gstreamer/subprojects/gst-plugins-bad/sys/qsv/gstqsvallocator_d3d11.cpp

526 lines
16 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 "gstqsvallocator_d3d11.h"
#include <string.h>
GST_DEBUG_CATEGORY_EXTERN (gst_qsv_allocator_debug);
#define GST_CAT_DEFAULT gst_qsv_allocator_debug
struct _GstQsvD3D11Allocator
{
GstQsvAllocator parent;
GstD3D11Device *device;
};
#define gst_qsv_d3d11_allocator_parent_class parent_class
G_DEFINE_TYPE (GstQsvD3D11Allocator, gst_qsv_d3d11_allocator,
GST_TYPE_QSV_ALLOCATOR);
static void gst_qsv_d3d11_allocator_dispose (GObject * object);
static mfxStatus gst_qsv_d3d11_allocator_alloc (GstQsvAllocator * allocator,
gboolean dummy_alloc, mfxFrameAllocRequest * request,
mfxFrameAllocResponse * response);
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 GstBuffer *gst_qsv_d3d11_allocator_upload (GstQsvAllocator * allocator,
const GstVideoInfo * info, GstBuffer * buffer, GstBufferPool * pool);
static GstBuffer *gst_qsv_d3d11_allocator_download (GstQsvAllocator * allocator,
const GstVideoInfo * info, gboolean force_copy, GstQsvFrame * frame,
GstBufferPool * 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
static void
gst_qsv_d3d11_allocator_class_init (GstQsvD3D11AllocatorClass * klass)
{
GObjectClass *object_class = G_OBJECT_CLASS (klass);
GstQsvAllocatorClass *alloc_class = GST_QSV_ALLOCATOR_CLASS (klass);
object_class->dispose = gst_qsv_d3d11_allocator_dispose;
alloc_class->alloc = GST_DEBUG_FUNCPTR (gst_qsv_d3d11_allocator_alloc);
alloc_class->upload = GST_DEBUG_FUNCPTR (gst_qsv_d3d11_allocator_upload);
alloc_class->download = GST_DEBUG_FUNCPTR (gst_qsv_d3d11_allocator_download);
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 void
gst_qsv_d3d11_allocator_init (GstQsvD3D11Allocator * self)
{
}
static void
gst_qsv_d3d11_allocator_dispose (GObject * object)
{
GstQsvD3D11Allocator *self = GST_QSV_D3D11_ALLOCATOR (object);
gst_clear_object (&self->device);
G_OBJECT_CLASS (parent_class)->dispose (object);
}
static mfxStatus
gst_qsv_d3d11_allocator_alloc (GstQsvAllocator * allocator,
gboolean dummy_alloc, mfxFrameAllocRequest * request,
mfxFrameAllocResponse * response)
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
{
GstQsvD3D11Allocator *self = GST_QSV_D3D11_ALLOCATOR (allocator);
DXGI_FORMAT dxgi_format = DXGI_FORMAT_UNKNOWN;
GstQsvFrame **mids = nullptr;
/* Something unexpected and went wrong */
if ((request->Type & MFX_MEMTYPE_SYSTEM_MEMORY) != 0) {
GST_ERROR_OBJECT (self,
"MFX is requesting system memory, type 0x%x", request->Type);
return MFX_ERR_UNSUPPORTED;
}
switch (request->Info.FourCC) {
case MFX_FOURCC_NV12:
dxgi_format = DXGI_FORMAT_NV12;
break;
case MFX_FOURCC_P010:
dxgi_format = DXGI_FORMAT_P010;
break;
case MFX_FOURCC_AYUV:
dxgi_format = DXGI_FORMAT_AYUV;
break;
case MFX_FOURCC_Y410:
dxgi_format = DXGI_FORMAT_Y410;
break;
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
default:
/* TODO: add more formats */
break;
}
if (dxgi_format == DXGI_FORMAT_UNKNOWN &&
request->Info.FourCC != MFX_FOURCC_P8) {
GST_ERROR_OBJECT (self, "Failed to convert %d to DXGI format",
request->Info.FourCC);
return MFX_ERR_UNSUPPORTED;
}
if (request->Info.FourCC == MFX_FOURCC_P8) {
GstD3D11Allocator *d3d11_alloc = nullptr;
D3D11_BUFFER_DESC desc;
GstVideoInfo info;
GstMemory *mem;
GstBuffer *buffer;
gsize offset[GST_VIDEO_MAX_PLANES] = { 0, };
gint stride[GST_VIDEO_MAX_PLANES] = { 0, };
guint size;
d3d11_alloc =
(GstD3D11Allocator *) gst_allocator_find (GST_D3D11_MEMORY_NAME);
if (!d3d11_alloc) {
GST_ERROR_OBJECT (self, "D3D11 allocator is unavailable");
return MFX_ERR_MEMORY_ALLOC;
}
memset (&desc, 0, sizeof (D3D11_BUFFER_DESC));
desc.ByteWidth = request->Info.Width * request->Info.Height;
desc.Usage = D3D11_USAGE_STAGING;
desc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
mem = gst_d3d11_allocator_alloc_buffer (d3d11_alloc, self->device, &desc);
gst_object_unref (d3d11_alloc);
if (!mem) {
GST_ERROR_OBJECT (self, "Failed to allocate buffer");
return MFX_ERR_MEMORY_ALLOC;
}
size = request->Info.Width * request->Info.Height;
stride[0] = size;
gst_video_info_set_format (&info, GST_VIDEO_FORMAT_GRAY8, size, 1);
buffer = gst_buffer_new ();
gst_buffer_append_memory (buffer, mem);
gst_buffer_add_video_meta_full (buffer, GST_VIDEO_FRAME_FLAG_NONE,
GST_VIDEO_FORMAT_GRAY8, size, 1, 1, offset, stride);
mids = g_new0 (GstQsvFrame *, 1);
response->NumFrameActual = 1;
mids[0] = gst_qsv_allocator_acquire_frame (allocator,
GST_QSV_VIDEO_MEMORY | GST_QSV_ENCODER_IN_MEMORY, &info, buffer,
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
} else {
GstBufferPool *pool;
GstVideoFormat format;
GstVideoInfo info;
GstCaps *caps;
GstStructure *config;
GstD3D11AllocationParams *params;
guint bind_flags = 0;
GstVideoAlignment align;
GstQsvMemoryType mem_type = GST_QSV_VIDEO_MEMORY;
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 ((request->Type & MFX_MEMTYPE_VIDEO_MEMORY_ENCODER_TARGET) != 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
bind_flags |= D3D11_BIND_VIDEO_ENCODER;
mem_type |= GST_QSV_ENCODER_IN_MEMORY;
}
if ((request->Type & MFX_MEMTYPE_VIDEO_MEMORY_DECODER_TARGET) != 0) {
bind_flags |= D3D11_BIND_DECODER;
mem_type |= GST_QSV_DECODER_OUT_MEMORY;
}
if (mem_type == GST_QSV_VIDEO_MEMORY) {
GST_ERROR_OBJECT (self, "Unknown read/write access");
return MFX_ERR_UNSUPPORTED;
}
mids = g_new0 (GstQsvFrame *, request->NumFrameSuggested);
response->NumFrameActual = request->NumFrameSuggested;
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
format = gst_d3d11_dxgi_format_to_gst (dxgi_format);
gst_video_info_set_format (&info,
format, request->Info.CropW, request->Info.CropH);
if (dummy_alloc) {
for (guint i = 0; i < request->NumFrameSuggested; i++) {
mids[i] = gst_qsv_allocator_acquire_frame (allocator,
mem_type, &info, nullptr, nullptr);
}
response->mids = (mfxMemId *) mids;
return MFX_ERR_NONE;
}
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
caps = gst_video_info_to_caps (&info);
gst_video_alignment_reset (&align);
align.padding_right = request->Info.Width - request->Info.CropW;
align.padding_bottom = request->Info.Height - request->Info.CropH;
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
pool = gst_d3d11_buffer_pool_new (self->device);
params = gst_d3d11_allocation_params_new (self->device, &info,
(GstD3D11AllocationFlags) 0, bind_flags);
gst_d3d11_allocation_params_alignment (params, &align);
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 (pool);
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 (&info), 0, 0);
gst_caps_unref (caps);
gst_buffer_pool_set_config (pool, config);
gst_buffer_pool_set_active (pool, TRUE);
for (guint i = 0; i < request->NumFrameSuggested; i++) {
GstBuffer *buffer;
if (gst_buffer_pool_acquire_buffer (pool, &buffer, nullptr) !=
GST_FLOW_OK) {
GST_ERROR_OBJECT (self, "Failed to allocate texture buffer");
gst_buffer_pool_set_active (pool, FALSE);
gst_object_unref (pool);
goto error;
}
mids[i] = gst_qsv_allocator_acquire_frame (allocator,
mem_type, &info, buffer, 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
}
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_active (pool, FALSE);
gst_object_unref (pool);
}
response->mids = (mfxMemId *) mids;
return MFX_ERR_NONE;
error:
if (mids) {
for (guint i = 0; i < response->NumFrameActual; i++)
gst_clear_qsv_frame (&mids[i]);
g_free (mids);
}
response->NumFrameActual = 0;
return MFX_ERR_MEMORY_ALLOC;
}
static GstBuffer *
gst_qsv_frame_copy_d3d11 (const GstVideoInfo * info, GstBuffer * src_buf,
GstBuffer * dst_buf)
{
D3D11_TEXTURE2D_DESC src_desc, dst_desc;
D3D11_BOX src_box;
guint subresource_idx;
GstMemory *src_mem, *dst_mem;
GstMapInfo src_info, dst_info;
ID3D11Texture2D *src_tex, *dst_tex;
GstD3D11Device *device;
ID3D11DeviceContext *device_context;
GST_TRACE ("Copying D3D11 buffer %" GST_PTR_FORMAT, src_buf);
src_mem = gst_buffer_peek_memory (src_buf, 0);
dst_mem = gst_buffer_peek_memory (dst_buf, 0);
device = GST_D3D11_MEMORY_CAST (dst_mem)->device;
device_context = gst_d3d11_device_get_device_context_handle (device);
if (!gst_memory_map (src_mem,
&src_info, (GstMapFlags) (GST_MAP_READ | GST_MAP_D3D11))) {
GST_WARNING ("Failed to map src memory");
gst_buffer_unref (dst_buf);
return nullptr;
}
if (!gst_memory_map (dst_mem,
&dst_info, (GstMapFlags) (GST_MAP_WRITE | GST_MAP_D3D11))) {
GST_WARNING ("Failed to map dst memory");
gst_memory_unmap (src_mem, &src_info);
gst_buffer_unref (dst_buf);
return nullptr;
}
src_tex = (ID3D11Texture2D *) src_info.data;
dst_tex = (ID3D11Texture2D *) dst_info.data;
src_tex->GetDesc (&src_desc);
dst_tex->GetDesc (&dst_desc);
subresource_idx =
gst_d3d11_memory_get_subresource_index (GST_D3D11_MEMORY_CAST (src_mem));
src_box.left = 0;
src_box.top = 0;
src_box.front = 0;
src_box.back = 1;
src_box.right = MIN (src_desc.Width, dst_desc.Width);
src_box.bottom = MIN (src_desc.Height, dst_desc.Height);
gst_d3d11_device_lock (device);
device_context->CopySubresourceRegion (dst_tex, 0,
0, 0, 0, src_tex, subresource_idx, &src_box);
gst_d3d11_device_unlock (device);
gst_memory_unmap (dst_mem, &dst_info);
gst_memory_unmap (src_mem, &src_info);
return dst_buf;
}
static GstBuffer *
gst_qsv_frame_upload_sysmem (const GstVideoInfo * info, GstBuffer * src_buf,
GstBuffer * dst_buf)
{
GstVideoFrame src_frame, dst_frame;
GST_TRACE ("Uploading sysmem buffer %" GST_PTR_FORMAT, src_buf);
if (!gst_video_frame_map (&src_frame, info, src_buf, GST_MAP_READ)) {
GST_WARNING ("Failed to map src frame");
gst_buffer_unref (dst_buf);
return nullptr;
}
if (!gst_video_frame_map (&dst_frame, info, dst_buf, GST_MAP_WRITE)) {
GST_WARNING ("Failed to map src frame");
gst_video_frame_unmap (&src_frame);
gst_buffer_unref (dst_buf);
return nullptr;
}
for (guint i = 0; i < GST_VIDEO_FRAME_N_PLANES (&src_frame); i++) {
guint src_width_in_bytes, src_height;
guint dst_width_in_bytes, dst_height;
guint width_in_bytes, height;
guint src_stride, dst_stride;
guint8 *src_data, *dst_data;
src_width_in_bytes = GST_VIDEO_FRAME_COMP_WIDTH (&src_frame, i) *
GST_VIDEO_FRAME_COMP_PSTRIDE (&src_frame, i);
src_height = GST_VIDEO_FRAME_COMP_HEIGHT (&src_frame, i);
src_stride = GST_VIDEO_FRAME_COMP_STRIDE (&src_frame, i);
dst_width_in_bytes = GST_VIDEO_FRAME_COMP_WIDTH (&dst_frame, i) *
GST_VIDEO_FRAME_COMP_PSTRIDE (&src_frame, i);
dst_height = GST_VIDEO_FRAME_COMP_HEIGHT (&src_frame, i);
dst_stride = GST_VIDEO_FRAME_COMP_STRIDE (&dst_frame, i);
width_in_bytes = MIN (src_width_in_bytes, dst_width_in_bytes);
height = MIN (src_height, dst_height);
src_data = (guint8 *) GST_VIDEO_FRAME_PLANE_DATA (&src_frame, i);
dst_data = (guint8 *) GST_VIDEO_FRAME_PLANE_DATA (&dst_frame, i);
for (guint j = 0; j < height; j++) {
memcpy (dst_data, src_data, width_in_bytes);
dst_data += dst_stride;
src_data += src_stride;
}
}
gst_video_frame_unmap (&dst_frame);
gst_video_frame_unmap (&src_frame);
return dst_buf;
}
static GstBuffer *
gst_qsv_d3d11_allocator_upload (GstQsvAllocator * allocator,
const GstVideoInfo * info, GstBuffer * buffer, GstBufferPool * pool)
{
GstMemory *mem;
GstD3D11Memory *dmem, *dst_dmem;
D3D11_TEXTURE2D_DESC desc, dst_desc;
GstBuffer *dst_buf;
GstFlowReturn flow_ret;
/* 1) D3D11 buffer from the same d3d11device with ours
* 1-1) Same resolution
* -> Increase refcount and wrap with GstQsvFrame
* 1-2) Different resolution
* -> GPU copy
* 2) non-D3D11 buffer or from other d3d11 device
* -> Always CPU copy
*/
if (!GST_IS_D3D11_BUFFER_POOL (pool)) {
GST_ERROR_OBJECT (allocator, "Not a d3d11 buffer pool");
return nullptr;
}
flow_ret = gst_buffer_pool_acquire_buffer (pool, &dst_buf, nullptr);
if (flow_ret != GST_FLOW_OK) {
GST_WARNING ("Failed to acquire buffer from pool, return %s",
gst_flow_get_name (flow_ret));
return nullptr;
}
mem = gst_buffer_peek_memory (buffer, 0);
if (!gst_is_d3d11_memory (mem) || gst_buffer_n_memory (buffer) > 1) {
/* d3d11 buffer should hold single memory object */
return gst_qsv_frame_upload_sysmem (info, buffer, dst_buf);
}
/* FIXME: Add support for shared texture for GPU copy or wrapping
* texture from different device */
dmem = GST_D3D11_MEMORY_CAST (mem);
if (dmem->device != GST_D3D11_BUFFER_POOL (pool)->device)
return gst_qsv_frame_upload_sysmem (info, buffer, dst_buf);
dst_dmem = (GstD3D11Memory *) gst_buffer_peek_memory (dst_buf, 0);
gst_d3d11_memory_get_texture_desc (dmem, &desc);
gst_d3d11_memory_get_texture_desc (dst_dmem, &dst_desc);
if (desc.Width == dst_desc.Width && desc.Height == dst_desc.Height &&
desc.Usage == D3D11_USAGE_DEFAULT) {
/* Identical size and non-staging texture, wrap without copying */
GST_TRACE ("Wrapping D3D11 buffer without copy");
gst_buffer_unref (dst_buf);
return gst_buffer_ref (buffer);
}
return gst_qsv_frame_copy_d3d11 (info, buffer, dst_buf);
}
static GstBuffer *
gst_qsv_d3d11_allocator_download (GstQsvAllocator * allocator,
const GstVideoInfo * info, gboolean force_copy, GstQsvFrame * frame,
GstBufferPool * pool)
{
GstBuffer *src_buf, *dst_buf;
GstMemory *mem;
GstD3D11Memory *dmem;
GstFlowReturn ret;
GST_TRACE_OBJECT (allocator, "Download");
src_buf = gst_qsv_frame_peek_buffer (frame);
if (!force_copy)
return gst_buffer_ref (src_buf);
mem = gst_buffer_peek_memory (src_buf, 0);
if (!gst_is_d3d11_memory (mem) || gst_buffer_n_memory (src_buf) != 1) {
GST_ERROR_OBJECT (allocator, "frame holds invalid d3d11 memory");
return nullptr;
}
if (!GST_IS_D3D11_BUFFER_POOL (pool) &&
!GST_IS_D3D11_STAGING_BUFFER_POOL (pool)) {
GST_TRACE_OBJECT (allocator, "Output is not d3d11 memory");
goto fallback;
}
dmem = GST_D3D11_MEMORY_CAST (mem);
/* both pool and qsvframe should hold the same d3d11 device already */
if (GST_IS_D3D11_BUFFER_POOL (pool)) {
GstD3D11BufferPool *d3d11_pool = GST_D3D11_BUFFER_POOL (pool);
if (d3d11_pool->device != dmem->device) {
GST_WARNING_OBJECT (allocator, "Pool holds different device");
goto fallback;
}
} else {
GstD3D11StagingBufferPool *d3d11_pool =
GST_D3D11_STAGING_BUFFER_POOL (pool);
if (d3d11_pool->device != dmem->device) {
GST_WARNING_OBJECT (allocator, "Staging pool holds different device");
goto fallback;
}
}
ret = gst_buffer_pool_acquire_buffer (pool, &dst_buf, nullptr);
if (ret != GST_FLOW_OK) {
GST_WARNING_OBJECT (allocator, "Failed to allocate output buffer");
return nullptr;
}
return gst_qsv_frame_copy_d3d11 (info, src_buf, dst_buf);
fallback:
GST_MINI_OBJECT_FLAG_SET (mem, GST_D3D11_MEMORY_TRANSFER_NEED_DOWNLOAD);
return GST_QSV_ALLOCATOR_CLASS (parent_class)->download (allocator,
info, TRUE, frame, 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
GstQsvAllocator *
gst_qsv_d3d11_allocator_new (GstD3D11Device * device)
{
GstQsvD3D11Allocator *self;
g_return_val_if_fail (GST_IS_D3D11_DEVICE (device), nullptr);
self = (GstQsvD3D11Allocator *)
g_object_new (GST_TYPE_QSV_D3D11_ALLOCATOR, nullptr);
self->device = (GstD3D11Device *) gst_object_ref (device);
gst_object_ref_sink (self);
return GST_QSV_ALLOCATOR (self);
}