptr, &size, nullptr, nullptr);
gst_structure_free (config);
gst_query_add_allocation_pool (query, pool, size, priv->surface_pool->len, 0);
gst_query_add_allocation_meta (query, GST_VIDEO_META_API_TYPE, nullptr);
gst_object_unref (pool);
return TRUE;
}
#else
static gboolean
gst_qsv_encoder_propose_allocation (GstVideoEncoder * encoder, GstQuery * query)
{
GstQsvEncoder *self = GST_QSV_ENCODER (encoder);
GstQsvEncoderPrivate *priv = self->priv;
GstVideoInfo info;
GstAllocator *allocator = nullptr;
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
GstBufferPool *pool;
GstCaps *caps;
guint size;
GstStructure *config;
GstVideoAlignment align;
GstAllocationParams params;
GArray *formats;
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_query_parse_allocation (query, &caps, nullptr);
if (!caps) {
GST_WARNING_OBJECT (self, "null caps in query");
return FALSE;
}
if (!gst_video_info_from_caps (&info, caps)) {
GST_WARNING_OBJECT (self, "Failed to convert caps into info");
return FALSE;
}
gst_allocation_params_init (&params);
formats = g_array_new (FALSE, FALSE, sizeof (GstVideoFormat));
g_array_append_val (formats, GST_VIDEO_INFO_FORMAT (&info));
allocator = gst_va_allocator_new (GST_VA_DISPLAY (priv->device), formats);
if (!allocator) {
GST_ERROR_OBJECT (self, "Failed to create allocator");
return FALSE;
}
pool = gst_va_pool_new_with_config (caps,
GST_VIDEO_INFO_SIZE (&info), priv->surface_pool->len, 0,
VA_SURFACE_ATTRIB_USAGE_HINT_GENERIC, GST_VA_FEATURE_AUTO,
allocator, &params);
if (!pool) {
GST_ERROR_OBJECT (self, "Failed to create va pool");
gst_object_unref (allocator);
return FALSE;
}
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_video_alignment_reset (&align);
align.padding_right = GST_VIDEO_INFO_WIDTH (&priv->aligned_info) -
GST_VIDEO_INFO_WIDTH (&info);
align.padding_bottom = GST_VIDEO_INFO_HEIGHT (&priv->aligned_info) -
GST_VIDEO_INFO_HEIGHT (&info);
config = gst_buffer_pool_get_config (pool);
gst_buffer_pool_config_add_option (config, GST_BUFFER_POOL_OPTION_VIDEO_META);
gst_buffer_pool_config_add_option (config,
GST_BUFFER_POOL_OPTION_VIDEO_ALIGNMENT);
gst_buffer_pool_config_set_video_alignment (config, &align);
gst_buffer_pool_config_set_params (config,
caps, GST_VIDEO_INFO_SIZE (&info), priv->surface_pool->len, 0);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
if (!gst_buffer_pool_set_config (pool, config)) {
GST_ERROR_OBJECT (self, "Failed to set pool config");
gst_clear_object (&allocator);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_object_unref (pool);
return FALSE;
}
if (allocator)
gst_query_add_allocation_param (query, allocator, &params);
config = gst_buffer_pool_get_config (pool);
gst_buffer_pool_config_get_params (config, nullptr, &size, nullptr, nullptr);
gst_structure_free (config);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_query_add_allocation_pool (query, pool, size, priv->surface_pool->len, 0);
gst_query_add_allocation_meta (query, GST_VIDEO_META_API_TYPE, nullptr);
gst_clear_object (&allocator);
qsv: Introduce H.264 Intel Quick Sync Video Encoder A new implementation of Intel Quick Sync Video plugin. This plugin supports both Windows and Linux but optimization for VA/DMABuf is not implemented yet. This new plugin has some notable differences compared with existing MSDK plugin. * Encoder will expose formats which can be natively supported without internal conversion. This will make encoder control/negotiation flow much simpler and cleaner than that of MSDK plugin. * This plugin includes QSV specific library loading helper, called dispatcher, with QSV SDK headers as a part of this plugin. So, there will be no more SDK version dependent #ifdef in the code and also there will be no more build-time MSDK/oneVPL SDK dependency. * Memory allocator interop between GStreamer and QSV is re-designed and decoupled. Instead of implementing QSV specific allocator/bufferpool, this plugin will make use of generic GStreamer memory allocator/bufferpool (e.g., GstD3D11Allocator and GstD3D11BufferPool). Specifically, GstQsvAllocator object will help interop between GstMemory and mfxFrameAllocator memory abstraction layers. Note that because of the design decision, VA/DMABuf support is not made as a part of this initial commit. We can add the optimization for Linux later once GstVA library exposes allocator/bufferpool implementation as an API like GstD3D11. * Initial encoder implementation supports interop with GstD3D11 infrastructure, including zero-copy encoding with upstream D3D11 element. Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1408>
2022-02-06 12:34:43 +00:00
gst_object_unref (pool);
return TRUE;
}
#endif