From patchwork Tue Dec 5 06:46:43 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44904 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145593pzc; Mon, 4 Dec 2023 22:51:38 -0800 (PST) X-Google-Smtp-Source: AGHT+IF7Lpo4BJogzWMWyOj9rFe8Qy8obk4CYK+C4KfOTaFtzjmtK/5MJ7AkKXgcRtZhcx/3uov3 X-Received: by 2002:a05:6512:48d5:b0:50b:f1b6:6337 with SMTP id er21-20020a05651248d500b0050bf1b66337mr1266517lfb.25.1701759098366; Mon, 04 Dec 2023 22:51:38 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759098; cv=none; d=google.com; s=arc-20160816; b=sBoOAuqYYw/OXfdE0hxUpBEfsj5ABf1fvKhc8icVG2IMZFraIOX73J/00MNdGm4hay OIDSd4o9nSidbiA0jrSQJcMnZ3usG8/Rf7g4jtr/Msj+InXlb66tptQuOuJ3/rBw67fO sIJ2ZyDNYXKzNhCHF/mb0ba9nXiaJvQrIETPe80dTjtVIh4YE32RaUU6FOhHp82S954N X240wINB/vxEdRl5P26GiNuElMpWzzn3VWhpvMhG669I5i871zM4vZrOrkZyC+ABQx4t U+Ob8fSOqFH9TmYzLjfnyqxz5+2w6rTwTdkCBd40bf3SXVKw29qdIUycSuZxlrBzKyRH JGjQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:message-id:date:to:from :dkim-signature:delivered-to; bh=+X9izM3Yq3Om36AjjAAn8De3OrmzUXJO2RLPo58Cq8w=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=UIppEVcPlsBCMVRBIbCm8z9laQcfqiEQWCw1sqhdcIs6/gp6q+JMJQHzJo4KiOxlkY ghQn7JfMCT8bDfCYTln36zboEOOb52K5LADoyUYbdWpiwnvtOT1hQmDpGkFlYMEFXZ9q mY5DLJLJCr8/94Dv3+HpfGW6bLw6CB3v5C+44uwNpk9NhxSwBEhxzFDhEUBAp/HbdXek tG+vi9U9NYyIOs2Flqk4Kc/hHpbkJuqWas21qXJtKz+zSiBPtMfW6BVFWe5DC1iZKcY1 zA8iX5nQSU0Wmf2U/13Wjeudd2FccesOUX7bpR1NVy5kZYySggr34ZXZO8UDMwN+M/4s sDqA== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b="X/B25wB0"; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id k23-20020a508ad7000000b0054cb7aa7101si567026edk.184.2023.12.04.22.51.37; Mon, 04 Dec 2023 22:51:38 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b="X/B25wB0"; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id D905068CD40; Tue, 5 Dec 2023 08:51:32 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 7D9C868CBFF for ; Tue, 5 Dec 2023 08:51:25 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759090; x=1733295090; h=from:to:cc:subject:date:message-id:mime-version: content-transfer-encoding; bh=o1kCIsdlfCBRSzmqyPnjc/glpClPknzCIsTKwOJfEfc=; b=X/B25wB0KGpHuwL7GU0P7NYgT8ri7OOZYQXZ+M/DSKspHHyqTFDEfRZ2 K50zbzNA/awIe5H//LoF6mKRMr2pyrIr/QwX7Q1i9e4t4g8j9A5Njnnc8 auAZhs7E9w5xqd+Jhx4TcZlsyUr18A2mDVyiq5DlPxX3CMwYrP8wdZbK8 svqrz2WrMIXSW9oWnldvtNnMqEsf/ZwgpRRBfdEdYkUNb/yV3Keal6+MJ 8l6C7XPlGKKeFMAbgy3fe+VrCFbvRBnDzy90bBlaF7YWWiNaBeuXOvwQe /ydGXZb/LFRs5Mx+9xeb+z2eQh3ZcglRVm/BoZklMKKFHaBO1J/mdUWf5 g==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012381" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012381" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:21 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853577" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853577" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:19 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:43 +0800 Message-ID: <20231205065034.1621-1-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 1/9] libavutil: add hwcontext_d3d12va and AV_PIX_FMT_D3D12 X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: swKkLJyBhyAz From: Wu Jianhua Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- Compared to v11, v12 set the initial value to 0 for the fence value, fixing a potential dynamic pool sync issue. Hence removed the non zero initial_pool_size for d3d12va decoder. Some other minor error message changes. configure | 5 + doc/APIchanges | 5 + libavutil/Makefile | 3 + libavutil/hwcontext.c | 4 + libavutil/hwcontext.h | 1 + libavutil/hwcontext_d3d12va.c | 703 +++++++++++++++++++++++++ libavutil/hwcontext_d3d12va.h | 134 +++++ libavutil/hwcontext_d3d12va_internal.h | 59 +++ libavutil/hwcontext_internal.h | 1 + libavutil/pixdesc.c | 4 + libavutil/pixfmt.h | 7 + libavutil/tests/hwdevice.c | 2 + libavutil/version.h | 2 +- 13 files changed, 929 insertions(+), 1 deletion(-) create mode 100644 libavutil/hwcontext_d3d12va.c create mode 100644 libavutil/hwcontext_d3d12va.h create mode 100644 libavutil/hwcontext_d3d12va_internal.h diff --git a/configure b/configure index d77c053226..1a343ac6b9 100755 --- a/configure +++ b/configure @@ -336,6 +336,7 @@ External library support: --disable-cuda-llvm disable CUDA compilation using clang [autodetect] --disable-cuvid disable Nvidia CUVID support [autodetect] --disable-d3d11va disable Microsoft Direct3D 11 video acceleration code [autodetect] + --disable-d3d12va disable Microsoft Direct3D 12 video acceleration code [autodetect] --disable-dxva2 disable Microsoft DirectX 9 video acceleration code [autodetect] --disable-ffnvcodec disable dynamically linked Nvidia code [autodetect] --enable-libdrm enable DRM code (Linux) [no] @@ -1926,6 +1927,7 @@ HWACCEL_AUTODETECT_LIBRARY_LIST=" cuda_llvm cuvid d3d11va + d3d12va dxva2 ffnvcodec nvdec @@ -3048,6 +3050,7 @@ crystalhd_deps="libcrystalhd_libcrystalhd_if_h" cuda_deps="ffnvcodec" cuvid_deps="ffnvcodec" d3d11va_deps="dxva_h ID3D11VideoDecoder ID3D11VideoContext" +d3d12va_deps="dxva_h ID3D12Device ID3D12VideoDecoder" dxva2_deps="dxva2api_h DXVA2_ConfigPictureDecode ole32 user32" ffnvcodec_deps_any="libdl LoadLibrary" mediacodec_deps="android" @@ -6575,6 +6578,8 @@ check_type "windows.h dxgi1_2.h" "IDXGIOutput1" check_type "windows.h dxgi1_5.h" "IDXGIOutput5" check_type "windows.h d3d11.h" "ID3D11VideoDecoder" check_type "windows.h d3d11.h" "ID3D11VideoContext" +check_type "windows.h d3d12.h" "ID3D12Device" +check_type "windows.h d3d12video.h" "ID3D12VideoDecoder" check_type "windows.h" "DPI_AWARENESS_CONTEXT" -D_WIN32_WINNT=0x0A00 check_type "d3d9.h dxva2api.h" DXVA2_ConfigPictureDecode -D_WIN32_WINNT=0x0602 check_func_headers mfapi.h MFCreateAlignedMemoryBuffer -lmfplat diff --git a/doc/APIchanges b/doc/APIchanges index 4a2dc1c44f..0435926d80 100644 --- a/doc/APIchanges +++ b/doc/APIchanges @@ -2,6 +2,11 @@ The last version increases of all libraries were on 2023-02-09 API changes, most recent first: +2023-11-07 - xxxxxxxxxx - lavu 58.33.100 - pixfmt.h hwcontext.h hwcontext_d3d12va.h + Add AV_HWDEVICE_TYPE_D3D12VA and AV_PIX_FMT_D3D12. + Add AVD3D12VADeviceContext, AVD3D12VASyncContext, AVD3D12VAFrame and + AVD3D12VAFramesContext. + 2023-11-08 - b82957a66a7 - lavu 58.32.100 - channel_layout.h Add AV_CH_LAYOUT_7POINT2POINT3 and AV_CHANNEL_LAYOUT_7POINT2POINT3. Add AV_CH_LAYOUT_9POINT1POINT4_BACK and AV_CHANNEL_LAYOUT_9POINT1POINT4_BACK. diff --git a/libavutil/Makefile b/libavutil/Makefile index 4711f8cde8..6a8566f1d9 100644 --- a/libavutil/Makefile +++ b/libavutil/Makefile @@ -42,6 +42,7 @@ HEADERS = adler32.h \ hwcontext.h \ hwcontext_cuda.h \ hwcontext_d3d11va.h \ + hwcontext_d3d12va.h \ hwcontext_drm.h \ hwcontext_dxva2.h \ hwcontext_qsv.h \ @@ -190,6 +191,7 @@ OBJS = adler32.o \ OBJS-$(CONFIG_CUDA) += hwcontext_cuda.o OBJS-$(CONFIG_D3D11VA) += hwcontext_d3d11va.o +OBJS-$(CONFIG_D3D12VA) += hwcontext_d3d12va.o OBJS-$(CONFIG_DXVA2) += hwcontext_dxva2.o OBJS-$(CONFIG_LIBDRM) += hwcontext_drm.o OBJS-$(CONFIG_MACOS_KPERF) += macos_kperf.o @@ -213,6 +215,7 @@ SKIPHEADERS-$(HAVE_CUDA_H) += hwcontext_cuda.h SKIPHEADERS-$(CONFIG_CUDA) += hwcontext_cuda_internal.h \ cuda_check.h SKIPHEADERS-$(CONFIG_D3D11VA) += hwcontext_d3d11va.h +SKIPHEADERS-$(CONFIG_D3D12VA) += hwcontext_d3d12va.h SKIPHEADERS-$(CONFIG_DXVA2) += hwcontext_dxva2.h SKIPHEADERS-$(CONFIG_QSV) += hwcontext_qsv.h SKIPHEADERS-$(CONFIG_OPENCL) += hwcontext_opencl.h diff --git a/libavutil/hwcontext.c b/libavutil/hwcontext.c index 3650d4653a..e23bad230f 100644 --- a/libavutil/hwcontext.c +++ b/libavutil/hwcontext.c @@ -36,6 +36,9 @@ static const HWContextType * const hw_table[] = { #if CONFIG_D3D11VA &ff_hwcontext_type_d3d11va, #endif +#if CONFIG_D3D12VA + &ff_hwcontext_type_d3d12va, +#endif #if CONFIG_LIBDRM &ff_hwcontext_type_drm, #endif @@ -71,6 +74,7 @@ static const char *const hw_type_names[] = { [AV_HWDEVICE_TYPE_DRM] = "drm", [AV_HWDEVICE_TYPE_DXVA2] = "dxva2", [AV_HWDEVICE_TYPE_D3D11VA] = "d3d11va", + [AV_HWDEVICE_TYPE_D3D12VA] = "d3d12va", [AV_HWDEVICE_TYPE_OPENCL] = "opencl", [AV_HWDEVICE_TYPE_QSV] = "qsv", [AV_HWDEVICE_TYPE_VAAPI] = "vaapi", diff --git a/libavutil/hwcontext.h b/libavutil/hwcontext.h index 7ff08c8608..2b33721a97 100644 --- a/libavutil/hwcontext.h +++ b/libavutil/hwcontext.h @@ -37,6 +37,7 @@ enum AVHWDeviceType { AV_HWDEVICE_TYPE_OPENCL, AV_HWDEVICE_TYPE_MEDIACODEC, AV_HWDEVICE_TYPE_VULKAN, + AV_HWDEVICE_TYPE_D3D12VA, }; typedef struct AVHWDeviceInternal AVHWDeviceInternal; diff --git a/libavutil/hwcontext_d3d12va.c b/libavutil/hwcontext_d3d12va.c new file mode 100644 index 0000000000..1600d94cb0 --- /dev/null +++ b/libavutil/hwcontext_d3d12va.c @@ -0,0 +1,703 @@ +/* + * Direct3D 12 HW acceleration. + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config.h" +#include "common.h" +#include "hwcontext.h" +#include "hwcontext_internal.h" +#include "hwcontext_d3d12va_internal.h" +#include "hwcontext_d3d12va.h" +#include "imgutils.h" +#include "pixdesc.h" +#include "pixfmt.h" +#include "thread.h" +#include "compat/w32dlfcn.h" +#include + +typedef HRESULT(WINAPI *PFN_CREATE_DXGI_FACTORY2)(UINT Flags, REFIID riid, void **ppFactory); + +typedef struct D3D12VAFramesContext { + ID3D12Resource *staging_download_buffer; + ID3D12Resource *staging_upload_buffer; + ID3D12CommandQueue *command_queue; + ID3D12CommandAllocator *command_allocator; + ID3D12GraphicsCommandList *command_list; + AVD3D12VASyncContext sync_ctx; + UINT luma_component_size; +} D3D12VAFramesContext; + +typedef struct D3D12VADevicePriv { + HANDLE d3d12lib; + HANDLE dxgilib; + PFN_CREATE_DXGI_FACTORY2 create_dxgi_factory2; + PFN_D3D12_CREATE_DEVICE create_device; + PFN_D3D12_GET_DEBUG_INTERFACE get_debug_interface; +} D3D12VADevicePriv; + +static const struct { + DXGI_FORMAT d3d_format; + enum AVPixelFormat pix_fmt; +} supported_formats[] = { + { DXGI_FORMAT_NV12, AV_PIX_FMT_NV12 }, + { DXGI_FORMAT_P010, AV_PIX_FMT_P010 }, +}; + +static void d3d12va_default_lock(void *ctx) +{ + WaitForSingleObjectEx(ctx, INFINITE, FALSE); +} + +static void d3d12va_default_unlock(void *ctx) +{ + ReleaseMutex(ctx); +} + +DXGI_FORMAT av_d3d12va_map_sw_to_hw_format(enum AVPixelFormat pix_fmt) +{ + switch (pix_fmt) { + case AV_PIX_FMT_NV12:return DXGI_FORMAT_NV12; + case AV_PIX_FMT_P010:return DXGI_FORMAT_P010; + default: return DXGI_FORMAT_UNKNOWN; + } +} + +static int d3d12va_fence_completion(AVD3D12VASyncContext *psync_ctx) +{ + uint64_t completion = ID3D12Fence_GetCompletedValue(psync_ctx->fence); + if (completion < psync_ctx->fence_value) { + if (FAILED(ID3D12Fence_SetEventOnCompletion(psync_ctx->fence, psync_ctx->fence_value, psync_ctx->event))) + return AVERROR(EINVAL); + + WaitForSingleObjectEx(psync_ctx->event, INFINITE, FALSE); + } + + return 0; +} + +static inline int d3d12va_wait_queue_idle(AVD3D12VASyncContext *psync_ctx, ID3D12CommandQueue *command_queue) +{ + DX_CHECK(ID3D12CommandQueue_Signal(command_queue, psync_ctx->fence, ++psync_ctx->fence_value)); + return d3d12va_fence_completion(psync_ctx); + +fail: + return AVERROR(EINVAL); +} + +static int d3d12va_create_staging_buffer_resource(AVHWFramesContext *ctx, D3D12_RESOURCE_STATES states, + ID3D12Resource **ppResource, int download) +{ + AVD3D12VADeviceContext *device_hwctx = ctx->device_ctx->hwctx; + D3D12VAFramesContext *s = ctx->internal->priv; + D3D12_HEAP_PROPERTIES props = { .Type = download ? D3D12_HEAP_TYPE_READBACK : D3D12_HEAP_TYPE_UPLOAD }; + D3D12_RESOURCE_DESC desc = { + .Dimension = D3D12_RESOURCE_DIMENSION_BUFFER, + .Alignment = 0, + .Width = s->luma_component_size + (s->luma_component_size >> 1), + .Height = 1, + .DepthOrArraySize = 1, + .MipLevels = 1, + .Format = DXGI_FORMAT_UNKNOWN, + .SampleDesc = { .Count = 1, .Quality = 0 }, + .Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR, + .Flags = D3D12_RESOURCE_FLAG_NONE, + }; + + if (FAILED(ID3D12Device_CreateCommittedResource(device_hwctx->device, &props, D3D12_HEAP_FLAG_NONE, &desc, + states, NULL, &IID_ID3D12Resource, (void **)ppResource))) { + av_log(ctx, AV_LOG_ERROR, "Could not create the staging buffer resource\n"); + return AVERROR_UNKNOWN; + } + + return 0; +} + +static int d3d12va_create_helper_objects(AVHWFramesContext *ctx) +{ + AVD3D12VADeviceContext *device_hwctx = ctx->device_ctx->hwctx; + AVD3D12VAFramesContext *frames_hwctx = ctx->hwctx; + D3D12VAFramesContext *s = ctx->internal->priv; + + D3D12_COMMAND_QUEUE_DESC queue_desc = { + .Type = D3D12_COMMAND_LIST_TYPE_COPY, + .Priority = 0, + .NodeMask = 0, + }; + + s->luma_component_size = FFALIGN(ctx->width * (frames_hwctx->format == DXGI_FORMAT_P010 ? 2 : 1), + D3D12_TEXTURE_DATA_PITCH_ALIGNMENT) * ctx->height; + + DX_CHECK(ID3D12Device_CreateFence(device_hwctx->device, 0, D3D12_FENCE_FLAG_NONE, + &IID_ID3D12Fence, (void **)&s->sync_ctx.fence)); + + s->sync_ctx.event = CreateEvent(NULL, FALSE, FALSE, NULL); + if (!s->sync_ctx.event) + goto fail; + + DX_CHECK(ID3D12Device_CreateCommandQueue(device_hwctx->device, &queue_desc, + &IID_ID3D12CommandQueue, (void **)&s->command_queue)); + + DX_CHECK(ID3D12Device_CreateCommandAllocator(device_hwctx->device, queue_desc.Type, + &IID_ID3D12CommandAllocator, (void **)&s->command_allocator)); + + DX_CHECK(ID3D12Device_CreateCommandList(device_hwctx->device, 0, queue_desc.Type, + s->command_allocator, NULL, &IID_ID3D12GraphicsCommandList, (void **)&s->command_list)); + + DX_CHECK(ID3D12GraphicsCommandList_Close(s->command_list)); + + ID3D12CommandQueue_ExecuteCommandLists(s->command_queue, 1, (ID3D12CommandList **)&s->command_list); + + return d3d12va_wait_queue_idle(&s->sync_ctx, s->command_queue); + +fail: + return AVERROR(EINVAL); +} + +static void d3d12va_frames_uninit(AVHWFramesContext *ctx) +{ + AVD3D12VAFramesContext *frames_hwctx = ctx->hwctx; + D3D12VAFramesContext *s = ctx->internal->priv; + + D3D12_OBJECT_RELEASE(s->sync_ctx.fence); + if (s->sync_ctx.event) + CloseHandle(s->sync_ctx.event); + + D3D12_OBJECT_RELEASE(s->staging_download_buffer); + D3D12_OBJECT_RELEASE(s->staging_upload_buffer); + D3D12_OBJECT_RELEASE(s->command_allocator); + D3D12_OBJECT_RELEASE(s->command_list); + D3D12_OBJECT_RELEASE(s->command_queue); +} + +static int d3d12va_frames_get_constraints(AVHWDeviceContext *ctx, const void *hwconfig, AVHWFramesConstraints *constraints) +{ + HRESULT hr; + int nb_sw_formats = 0; + AVD3D12VADeviceContext *device_hwctx = ctx->hwctx; + + constraints->valid_sw_formats = av_malloc_array(FF_ARRAY_ELEMS(supported_formats) + 1, + sizeof(*constraints->valid_sw_formats)); + if (!constraints->valid_sw_formats) + return AVERROR(ENOMEM); + + for (int i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) { + D3D12_FEATURE_DATA_FORMAT_SUPPORT format_support = { supported_formats[i].d3d_format }; + hr = ID3D12Device_CheckFeatureSupport(device_hwctx->device, D3D12_FEATURE_FORMAT_SUPPORT, &format_support, sizeof(format_support)); + if (SUCCEEDED(hr) && (format_support.Support1 & D3D12_FORMAT_SUPPORT1_TEXTURE2D)) + constraints->valid_sw_formats[nb_sw_formats++] = supported_formats[i].pix_fmt; + } + constraints->valid_sw_formats[nb_sw_formats] = AV_PIX_FMT_NONE; + + constraints->valid_hw_formats = av_malloc_array(2, sizeof(*constraints->valid_hw_formats)); + if (!constraints->valid_hw_formats) + return AVERROR(ENOMEM); + + constraints->valid_hw_formats[0] = AV_PIX_FMT_D3D12; + constraints->valid_hw_formats[1] = AV_PIX_FMT_NONE; + + return 0; +} + +static void free_texture(void *opaque, uint8_t *data) +{ + AVD3D12VAFrame *frame = (AVD3D12VAFrame *)data; + + D3D12_OBJECT_RELEASE(frame->texture); + D3D12_OBJECT_RELEASE(frame->sync_ctx.fence); + if (frame->sync_ctx.event) + CloseHandle(frame->sync_ctx.event); + + av_freep(&data); +} + +static AVBufferRef *d3d12va_pool_alloc(void *opaque, size_t size) +{ + AVHWFramesContext *ctx = (AVHWFramesContext *)opaque; + AVD3D12VAFramesContext *hwctx = ctx->hwctx; + AVD3D12VADeviceContext *device_hwctx = ctx->device_ctx->hwctx; + + AVBufferRef *buf; + AVD3D12VAFrame *frame; + D3D12_HEAP_PROPERTIES props = { .Type = D3D12_HEAP_TYPE_DEFAULT }; + D3D12_RESOURCE_DESC desc = { + .Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D, + .Alignment = 0, + .Width = ctx->width, + .Height = ctx->height, + .DepthOrArraySize = 1, + .MipLevels = 1, + .Format = hwctx->format, + .SampleDesc = {.Count = 1, .Quality = 0 }, + .Layout = D3D12_TEXTURE_LAYOUT_UNKNOWN, + .Flags = D3D12_RESOURCE_FLAG_NONE, + }; + + frame = av_mallocz(sizeof(AVD3D12VAFrame)); + if (!frame) + return NULL; + + if (FAILED(ID3D12Device_CreateCommittedResource(device_hwctx->device, &props, D3D12_HEAP_FLAG_NONE, &desc, + D3D12_RESOURCE_STATE_COMMON, NULL, &IID_ID3D12Resource, (void **)&frame->texture))) { + av_log(ctx, AV_LOG_ERROR, "Could not create the texture\n"); + goto fail; + } + + DX_CHECK(ID3D12Device_CreateFence(device_hwctx->device, 0, D3D12_FENCE_FLAG_NONE, + &IID_ID3D12Fence, (void **)&frame->sync_ctx.fence)); + + frame->sync_ctx.event = CreateEvent(NULL, FALSE, FALSE, NULL); + if (!frame->sync_ctx.event) + goto fail; + + buf = av_buffer_create((uint8_t *)frame, sizeof(frame), free_texture, NULL, 0); + if (!buf) + goto fail; + + return buf; + +fail: + free_texture(NULL, (uint8_t *)frame); + return NULL; +} + +static int d3d12va_frames_init(AVHWFramesContext *ctx) +{ + AVD3D12VAFramesContext *hwctx = ctx->hwctx; + AVD3D12VADeviceContext *device_hwctx = ctx->device_ctx->hwctx; + D3D12VAFramesContext *s = ctx->internal->priv; + int i; + + for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) { + if (ctx->sw_format == supported_formats[i].pix_fmt) { + if (hwctx->format != DXGI_FORMAT_UNKNOWN && + hwctx->format != supported_formats[i].d3d_format) + av_log(ctx, AV_LOG_WARNING, "Incompatible DXGI format provided by user, will be overided\n"); + hwctx->format = supported_formats[i].d3d_format; + break; + } + } + if (i == FF_ARRAY_ELEMS(supported_formats)) { + av_log(ctx, AV_LOG_ERROR, "Unsupported pixel format: %s\n", + av_get_pix_fmt_name(ctx->sw_format)); + return AVERROR(EINVAL); + } + + ctx->internal->pool_internal = av_buffer_pool_init2(sizeof(AVD3D12VAFrame), + ctx, d3d12va_pool_alloc, NULL); + + if (!ctx->internal->pool_internal) + return AVERROR(ENOMEM); + + return 0; +} + +static int d3d12va_get_buffer(AVHWFramesContext *ctx, AVFrame *frame) +{ + int ret; + + frame->buf[0] = av_buffer_pool_get(ctx->pool); + if (!frame->buf[0]) + return AVERROR(ENOMEM); + + ret = av_image_fill_arrays(frame->data, frame->linesize, NULL, + ctx->sw_format, ctx->width, ctx->height, + D3D12_TEXTURE_DATA_PITCH_ALIGNMENT); + if (ret < 0) + return ret; + + frame->data[0] = frame->buf[0]->data; + frame->format = AV_PIX_FMT_D3D12; + frame->width = ctx->width; + frame->height = ctx->height; + + return 0; +} + +static int d3d12va_transfer_get_formats(AVHWFramesContext *ctx, + enum AVHWFrameTransferDirection dir, + enum AVPixelFormat **formats) +{ + D3D12VAFramesContext *s = ctx->internal->priv; + enum AVPixelFormat *fmts; + + fmts = av_malloc_array(2, sizeof(*fmts)); + if (!fmts) + return AVERROR(ENOMEM); + + fmts[0] = ctx->sw_format; + fmts[1] = AV_PIX_FMT_NONE; + + *formats = fmts; + + return 0; +} + +static int d3d12va_transfer_data(AVHWFramesContext *ctx, AVFrame *dst, + const AVFrame *src) +{ + AVD3D12VADeviceContext *hwctx = ctx->device_ctx->hwctx; + AVD3D12VAFramesContext *frames_hwctx = ctx->hwctx; + D3D12VAFramesContext *s = ctx->internal->priv; + + int ret; + int download = src->format == AV_PIX_FMT_D3D12; + const AVFrame *frame = download ? src : dst; + const AVFrame *other = download ? dst : src; + + AVD3D12VAFrame *f = (AVD3D12VAFrame *)frame->data[0]; + ID3D12Resource *texture = (ID3D12Resource *)f->texture; + + uint8_t *mapped_data; + uint8_t *data[4]; + int linesizes[4]; + + D3D12_TEXTURE_COPY_LOCATION staging_y_location = { 0 }; + D3D12_TEXTURE_COPY_LOCATION staging_uv_location = { 0 }; + + D3D12_TEXTURE_COPY_LOCATION texture_y_location = { + .pResource = texture, + .Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX, + .SubresourceIndex = 0, + }; + + D3D12_TEXTURE_COPY_LOCATION texture_uv_location = { + .pResource = texture, + .Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX, + .SubresourceIndex = 1, + }; + + D3D12_RESOURCE_BARRIER barrier = { + .Type = D3D12_RESOURCE_BARRIER_TYPE_TRANSITION, + .Flags = D3D12_RESOURCE_BARRIER_FLAG_NONE, + .Transition = { + .pResource = texture, + .StateBefore = D3D12_RESOURCE_STATE_COMMON, + .StateAfter = download ? D3D12_RESOURCE_STATE_COPY_SOURCE : D3D12_RESOURCE_STATE_COPY_DEST, + .Subresource = D3D12_RESOURCE_BARRIER_ALL_SUBRESOURCES, + }, + }; + + if (frame->hw_frames_ctx->data != (uint8_t *)ctx || other->format != ctx->sw_format) + return AVERROR(EINVAL); + + hwctx->lock(hwctx->lock_ctx); + + if (!s->command_queue) { + ret = d3d12va_create_helper_objects(ctx); + if (ret < 0) + goto fail; + } + + for (int i = 0; i < 4; i++) + linesizes[i] = FFALIGN(frame->width * (frames_hwctx->format == DXGI_FORMAT_P010 ? 2 : 1), + D3D12_TEXTURE_DATA_PITCH_ALIGNMENT); + + staging_y_location = (D3D12_TEXTURE_COPY_LOCATION) { + .Type = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT, + .PlacedFootprint = { + .Offset = 0, + .Footprint = { + .Format = frames_hwctx->format == DXGI_FORMAT_P010 ? + DXGI_FORMAT_R16_UNORM : DXGI_FORMAT_R8_UNORM, + .Width = ctx->width, + .Height = ctx->height, + .Depth = 1, + .RowPitch = linesizes[0], + }, + }, + }; + + staging_uv_location = (D3D12_TEXTURE_COPY_LOCATION) { + .Type = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT, + .PlacedFootprint = { + .Offset = s->luma_component_size, + .Footprint = { + .Format = frames_hwctx->format == DXGI_FORMAT_P010 ? + DXGI_FORMAT_R16G16_UNORM : DXGI_FORMAT_R8G8_UNORM, + .Width = ctx->width >> 1, + .Height = ctx->height >> 1, + .Depth = 1, + .RowPitch = linesizes[0], + }, + }, + }; + + DX_CHECK(ID3D12CommandAllocator_Reset(s->command_allocator)); + + DX_CHECK(ID3D12GraphicsCommandList_Reset(s->command_list, s->command_allocator, NULL)); + + if (download) { + if (!s->staging_download_buffer) { + ret = d3d12va_create_staging_buffer_resource(ctx, D3D12_RESOURCE_STATE_COPY_DEST, + &s->staging_download_buffer, 1); + if (ret < 0) { + goto fail; + } + } + + staging_y_location.pResource = staging_uv_location.pResource = s->staging_download_buffer; + + ID3D12GraphicsCommandList_ResourceBarrier(s->command_list, 1, &barrier); + + ID3D12GraphicsCommandList_CopyTextureRegion(s->command_list, + &staging_y_location, 0, 0, 0, + &texture_y_location, NULL); + + ID3D12GraphicsCommandList_CopyTextureRegion(s->command_list, + &staging_uv_location, 0, 0, 0, + &texture_uv_location, NULL); + + barrier.Transition.StateBefore = barrier.Transition.StateAfter; + barrier.Transition.StateAfter = D3D12_RESOURCE_STATE_COMMON; + ID3D12GraphicsCommandList_ResourceBarrier(s->command_list, 1, &barrier); + + DX_CHECK(ID3D12GraphicsCommandList_Close(s->command_list)); + + DX_CHECK(ID3D12CommandQueue_Wait(s->command_queue, f->sync_ctx.fence, f->sync_ctx.fence_value)); + + ID3D12CommandQueue_ExecuteCommandLists(s->command_queue, 1, (ID3D12CommandList **)&s->command_list); + + ret = d3d12va_wait_queue_idle(&s->sync_ctx, s->command_queue); + if (ret < 0) + goto fail; + + DX_CHECK(ID3D12Resource_Map(s->staging_download_buffer, 0, NULL, (void **)&mapped_data)); + av_image_fill_pointers(data, ctx->sw_format, ctx->height, mapped_data, linesizes); + + av_image_copy(dst->data, dst->linesize, data, linesizes, + ctx->sw_format, ctx->width, ctx->height); + + ID3D12Resource_Unmap(s->staging_download_buffer, 0, NULL); + } else { + if (!s->staging_upload_buffer) { + ret = d3d12va_create_staging_buffer_resource(ctx, D3D12_RESOURCE_STATE_GENERIC_READ, + &s->staging_upload_buffer, 0); + if (ret < 0) { + goto fail; + } + } + + staging_y_location.pResource = staging_uv_location.pResource = s->staging_upload_buffer; + + DX_CHECK(ID3D12Resource_Map(s->staging_upload_buffer, 0, NULL, (void **)&mapped_data)); + av_image_fill_pointers(data, ctx->sw_format, ctx->height, mapped_data, linesizes); + + av_image_copy(data, linesizes, src->data, src->linesize, + ctx->sw_format, ctx->width, ctx->height); + + ID3D12Resource_Unmap(s->staging_upload_buffer, 0, NULL); + + ID3D12GraphicsCommandList_ResourceBarrier(s->command_list, 1, &barrier); + + ID3D12GraphicsCommandList_CopyTextureRegion(s->command_list, + &texture_y_location, 0, 0, 0, + &staging_y_location, NULL); + + ID3D12GraphicsCommandList_CopyTextureRegion(s->command_list, + &texture_uv_location, 0, 0, 0, + &staging_uv_location, NULL); + + barrier.Transition.StateBefore = barrier.Transition.StateAfter; + barrier.Transition.StateAfter = D3D12_RESOURCE_STATE_COMMON; + ID3D12GraphicsCommandList_ResourceBarrier(s->command_list, 1, &barrier); + + DX_CHECK(ID3D12GraphicsCommandList_Close(s->command_list)); + + ID3D12CommandQueue_ExecuteCommandLists(s->command_queue, 1, (ID3D12CommandList **)&s->command_list); + + ret = d3d12va_wait_queue_idle(&s->sync_ctx, s->command_queue); + if (ret < 0) + goto fail; + } + + hwctx->unlock(hwctx->lock_ctx); + + return 0; + +fail: + hwctx->unlock(hwctx->lock_ctx); + return AVERROR(EINVAL); +} + +static int d3d12va_load_functions(AVHWDeviceContext *hwdev) +{ + D3D12VADevicePriv *priv = hwdev->internal->priv; + +#if !HAVE_UWP + priv->d3d12lib = dlopen("d3d12.dll", 0); + priv->dxgilib = dlopen("dxgi.dll", 0); + + if (!priv->d3d12lib || !priv->dxgilib) + goto fail; + + priv->create_device = (PFN_D3D12_CREATE_DEVICE)GetProcAddress(priv->d3d12lib, "D3D12CreateDevice"); + if (!priv->create_device) + goto fail; + + priv->create_dxgi_factory2 = (PFN_CREATE_DXGI_FACTORY2)GetProcAddress(priv->dxgilib, "CreateDXGIFactory2"); + if (!priv->create_dxgi_factory2) + goto fail; + + priv->get_debug_interface = (PFN_D3D12_GET_DEBUG_INTERFACE)GetProcAddress(priv->d3d12lib, "D3D12GetDebugInterface"); +#else + priv->create_device = (PFN_D3D12_CREATE_DEVICE) D3D12CreateDevice; + priv->create_dxgi_factory2 = (PFN_CREATE_DXGI_FACTORY2) CreateDXGIFactory2; + priv->get_debug_interface = (PFN_D3D12_GET_DEBUG_INTERFACE) D3D12GetDebugInterface; +#endif + return 0; + +fail: + av_log(hwdev, AV_LOG_ERROR, "Failed to load D3D12 library or its functions\n"); + return AVERROR_UNKNOWN; +} + +static void d3d12va_device_free(AVHWDeviceContext *hwdev) +{ + AVD3D12VADeviceContext *ctx = hwdev->hwctx; + D3D12VADevicePriv *priv = hwdev->internal->priv; + + D3D12_OBJECT_RELEASE(ctx->device); + + if (priv->d3d12lib) + dlclose(priv->d3d12lib); + + if (priv->dxgilib) + dlclose(priv->dxgilib); +} + +static int d3d12va_device_init(AVHWDeviceContext *hwdev) +{ + AVD3D12VADeviceContext *ctx = hwdev->hwctx; + + if (!ctx->lock) { + ctx->lock_ctx = CreateMutex(NULL, 0, NULL); + if (ctx->lock_ctx == INVALID_HANDLE_VALUE) { + av_log(NULL, AV_LOG_ERROR, "Failed to create a mutex\n"); + return AVERROR(EINVAL); + } + ctx->lock = d3d12va_default_lock; + ctx->unlock = d3d12va_default_unlock; + } + + if (!ctx->video_device) + DX_CHECK(ID3D12Device_QueryInterface(ctx->device, &IID_ID3D12VideoDevice, (void **)&ctx->video_device)); + + return 0; + +fail: + return AVERROR(EINVAL); +} + +static void d3d12va_device_uninit(AVHWDeviceContext *hwdev) +{ + AVD3D12VADeviceContext *device_hwctx = hwdev->hwctx; + + D3D12_OBJECT_RELEASE(device_hwctx->video_device); + + if (device_hwctx->lock == d3d12va_default_lock) { + CloseHandle(device_hwctx->lock_ctx); + device_hwctx->lock_ctx = INVALID_HANDLE_VALUE; + device_hwctx->lock = NULL; + } +} + +static int d3d12va_device_create(AVHWDeviceContext *hwdev, const char *device, + AVDictionary *opts, int flags) +{ + AVD3D12VADeviceContext *ctx = hwdev->hwctx; + D3D12VADevicePriv *priv = hwdev->internal->priv; + + HRESULT hr; + UINT create_flags = 0; + IDXGIAdapter *pAdapter = NULL; + + int ret; + int is_debug = !!av_dict_get(opts, "debug", NULL, 0); + + hwdev->free = d3d12va_device_free; + + ret = d3d12va_load_functions(hwdev); + if (ret < 0) + return ret; + + if (is_debug) { + ID3D12Debug *pDebug; + if (priv->get_debug_interface && SUCCEEDED(priv->get_debug_interface(&IID_ID3D12Debug, (void **)&pDebug))) { + create_flags |= DXGI_CREATE_FACTORY_DEBUG; + ID3D12Debug_EnableDebugLayer(pDebug); + D3D12_OBJECT_RELEASE(pDebug); + av_log(hwdev, AV_LOG_INFO, "D3D12 debug layer is enabled!\n"); + } + } + + if (!ctx->device) { + IDXGIFactory2 *pDXGIFactory = NULL; + + hr = priv->create_dxgi_factory2(create_flags, &IID_IDXGIFactory2, (void **)&pDXGIFactory); + if (SUCCEEDED(hr)) { + int adapter = device ? atoi(device) : 0; + if (FAILED(IDXGIFactory2_EnumAdapters(pDXGIFactory, adapter, &pAdapter))) + pAdapter = NULL; + IDXGIFactory2_Release(pDXGIFactory); + } + + if (pAdapter) { + DXGI_ADAPTER_DESC desc; + hr = IDXGIAdapter2_GetDesc(pAdapter, &desc); + if (!FAILED(hr)) { + av_log(ctx, AV_LOG_INFO, "Using device %04x:%04x (%ls).\n", + desc.VendorId, desc.DeviceId, desc.Description); + } + } + + hr = priv->create_device((IUnknown *)pAdapter, D3D_FEATURE_LEVEL_12_0, &IID_ID3D12Device, (void **)&ctx->device); + D3D12_OBJECT_RELEASE(pAdapter); + if (FAILED(hr)) { + av_log(ctx, AV_LOG_ERROR, "Failed to create Direct 3D 12 device (%lx)\n", (long)hr); + return AVERROR_UNKNOWN; + } + } + + return 0; +} + +const HWContextType ff_hwcontext_type_d3d12va = { + .type = AV_HWDEVICE_TYPE_D3D12VA, + .name = "D3D12VA", + + .device_hwctx_size = sizeof(AVD3D12VADeviceContext), + .device_priv_size = sizeof(D3D12VADevicePriv), + .frames_hwctx_size = sizeof(AVD3D12VAFramesContext), + .frames_priv_size = sizeof(D3D12VAFramesContext), + + .device_create = d3d12va_device_create, + .device_init = d3d12va_device_init, + .device_uninit = d3d12va_device_uninit, + .frames_get_constraints = d3d12va_frames_get_constraints, + .frames_init = d3d12va_frames_init, + .frames_uninit = d3d12va_frames_uninit, + .frames_get_buffer = d3d12va_get_buffer, + .transfer_get_formats = d3d12va_transfer_get_formats, + .transfer_data_to = d3d12va_transfer_data, + .transfer_data_from = d3d12va_transfer_data, + + .pix_fmts = (const enum AVPixelFormat[]){ AV_PIX_FMT_D3D12, AV_PIX_FMT_NONE }, +}; diff --git a/libavutil/hwcontext_d3d12va.h b/libavutil/hwcontext_d3d12va.h new file mode 100644 index 0000000000..ff06e6f2ef --- /dev/null +++ b/libavutil/hwcontext_d3d12va.h @@ -0,0 +1,134 @@ +/* + * Direct3D 12 HW acceleration. + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVUTIL_HWCONTEXT_D3D12VA_H +#define AVUTIL_HWCONTEXT_D3D12VA_H + +/** + * @file + * An API-specific header for AV_HWDEVICE_TYPE_D3D12VA. + * + * AVHWFramesContext.pool must contain AVBufferRefs whose + * data pointer points to an AVD3D12VAFrame struct. + */ +#include +#include +#include +#include +#include + +/** + * @brief This struct is allocated as AVHWDeviceContext.hwctx + * + */ +typedef struct AVD3D12VADeviceContext { + /** + * Device used for objects creation and access. This can also be + * used to set the libavcodec decoding device. + * + * Can be set by the user. This is the only mandatory field - the other + * device context fields are set from this and are available for convenience. + * + * Deallocating the AVHWDeviceContext will always release this interface, + * and it does not matter whether it was user-allocated. + */ + ID3D12Device *device; + + /** + * If unset, this will be set from the device field on init. + * + * Deallocating the AVHWDeviceContext will always release this interface, + * and it does not matter whether it was user-allocated. + */ + ID3D12VideoDevice *video_device; + + /** + * Callbacks for locking. They protect access to the internal staging + * texture (for av_hwframe_transfer_data() calls). They do NOT protect + * access to hwcontext or decoder state in general. + * + * If unset on init, the hwcontext implementation will set them to use an + * internal mutex. + * + * The underlying lock must be recursive. lock_ctx is for free use by the + * locking implementation. + */ + void (*lock)(void *lock_ctx); + void (*unlock)(void *lock_ctx); + void *lock_ctx; +} AVD3D12VADeviceContext; + +/** + * @brief This struct is used to sync d3d12 execution + * + */ +typedef struct AVD3D12VASyncContext { + /** + * D3D12 fence object + */ + ID3D12Fence *fence; + + /** + * A handle to the event object that's raised when the fence + * reaches a certain value. + */ + HANDLE event; + + /** + * The fence value used for sync + */ + uint64_t fence_value; +} AVD3D12VASyncContext; + +/** + * @brief D3D12VA frame descriptor for pool allocation. + * + */ +typedef struct AVD3D12VAFrame { + /** + * The texture in which the frame is located. The reference count is + * managed by the AVBufferRef, and destroying the reference will release + * the interface. + */ + ID3D12Resource *texture; + + /** + * The sync context for the texture + * + * @see: https://learn.microsoft.com/en-us/windows/win32/medfound/direct3d-12-video-overview#directx-12-fences + */ + AVD3D12VASyncContext sync_ctx; +} AVD3D12VAFrame; + +/** + * @brief This struct is allocated as AVHWFramesContext.hwctx + * + */ +typedef struct AVD3D12VAFramesContext { + /** + * DXGI_FORMAT format. MUST be compatible with the pixel format. + * If unset, will be automatically set. + */ + DXGI_FORMAT format; +} AVD3D12VAFramesContext; + +#endif /* AVUTIL_HWCONTEXT_D3D12VA_H */ diff --git a/libavutil/hwcontext_d3d12va_internal.h b/libavutil/hwcontext_d3d12va_internal.h new file mode 100644 index 0000000000..bfd89b3545 --- /dev/null +++ b/libavutil/hwcontext_d3d12va_internal.h @@ -0,0 +1,59 @@ +/* + * Direct3D 12 HW acceleration. + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVUTIL_HWCONTEXT_D3D12VA_INTERNAL_H +#define AVUTIL_HWCONTEXT_D3D12VA_INTERNAL_H + +/** + * @def COBJMACROS + * + * @brief Enable C style interface for D3D12 + */ +#ifndef COBJMACROS +#define COBJMACROS +#endif + +/** + * @def DX_CHECK + * + * @brief A check macro used by D3D12 functions highly frequently + */ +#define DX_CHECK(hr) \ + do { \ + if (FAILED(hr)) \ + goto fail; \ + } while (0) + +/** + * @def D3D12_OBJECT_RELEASE + * + * @brief A release macro used by D3D12 objects highly frequently + */ +#define D3D12_OBJECT_RELEASE(pInterface) \ + do { \ + if (pInterface) { \ + IUnknown_Release((IUnknown *)pInterface); \ + pInterface = NULL; \ + } \ + } while (0) + +#endif /* AVUTIL_HWCONTEXT_D3D12VA_INTERNAL_H */ \ No newline at end of file diff --git a/libavutil/hwcontext_internal.h b/libavutil/hwcontext_internal.h index e6266494ac..4df516ee6a 100644 --- a/libavutil/hwcontext_internal.h +++ b/libavutil/hwcontext_internal.h @@ -165,6 +165,7 @@ int ff_hwframe_map_replace(AVFrame *dst, const AVFrame *src); extern const HWContextType ff_hwcontext_type_cuda; extern const HWContextType ff_hwcontext_type_d3d11va; +extern const HWContextType ff_hwcontext_type_d3d12va; extern const HWContextType ff_hwcontext_type_drm; extern const HWContextType ff_hwcontext_type_dxva2; extern const HWContextType ff_hwcontext_type_opencl; diff --git a/libavutil/pixdesc.c b/libavutil/pixdesc.c index 4e4a63e287..0db4167934 100644 --- a/libavutil/pixdesc.c +++ b/libavutil/pixdesc.c @@ -2311,6 +2311,10 @@ static const AVPixFmtDescriptor av_pix_fmt_descriptors[AV_PIX_FMT_NB] = { .name = "d3d11", .flags = AV_PIX_FMT_FLAG_HWACCEL, }, + [AV_PIX_FMT_D3D12] = { + .name = "d3d12", + .flags = AV_PIX_FMT_FLAG_HWACCEL, + }, [AV_PIX_FMT_GBRPF32BE] = { .name = "gbrpf32be", .nb_components = 3, diff --git a/libavutil/pixfmt.h b/libavutil/pixfmt.h index a26c72d56b..58f9ad28bd 100644 --- a/libavutil/pixfmt.h +++ b/libavutil/pixfmt.h @@ -429,6 +429,13 @@ enum AVPixelFormat { AV_PIX_FMT_GBRAP14BE, ///< planar GBR 4:4:4:4 56bpp, big-endian AV_PIX_FMT_GBRAP14LE, ///< planar GBR 4:4:4:4 56bpp, little-endian + /** + * Hardware surfaces for Direct3D 12. + * + * data[0] points to an AVD3D12VAFrame + */ + AV_PIX_FMT_D3D12, + AV_PIX_FMT_NB ///< number of pixel formats, DO NOT USE THIS if you want to link with shared libav* because the number of formats might differ between versions }; diff --git a/libavutil/tests/hwdevice.c b/libavutil/tests/hwdevice.c index c57586613a..9d7964f9ee 100644 --- a/libavutil/tests/hwdevice.c +++ b/libavutil/tests/hwdevice.c @@ -137,6 +137,8 @@ static const struct { { "0", "1", "2" } }, { AV_HWDEVICE_TYPE_D3D11VA, { "0", "1", "2" } }, + { AV_HWDEVICE_TYPE_D3D12VA, + { "0", "1", "2" } }, { AV_HWDEVICE_TYPE_OPENCL, { "0.0", "0.1", "1.0", "1.1" } }, { AV_HWDEVICE_TYPE_VAAPI, diff --git a/libavutil/version.h b/libavutil/version.h index c5fa7c3692..0684996bf2 100644 --- a/libavutil/version.h +++ b/libavutil/version.h @@ -79,7 +79,7 @@ */ #define LIBAVUTIL_VERSION_MAJOR 58 -#define LIBAVUTIL_VERSION_MINOR 32 +#define LIBAVUTIL_VERSION_MINOR 33 #define LIBAVUTIL_VERSION_MICRO 100 #define LIBAVUTIL_VERSION_INT AV_VERSION_INT(LIBAVUTIL_VERSION_MAJOR, \ From patchwork Tue Dec 5 06:46:44 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44905 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145667pzc; Mon, 4 Dec 2023 22:51:50 -0800 (PST) X-Google-Smtp-Source: AGHT+IEsRz/Em8M4fud5jlVPibVuUrNlXcHHuMa36qfOGwlrm86tghMXgIN/zScjlkOOaEcmSP8M X-Received: by 2002:a17:906:7491:b0:a19:a19a:eaa2 with SMTP id e17-20020a170906749100b00a19a19aeaa2mr179788ejl.91.1701759110717; Mon, 04 Dec 2023 22:51:50 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759110; cv=none; d=google.com; s=arc-20160816; b=VFUbC68QDSO/426Jinrr8iD945rlF3t3rVr/ncSj8ftI2XTr/rTAX2+AAipRpm+VvJ fFP+hRaQclZKmses/ZvF/XV5IWkW8kEir4bKFezF/HnqD3YhfAkGLgUs7Q/boBSF4dWl 9GnIiyO3JlZee1ccrsOgyUkU7wGrewUzXYayuSXMahn1lfquOlOs+brbToMEa/1S08F0 TVpJ+L58Ur7aAUBNLCYLQZ36Gg6f/l/KLcKlhE0eKB7ugXpCRfpVICxch8YAi3uhCUt1 tyhUQls7NG5QDWnWZms3pD9okIrzgdm2gDCto+9JMI2QAvw52B62lkcL8+DVhibPfyiU rS7w== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=woZZYNCpQjQNAGJZviAUjXPKMmBM6FGcKobuiKG83fQ=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=f1tWQx+BZDHqFmbZZDVtOOmkFmyeguPq/7fpTic6DFCX5MVsF6gNY+crkvirvBU3B8 WGwSXdPDp7gATlB9+a7fmC0TLio9nSPQ4Snlf+nxfb428xDKziSamsjzLC4KaAUnSKUX 3UeUHhnPEONEPM8OFtIs+5IDb6Cno33Pdd7NILBhikx6yB+DkAU8otzAWw1OsFLNDRi1 Pif/kRQY/Axvl/EFjcw1Wa1kAUQ1nyw0zb0v1uCPsD93GViMFBIie3O7U4JXYY/7IOuZ cs1OquABgKyRguIsZs9EXa12IeHLl8aGLSPDCBWrUfEyhYdZXkmcpCjkM5bs2kO6UVzi l/7g== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=VuKoBQnH; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id mt17-20020a170907619100b00a1c936d33aasi144462ejc.235.2023.12.04.22.51.50; Mon, 04 Dec 2023 22:51:50 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=VuKoBQnH; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 31F6068CDEB; Tue, 5 Dec 2023 08:51:35 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 7590068CA67 for ; Tue, 5 Dec 2023 08:51:27 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759092; x=1733295092; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=NtCqfu8tC6fCJiQ9wpALD7emt8+auq2rlJNMBIFW3dQ=; b=VuKoBQnH+GZ7GJoCol41Y/09N6bhuCSX7pEMI/VXkE3+SuFm6OnycOMV YI7BvLwYtdysmyn78Aa2af8nFzN9DqJsd+sYT2IMmi0dOXO10YVfYHGP5 TrcnUNIqT2Dt++3RVX22wmjzTzD2yCi2zvVhZzDj/QuxDqZIpmK/l4Rsr 11hpyhniL0YGNRZWhVG+2jalGcZufbFjGs6/nA3KGY3sodqLQkx4qVedn rScn0bQgBIKwS1OaYkEoQIH4ao0ezeBk3wX2tWvrcSNcnogQKm2fYyqRb Ihqs40Qmhj2fAOUKbxLhDnzcXNZWvQ3hp0frpWxHLs7AISqQaj3KAM90K w==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012383" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012383" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:22 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853589" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853589" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:21 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:44 +0800 Message-ID: <20231205065034.1621-2-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 2/9] avcodec: add D3D12VA hardware accelerated H264 decoding X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: DspTkaGRRf24 From: Wu Jianhua The implementation is based on: https://learn.microsoft.com/en-us/windows/win32/medfound/direct3d-12-video-overview With the Direct3D 12 video decoding support, we can render or process the decoded images by the pixel shaders or compute shaders directly without the extra copy overhead, which is beneficial especially if you are trying to render or post-process a 4K or 8K video. The command below is how to enable d3d12va: ffmpeg -hwaccel d3d12va -i input.mp4 output.mp4 Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- configure | 2 + libavcodec/Makefile | 3 + libavcodec/d3d11va.h | 3 - libavcodec/d3d12va_decode.c | 538 ++++++++++++++++++++++++++++++++++++ libavcodec/d3d12va_decode.h | 179 ++++++++++++ libavcodec/d3d12va_h264.c | 207 ++++++++++++++ libavcodec/dxva2.c | 29 +- libavcodec/dxva2.h | 3 - libavcodec/dxva2_av1.c | 4 +- libavcodec/dxva2_h264.c | 26 +- libavcodec/dxva2_hevc.c | 6 +- libavcodec/dxva2_internal.h | 74 ++--- libavcodec/dxva2_mpeg2.c | 6 +- libavcodec/dxva2_vc1.c | 6 +- libavcodec/dxva2_vp9.c | 6 +- libavcodec/h264_slice.c | 4 + libavcodec/h264dec.c | 3 + libavcodec/hwaccels.h | 1 + libavcodec/hwconfig.h | 2 + 19 files changed, 1036 insertions(+), 66 deletions(-) create mode 100644 libavcodec/d3d12va_decode.c create mode 100644 libavcodec/d3d12va_decode.h create mode 100644 libavcodec/d3d12va_h264.c diff --git a/configure b/configure index 1a343ac6b9..90b20da49a 100755 --- a/configure +++ b/configure @@ -3082,6 +3082,8 @@ h264_d3d11va_hwaccel_deps="d3d11va" h264_d3d11va_hwaccel_select="h264_decoder" h264_d3d11va2_hwaccel_deps="d3d11va" h264_d3d11va2_hwaccel_select="h264_decoder" +h264_d3d12va_hwaccel_deps="d3d12va" +h264_d3d12va_hwaccel_select="h264_decoder" h264_dxva2_hwaccel_deps="dxva2" h264_dxva2_hwaccel_select="h264_decoder" h264_nvdec_hwaccel_deps="nvdec" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index fd9883d2ca..ead3a4480e 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -986,6 +986,7 @@ OBJS-$(CONFIG_ADPCM_ZORK_DECODER) += adpcm.o adpcm_data.o # hardware accelerators OBJS-$(CONFIG_D3D11VA) += dxva2.o +OBJS-$(CONFIG_D3D12VA) += dxva2.o d3d12va_decode.o OBJS-$(CONFIG_DXVA2) += dxva2.o OBJS-$(CONFIG_NVDEC) += nvdec.o OBJS-$(CONFIG_VAAPI) += vaapi_decode.o @@ -1003,6 +1004,7 @@ OBJS-$(CONFIG_H263_VAAPI_HWACCEL) += vaapi_mpeg4.o OBJS-$(CONFIG_H263_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o OBJS-$(CONFIG_H264_D3D11VA_HWACCEL) += dxva2_h264.o OBJS-$(CONFIG_H264_DXVA2_HWACCEL) += dxva2_h264.o +OBJS-$(CONFIG_H264_D3D12VA_HWACCEL) += dxva2_h264.o d3d12va_h264.o OBJS-$(CONFIG_H264_NVDEC_HWACCEL) += nvdec_h264.o OBJS-$(CONFIG_H264_QSV_HWACCEL) += qsvdec.o OBJS-$(CONFIG_H264_VAAPI_HWACCEL) += vaapi_h264.o @@ -1296,6 +1298,7 @@ SKIPHEADERS += %_tablegen.h \ SKIPHEADERS-$(CONFIG_AMF) += amfenc.h SKIPHEADERS-$(CONFIG_D3D11VA) += d3d11va.h dxva2_internal.h +SKIPHEADERS-$(CONFIG_D3D12VA) += d3d12va_decode.h SKIPHEADERS-$(CONFIG_DXVA2) += dxva2.h dxva2_internal.h SKIPHEADERS-$(CONFIG_JNI) += ffjni.h SKIPHEADERS-$(CONFIG_LCMS2) += fflcms2.h diff --git a/libavcodec/d3d11va.h b/libavcodec/d3d11va.h index 6816b6c1e6..27f40e5519 100644 --- a/libavcodec/d3d11va.h +++ b/libavcodec/d3d11va.h @@ -45,9 +45,6 @@ * @{ */ -#define FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG 1 ///< Work around for Direct3D11 and old UVD/UVD+ ATI video cards -#define FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO 2 ///< Work around for Direct3D11 and old Intel GPUs with ClearVideo interface - /** * This structure is used to provides the necessary configurations and data * to the Direct3D11 FFmpeg HWAccel implementation. diff --git a/libavcodec/d3d12va_decode.c b/libavcodec/d3d12va_decode.c new file mode 100644 index 0000000000..03e565066c --- /dev/null +++ b/libavcodec/d3d12va_decode.c @@ -0,0 +1,538 @@ +/* + * Direct3D 12 HW acceleration video decoder + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include +#include + +#include "libavutil/common.h" +#include "libavutil/log.h" +#include "libavutil/time.h" +#include "libavutil/imgutils.h" +#include "libavutil/hwcontext_d3d12va_internal.h" +#include "libavutil/hwcontext_d3d12va.h" +#include "avcodec.h" +#include "decode.h" +#include "d3d12va_decode.h" + +typedef struct HelperObjects { + ID3D12CommandAllocator *command_allocator; + ID3D12Resource *buffer; + uint64_t fence_value; +} HelperObjects; + +int ff_d3d12va_get_suitable_max_bitstream_size(AVCodecContext *avctx) +{ + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + return av_image_get_buffer_size(frames_ctx->sw_format, avctx->coded_width, avctx->coded_height, 1); +} + +unsigned ff_d3d12va_get_surface_index(const AVCodecContext *avctx, + D3D12VADecodeContext *ctx, const AVFrame *frame, + int curr) +{ + AVD3D12VAFrame *f; + ID3D12Resource *res; + unsigned i; + + f = (AVD3D12VAFrame *)frame->data[0]; + if (!f) + goto fail; + + res = f->texture; + if (!res) + goto fail; + + if (!curr) { + for (i = 0; i < ctx->max_num_ref; i++) { + if (ctx->ref_resources[i] && res == ctx->ref_resources[i]) { + ctx->used_mask |= 1 << i; + return i; + } + } + } else { + for (i = 0; i < ctx->max_num_ref; i++) { + if (!((ctx->used_mask >> i) & 0x1)) { + ctx->ref_resources[i] = res; + return i; + } + } + } + +fail: + assert(0); + return 0; +} + +static int d3d12va_get_valid_helper_objects(AVCodecContext *avctx, ID3D12CommandAllocator **ppAllocator, + ID3D12Resource **ppBuffer) +{ + HRESULT hr; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + HelperObjects obj = { 0 }; + D3D12_HEAP_PROPERTIES heap_props = { .Type = D3D12_HEAP_TYPE_UPLOAD }; + + D3D12_RESOURCE_DESC desc = { + .Dimension = D3D12_RESOURCE_DIMENSION_BUFFER, + .Alignment = D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT, + .Width = ctx->bitstream_size, + .Height = 1, + .DepthOrArraySize = 1, + .MipLevels = 1, + .Format = DXGI_FORMAT_UNKNOWN, + .SampleDesc = { .Count = 1, .Quality = 0 }, + .Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR, + .Flags = D3D12_RESOURCE_FLAG_NONE, + }; + + if (av_fifo_peek(ctx->objects_queue, &obj, 1, 0) >= 0) { + uint64_t completion = ID3D12Fence_GetCompletedValue(ctx->sync_ctx.fence); + if (completion >= obj.fence_value) { + *ppAllocator = obj.command_allocator; + *ppBuffer = obj.buffer; + av_fifo_read(ctx->objects_queue, &obj, 1); + return 0; + } + } + + hr = ID3D12Device_CreateCommandAllocator(ctx->device_ctx->device, D3D12_COMMAND_LIST_TYPE_VIDEO_DECODE, + &IID_ID3D12CommandAllocator, (void **)ppAllocator); + if (FAILED(hr)) { + av_log(avctx, AV_LOG_ERROR, "Failed to create a new command allocator!\n"); + return AVERROR(EINVAL); + } + + hr = ID3D12Device_CreateCommittedResource(ctx->device_ctx->device, &heap_props, D3D12_HEAP_FLAG_NONE, + &desc, D3D12_RESOURCE_STATE_GENERIC_READ, NULL, + &IID_ID3D12Resource, (void **)ppBuffer); + + if (FAILED(hr)) { + av_log(avctx, AV_LOG_ERROR, "Failed to create a new d3d12 buffer!\n"); + return AVERROR(EINVAL); + } + + return 0; +} + +static int d3d12va_discard_helper_objects(AVCodecContext *avctx, ID3D12CommandAllocator *pAllocator, + ID3D12Resource *pBuffer, uint64_t fence_value) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + + HelperObjects obj = { + .command_allocator = pAllocator, + .buffer = pBuffer, + .fence_value = fence_value, + }; + + if (av_fifo_write(ctx->objects_queue, &obj, 1) < 0) { + D3D12_OBJECT_RELEASE(pAllocator); + D3D12_OBJECT_RELEASE(pBuffer); + return AVERROR(ENOMEM); + } + + return 0; +} + +static int d3d12va_fence_completion(AVD3D12VASyncContext *psync_ctx) +{ + uint64_t completion = ID3D12Fence_GetCompletedValue(psync_ctx->fence); + if (completion < psync_ctx->fence_value) { + if (FAILED(ID3D12Fence_SetEventOnCompletion(psync_ctx->fence, psync_ctx->fence_value, psync_ctx->event))) + return AVERROR(EINVAL); + + WaitForSingleObjectEx(psync_ctx->event, INFINITE, FALSE); + } + + return 0; +} + +static void bufref_free_interface(void *opaque, uint8_t *data) +{ + D3D12_OBJECT_RELEASE(opaque); +} + +static AVBufferRef *bufref_wrap_interface(IUnknown *iface) +{ + return av_buffer_create((uint8_t*)iface, 1, bufref_free_interface, iface, 0); +} + +static int d3d12va_sync_with_gpu(AVCodecContext *avctx) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + + DX_CHECK(ID3D12CommandQueue_Signal(ctx->command_queue, ctx->sync_ctx.fence, ++ctx->sync_ctx.fence_value)); + return d3d12va_fence_completion(&ctx->sync_ctx); + +fail: + return AVERROR(EINVAL); +} + +static int d3d12va_create_decoder_heap(AVCodecContext *avctx) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VADeviceContext *device_hwctx = ctx->device_ctx; + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + + D3D12_VIDEO_DECODER_HEAP_DESC desc = { + .NodeMask = 0, + .Configuration = ctx->cfg, + .DecodeWidth = frames_ctx->width, + .DecodeHeight = frames_ctx->height, + .Format = frames_hwctx->format, + .FrameRate = { avctx->framerate.num, avctx->framerate.den }, + .BitRate = avctx->bit_rate, + .MaxDecodePictureBufferCount = ctx->max_num_ref, + }; + + DX_CHECK(ID3D12VideoDevice_CreateVideoDecoderHeap(device_hwctx->video_device, &desc, + &IID_ID3D12VideoDecoderHeap, (void **)&ctx->decoder_heap)); + + return 0; + +fail: + if (ctx->decoder) { + av_log(avctx, AV_LOG_ERROR, "D3D12 doesn't support decoding frames with an extent " + "[width(%d), height(%d)], on your device!\n", frames_ctx->width, frames_ctx->height); + } + + return AVERROR(EINVAL); +} + +static int d3d12va_create_decoder(AVCodecContext *avctx) +{ + D3D12_VIDEO_DECODER_DESC desc; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VADeviceContext *device_hwctx = ctx->device_ctx; + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + + D3D12_FEATURE_DATA_VIDEO_DECODE_SUPPORT feature = { + .NodeIndex = 0, + .Configuration = ctx->cfg, + .Width = frames_ctx->width, + .Height = frames_ctx->height, + .DecodeFormat = frames_hwctx->format, + .FrameRate = { avctx->framerate.num, avctx->framerate.den }, + .BitRate = avctx->bit_rate, + }; + + DX_CHECK(ID3D12VideoDevice_CheckFeatureSupport(device_hwctx->video_device, D3D12_FEATURE_VIDEO_DECODE_SUPPORT, + &feature, sizeof(feature))); + if (!(feature.SupportFlags & D3D12_VIDEO_DECODE_SUPPORT_FLAG_SUPPORTED) || + !(feature.DecodeTier >= D3D12_VIDEO_DECODE_TIER_2)) { + av_log(avctx, AV_LOG_ERROR, "D3D12 decoder doesn't support on this device\n"); + return AVERROR(EINVAL); + } + + desc = (D3D12_VIDEO_DECODER_DESC) { + .NodeMask = 0, + .Configuration = ctx->cfg, + }; + + DX_CHECK(ID3D12VideoDevice_CreateVideoDecoder(device_hwctx->video_device, &desc, &IID_ID3D12VideoDecoder, + (void **)&ctx->decoder)); + + ctx->decoder_ref = bufref_wrap_interface((IUnknown *)ctx->decoder); + if (!ctx->decoder_ref) + return AVERROR(ENOMEM); + + return 0; + +fail: + return AVERROR(EINVAL); +} + +int ff_d3d12va_common_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx) +{ + AVHWFramesContext *frames_ctx = (AVHWFramesContext *)hw_frames_ctx->data; + AVHWDeviceContext *device_ctx = frames_ctx->device_ctx; + + frames_ctx->format = AV_PIX_FMT_D3D12; + frames_ctx->sw_format = avctx->sw_pix_fmt == AV_PIX_FMT_YUV420P10 ? AV_PIX_FMT_P010 : AV_PIX_FMT_NV12; + frames_ctx->width = avctx->width; + frames_ctx->height = avctx->height; + + return 0; +} + +int ff_d3d12va_decode_init(AVCodecContext *avctx) +{ + int ret; + AVHWFramesContext *frames_ctx; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + ID3D12Resource *buffer = NULL; + ID3D12CommandAllocator *command_allocator = NULL; + D3D12_COMMAND_QUEUE_DESC queue_desc = { + .Type = D3D12_COMMAND_LIST_TYPE_VIDEO_DECODE, + .Priority = 0, + .Flags = D3D12_COMMAND_QUEUE_FLAG_NONE, + .NodeMask = 0, + }; + + ctx->pix_fmt = avctx->hwaccel->pix_fmt; + + ret = ff_decode_get_hw_frames_ctx(avctx, AV_HWDEVICE_TYPE_D3D12VA); + if (ret < 0) + return ret; + + frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + ctx->device_ctx = (AVD3D12VADeviceContext *)frames_ctx->device_ctx->hwctx; + + if (frames_ctx->format != ctx->pix_fmt) { + av_log(avctx, AV_LOG_ERROR, "Invalid pixfmt for hwaccel!\n"); + goto fail; + } + + ret = d3d12va_create_decoder(avctx); + if (ret < 0) + goto fail; + + ret = d3d12va_create_decoder_heap(avctx); + if (ret < 0) + goto fail; + + ctx->bitstream_size = ff_d3d12va_get_suitable_max_bitstream_size(avctx); + + ctx->ref_resources = av_calloc(ctx->max_num_ref, sizeof(*ctx->ref_resources)); + if (!ctx->ref_resources) + return AVERROR(ENOMEM); + + ctx->ref_subresources = av_calloc(ctx->max_num_ref, sizeof(*ctx->ref_subresources)); + if (!ctx->ref_subresources) + return AVERROR(ENOMEM); + + ctx->objects_queue = av_fifo_alloc2(D3D12VA_VIDEO_DEC_ASYNC_DEPTH, + sizeof(HelperObjects), AV_FIFO_FLAG_AUTO_GROW); + if (!ctx->objects_queue) + return AVERROR(ENOMEM); + + + DX_CHECK(ID3D12Device_CreateFence(ctx->device_ctx->device, 0, D3D12_FENCE_FLAG_NONE, + &IID_ID3D12Fence, (void **)&ctx->sync_ctx.fence)); + + ctx->sync_ctx.event = CreateEvent(NULL, FALSE, FALSE, NULL); + if (!ctx->sync_ctx.event) + goto fail; + + ret = d3d12va_get_valid_helper_objects(avctx, &command_allocator, &buffer); + if (ret < 0) + goto fail; + + DX_CHECK(ID3D12Device_CreateCommandQueue(ctx->device_ctx->device, &queue_desc, + &IID_ID3D12CommandQueue, (void **)&ctx->command_queue)); + + DX_CHECK(ID3D12Device_CreateCommandList(ctx->device_ctx->device, 0, queue_desc.Type, + command_allocator, NULL, &IID_ID3D12CommandList, (void **)&ctx->command_list)); + + DX_CHECK(ID3D12VideoDecodeCommandList_Close(ctx->command_list)); + + ID3D12CommandQueue_ExecuteCommandLists(ctx->command_queue, 1, (ID3D12CommandList **)&ctx->command_list); + + ret = d3d12va_sync_with_gpu(avctx); + if (ret < 0) + goto fail; + + d3d12va_discard_helper_objects(avctx, command_allocator, buffer, ctx->sync_ctx.fence_value); + if (ret < 0) + goto fail; + + return 0; + +fail: + D3D12_OBJECT_RELEASE(command_allocator); + D3D12_OBJECT_RELEASE(buffer); + ff_d3d12va_decode_uninit(avctx); + + return AVERROR(EINVAL); +} + +int ff_d3d12va_decode_uninit(AVCodecContext *avctx) +{ + int num_allocator = 0; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + HelperObjects obj; + + if (ctx->sync_ctx.fence) + d3d12va_sync_with_gpu(avctx); + + av_freep(&ctx->ref_resources); + av_freep(&ctx->ref_subresources); + + D3D12_OBJECT_RELEASE(ctx->command_list); + D3D12_OBJECT_RELEASE(ctx->command_queue); + + if (ctx->objects_queue) { + while (av_fifo_read(ctx->objects_queue, &obj, 1) >= 0) { + num_allocator++; + D3D12_OBJECT_RELEASE(obj.buffer); + D3D12_OBJECT_RELEASE(obj.command_allocator); + } + + av_log(avctx, AV_LOG_VERBOSE, "Total number of command allocators reused: %d\n", num_allocator); + } + + av_fifo_freep2(&ctx->objects_queue); + + D3D12_OBJECT_RELEASE(ctx->sync_ctx.fence); + if (ctx->sync_ctx.event) + CloseHandle(ctx->sync_ctx.event); + + D3D12_OBJECT_RELEASE(ctx->decoder_heap); + + av_buffer_unref(&ctx->decoder_ref); + + return 0; +} + +static inline int d3d12va_update_reference_frames_state(AVCodecContext *avctx, D3D12_RESOURCE_BARRIER *barriers, int state_before, int state_end) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + + int num_barrier = 0; + for (int i = 0; i < ctx->max_num_ref; i++) { + if (((ctx->used_mask >> i) & 0x1) && ctx->ref_resources[i]) { + barriers[num_barrier].Type = D3D12_RESOURCE_BARRIER_TYPE_TRANSITION; + barriers[num_barrier].Flags = D3D12_RESOURCE_BARRIER_FLAG_NONE; + barriers[num_barrier].Transition = (D3D12_RESOURCE_TRANSITION_BARRIER){ + .pResource = ctx->ref_resources[i], + .Subresource = D3D12_RESOURCE_BARRIER_ALL_SUBRESOURCES, + .StateBefore = state_before, + .StateAfter = state_end, + }; + num_barrier++; + } + } + + return num_barrier; +} + +int ff_d3d12va_common_end_frame(AVCodecContext *avctx, AVFrame *frame, + const void *pp, unsigned pp_size, + const void *qm, unsigned qm_size, + int(*update_input_arguments)(AVCodecContext *, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *, ID3D12Resource *)) +{ + int ret; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + ID3D12Resource *buffer = NULL; + ID3D12CommandAllocator *command_allocator = NULL; + AVD3D12VAFrame *f = (AVD3D12VAFrame *)frame->data[0]; + ID3D12Resource *resource = (ID3D12Resource *)f->texture; + + ID3D12VideoDecodeCommandList *cmd_list = ctx->command_list; + D3D12_RESOURCE_BARRIER barriers[32] = { 0 }; + + D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS input_args = { + .NumFrameArguments = 2, + .FrameArguments = { + [0] = { + .Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_PICTURE_PARAMETERS, + .Size = pp_size, + .pData = (void *)pp, + }, + [1] = { + .Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_INVERSE_QUANTIZATION_MATRIX, + .Size = qm_size, + .pData = (void *)qm, + }, + }, + .pHeap = ctx->decoder_heap, + }; + + D3D12_VIDEO_DECODE_OUTPUT_STREAM_ARGUMENTS output_args = { + .ConversionArguments = { 0 }, + .OutputSubresource = 0, + .pOutputTexture2D = resource, + }; + + UINT num_barrier = 1; + barriers[0] = (D3D12_RESOURCE_BARRIER) { + .Type = D3D12_RESOURCE_BARRIER_TYPE_TRANSITION, + .Flags = D3D12_RESOURCE_BARRIER_FLAG_NONE, + .Transition = { + .pResource = resource, + .Subresource = D3D12_RESOURCE_BARRIER_ALL_SUBRESOURCES, + .StateBefore = D3D12_RESOURCE_STATE_COMMON, + .StateAfter = D3D12_RESOURCE_STATE_VIDEO_DECODE_WRITE, + }, + }; + + memset(ctx->ref_subresources, 0, sizeof(UINT) * ctx->max_num_ref); + input_args.ReferenceFrames.NumTexture2Ds = ctx->max_num_ref; + input_args.ReferenceFrames.ppTexture2Ds = ctx->ref_resources; + input_args.ReferenceFrames.pSubresources = ctx->ref_subresources; + + ret = d3d12va_fence_completion(&f->sync_ctx); + if (ret < 0) + goto fail; + + if (!qm) + input_args.NumFrameArguments = 1; + + ret = d3d12va_get_valid_helper_objects(avctx, &command_allocator, &buffer); + if (ret < 0) + goto fail; + + ret = update_input_arguments(avctx, &input_args, buffer); + if (ret < 0) + goto fail; + + DX_CHECK(ID3D12CommandAllocator_Reset(command_allocator)); + + DX_CHECK(ID3D12VideoDecodeCommandList_Reset(cmd_list, command_allocator)); + + num_barrier += d3d12va_update_reference_frames_state(avctx, &barriers[1], D3D12_RESOURCE_STATE_COMMON, D3D12_RESOURCE_STATE_VIDEO_DECODE_READ); + + ID3D12VideoDecodeCommandList_ResourceBarrier(cmd_list, num_barrier, barriers); + + ID3D12VideoDecodeCommandList_DecodeFrame(cmd_list, ctx->decoder, &output_args, &input_args); + + barriers[0].Transition.StateBefore = barriers[0].Transition.StateAfter; + barriers[0].Transition.StateAfter = D3D12_RESOURCE_STATE_COMMON; + d3d12va_update_reference_frames_state(avctx, &barriers[1], D3D12_RESOURCE_STATE_VIDEO_DECODE_READ, D3D12_RESOURCE_STATE_COMMON); + + ID3D12VideoDecodeCommandList_ResourceBarrier(cmd_list, num_barrier, barriers); + + DX_CHECK(ID3D12VideoDecodeCommandList_Close(cmd_list)); + + ID3D12CommandQueue_ExecuteCommandLists(ctx->command_queue, 1, (ID3D12CommandList **)&ctx->command_list); + + DX_CHECK(ID3D12CommandQueue_Signal(ctx->command_queue, f->sync_ctx.fence, ++f->sync_ctx.fence_value)); + + DX_CHECK(ID3D12CommandQueue_Signal(ctx->command_queue, ctx->sync_ctx.fence, ++ctx->sync_ctx.fence_value)); + + ret = d3d12va_discard_helper_objects(avctx, command_allocator, buffer, ctx->sync_ctx.fence_value); + if (ret < 0) + return ret; + + return 0; + +fail: + if (command_allocator) + d3d12va_discard_helper_objects(avctx, command_allocator, buffer, ctx->sync_ctx.fence_value); + return AVERROR(EINVAL); +} diff --git a/libavcodec/d3d12va_decode.h b/libavcodec/d3d12va_decode.h new file mode 100644 index 0000000000..b64994760a --- /dev/null +++ b/libavcodec/d3d12va_decode.h @@ -0,0 +1,179 @@ +/* + * Direct3D 12 HW acceleration video decoder + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_D3D12VA_DECODE_H +#define AVCODEC_D3D12VA_DECODE_H + +#include "libavutil/fifo.h" +#include "libavutil/hwcontext.h" +#include "libavutil/hwcontext_d3d12va.h" +#include "avcodec.h" +#include "internal.h" +#include "hwaccel_internal.h" + +/** + * @brief This structure is used to provide the necessary configurations and data + * to the FFmpeg Direct3D 12 HWAccel implementation for video decoder. + */ +typedef struct D3D12VADecodeContext { + AVBufferRef *decoder_ref; + + /** + * D3D12 video decoder + */ + ID3D12VideoDecoder *decoder; + + /** + * D3D12 video decoder heap + */ + ID3D12VideoDecoderHeap *decoder_heap; + + /** + * D3D12 configuration used to create the decoder + * + * Specified by decoders + */ + D3D12_VIDEO_DECODE_CONFIGURATION cfg; + + /** + * A cached queue for reusing the D3D12 command allocators and upload buffers + * + * @see https://learn.microsoft.com/en-us/windows/win32/direct3d12/recording-command-lists-and-bundles#id3d12commandallocator + */ + AVFifo *objects_queue; + + /** + * D3D12 command queue + */ + ID3D12CommandQueue *command_queue; + + /** + * D3D12 video decode command list + */ + ID3D12VideoDecodeCommandList *command_list; + + /** + * The array of resources used for reference frames + * + * The ref_resources.length is the same as D3D12VADecodeContext.max_num_ref + */ + ID3D12Resource **ref_resources; + + /** + * The array of subresources used for reference frames + * + * The ref_subresources.length is the same as D3D12VADecodeContext.max_num_ref + */ + UINT *ref_subresources; + + /** + * Maximum number of reference frames + */ + UINT max_num_ref; + + /** + * Used mask used to record reference frames indices + */ + UINT used_mask; + + /** + * Bitstream size for each frame + */ + UINT bitstream_size; + + /** + * The sync context used to sync command queue + */ + AVD3D12VASyncContext sync_ctx; + + /** + * A pointer to AVD3D12VADeviceContext used to create D3D12 objects + */ + AVD3D12VADeviceContext *device_ctx; + + /** + * Pixel format + */ + enum AVPixelFormat pix_fmt; + + /** + * Private to the FFmpeg AVHWAccel implementation + */ + unsigned report_id; +} D3D12VADecodeContext; + +/** + * @} + */ +#define D3D12VA_VIDEO_DEC_ASYNC_DEPTH 36 +#define D3D12VA_DECODE_CONTEXT(avctx) ((D3D12VADecodeContext *)((avctx)->internal->hwaccel_priv_data)) +#define D3D12VA_FRAMES_CONTEXT(avctx) ((AVHWFramesContext *)(avctx)->hw_frames_ctx->data) + +/** + * @brief Get a suitable maximum bitstream size + * + * Creating and destroying a resource on d3d12 needs sync and reallocation, so use this function + * to help allocate a big enough bitstream buffer to avoid recreating resources when decoding. + * + * @return the suitable size + */ +int ff_d3d12va_get_suitable_max_bitstream_size(AVCodecContext *avctx); + +/** + * @brief init D3D12VADecodeContext + * + * @return Error code (ret < 0 if failed) + */ +int ff_d3d12va_decode_init(AVCodecContext *avctx); + +/** + * @brief uninit D3D12VADecodeContext + * + * @return Error code (ret < 0 if failed) + */ +int ff_d3d12va_decode_uninit(AVCodecContext *avctx); + +/** + * @brief d3d12va common frame params + * + * @return Error code (ret < 0 if failed) + */ +int ff_d3d12va_common_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx); + +/** + * @brief d3d12va common end frame + * + * @param avctx codec context + * @param frame current output frame + * @param pp picture parameters + * @param pp_size the size of the picture parameters + * @param qm quantization matrix + * @param qm_size the size of the quantization matrix + * @param callback update decoder-specified input stream arguments + * @return Error code (ret < 0 if failed) + */ +int ff_d3d12va_common_end_frame(AVCodecContext *avctx, AVFrame *frame, + const void *pp, unsigned pp_size, + const void *qm, unsigned qm_size, + int(*)(AVCodecContext *, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *, ID3D12Resource *)); + +#endif /* AVCODEC_D3D12VA_DEC_H */ diff --git a/libavcodec/d3d12va_h264.c b/libavcodec/d3d12va_h264.c new file mode 100644 index 0000000000..5a6d6852c8 --- /dev/null +++ b/libavcodec/d3d12va_h264.c @@ -0,0 +1,207 @@ +/* + * Direct3D 12 h264 HW acceleration + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config_components.h" +#include "libavutil/avassert.h" +#include "h264dec.h" +#include "h264data.h" +#include "h264_ps.h" +#include "mpegutils.h" +#include "dxva2_internal.h" +#include "d3d12va_decode.h" +#include "libavutil/hwcontext_d3d12va_internal.h" +#include + +typedef struct H264DecodePictureContext { + DXVA_PicParams_H264 pp; + DXVA_Qmatrix_H264 qm; + unsigned slice_count; + DXVA_Slice_H264_Short slice_short[MAX_SLICES]; + const uint8_t *bitstream; + unsigned bitstream_size; +} H264DecodePictureContext; + +static void fill_slice_short(DXVA_Slice_H264_Short *slice, + unsigned position, unsigned size) +{ + memset(slice, 0, sizeof(*slice)); + slice->BSNALunitDataLocation = position; + slice->SliceBytesInBuffer = size; + slice->wBadSliceChopping = 0; +} + +static int d3d12va_h264_start_frame(AVCodecContext *avctx, + av_unused const uint8_t *buffer, + av_unused uint32_t size) +{ + const H264Context *h = avctx->priv_data; + H264DecodePictureContext *ctx_pic = h->cur_pic_ptr->hwaccel_picture_private; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + + if (!ctx) + return -1; + + assert(ctx_pic); + + ctx->used_mask = 0; + + ff_dxva2_h264_fill_picture_parameters(avctx, (AVDXVAContext *)ctx, &ctx_pic->pp); + + ff_dxva2_h264_fill_scaling_lists(avctx, (AVDXVAContext *)ctx, &ctx_pic->qm); + + ctx_pic->slice_count = 0; + ctx_pic->bitstream_size = 0; + ctx_pic->bitstream = NULL; + + return 0; +} + +static int d3d12va_h264_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + unsigned position; + const H264Context *h = avctx->priv_data; + const H264SliceContext *sl = &h->slice_ctx[0]; + const H264Picture *current_picture = h->cur_pic_ptr; + H264DecodePictureContext *ctx_pic = current_picture->hwaccel_picture_private; + + if (ctx_pic->slice_count >= MAX_SLICES) + return AVERROR(ERANGE); + + if (!ctx_pic->bitstream) + ctx_pic->bitstream = buffer; + ctx_pic->bitstream_size += size; + + position = buffer - ctx_pic->bitstream; + fill_slice_short(&ctx_pic->slice_short[ctx_pic->slice_count], position, size); + ctx_pic->slice_count++; + + if (sl->slice_type != AV_PICTURE_TYPE_I && sl->slice_type != AV_PICTURE_TYPE_SI) + ctx_pic->pp.wBitFields &= ~(1 << 15); /* Set IntraPicFlag to 0 */ + + return 0; +} + +#define START_CODE 65536 +#define START_CODE_SIZE 3 +static int update_input_arguments(AVCodecContext *avctx, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *input_args, ID3D12Resource *buffer) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + + const H264Context *h = avctx->priv_data; + const H264Picture *current_picture = h->cur_pic_ptr; + H264DecodePictureContext *ctx_pic = current_picture->hwaccel_picture_private; + + int i; + uint8_t *mapped_data, *mapped_ptr; + DXVA_Slice_H264_Short *slice; + D3D12_VIDEO_DECODE_FRAME_ARGUMENT *args; + + if (FAILED(ID3D12Resource_Map(buffer, 0, NULL, &mapped_data))) { + av_log(avctx, AV_LOG_ERROR, "Failed to map D3D12 Buffer resource!\n"); + return AVERROR(EINVAL); + } + + mapped_ptr = mapped_data; + for (i = 0; i < ctx_pic->slice_count; i++) { + UINT position, size; + slice = &ctx_pic->slice_short[i]; + + position = slice->BSNALunitDataLocation; + size = slice->SliceBytesInBuffer; + + slice->SliceBytesInBuffer += START_CODE_SIZE; + slice->BSNALunitDataLocation = mapped_ptr - mapped_data; + + *(uint32_t *)mapped_ptr = START_CODE; + mapped_ptr += START_CODE_SIZE; + + memcpy(mapped_ptr, &ctx_pic->bitstream[position], size); + mapped_ptr += size; + } + + ID3D12Resource_Unmap(buffer, 0, NULL); + + input_args->CompressedBitstream = (D3D12_VIDEO_DECODE_COMPRESSED_BITSTREAM){ + .pBuffer = buffer, + .Offset = 0, + .Size = mapped_ptr - mapped_data, + }; + + args = &input_args->FrameArguments[input_args->NumFrameArguments++]; + args->Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_SLICE_CONTROL; + args->Size = sizeof(DXVA_Slice_H264_Short) * ctx_pic->slice_count; + args->pData = ctx_pic->slice_short; + + return 0; +} + +static int d3d12va_h264_end_frame(AVCodecContext *avctx) +{ + H264Context *h = avctx->priv_data; + H264DecodePictureContext *ctx_pic = h->cur_pic_ptr->hwaccel_picture_private; + H264SliceContext *sl = &h->slice_ctx[0]; + + int ret; + + if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0) + return -1; + + ret = ff_d3d12va_common_end_frame(avctx, h->cur_pic_ptr->f, + &ctx_pic->pp, sizeof(ctx_pic->pp), + &ctx_pic->qm, sizeof(ctx_pic->qm), + update_input_arguments); + if (!ret) + ff_h264_draw_horiz_band(h, sl, 0, h->avctx->height); + + return ret; +} + +static int d3d12va_h264_decode_init(AVCodecContext *avctx) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + DXVA_PicParams_H264 pp; + + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_H264; + + ctx->max_num_ref = FF_ARRAY_ELEMS(pp.RefFrameList) + 1; + + return ff_d3d12va_decode_init(avctx); +} + +#if CONFIG_H264_D3D12VA_HWACCEL +const FFHWAccel ff_h264_d3d12va_hwaccel = { + .p.name = "h264_d3d12va", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_H264, + .p.pix_fmt = AV_PIX_FMT_D3D12, + .init = d3d12va_h264_decode_init, + .uninit = ff_d3d12va_decode_uninit, + .start_frame = d3d12va_h264_start_frame, + .decode_slice = d3d12va_h264_decode_slice, + .end_frame = d3d12va_h264_end_frame, + .frame_params = ff_d3d12va_common_frame_params, + .frame_priv_data_size = sizeof(H264DecodePictureContext), + .priv_data_size = sizeof(D3D12VADecodeContext), +}; +#endif diff --git a/libavcodec/dxva2.c b/libavcodec/dxva2.c index d7bc587562..7160a0008b 100644 --- a/libavcodec/dxva2.c +++ b/libavcodec/dxva2.c @@ -768,12 +768,17 @@ static void *get_surface(const AVCodecContext *avctx, const AVFrame *frame) } unsigned ff_dxva2_get_surface_index(const AVCodecContext *avctx, - const AVDXVAContext *ctx, - const AVFrame *frame) + AVDXVAContext *ctx, const AVFrame *frame, + int curr) { void *surface = get_surface(avctx, frame); unsigned i; +#if CONFIG_D3D12VA + if (avctx->pix_fmt == AV_PIX_FMT_D3D12) { + return ff_d3d12va_get_surface_index(avctx, (D3D12VADecodeContext *)ctx, frame, curr); + } +#endif #if CONFIG_D3D11VA if (avctx->pix_fmt == AV_PIX_FMT_D3D11) return (intptr_t)frame->data[1]; @@ -1056,3 +1061,23 @@ int ff_dxva2_is_d3d11(const AVCodecContext *avctx) else return 0; } + +unsigned *ff_dxva2_get_report_id(const AVCodecContext *avctx, AVDXVAContext *ctx) +{ + unsigned *report_id = NULL; + +#if CONFIG_D3D12VA + if (avctx->pix_fmt == AV_PIX_FMT_D3D12) + report_id = &ctx->d3d12va.report_id; +#endif +#if CONFIG_D3D11VA + if (ff_dxva2_is_d3d11(avctx)) + report_id = &ctx->d3d11va.report_id; +#endif +#if CONFIG_DXVA2 + if (avctx->pix_fmt == AV_PIX_FMT_DXVA2_VLD) + report_id = &ctx->dxva2.report_id; +#endif + + return report_id; +} diff --git a/libavcodec/dxva2.h b/libavcodec/dxva2.h index 22c93992f2..bdec6112e9 100644 --- a/libavcodec/dxva2.h +++ b/libavcodec/dxva2.h @@ -45,9 +45,6 @@ * @{ */ -#define FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG 1 ///< Work around for DXVA2 and old UVD/UVD+ ATI video cards -#define FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO 2 ///< Work around for DXVA2 and old Intel GPUs with ClearVideo interface - /** * This structure is used to provides the necessary configurations and data * to the DXVA2 FFmpeg HWAccel implementation. diff --git a/libavcodec/dxva2_av1.c b/libavcodec/dxva2_av1.c index ab118a4356..12a895b791 100644 --- a/libavcodec/dxva2_av1.c +++ b/libavcodec/dxva2_av1.c @@ -75,7 +75,7 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c pp->max_width = seq->max_frame_width_minus_1 + 1; pp->max_height = seq->max_frame_height_minus_1 + 1; - pp->CurrPicTextureIndex = ff_dxva2_get_surface_index(avctx, ctx, h->cur_frame.f); + pp->CurrPicTextureIndex = ff_dxva2_get_surface_index(avctx, ctx, h->cur_frame.f, 1); pp->superres_denom = frame_header->use_superres ? frame_header->coded_denom + AV1_SUPERRES_DENOM_MIN : AV1_SUPERRES_NUM; pp->bitdepth = get_bit_depth_from_seq(seq); pp->seq_profile = seq->seq_profile; @@ -151,7 +151,7 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c for (i = 0; i < AV1_NUM_REF_FRAMES; i++) { AVFrame *ref_frame = h->ref[i].f; if (ref_frame->buf[0]) - pp->RefFrameMapTextureIndex[i] = ff_dxva2_get_surface_index(avctx, ctx, ref_frame); + pp->RefFrameMapTextureIndex[i] = ff_dxva2_get_surface_index(avctx, ctx, ref_frame, 0); } /* Loop filter parameters */ diff --git a/libavcodec/dxva2_h264.c b/libavcodec/dxva2_h264.c index 20e64f848d..e0ec4878a7 100644 --- a/libavcodec/dxva2_h264.c +++ b/libavcodec/dxva2_h264.c @@ -48,19 +48,16 @@ static void fill_picture_entry(DXVA_PicEntry_H264 *pic, pic->bPicEntry = index | (flag << 7); } -static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, const H264Context *h, +void ff_dxva2_h264_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_H264 *pp) { + const H264Context *h = avctx->priv_data; const H264Picture *current_picture = h->cur_pic_ptr; const SPS *sps = h->ps.sps; const PPS *pps = h->ps.pps; int i, j; memset(pp, 0, sizeof(*pp)); - /* Configure current picture */ - fill_picture_entry(&pp->CurrPic, - ff_dxva2_get_surface_index(avctx, ctx, current_picture->f), - h->picture_structure == PICT_BOTTOM_FIELD); /* Configure the set of references */ pp->UsedForReferenceFlags = 0; pp->NonExistingFrameFlags = 0; @@ -75,7 +72,7 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * } if (r) { fill_picture_entry(&pp->RefFrameList[i], - ff_dxva2_get_surface_index(avctx, ctx, r->f), + ff_dxva2_get_surface_index(avctx, ctx, r->f, 0), r->long_ref != 0); if ((r->reference & PICT_TOP_FIELD) && r->field_poc[0] != INT_MAX) @@ -95,6 +92,10 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * pp->FrameNumList[i] = 0; } } + /* Configure current picture */ + fill_picture_entry(&pp->CurrPic, + ff_dxva2_get_surface_index(avctx, ctx, current_picture->f, 1), + h->picture_structure == PICT_BOTTOM_FIELD); pp->wFrameWidthInMbsMinus1 = h->mb_width - 1; pp->wFrameHeightInMbsMinus1 = h->mb_height - 1; @@ -164,9 +165,10 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * //pp->SliceGroupMap[810]; /* XXX not implemented by FFmpeg */ } -static void fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, const H264Context *h, DXVA_Qmatrix_H264 *qm) +void ff_dxva2_h264_fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_Qmatrix_H264 *qm) { - const PPS *pps = h->ps.pps; + const H264Context *h = avctx->priv_data; + const PPS *pps = h->ps.pps; unsigned i, j; memset(qm, 0, sizeof(*qm)); if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG) { @@ -253,9 +255,9 @@ static void fill_slice_long(AVCodecContext *avctx, DXVA_Slice_H264_Long *slice, unsigned plane; unsigned index; if (DXVA_CONTEXT_WORKAROUND(avctx, ctx) & FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO) - index = ff_dxva2_get_surface_index(avctx, ctx, r->f); + index = ff_dxva2_get_surface_index(avctx, ctx, r->f, 0); else - index = get_refpic_index(pp, ff_dxva2_get_surface_index(avctx, ctx, r->f)); + index = get_refpic_index(pp, ff_dxva2_get_surface_index(avctx, ctx, r->f, 0)); fill_picture_entry(&slice->RefPicList[list][i], index, sl->ref_list[list][i].reference == PICT_BOTTOM_FIELD); for (plane = 0; plane < 3; plane++) { @@ -454,10 +456,10 @@ static int dxva2_h264_start_frame(AVCodecContext *avctx, assert(ctx_pic); /* Fill up DXVA_PicParams_H264 */ - fill_picture_parameters(avctx, ctx, h, &ctx_pic->pp); + ff_dxva2_h264_fill_picture_parameters(avctx, ctx, &ctx_pic->pp); /* Fill up DXVA_Qmatrix_H264 */ - fill_scaling_lists(avctx, ctx, h, &ctx_pic->qm); + ff_dxva2_h264_fill_scaling_lists(avctx, ctx, &ctx_pic->qm); ctx_pic->slice_count = 0; ctx_pic->bitstream_size = 0; diff --git a/libavcodec/dxva2_hevc.c b/libavcodec/dxva2_hevc.c index b6c08943f0..9e1d081412 100644 --- a/libavcodec/dxva2_hevc.c +++ b/libavcodec/dxva2_hevc.c @@ -79,7 +79,7 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * (0 << 14) | (0 << 15); - fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, current_picture->frame), 0); + fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, current_picture->frame, 1), 0); pp->sps_max_dec_pic_buffering_minus1 = sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering - 1; pp->log2_min_luma_coding_block_size_minus3 = sps->log2_min_cb_size - 3; @@ -171,7 +171,7 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * } if (frame) { - fill_picture_entry(&pp->RefPicList[i], ff_dxva2_get_surface_index(avctx, ctx, frame->frame), !!(frame->flags & HEVC_FRAME_FLAG_LONG_REF)); + fill_picture_entry(&pp->RefPicList[i], ff_dxva2_get_surface_index(avctx, ctx, frame->frame, 0), !!(frame->flags & HEVC_FRAME_FLAG_LONG_REF)); pp->PicOrderCntValList[i] = frame->poc; } else { pp->RefPicList[i].bPicEntry = 0xff; @@ -186,7 +186,7 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * while (!frame && j < rpl->nb_refs) \ frame = rpl->ref[j++]; \ if (frame && frame->flags & (HEVC_FRAME_FLAG_LONG_REF | HEVC_FRAME_FLAG_SHORT_REF)) \ - pp->ref_list[i] = get_refpic_index(pp, ff_dxva2_get_surface_index(avctx, ctx, frame->frame)); \ + pp->ref_list[i] = get_refpic_index(pp, ff_dxva2_get_surface_index(avctx, ctx, frame->frame, 0)); \ else \ pp->ref_list[i] = 0xff; \ } \ diff --git a/libavcodec/dxva2_internal.h b/libavcodec/dxva2_internal.h index b822af59cd..73bb954c6c 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -26,18 +26,34 @@ #define COBJMACROS #include "config.h" +#include "config_components.h" /* define the proper COM entries before forcing desktop APIs */ #include +#define FF_DXVA2_WORKAROUND_SCALING_LIST_ZIGZAG 1 ///< Work around for DXVA2/Direct3D11 and old UVD/UVD+ ATI video cards +#define FF_DXVA2_WORKAROUND_INTEL_CLEARVIDEO 2 ///< Work around for DXVA2/Direct3D11 and old Intel GPUs with ClearVideo interface + #if CONFIG_DXVA2 #include "dxva2.h" #include "libavutil/hwcontext_dxva2.h" +#define DXVA2_VAR(ctx, var) ctx->dxva2.var +#else +#define DXVA2_VAR(ctx, var) 0 #endif + #if CONFIG_D3D11VA #include "d3d11va.h" #include "libavutil/hwcontext_d3d11va.h" +#define D3D11VA_VAR(ctx, var) ctx->d3d11va.var +#else +#define D3D11VA_VAR(ctx, var) 0 +#endif + +#if CONFIG_D3D12VA +#include "d3d12va_decode.h" #endif + #if HAVE_DXVA_H /* When targeting WINAPI_FAMILY_PHONE_APP or WINAPI_FAMILY_APP, dxva.h * defines nothing. Force the struct definitions to be visible. */ @@ -62,6 +78,9 @@ typedef union { #if CONFIG_DXVA2 struct dxva_context dxva2; #endif +#if CONFIG_D3D12VA + struct D3D12VADecodeContext d3d12va; +#endif } AVDXVAContext; typedef struct FFDXVASharedContext { @@ -101,43 +120,28 @@ typedef struct FFDXVASharedContext { #define D3D11VA_CONTEXT(ctx) (&ctx->d3d11va) #define DXVA2_CONTEXT(ctx) (&ctx->dxva2) -#if CONFIG_D3D11VA && CONFIG_DXVA2 -#define DXVA_CONTEXT_WORKAROUND(avctx, ctx) (ff_dxva2_is_d3d11(avctx) ? ctx->d3d11va.workaround : ctx->dxva2.workaround) -#define DXVA_CONTEXT_COUNT(avctx, ctx) (ff_dxva2_is_d3d11(avctx) ? ctx->d3d11va.surface_count : ctx->dxva2.surface_count) -#define DXVA_CONTEXT_DECODER(avctx, ctx) (ff_dxva2_is_d3d11(avctx) ? (void *)ctx->d3d11va.decoder : (void *)ctx->dxva2.decoder) -#define DXVA_CONTEXT_REPORT_ID(avctx, ctx) (*(ff_dxva2_is_d3d11(avctx) ? &ctx->d3d11va.report_id : &ctx->dxva2.report_id)) -#define DXVA_CONTEXT_CFG(avctx, ctx) (ff_dxva2_is_d3d11(avctx) ? (void *)ctx->d3d11va.cfg : (void *)ctx->dxva2.cfg) -#define DXVA_CONTEXT_CFG_BITSTREAM(avctx, ctx) (ff_dxva2_is_d3d11(avctx) ? ctx->d3d11va.cfg->ConfigBitstreamRaw : ctx->dxva2.cfg->ConfigBitstreamRaw) -#define DXVA_CONTEXT_CFG_INTRARESID(avctx, ctx) (ff_dxva2_is_d3d11(avctx) ? ctx->d3d11va.cfg->ConfigIntraResidUnsigned : ctx->dxva2.cfg->ConfigIntraResidUnsigned) -#define DXVA_CONTEXT_CFG_RESIDACCEL(avctx, ctx) (ff_dxva2_is_d3d11(avctx) ? ctx->d3d11va.cfg->ConfigResidDiffAccelerator : ctx->dxva2.cfg->ConfigResidDiffAccelerator) +#define DXVA2_CONTEXT_VAR(avctx, ctx, var) (avctx->pix_fmt == AV_PIX_FMT_D3D12 ? 0 : (ff_dxva2_is_d3d11(avctx) ? D3D11VA_VAR(ctx, var) : DXVA2_VAR(ctx, var))) + +#define DXVA_CONTEXT_REPORT_ID(avctx, ctx) (*ff_dxva2_get_report_id(avctx, ctx)) +#define DXVA_CONTEXT_WORKAROUND(avctx, ctx) DXVA2_CONTEXT_VAR(avctx, ctx, workaround) +#define DXVA_CONTEXT_COUNT(avctx, ctx) DXVA2_CONTEXT_VAR(avctx, ctx, surface_count) +#define DXVA_CONTEXT_DECODER(avctx, ctx) (avctx->pix_fmt == AV_PIX_FMT_D3D12 ? 0 : (ff_dxva2_is_d3d11(avctx) ? (void *)D3D11VA_VAR(ctx, decoder) : (void *)DXVA2_VAR(ctx, decoder))) +#define DXVA_CONTEXT_CFG(avctx, ctx) (avctx->pix_fmt == AV_PIX_FMT_D3D12 ? 0 : (ff_dxva2_is_d3d11(avctx) ? (void *)D3D11VA_VAR(ctx, cfg) : (void *)DXVA2_VAR(ctx, cfg))) +#define DXVA_CONTEXT_CFG_BITSTREAM(avctx, ctx) DXVA2_CONTEXT_VAR(avctx, ctx, cfg->ConfigBitstreamRaw) +#define DXVA_CONTEXT_CFG_INTRARESID(avctx, ctx) DXVA2_CONTEXT_VAR(avctx, ctx, cfg->ConfigIntraResidUnsigned) +#define DXVA_CONTEXT_CFG_RESIDACCEL(avctx, ctx) DXVA2_CONTEXT_VAR(avctx, ctx, cfg->ConfigResidDiffAccelerator) #define DXVA_CONTEXT_VALID(avctx, ctx) (DXVA_CONTEXT_DECODER(avctx, ctx) && \ DXVA_CONTEXT_CFG(avctx, ctx) && \ - (ff_dxva2_is_d3d11(avctx) || ctx->dxva2.surface_count)) -#elif CONFIG_DXVA2 -#define DXVA_CONTEXT_WORKAROUND(avctx, ctx) (ctx->dxva2.workaround) -#define DXVA_CONTEXT_COUNT(avctx, ctx) (ctx->dxva2.surface_count) -#define DXVA_CONTEXT_DECODER(avctx, ctx) (ctx->dxva2.decoder) -#define DXVA_CONTEXT_REPORT_ID(avctx, ctx) (*(&ctx->dxva2.report_id)) -#define DXVA_CONTEXT_CFG(avctx, ctx) (ctx->dxva2.cfg) -#define DXVA_CONTEXT_CFG_BITSTREAM(avctx, ctx) (ctx->dxva2.cfg->ConfigBitstreamRaw) -#define DXVA_CONTEXT_CFG_INTRARESID(avctx, ctx) (ctx->dxva2.cfg->ConfigIntraResidUnsigned) -#define DXVA_CONTEXT_CFG_RESIDACCEL(avctx, ctx) (ctx->dxva2.cfg->ConfigResidDiffAccelerator) -#define DXVA_CONTEXT_VALID(avctx, ctx) (ctx->dxva2.decoder && ctx->dxva2.cfg && ctx->dxva2.surface_count) -#elif CONFIG_D3D11VA -#define DXVA_CONTEXT_WORKAROUND(avctx, ctx) (ctx->d3d11va.workaround) -#define DXVA_CONTEXT_COUNT(avctx, ctx) (ctx->d3d11va.surface_count) -#define DXVA_CONTEXT_DECODER(avctx, ctx) (ctx->d3d11va.decoder) -#define DXVA_CONTEXT_REPORT_ID(avctx, ctx) (*(&ctx->d3d11va.report_id)) -#define DXVA_CONTEXT_CFG(avctx, ctx) (ctx->d3d11va.cfg) -#define DXVA_CONTEXT_CFG_BITSTREAM(avctx, ctx) (ctx->d3d11va.cfg->ConfigBitstreamRaw) -#define DXVA_CONTEXT_CFG_INTRARESID(avctx, ctx) (ctx->d3d11va.cfg->ConfigIntraResidUnsigned) -#define DXVA_CONTEXT_CFG_RESIDACCEL(avctx, ctx) (ctx->d3d11va.cfg->ConfigResidDiffAccelerator) -#define DXVA_CONTEXT_VALID(avctx, ctx) (ctx->d3d11va.decoder && ctx->d3d11va.cfg) + (ff_dxva2_is_d3d11(avctx) || DXVA2_VAR(ctx, surface_count))) + +#if CONFIG_D3D12VA +unsigned ff_d3d12va_get_surface_index(const AVCodecContext *avctx, + D3D12VADecodeContext *ctx, const AVFrame *frame, + int curr); #endif unsigned ff_dxva2_get_surface_index(const AVCodecContext *avctx, - const AVDXVAContext *, - const AVFrame *frame); + AVDXVAContext *, const AVFrame *frame, int curr); int ff_dxva2_commit_buffer(AVCodecContext *, AVDXVAContext *, DECODER_BUFFER_DESC *, @@ -161,4 +165,10 @@ int ff_dxva2_common_frame_params(AVCodecContext *avctx, int ff_dxva2_is_d3d11(const AVCodecContext *avctx); +unsigned *ff_dxva2_get_report_id(const AVCodecContext *avctx, AVDXVAContext *ctx); + +void ff_dxva2_h264_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_H264 *pp); + +void ff_dxva2_h264_fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_Qmatrix_H264 *qm); + #endif /* AVCODEC_DXVA2_INTERNAL_H */ diff --git a/libavcodec/dxva2_mpeg2.c b/libavcodec/dxva2_mpeg2.c index 75c416654f..b739dc1e54 100644 --- a/libavcodec/dxva2_mpeg2.c +++ b/libavcodec/dxva2_mpeg2.c @@ -49,14 +49,14 @@ static void fill_picture_parameters(AVCodecContext *avctx, int is_field = s->picture_structure != PICT_FRAME; memset(pp, 0, sizeof(*pp)); - pp->wDecodedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f); + pp->wDecodedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f, 1); pp->wDeblockedPictureIndex = 0; if (s->pict_type != AV_PICTURE_TYPE_I) - pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f); + pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f, 0); else pp->wForwardRefPictureIndex = 0xffff; if (s->pict_type == AV_PICTURE_TYPE_B) - pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f); + pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f, 0); else pp->wBackwardRefPictureIndex = 0xffff; pp->wPicWidthInMBminus1 = s->mb_width - 1; diff --git a/libavcodec/dxva2_vc1.c b/libavcodec/dxva2_vc1.c index b35fb115f7..8ee23feabf 100644 --- a/libavcodec/dxva2_vc1.c +++ b/libavcodec/dxva2_vc1.c @@ -58,13 +58,13 @@ static void fill_picture_parameters(AVCodecContext *avctx, memset(pp, 0, sizeof(*pp)); pp->wDecodedPictureIndex = - pp->wDeblockedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f); + pp->wDeblockedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f, 1); if (s->pict_type != AV_PICTURE_TYPE_I && !v->bi_type) - pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f); + pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f, 0); else pp->wForwardRefPictureIndex = 0xffff; if (s->pict_type == AV_PICTURE_TYPE_B && !v->bi_type) - pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f); + pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f, 0); else pp->wBackwardRefPictureIndex = 0xffff; if (v->profile == PROFILE_ADVANCED) { diff --git a/libavcodec/dxva2_vp9.c b/libavcodec/dxva2_vp9.c index eba4df9031..21699eb3f4 100644 --- a/libavcodec/dxva2_vp9.c +++ b/libavcodec/dxva2_vp9.c @@ -54,7 +54,7 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c memset(pp, 0, sizeof(*pp)); - fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, h->frames[CUR_FRAME].tf.f), 0); + fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, h->frames[CUR_FRAME].tf.f, 1), 0); pp->profile = h->h.profile; pp->wFormatAndPictureInfoFlags = ((h->h.keyframe == 0) << 0) | @@ -81,7 +81,7 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c for (i = 0; i < 8; i++) { if (h->refs[i].f->buf[0]) { - fill_picture_entry(&pp->ref_frame_map[i], ff_dxva2_get_surface_index(avctx, ctx, h->refs[i].f), 0); + fill_picture_entry(&pp->ref_frame_map[i], ff_dxva2_get_surface_index(avctx, ctx, h->refs[i].f, 0), 0); pp->ref_frame_coded_width[i] = h->refs[i].f->width; pp->ref_frame_coded_height[i] = h->refs[i].f->height; } else @@ -91,7 +91,7 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c for (i = 0; i < 3; i++) { uint8_t refidx = h->h.refidx[i]; if (h->refs[refidx].f->buf[0]) - fill_picture_entry(&pp->frame_refs[i], ff_dxva2_get_surface_index(avctx, ctx, h->refs[refidx].f), 0); + fill_picture_entry(&pp->frame_refs[i], ff_dxva2_get_surface_index(avctx, ctx, h->refs[refidx].f, 0), 0); else pp->frame_refs[i].bPicEntry = 0xFF; diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c index 3b6c6c3592..8464a0b34c 100644 --- a/libavcodec/h264_slice.c +++ b/libavcodec/h264_slice.c @@ -784,6 +784,7 @@ static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) { #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \ (CONFIG_H264_D3D11VA_HWACCEL * 2) + \ + CONFIG_H264_D3D12VA_HWACCEL + \ CONFIG_H264_NVDEC_HWACCEL + \ CONFIG_H264_VAAPI_HWACCEL + \ CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \ @@ -887,6 +888,9 @@ static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback) *fmt++ = AV_PIX_FMT_D3D11VA_VLD; *fmt++ = AV_PIX_FMT_D3D11; #endif +#if CONFIG_H264_D3D12VA_HWACCEL + *fmt++ = AV_PIX_FMT_D3D12; +#endif #if CONFIG_H264_VAAPI_HWACCEL *fmt++ = AV_PIX_FMT_VAAPI; #endif diff --git a/libavcodec/h264dec.c b/libavcodec/h264dec.c index 0bf05b2cfe..9f5893c512 100644 --- a/libavcodec/h264dec.c +++ b/libavcodec/h264dec.c @@ -1131,6 +1131,9 @@ const FFCodec ff_h264_decoder = { #if CONFIG_H264_D3D11VA2_HWACCEL HWACCEL_D3D11VA2(h264), #endif +#if CONFIG_H264_D3D12VA_HWACCEL + HWACCEL_D3D12VA(h264), +#endif #if CONFIG_H264_NVDEC_HWACCEL HWACCEL_NVDEC(h264), #endif diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index c4630718cf..90b8beb0f5 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -30,6 +30,7 @@ extern const struct FFHWAccel ff_h263_vaapi_hwaccel; extern const struct FFHWAccel ff_h263_videotoolbox_hwaccel; extern const struct FFHWAccel ff_h264_d3d11va_hwaccel; extern const struct FFHWAccel ff_h264_d3d11va2_hwaccel; +extern const struct FFHWAccel ff_h264_d3d12va_hwaccel; extern const struct FFHWAccel ff_h264_dxva2_hwaccel; extern const struct FFHWAccel ff_h264_nvdec_hwaccel; extern const struct FFHWAccel ff_h264_vaapi_hwaccel; diff --git a/libavcodec/hwconfig.h b/libavcodec/hwconfig.h index e164722a94..ee29ca631d 100644 --- a/libavcodec/hwconfig.h +++ b/libavcodec/hwconfig.h @@ -77,6 +77,8 @@ void ff_hwaccel_uninit(AVCodecContext *avctx); HW_CONFIG_HWACCEL(1, 1, 1, VULKAN, VULKAN, ff_ ## codec ## _vulkan_hwaccel) #define HWACCEL_D3D11VA(codec) \ HW_CONFIG_HWACCEL(0, 0, 1, D3D11VA_VLD, NONE, ff_ ## codec ## _d3d11va_hwaccel) +#define HWACCEL_D3D12VA(codec) \ + HW_CONFIG_HWACCEL(1, 1, 0, D3D12, D3D12VA, ff_ ## codec ## _d3d12va_hwaccel) #define HW_CONFIG_ENCODER(device, frames, ad_hoc, format, device_type_) \ &(const AVCodecHWConfigInternal) { \ From patchwork Tue Dec 5 06:46:45 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44906 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145709pzc; Mon, 4 Dec 2023 22:52:01 -0800 (PST) X-Google-Smtp-Source: AGHT+IFAoRbY5JKs6mCyhmFZW6kHRT16QjRpt2sAHp6P98E+8ugAs2u3R6K+ra/neZ2snXEGp+q8 X-Received: by 2002:a17:906:4:b0:a19:9b79:8b4b with SMTP id 4-20020a170906000400b00a199b798b4bmr100040eja.92.1701759121005; Mon, 04 Dec 2023 22:52:01 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759120; cv=none; d=google.com; s=arc-20160816; b=Nac5G+dS03k9QbP6MBTYyHCYzHle/81yMyGdELzkVY+MUV0OjL4OKwot+8xEz+XyAZ 03t2M7ki3aB84rWOvIVJ2lgYCWZWgNAIaPwQO3Ji8eT54Fs9ieva4eOe3/7rhdU3Cn66 uRrjMWz3KCZH2vdUi/Mf5mOgMGpP68M2FgQEwgItblsdZ0nX02WaLjA5oYIyGRvzgy2f lbV7Wgiy6f/FcVRhJMgp3uj1Ht2+IqfgIDuQ0K+dDoLuv7mFmcgo5n9u22UsOI3sCK/M 6eHQvvr4nlcDiyPefU+aQ4a6os7lOIpMsd/QiJPQ9PrcQhG9molTW5R6yWFMtGYZszVP MKlg== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=UwtK+iPD9jv5FlRHHnnNG5YITFue8yNEBgfZmGOa9Uw=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=trjKUxmrR85Nz8iVtcKgawTFaj+tseiABDx6xNqUd5PZVNsSFrGt8SdDcTEpKI6GSo qYt/Ipnif+YQWia5/7fb3TqcLtuP/Unqm6mJRJuBgJu+a++7CdcoVVLrLjv8zm2HsiHa dY7AUpjMY/F8/78biBB+U5e4LPO+2+to/TJ0u52yE/TkCuXbS23QUjGhyXZnUY9iKd8N IVWtjvE69++nv2f3k6mgIGPea61JrM/TvTZyYGpX2L7skA4QrJSs75zKQJ72mf2E5+nc 1CedognQdFUZGZCjXG8RcmMERs52gVSeayMPio95fqUgiiX4A25YHKwBl08qGczYKiEj ljHw== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=B7HUehYS; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id i21-20020a170906251500b00a1872f23b68si5084853ejb.891.2023.12.04.22.52.00; Mon, 04 Dec 2023 22:52:00 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=B7HUehYS; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 4E16768CED9; Tue, 5 Dec 2023 08:51:36 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id BE1FC68CA67 for ; Tue, 5 Dec 2023 08:51:28 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759094; x=1733295094; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=gV7LP+VXPs7qIr4qBr310QNz48OEFl+mhbvO9TCXaGs=; b=B7HUehYSKWeUr7HiMDIrFgYK9MFyUOKgP6GL4mSJx9lBVKVJvZNorFpM E1OWKmRXm3dwtPE7eVPFA+h1nS79Gqt6si8pBaxWTk74jIFJhl14ae0x5 mGdpaQtHkbsI84Pqr21A+tKThWuneJyTJVfQZDpUZzH/WrDu89QdOm9Vp zHWNQJbW5V7P/+K31CGeFymq+En1+qFsfGH/9CvT3W2tJSbixPpKk5VAt aAK/3lQ9a1GXeh//ZHitDDsC3g+oL1vAQbz+fS2Voqn3TAfodk30nnchz SAB+45gx5RIi4TBs6Ramr6vbEe9/ZOPt01AMiFhOMifzLw7gpImDdreTO w==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012385" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012385" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:24 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853599" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853599" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:22 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:45 +0800 Message-ID: <20231205065034.1621-3-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 3/9] avcodec: add D3D12VA hardware accelerated HEVC decoding X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: p3R6LCvIRec2 From: Wu Jianhua The command below is how to enable d3d12va: ffmpeg -hwaccel d3d12va -i input.mp4 output.mp4 Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- configure | 2 + libavcodec/Makefile | 1 + libavcodec/d3d12va_hevc.c | 208 ++++++++++++++++++++++++++++++++++++ libavcodec/dxva2_hevc.c | 14 +-- libavcodec/dxva2_internal.h | 4 + libavcodec/hevcdec.c | 10 ++ libavcodec/hwaccels.h | 1 + 7 files changed, 234 insertions(+), 6 deletions(-) create mode 100644 libavcodec/d3d12va_hevc.c diff --git a/configure b/configure index 90b20da49a..f1a9406123 100755 --- a/configure +++ b/configure @@ -3100,6 +3100,8 @@ hevc_d3d11va_hwaccel_deps="d3d11va DXVA_PicParams_HEVC" hevc_d3d11va_hwaccel_select="hevc_decoder" hevc_d3d11va2_hwaccel_deps="d3d11va DXVA_PicParams_HEVC" hevc_d3d11va2_hwaccel_select="hevc_decoder" +hevc_d3d12va_hwaccel_deps="d3d12va DXVA_PicParams_HEVC" +hevc_d3d12va_hwaccel_select="hevc_decoder" hevc_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_HEVC" hevc_dxva2_hwaccel_select="hevc_decoder" hevc_nvdec_hwaccel_deps="nvdec" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index ead3a4480e..66284b3420 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1013,6 +1013,7 @@ OBJS-$(CONFIG_H264_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o OBJS-$(CONFIG_H264_VULKAN_HWACCEL) += vulkan_decode.o vulkan_h264.o OBJS-$(CONFIG_HEVC_D3D11VA_HWACCEL) += dxva2_hevc.o OBJS-$(CONFIG_HEVC_DXVA2_HWACCEL) += dxva2_hevc.o +OBJS-$(CONFIG_HEVC_D3D12VA_HWACCEL) += dxva2_hevc.o d3d12va_hevc.o OBJS-$(CONFIG_HEVC_NVDEC_HWACCEL) += nvdec_hevc.o OBJS-$(CONFIG_HEVC_QSV_HWACCEL) += qsvdec.o OBJS-$(CONFIG_HEVC_VAAPI_HWACCEL) += vaapi_hevc.o h265_profile_level.o diff --git a/libavcodec/d3d12va_hevc.c b/libavcodec/d3d12va_hevc.c new file mode 100644 index 0000000000..6537e46065 --- /dev/null +++ b/libavcodec/d3d12va_hevc.c @@ -0,0 +1,208 @@ +/* + * Direct3D 12 HEVC HW acceleration + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config_components.h" + +#include "libavutil/avassert.h" +#include "libavutil/hwcontext_d3d12va_internal.h" +#include "hevc_data.h" +#include "hevcdec.h" +#include "dxva2_internal.h" +#include "d3d12va_decode.h" +#include + +#define MAX_SLICES 256 + +typedef struct HEVCDecodePictureContext { + DXVA_PicParams_HEVC pp; + DXVA_Qmatrix_HEVC qm; + unsigned slice_count; + DXVA_Slice_HEVC_Short slice_short[MAX_SLICES]; + const uint8_t *bitstream; + unsigned bitstream_size; +} HEVCDecodePictureContext; + +static void fill_slice_short(DXVA_Slice_HEVC_Short *slice, unsigned position, unsigned size) +{ + memset(slice, 0, sizeof(*slice)); + slice->BSNALunitDataLocation = position; + slice->SliceBytesInBuffer = size; + slice->wBadSliceChopping = 0; +} + +static int d3d12va_hevc_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size) +{ + const HEVCContext *h = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + HEVCDecodePictureContext *ctx_pic = h->ref->hwaccel_picture_private; + + if (!ctx) + return -1; + + av_assert0(ctx_pic); + + ctx->used_mask = 0; + + ff_dxva2_hevc_fill_picture_parameters(avctx, (AVDXVAContext *)ctx, &ctx_pic->pp); + + ff_dxva2_hevc_fill_scaling_lists(avctx, (AVDXVAContext *)ctx, &ctx_pic->qm); + + ctx_pic->slice_count = 0; + ctx_pic->bitstream_size = 0; + ctx_pic->bitstream = NULL; + + return 0; +} + +static int d3d12va_hevc_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const HEVCContext *h = avctx->priv_data; + const HEVCFrame *current_picture = h->ref; + HEVCDecodePictureContext *ctx_pic = current_picture->hwaccel_picture_private; + unsigned position; + + if (ctx_pic->slice_count >= MAX_SLICES) + return AVERROR(ERANGE); + + if (!ctx_pic->bitstream) + ctx_pic->bitstream = buffer; + ctx_pic->bitstream_size += size; + + position = buffer - ctx_pic->bitstream; + fill_slice_short(&ctx_pic->slice_short[ctx_pic->slice_count], position, size); + ctx_pic->slice_count++; + + return 0; +} + +#define START_CODE 65536 +#define START_CODE_SIZE 3 +static int update_input_arguments(AVCodecContext *avctx, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *input_args, ID3D12Resource *buffer) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + + const HEVCContext *h = avctx->priv_data; + const HEVCFrame *current_picture = h->ref; + HEVCDecodePictureContext *ctx_pic = current_picture->hwaccel_picture_private; + + int i; + uint8_t *mapped_data, *mapped_ptr; + DXVA_Slice_HEVC_Short *slice; + D3D12_VIDEO_DECODE_FRAME_ARGUMENT *args; + + if (FAILED(ID3D12Resource_Map(buffer, 0, NULL, &mapped_data))) { + av_log(avctx, AV_LOG_ERROR, "Failed to map D3D12 Buffer resource!\n"); + return AVERROR(EINVAL); + } + + mapped_ptr = mapped_data; + for (i = 0; i < ctx_pic->slice_count; i++) { + UINT position, size; + slice = &ctx_pic->slice_short[i]; + + position = slice->BSNALunitDataLocation; + size = slice->SliceBytesInBuffer; + + slice->SliceBytesInBuffer += START_CODE_SIZE; + slice->BSNALunitDataLocation = mapped_ptr - mapped_data; + + *(uint32_t *)mapped_ptr = START_CODE; + mapped_ptr += START_CODE_SIZE; + + memcpy(mapped_ptr, &ctx_pic->bitstream[position], size); + mapped_ptr += size; + } + + ID3D12Resource_Unmap(buffer, 0, NULL); + + input_args->CompressedBitstream = (D3D12_VIDEO_DECODE_COMPRESSED_BITSTREAM){ + .pBuffer = buffer, + .Offset = 0, + .Size = mapped_ptr - mapped_data, + }; + + args = &input_args->FrameArguments[input_args->NumFrameArguments++]; + args->Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_SLICE_CONTROL; + args->Size = sizeof(DXVA_Slice_HEVC_Short) * ctx_pic->slice_count; + args->pData = ctx_pic->slice_short; + + return 0; +} + +static int d3d12va_hevc_end_frame(AVCodecContext *avctx) +{ + HEVCContext *h = avctx->priv_data; + HEVCDecodePictureContext *ctx_pic = h->ref->hwaccel_picture_private; + + int scale = ctx_pic->pp.dwCodingParamToolFlags & 1; + + if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0) + return -1; + + return ff_d3d12va_common_end_frame(avctx, h->ref->frame, &ctx_pic->pp, sizeof(ctx_pic->pp), + scale ? &ctx_pic->qm : NULL, scale ? sizeof(ctx_pic->qm) : 0, update_input_arguments); +} + +static int d3d12va_hevc_decode_init(AVCodecContext *avctx) +{ + HEVCContext *h = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + DXVA_PicParams_HEVC pp; + + switch (avctx->profile) { + case FF_PROFILE_HEVC_MAIN_10: + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_HEVC_MAIN10; + break; + + case FF_PROFILE_HEVC_MAIN_STILL_PICTURE: + av_log(avctx, AV_LOG_ERROR, "D3D12 doesn't support PROFILE_HEVC_MAIN_STILL_PICTURE!\n"); + return AVERROR(EINVAL); + + case FF_PROFILE_HEVC_MAIN: + default: + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_HEVC_MAIN; + break; + }; + + ctx->max_num_ref = FF_ARRAY_ELEMS(pp.RefPicList) + 1; + + return ff_d3d12va_decode_init(avctx); +} + +#if CONFIG_HEVC_D3D12VA_HWACCEL +const FFHWAccel ff_hevc_d3d12va_hwaccel = { + .p.name = "hevc_d3d12va", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_HEVC, + .p.pix_fmt = AV_PIX_FMT_D3D12, + .init = d3d12va_hevc_decode_init, + .uninit = ff_d3d12va_decode_uninit, + .start_frame = d3d12va_hevc_start_frame, + .decode_slice = d3d12va_hevc_decode_slice, + .end_frame = d3d12va_hevc_end_frame, + .frame_params = ff_d3d12va_common_frame_params, + .frame_priv_data_size = sizeof(HEVCDecodePictureContext), + .priv_data_size = sizeof(D3D12VADecodeContext), +}; +#endif diff --git a/libavcodec/dxva2_hevc.c b/libavcodec/dxva2_hevc.c index 9e1d081412..31d74a7164 100644 --- a/libavcodec/dxva2_hevc.c +++ b/libavcodec/dxva2_hevc.c @@ -57,9 +57,10 @@ static int get_refpic_index(const DXVA_PicParams_HEVC *pp, int surface_index) return 0xff; } -static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, const HEVCContext *h, +void ff_dxva2_hevc_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_HEVC *pp) { + const HEVCContext *h = avctx->priv_data; const HEVCFrame *current_picture = h->ref; const HEVCSPS *sps = h->ps.sps; const HEVCPPS *pps = h->ps.pps; @@ -79,8 +80,6 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * (0 << 14) | (0 << 15); - fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, current_picture->frame, 1), 0); - pp->sps_max_dec_pic_buffering_minus1 = sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering - 1; pp->log2_min_luma_coding_block_size_minus3 = sps->log2_min_cb_size - 3; pp->log2_diff_max_min_luma_coding_block_size = sps->log2_diff_max_min_coding_block_size; @@ -179,6 +178,8 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * } } + fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, current_picture->frame, 1), 0); + #define DO_REF_LIST(ref_idx, ref_list) { \ const RefPicList *rpl = &h->rps[ref_idx]; \ for (i = 0, j = 0; i < FF_ARRAY_ELEMS(pp->ref_list); i++) { \ @@ -200,8 +201,9 @@ static void fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext * pp->StatusReportFeedbackNumber = 1 + DXVA_CONTEXT_REPORT_ID(avctx, ctx)++; } -static void fill_scaling_lists(AVDXVAContext *ctx, const HEVCContext *h, DXVA_Qmatrix_HEVC *qm) +void ff_dxva2_hevc_fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_Qmatrix_HEVC *qm) { + const HEVCContext *h = avctx->priv_data; unsigned i, j, pos; const ScalingList *sl = h->ps.pps->scaling_list_data_present_flag ? &h->ps.pps->scaling_list : &h->ps.sps->scaling_list; @@ -369,10 +371,10 @@ static int dxva2_hevc_start_frame(AVCodecContext *avctx, av_assert0(ctx_pic); /* Fill up DXVA_PicParams_HEVC */ - fill_picture_parameters(avctx, ctx, h, &ctx_pic->pp); + ff_dxva2_hevc_fill_picture_parameters(avctx, ctx, &ctx_pic->pp); /* Fill up DXVA_Qmatrix_HEVC */ - fill_scaling_lists(ctx, h, &ctx_pic->qm); + ff_dxva2_hevc_fill_scaling_lists(avctx, ctx, &ctx_pic->qm); ctx_pic->slice_count = 0; ctx_pic->bitstream_size = 0; diff --git a/libavcodec/dxva2_internal.h b/libavcodec/dxva2_internal.h index 73bb954c6c..49c696068a 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -171,4 +171,8 @@ void ff_dxva2_h264_fill_picture_parameters(const AVCodecContext *avctx, AVDXVACo void ff_dxva2_h264_fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_Qmatrix_H264 *qm); +void ff_dxva2_hevc_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_HEVC *pp); + +void ff_dxva2_hevc_fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_Qmatrix_HEVC *qm); + #endif /* AVCODEC_DXVA2_INTERNAL_H */ diff --git a/libavcodec/hevcdec.c b/libavcodec/hevcdec.c index 3ce845dddb..b5311ae510 100644 --- a/libavcodec/hevcdec.c +++ b/libavcodec/hevcdec.c @@ -398,6 +398,7 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps) { #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + \ CONFIG_HEVC_D3D11VA_HWACCEL * 2 + \ + CONFIG_HEVC_D3D12VA_HWACCEL + \ CONFIG_HEVC_NVDEC_HWACCEL + \ CONFIG_HEVC_VAAPI_HWACCEL + \ CONFIG_HEVC_VIDEOTOOLBOX_HWACCEL + \ @@ -415,6 +416,9 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps) *fmt++ = AV_PIX_FMT_D3D11VA_VLD; *fmt++ = AV_PIX_FMT_D3D11; #endif +#if CONFIG_HEVC_D3D12VA_HWACCEL + *fmt++ = AV_PIX_FMT_D3D12; +#endif #if CONFIG_HEVC_VAAPI_HWACCEL *fmt++ = AV_PIX_FMT_VAAPI; #endif @@ -439,6 +443,9 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps) *fmt++ = AV_PIX_FMT_D3D11VA_VLD; *fmt++ = AV_PIX_FMT_D3D11; #endif +#if CONFIG_HEVC_D3D12VA_HWACCEL + *fmt++ = AV_PIX_FMT_D3D12; +#endif #if CONFIG_HEVC_VAAPI_HWACCEL *fmt++ = AV_PIX_FMT_VAAPI; #endif @@ -3701,6 +3708,9 @@ const FFCodec ff_hevc_decoder = { #if CONFIG_HEVC_D3D11VA2_HWACCEL HWACCEL_D3D11VA2(hevc), #endif +#if CONFIG_HEVC_D3D12VA_HWACCEL + HWACCEL_D3D12VA(hevc), +#endif #if CONFIG_HEVC_NVDEC_HWACCEL HWACCEL_NVDEC(hevc), #endif diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index 90b8beb0f5..6b71919b0a 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -39,6 +39,7 @@ extern const struct FFHWAccel ff_h264_videotoolbox_hwaccel; extern const struct FFHWAccel ff_h264_vulkan_hwaccel; extern const struct FFHWAccel ff_hevc_d3d11va_hwaccel; extern const struct FFHWAccel ff_hevc_d3d11va2_hwaccel; +extern const struct FFHWAccel ff_hevc_d3d12va_hwaccel; extern const struct FFHWAccel ff_hevc_dxva2_hwaccel; extern const struct FFHWAccel ff_hevc_nvdec_hwaccel; extern const struct FFHWAccel ff_hevc_vaapi_hwaccel; From patchwork Tue Dec 5 06:46:46 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44907 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145774pzc; Mon, 4 Dec 2023 22:52:10 -0800 (PST) X-Google-Smtp-Source: AGHT+IEZfHbdArpR557vvCtkVEgq4fIzyJOzcOSLkqdltPJZvqIFDRrUDn/+QH+UfMOfvMp1RUh6 X-Received: by 2002:a50:cd97:0:b0:54c:81f4:2526 with SMTP id p23-20020a50cd97000000b0054c81f42526mr2336377edi.36.1701759130536; Mon, 04 Dec 2023 22:52:10 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759130; cv=none; d=google.com; s=arc-20160816; b=HbByfEA+uticUm3ScFUBWJ7rCenxWDDwUSQtHl6zb7p5X7AEmgX44wNNJho0ogBQ3i jtBfnYjxksVozzlbNRohHo/Jp2PFzpLRT1Pe4ygNSY7WwnL33pMUScxt+xmgaQE1pGvn BUbR/QpGIaIVZyMWudL4LC1yCYmXbT3kkiSHyQncZD+3GIV37vkPZUTdDRO3qYJRrcfe chjIRBoLz4reeKWSWf5UQnE1H8QxDvD+5+yQ+dAQ2ss9Hn9g86+cuFRN0dNrchGnVorw lFOPJk9X8NM11pdsYisbgTBR6JNBtoupUGr01jvpxrV5sclCwb/WwvjI6YIlWBmT2KRG t+Zw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=6joVLQGpRh0Ed2zFfE8Q+qg4c5zOco+jGfovAkf/yDo=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=tFItJorIp+3Z8+E/sJauMM+y5HxpJHzhoXg+XNnyFD4OlHnqzjA+Tz06lVeoOaTm9x ctn7Q8zc1DdZxrYIagubiAfZHl7Z1HUk9YyExz3Y1GXeXlFTRs1RPL5kCj2JmiXIl3SK K08cU+ggFcGMpEyeF+CNvQvgYHlfLjXkiswHqD6Sl2tid32rpHEIJYG4dyZrvpWgGuwV 2oBfUD+OFvHBdu4XSc1r4FcOUhUwAcdWX0/1IdnMHagjR6KS4GUzPXJyYYLWbO5gwZon Eq19b/UMn8S5t1XJlY7F8YiwFroicq1dkc9MkkwV6Lz0mp7FZONq8mhJOZRx48LQmRJg mpUQ== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=T+UPEYf6; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id u26-20020a50d51a000000b00542ecfc2aa2si595356edi.360.2023.12.04.22.52.10; Mon, 04 Dec 2023 22:52:10 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=T+UPEYf6; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 6AEAF68CEF1; Tue, 5 Dec 2023 08:51:39 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 1296768CE3C for ; Tue, 5 Dec 2023 08:51:31 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759097; x=1733295097; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=q4ZUTtmCq2zl8Yf43gnsiQclB3ZDhOc/ToEvf2STVkE=; b=T+UPEYf6IkRNmtRWZ4LD5E6u8WoDFrL0uLtW1pDnKauK0ZvFheNJrno2 VrsFveC47igiBeWtP9g1QBlFFykZESLRgfQD4z5iXB0TJc8mtULCP2U2y xviLZ6/3bkNta3scSWhP3n2GVSJbuaEESYZRC2htOXOS/G/3AfLOMCScl R/Ef70mnByosGbnHEPsfzp3tjNpZO0HZOmdYuIWYyeKM4MgZDYyDD8tie dkk2qtn9nbY96qZQpILt0o4nFNPNWH53LZuhmgQoYXmT7+Bru9eefNlG4 TC1uEI09y4dvfADz5rrwMjGDlV0YLzZO61pwlY1eEwOeaJlIcSY1dwIxt g==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012387" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012387" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:25 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853605" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853605" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:24 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:46 +0800 Message-ID: <20231205065034.1621-4-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 4/9] avcodec: add D3D12VA hardware accelerated VP9 decoding X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: XYK8VOFq2+/g From: Wu Jianhua The command below is how to enable d3d12va: ffmpeg -hwaccel d3d12va -i input.mp4 output.mp4 Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- configure | 2 + libavcodec/Makefile | 1 + libavcodec/d3d12va_vp9.c | 171 ++++++++++++++++++++++++++++++++++++ libavcodec/dxva2_internal.h | 2 + libavcodec/dxva2_vp9.c | 11 +-- libavcodec/hwaccels.h | 1 + libavcodec/vp9.c | 7 ++ 7 files changed, 190 insertions(+), 5 deletions(-) create mode 100644 libavcodec/d3d12va_vp9.c diff --git a/configure b/configure index f1a9406123..cb4d74a321 100755 --- a/configure +++ b/configure @@ -3168,6 +3168,8 @@ vp9_d3d11va_hwaccel_deps="d3d11va DXVA_PicParams_VP9" vp9_d3d11va_hwaccel_select="vp9_decoder" vp9_d3d11va2_hwaccel_deps="d3d11va DXVA_PicParams_VP9" vp9_d3d11va2_hwaccel_select="vp9_decoder" +vp9_d3d12va_hwaccel_deps="d3d12va DXVA_PicParams_VP9" +vp9_d3d12va_hwaccel_select="vp9_decoder" vp9_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_VP9" vp9_dxva2_hwaccel_select="vp9_decoder" vp9_nvdec_hwaccel_deps="nvdec" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 66284b3420..dd193115d4 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1045,6 +1045,7 @@ OBJS-$(CONFIG_VP8_NVDEC_HWACCEL) += nvdec_vp8.o OBJS-$(CONFIG_VP8_VAAPI_HWACCEL) += vaapi_vp8.o OBJS-$(CONFIG_VP9_D3D11VA_HWACCEL) += dxva2_vp9.o OBJS-$(CONFIG_VP9_DXVA2_HWACCEL) += dxva2_vp9.o +OBJS-$(CONFIG_VP9_D3D12VA_HWACCEL) += dxva2_vp9.o d3d12va_vp9.o OBJS-$(CONFIG_VP9_NVDEC_HWACCEL) += nvdec_vp9.o OBJS-$(CONFIG_VP9_VAAPI_HWACCEL) += vaapi_vp9.o OBJS-$(CONFIG_VP9_VDPAU_HWACCEL) += vdpau_vp9.o diff --git a/libavcodec/d3d12va_vp9.c b/libavcodec/d3d12va_vp9.c new file mode 100644 index 0000000000..bb94e18781 --- /dev/null +++ b/libavcodec/d3d12va_vp9.c @@ -0,0 +1,171 @@ +/* + * Direct3D 12 VP9 HW acceleration + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config_components.h" + +#include "libavutil/avassert.h" +#include "libavutil/pixdesc.h" +#include "libavutil/hwcontext_d3d12va_internal.h" + +#include "vp9shared.h" +#include "dxva2_internal.h" +#include "d3d12va_decode.h" + +typedef struct VP9DecodePictureContext { + DXVA_PicParams_VP9 pp; + DXVA_Slice_VPx_Short slice; + const uint8_t *bitstream; + unsigned bitstream_size; +} VP9DecodePictureContext; + +static void fill_slice_short(DXVA_Slice_VPx_Short *slice, unsigned position, unsigned size) +{ + memset(slice, 0, sizeof(*slice)); + slice->BSNALunitDataLocation = position; + slice->SliceBytesInBuffer = size; + slice->wBadSliceChopping = 0; +} + +static int d3d12va_vp9_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size) +{ + const VP9SharedContext *h = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + VP9DecodePictureContext *ctx_pic = h->frames[CUR_FRAME].hwaccel_picture_private; + + if (!ctx) + return -1; + + av_assert0(ctx_pic); + + ctx->used_mask = 0; + + if (ff_dxva2_vp9_fill_picture_parameters(avctx, (AVDXVAContext *)ctx, &ctx_pic->pp) < 0) + return -1; + + ctx_pic->bitstream_size = 0; + ctx_pic->bitstream = NULL; + + return 0; +} + +static int d3d12va_vp9_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const VP9SharedContext *h = avctx->priv_data; + VP9DecodePictureContext *ctx_pic = h->frames[CUR_FRAME].hwaccel_picture_private; + unsigned position; + + if (!ctx_pic->bitstream) + ctx_pic->bitstream = buffer; + ctx_pic->bitstream_size += size; + + position = buffer - ctx_pic->bitstream; + fill_slice_short(&ctx_pic->slice, position, size); + + return 0; +} + +static int update_input_arguments(AVCodecContext *avctx, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *input_args, ID3D12Resource *buffer) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + + const VP9SharedContext *h = avctx->priv_data; + VP9DecodePictureContext *ctx_pic = h->frames[CUR_FRAME].hwaccel_picture_private; + + uint8_t *mapped_data; + D3D12_VIDEO_DECODE_FRAME_ARGUMENT *args; + + if (FAILED(ID3D12Resource_Map(buffer, 0, NULL, &mapped_data))) { + av_log(avctx, AV_LOG_ERROR, "Failed to map D3D12 Buffer resource!\n"); + return AVERROR(EINVAL); + } + + args = &input_args->FrameArguments[input_args->NumFrameArguments++]; + args->Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_SLICE_CONTROL; + args->Size = sizeof(ctx_pic->slice); + args->pData = &ctx_pic->slice; + + memcpy(mapped_data, ctx_pic->bitstream, ctx_pic->slice.SliceBytesInBuffer); + + ID3D12Resource_Unmap(buffer, 0, NULL); + + input_args->CompressedBitstream = (D3D12_VIDEO_DECODE_COMPRESSED_BITSTREAM){ + .pBuffer = buffer, + .Offset = 0, + .Size = ctx_pic->slice.SliceBytesInBuffer, + }; + + return 0; +} + +static int d3d12va_vp9_end_frame(AVCodecContext *avctx) +{ + VP9SharedContext *h = avctx->priv_data; + VP9DecodePictureContext *ctx_pic = h->frames[CUR_FRAME].hwaccel_picture_private; + + if (ctx_pic->bitstream_size <= 0) + return -1; + + return ff_d3d12va_common_end_frame(avctx, h->frames[CUR_FRAME].tf.f, + &ctx_pic->pp, sizeof(ctx_pic->pp), NULL, 0, update_input_arguments); +} + +static int d3d12va_vp9_decode_init(AVCodecContext *avctx) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + DXVA_PicParams_VP9 pp; + + switch (avctx->profile) { + case FF_PROFILE_VP9_2: + case FF_PROFILE_VP9_3: + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_VP9_10BIT_PROFILE2; + break; + + case FF_PROFILE_VP9_0: + case FF_PROFILE_VP9_1: + default: + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_VP9; + break; + }; + + ctx->max_num_ref = FF_ARRAY_ELEMS(pp.frame_refs) + 1; + + return ff_d3d12va_decode_init(avctx); +} + +#if CONFIG_VP9_D3D12VA_HWACCEL +const FFHWAccel ff_vp9_d3d12va_hwaccel = { + .p.name = "vp9_d3d12va", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_VP9, + .p.pix_fmt = AV_PIX_FMT_D3D12, + .init = d3d12va_vp9_decode_init, + .uninit = ff_d3d12va_decode_uninit, + .start_frame = d3d12va_vp9_start_frame, + .decode_slice = d3d12va_vp9_decode_slice, + .end_frame = d3d12va_vp9_end_frame, + .frame_params = ff_d3d12va_common_frame_params, + .frame_priv_data_size = sizeof(VP9DecodePictureContext), + .priv_data_size = sizeof(D3D12VADecodeContext), +}; +#endif diff --git a/libavcodec/dxva2_internal.h b/libavcodec/dxva2_internal.h index 49c696068a..7c712b31f3 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -175,4 +175,6 @@ void ff_dxva2_hevc_fill_picture_parameters(const AVCodecContext *avctx, AVDXVACo void ff_dxva2_hevc_fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_Qmatrix_HEVC *qm); +int ff_dxva2_vp9_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_VP9 *pp); + #endif /* AVCODEC_DXVA2_INTERNAL_H */ diff --git a/libavcodec/dxva2_vp9.c b/libavcodec/dxva2_vp9.c index 21699eb3f4..1498deb3c8 100644 --- a/libavcodec/dxva2_vp9.c +++ b/libavcodec/dxva2_vp9.c @@ -43,19 +43,18 @@ static void fill_picture_entry(DXVA_PicEntry_VPx *pic, pic->bPicEntry = index | (flag << 7); } -static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, const VP9SharedContext *h, +int ff_dxva2_vp9_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_VP9 *pp) { + const VP9SharedContext *h = avctx->priv_data; + const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt); int i; - const AVPixFmtDescriptor * pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt); if (!pixdesc) return -1; memset(pp, 0, sizeof(*pp)); - fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, h->frames[CUR_FRAME].tf.f, 1), 0); - pp->profile = h->h.profile; pp->wFormatAndPictureInfoFlags = ((h->h.keyframe == 0) << 0) | ((h->h.invisible == 0) << 1) | @@ -98,6 +97,8 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c pp->ref_frame_sign_bias[i + 1] = h->h.signbias[i]; } + fill_picture_entry(&pp->CurrPic, ff_dxva2_get_surface_index(avctx, ctx, h->frames[CUR_FRAME].tf.f, 1), 0); + pp->filter_level = h->h.filter.level; pp->sharpness_level = h->h.filter.sharpness; @@ -265,7 +266,7 @@ static int dxva2_vp9_start_frame(AVCodecContext *avctx, av_assert0(ctx_pic); /* Fill up DXVA_PicParams_VP9 */ - if (fill_picture_parameters(avctx, ctx, h, &ctx_pic->pp) < 0) + if (ff_dxva2_vp9_fill_picture_parameters(avctx, ctx, &ctx_pic->pp) < 0) return -1; ctx_pic->bitstream_size = 0; diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index 6b71919b0a..1344b84326 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -73,6 +73,7 @@ extern const struct FFHWAccel ff_vp8_nvdec_hwaccel; extern const struct FFHWAccel ff_vp8_vaapi_hwaccel; extern const struct FFHWAccel ff_vp9_d3d11va_hwaccel; extern const struct FFHWAccel ff_vp9_d3d11va2_hwaccel; +extern const struct FFHWAccel ff_vp9_d3d12va_hwaccel; extern const struct FFHWAccel ff_vp9_dxva2_hwaccel; extern const struct FFHWAccel ff_vp9_nvdec_hwaccel; extern const struct FFHWAccel ff_vp9_vaapi_hwaccel; diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c index 75e2e1d329..855936cdc1 100644 --- a/libavcodec/vp9.c +++ b/libavcodec/vp9.c @@ -170,6 +170,7 @@ static int update_size(AVCodecContext *avctx, int w, int h) { #define HWACCEL_MAX (CONFIG_VP9_DXVA2_HWACCEL + \ CONFIG_VP9_D3D11VA_HWACCEL * 2 + \ + CONFIG_VP9_D3D12VA_HWACCEL + \ CONFIG_VP9_NVDEC_HWACCEL + \ CONFIG_VP9_VAAPI_HWACCEL + \ CONFIG_VP9_VDPAU_HWACCEL + \ @@ -196,6 +197,9 @@ static int update_size(AVCodecContext *avctx, int w, int h) *fmtp++ = AV_PIX_FMT_D3D11VA_VLD; *fmtp++ = AV_PIX_FMT_D3D11; #endif +#if CONFIG_VP9_D3D12VA_HWACCEL + *fmtp++ = AV_PIX_FMT_D3D12; +#endif #if CONFIG_VP9_NVDEC_HWACCEL *fmtp++ = AV_PIX_FMT_CUDA; #endif @@ -1899,6 +1903,9 @@ const FFCodec ff_vp9_decoder = { #if CONFIG_VP9_D3D11VA2_HWACCEL HWACCEL_D3D11VA2(vp9), #endif +#if CONFIG_VP9_D3D12VA_HWACCEL + HWACCEL_D3D12VA(vp9), +#endif #if CONFIG_VP9_NVDEC_HWACCEL HWACCEL_NVDEC(vp9), #endif From patchwork Tue Dec 5 06:46:47 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44908 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145812pzc; Mon, 4 Dec 2023 22:52:20 -0800 (PST) X-Google-Smtp-Source: AGHT+IEC7d69KXk4mz4W0jRG+uSMKZnHgBNdPIfJe89L7z0Yiy9dTZwo3VgJcfVTWJLCS56goWLz X-Received: by 2002:a50:d4d4:0:b0:54b:5a7:1435 with SMTP id e20-20020a50d4d4000000b0054b05a71435mr934093edj.4.1701759139983; Mon, 04 Dec 2023 22:52:19 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759139; cv=none; d=google.com; s=arc-20160816; b=KzKTIR0LGvQ9hanEIX4jEGxHIQBrC9/y7H/loMImoe2v7CH8E0IhFSxTbs5v9V3ebw sEk/CV1nXfqi4Ak7o+rMkOMN3UKZchvHJcdQ0BpaXeMKy1qsGo5MjqGYst4rWccTrHvw yHeVCdYXRI3F6NaPs+m5tt30/Sx3glC97HR9v1qnih8o/xDucZYQH/TZiH1R5BGSAEm3 kkeLQXubOgnK8SbrGGoSp822j3yGN9WZ5L1NGOXj+XtZrYu7XkyH4exNz1A0U7UxJDUP UeKkltdh3G9T3seeZBwWHQnN75Qgx139rFIXiWwuvP1k8pQnpTWHaPrtktPQE9C9GMMy LKYw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=b/29MieKulxJVGO4rgVTcNYMeoFJHICnDzLpnNQH0Cg=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=ykbIzoz2ltPPzl54nOqdd1q/5pZXcJXkK/zeVXU5OOgM72TtJNEhpqFw59/llhiCWG WWG048NFleP36+J6HPum3KTKN7CPN4YD6Fp2doNz/wEhBZYXGOS1VmCxU17+h6GBy98a I3cg9mpki/p+xGuyoOYIBMUaK9L4SP+EGerBDpCJh6ZYu2l/GMs4sgVdH7SG50nHKFeN C3Dlmkz/Nxwx9wQ6dqeiwN2A5NE7jJupNtJurrmyhM9vL8vxDHOzqbNVuahK1ZItP7TH UTs/0EPQMcKOeiEJbCtG/b+PrMLzLMlYJYWMeSF4/DBzpBRjFNyMfLPsNxVigXkfbW1W MFjg== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=aUk4+24t; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id y4-20020a056402358400b0054c738b34ccsi574264edc.386.2023.12.04.22.52.19; Mon, 04 Dec 2023 22:52:19 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=aUk4+24t; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 69FDC68CF04; Tue, 5 Dec 2023 08:51:40 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 0143268CEAF for ; Tue, 5 Dec 2023 08:51:33 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759099; x=1733295099; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=+AIAOS7/ifOb78ayPC0y/Ebj6wjfEwT0PkvDnDBaG2k=; b=aUk4+24t75SxTaQE/i81W1F+n7s4xfCwbXCHNiBjg0UvYP70UhecsTOU 34sa9ecEMyhKDK58weEag1bB6/AzLJU5s3GOKz3YtXwWvgZyn4lRzwNNI uI59gY6Gm+A0Qz6evWGh9IXj1UJMQHHNfIvvd9TEcQBcg1jSuSif6X8Cg s/BTMQtWgMoGVpLVV+DUbPlMZSl+O3fwYvbCKIfP76rSg6p32gS2gd94Z fR0gfgfUBbnWbazB+Osx59ZBWTYIAenSpsXnq1Iar70/xKsOO6DSPgFCT ldRUoqxfu5Z2n77iUQFRDAL+gwSfAx2Xyc5rPfb+0uN84wEpoNKTqtGCh g==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012389" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012389" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:26 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853620" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853620" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:25 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:47 +0800 Message-ID: <20231205065034.1621-5-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 5/9] avcodec: add D3D12VA hardware accelerated AV1 decoding X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: TNsgGYSDG/Ku From: Wu Jianhua The command below is how to enable d3d12va: ffmpeg -hwaccel d3d12va -i input.mp4 output.mp4 Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- configure | 2 + libavcodec/Makefile | 1 + libavcodec/av1dec.c | 10 ++ libavcodec/d3d12va_av1.c | 214 ++++++++++++++++++++++++++++++++++++ libavcodec/dxva2_av1.c | 8 +- libavcodec/dxva2_internal.h | 4 + libavcodec/hwaccels.h | 1 + 7 files changed, 237 insertions(+), 3 deletions(-) create mode 100644 libavcodec/d3d12va_av1.c diff --git a/configure b/configure index cb4d74a321..b5a096de7d 100755 --- a/configure +++ b/configure @@ -3064,6 +3064,8 @@ av1_d3d11va_hwaccel_deps="d3d11va DXVA_PicParams_AV1" av1_d3d11va_hwaccel_select="av1_decoder" av1_d3d11va2_hwaccel_deps="d3d11va DXVA_PicParams_AV1" av1_d3d11va2_hwaccel_select="av1_decoder" +av1_d3d12va_hwaccel_deps="d3d12va DXVA_PicParams_AV1" +av1_d3d12va_hwaccel_select="av1_decoder" av1_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_AV1" av1_dxva2_hwaccel_select="av1_decoder" av1_nvdec_hwaccel_deps="nvdec CUVIDAV1PICPARAMS" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index dd193115d4..730dcf373f 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -996,6 +996,7 @@ OBJS-$(CONFIG_VULKAN) += vulkan.o vulkan_video.o OBJS-$(CONFIG_AV1_D3D11VA_HWACCEL) += dxva2_av1.o OBJS-$(CONFIG_AV1_DXVA2_HWACCEL) += dxva2_av1.o +OBJS-$(CONFIG_AV1_D3D12VA_HWACCEL) += dxva2_av1.o d3d12va_av1.o OBJS-$(CONFIG_AV1_NVDEC_HWACCEL) += nvdec_av1.o OBJS-$(CONFIG_AV1_VAAPI_HWACCEL) += vaapi_av1.o OBJS-$(CONFIG_AV1_VDPAU_HWACCEL) += vdpau_av1.o diff --git a/libavcodec/av1dec.c b/libavcodec/av1dec.c index 6114cb78e6..d23627f184 100644 --- a/libavcodec/av1dec.c +++ b/libavcodec/av1dec.c @@ -511,6 +511,7 @@ static int get_pixel_format(AVCodecContext *avctx) enum AVPixelFormat pix_fmt = get_sw_pixel_format(avctx, seq); #define HWACCEL_MAX (CONFIG_AV1_DXVA2_HWACCEL + \ CONFIG_AV1_D3D11VA_HWACCEL * 2 + \ + CONFIG_AV1_D3D12VA_HWACCEL + \ CONFIG_AV1_NVDEC_HWACCEL + \ CONFIG_AV1_VAAPI_HWACCEL + \ CONFIG_AV1_VDPAU_HWACCEL + \ @@ -529,6 +530,9 @@ static int get_pixel_format(AVCodecContext *avctx) *fmtp++ = AV_PIX_FMT_D3D11VA_VLD; *fmtp++ = AV_PIX_FMT_D3D11; #endif +#if CONFIG_AV1_D3D12VA_HWACCEL + *fmtp++ = AV_PIX_FMT_D3D12; +#endif #if CONFIG_AV1_NVDEC_HWACCEL *fmtp++ = AV_PIX_FMT_CUDA; #endif @@ -550,6 +554,9 @@ static int get_pixel_format(AVCodecContext *avctx) *fmtp++ = AV_PIX_FMT_D3D11VA_VLD; *fmtp++ = AV_PIX_FMT_D3D11; #endif +#if CONFIG_AV1_D3D12VA_HWACCEL + *fmtp++ = AV_PIX_FMT_D3D12; +#endif #if CONFIG_AV1_NVDEC_HWACCEL *fmtp++ = AV_PIX_FMT_CUDA; #endif @@ -1507,6 +1514,9 @@ const FFCodec ff_av1_decoder = { #if CONFIG_AV1_D3D11VA2_HWACCEL HWACCEL_D3D11VA2(av1), #endif +#if CONFIG_AV1_D3D12VA_HWACCEL + HWACCEL_D3D12VA(av1), +#endif #if CONFIG_AV1_NVDEC_HWACCEL HWACCEL_NVDEC(av1), #endif diff --git a/libavcodec/d3d12va_av1.c b/libavcodec/d3d12va_av1.c new file mode 100644 index 0000000000..6a3543353f --- /dev/null +++ b/libavcodec/d3d12va_av1.c @@ -0,0 +1,214 @@ +/* + * Direct3D 12 AV1 HW acceleration + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config_components.h" +#include "libavutil/avassert.h" +#include "libavutil/hwcontext_d3d12va_internal.h" +#include "av1dec.h" +#include "dxva2_internal.h" +#include "d3d12va_decode.h" + +#define MAX_TILES 256 + +typedef struct D3D12AV1DecodeContext { + D3D12VADecodeContext ctx; + uint8_t *bitstream_buffer; +} D3D12AV1DecodeContext; + +#define D3D12_AV1_DECODE_CONTEXT(avctx) ((D3D12AV1DecodeContext *)D3D12VA_DECODE_CONTEXT(avctx)) + +typedef struct AV1DecodePictureContext { + DXVA_PicParams_AV1 pp; + unsigned tile_count; + DXVA_Tile_AV1 tiles[MAX_TILES]; + uint8_t *bitstream; + unsigned bitstream_size; +} AV1DecodePictureContext; + +static int d3d12va_av1_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size) +{ + const AV1DecContext *h = avctx->priv_data; + AV1DecodePictureContext *ctx_pic = h->cur_frame.hwaccel_picture_private; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + if (!ctx) + return -1; + + av_assert0(ctx_pic); + + ctx->used_mask = 0; + + if (ff_dxva2_av1_fill_picture_parameters(avctx, (AVDXVAContext *)ctx, &ctx_pic->pp) < 0) + return -1; + + ctx_pic->bitstream = NULL; + ctx_pic->bitstream_size = 0; + ctx_pic->tile_count = 0; + + return 0; +} + +static int d3d12va_av1_decode_slice(AVCodecContext *avctx, + const uint8_t *buffer, + uint32_t size) +{ + const AV1DecContext *h = avctx->priv_data; + const AV1RawFrameHeader *frame_header = h->raw_frame_header; + AV1DecodePictureContext *ctx_pic = h->cur_frame.hwaccel_picture_private; + int offset = 0; + uint32_t tg_start, tg_end; + + ctx_pic->tile_count = frame_header->tile_cols * frame_header->tile_rows; + + if (ctx_pic->tile_count > MAX_TILES) + return AVERROR(ENOSYS); + + if (ctx_pic->tile_count == h->tg_end - h->tg_start + 1) { + tg_start = 0; + tg_end = ctx_pic->tile_count - 1; + ctx_pic->bitstream = (uint8_t *)buffer; + ctx_pic->bitstream_size = size; + } else { + ctx_pic->bitstream = D3D12_AV1_DECODE_CONTEXT(avctx)->bitstream_buffer; + memcpy(ctx_pic->bitstream + ctx_pic->bitstream_size, buffer, size); + tg_start = h->tg_start; + tg_end = h->tg_end; + offset = ctx_pic->bitstream_size; + ctx_pic->bitstream_size += size; + } + + for (uint32_t tile_num = tg_start; tile_num <= tg_end; tile_num++) { + ctx_pic->tiles[tile_num].DataOffset = offset + h->tile_group_info[tile_num].tile_offset; + ctx_pic->tiles[tile_num].DataSize = h->tile_group_info[tile_num].tile_size; + ctx_pic->tiles[tile_num].row = h->tile_group_info[tile_num].tile_row; + ctx_pic->tiles[tile_num].column = h->tile_group_info[tile_num].tile_column; + ctx_pic->tiles[tile_num].anchor_frame = 0xFF; + } + + return 0; +} + +static int update_input_arguments(AVCodecContext *avctx, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *input_args, ID3D12Resource *buffer) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + const AV1DecContext *h = avctx->priv_data; + AV1DecodePictureContext *ctx_pic = h->cur_frame.hwaccel_picture_private; + uint8_t *mapped_data; + + D3D12_VIDEO_DECODE_FRAME_ARGUMENT *args = &input_args->FrameArguments[input_args->NumFrameArguments++]; + args->Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_SLICE_CONTROL; + args->Size = sizeof(DXVA_Tile_AV1) * ctx_pic->tile_count; + args->pData = ctx_pic->tiles; + + input_args->CompressedBitstream = (D3D12_VIDEO_DECODE_COMPRESSED_BITSTREAM){ + .pBuffer = buffer, + .Offset = 0, + .Size = ctx_pic->bitstream_size, + }; + + if (FAILED(ID3D12Resource_Map(buffer, 0, NULL, &mapped_data))) { + av_log(avctx, AV_LOG_ERROR, "Failed to map D3D12 Buffer resource!\n"); + return AVERROR(EINVAL); + } + + memcpy(mapped_data, ctx_pic->bitstream, ctx_pic->bitstream_size); + + ID3D12Resource_Unmap(buffer, 0, NULL); + + return 0; +} + +static int d3d12va_av1_end_frame(AVCodecContext *avctx) +{ + int ret; + const AV1DecContext *h = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AV1DecodePictureContext *ctx_pic = h->cur_frame.hwaccel_picture_private; + + if (ctx_pic->tiles <= 0 || ctx_pic->bitstream_size <= 0) + return -1; + + ret = ff_d3d12va_common_end_frame(avctx, h->cur_frame.f, &ctx_pic->pp, sizeof(ctx_pic->pp), + NULL, 0, update_input_arguments); + + return ret; +} + +static int d3d12va_av1_decode_init(AVCodecContext *avctx) +{ + const AV1DecContext *h = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + D3D12AV1DecodeContext *av1_ctx = D3D12_AV1_DECODE_CONTEXT(avctx); + AV1DecodePictureContext *ctx_pic = h->cur_frame.hwaccel_picture_private; + DXVA_PicParams_AV1 pp; + + int ret; + + if (avctx->profile != FF_PROFILE_AV1_MAIN) + return AVERROR(EINVAL); + + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_AV1_PROFILE0; + + ctx->max_num_ref = FF_ARRAY_ELEMS(pp.RefFrameMapTextureIndex) + 1; + + ret = ff_d3d12va_decode_init(avctx); + if (ret < 0) + return ret; + + if (!av1_ctx->bitstream_buffer) { + av1_ctx->bitstream_buffer = av_malloc(ff_d3d12va_get_suitable_max_bitstream_size(avctx)); + if (!av1_ctx->bitstream_buffer) + return AVERROR(ENOMEM); + } + + return 0; +} + +static int d3d12va_av1_decode_uninit(AVCodecContext *avctx) +{ + const AV1DecContext *h = avctx->priv_data; + D3D12AV1DecodeContext *ctx = D3D12_AV1_DECODE_CONTEXT(avctx); + AV1DecodePictureContext *ctx_pic = h->cur_frame.hwaccel_picture_private; + + if (ctx->bitstream_buffer) + av_freep(&ctx->bitstream_buffer); + + return ff_d3d12va_decode_uninit(avctx); +} + +#if CONFIG_AV1_D3D12VA_HWACCEL +const FFHWAccel ff_av1_d3d12va_hwaccel = { + .p.name = "av1_d3d12va", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_AV1, + .p.pix_fmt = AV_PIX_FMT_D3D12, + .init = d3d12va_av1_decode_init, + .uninit = d3d12va_av1_decode_uninit, + .start_frame = d3d12va_av1_start_frame, + .decode_slice = d3d12va_av1_decode_slice, + .end_frame = d3d12va_av1_end_frame, + .frame_params = ff_d3d12va_common_frame_params, + .frame_priv_data_size = sizeof(AV1DecodePictureContext), + .priv_data_size = sizeof(D3D12AV1DecodeContext), +}; +#endif diff --git a/libavcodec/dxva2_av1.c b/libavcodec/dxva2_av1.c index 12a895b791..85d747bca8 100644 --- a/libavcodec/dxva2_av1.c +++ b/libavcodec/dxva2_av1.c @@ -56,10 +56,11 @@ static int get_bit_depth_from_seq(const AV1RawSequenceHeader *seq) return 8; } -static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, const AV1DecContext *h, +int ff_dxva2_av1_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_AV1 *pp) { int i,j, uses_lr; + const AV1DecContext *h = avctx->priv_data; const AV1RawSequenceHeader *seq = h->raw_seq; const AV1RawFrameHeader *frame_header = h->raw_frame_header; const AV1RawFilmGrainParams *film_grain = &h->cur_frame.film_grain; @@ -75,7 +76,6 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c pp->max_width = seq->max_frame_width_minus_1 + 1; pp->max_height = seq->max_frame_height_minus_1 + 1; - pp->CurrPicTextureIndex = ff_dxva2_get_surface_index(avctx, ctx, h->cur_frame.f, 1); pp->superres_denom = frame_header->use_superres ? frame_header->coded_denom + AV1_SUPERRES_DENOM_MIN : AV1_SUPERRES_NUM; pp->bitdepth = get_bit_depth_from_seq(seq); pp->seq_profile = seq->seq_profile; @@ -154,6 +154,8 @@ static int fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *c pp->RefFrameMapTextureIndex[i] = ff_dxva2_get_surface_index(avctx, ctx, ref_frame, 0); } + pp->CurrPicTextureIndex = ff_dxva2_get_surface_index(avctx, ctx, h->cur_frame.f, 1); + /* Loop filter parameters */ pp->loop_filter.filter_level[0] = frame_header->loop_filter_level[0]; pp->loop_filter.filter_level[1] = frame_header->loop_filter_level[1]; @@ -281,7 +283,7 @@ static int dxva2_av1_start_frame(AVCodecContext *avctx, av_assert0(ctx_pic); /* Fill up DXVA_PicParams_AV1 */ - if (fill_picture_parameters(avctx, ctx, h, &ctx_pic->pp) < 0) + if (ff_dxva2_av1_fill_picture_parameters(avctx, ctx, &ctx_pic->pp) < 0) return -1; ctx_pic->bitstream_size = 0; diff --git a/libavcodec/dxva2_internal.h b/libavcodec/dxva2_internal.h index 7c712b31f3..7a30672ecd 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -177,4 +177,8 @@ void ff_dxva2_hevc_fill_scaling_lists(const AVCodecContext *avctx, AVDXVAContext int ff_dxva2_vp9_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_VP9 *pp); +#if CONFIG_AV1_D3D12VA_HWACCEL || CONFIG_AV1_D3D11VA_HWACCEL || CONFIG_AV1_D3D11VA2_HWACCEL || CONFIG_AV1_DXVA2_HWACCEL +int ff_dxva2_av1_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_AV1 *pp); +#endif + #endif /* AVCODEC_DXVA2_INTERNAL_H */ diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index 1344b84326..1c4825f2fb 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -21,6 +21,7 @@ extern const struct FFHWAccel ff_av1_d3d11va_hwaccel; extern const struct FFHWAccel ff_av1_d3d11va2_hwaccel; +extern const struct FFHWAccel ff_av1_d3d12va_hwaccel; extern const struct FFHWAccel ff_av1_dxva2_hwaccel; extern const struct FFHWAccel ff_av1_nvdec_hwaccel; extern const struct FFHWAccel ff_av1_vaapi_hwaccel; From patchwork Tue Dec 5 06:46:48 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44909 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145844pzc; Mon, 4 Dec 2023 22:52:29 -0800 (PST) X-Google-Smtp-Source: AGHT+IFQSZI3NSiXiiz74TznMKINl4JLSveqBMG4OFHVNI59Ek3u/LScRuJfFVjKB3onLpVwLeie X-Received: by 2002:a05:600c:1384:b0:40b:5e21:dd4e with SMTP id u4-20020a05600c138400b0040b5e21dd4emr159277wmf.124.1701759149454; Mon, 04 Dec 2023 22:52:29 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759149; cv=none; d=google.com; s=arc-20160816; b=GqtTSWrTmEz1Cea4WNJH8GCy4S2dwDWhUNyMjjjZ1yFZ31xFdlyDx2CifD85FKaROU W2R3xW3ic4D7jpPjzIOs6edRJo58y5H3SvpqcbrY2Hu3ul6bW/5fZwtQ3RV8gF4HSxtj JFG6Y4AxmCkh6+nxY+8Kr7xH13guBoeQj6q2YZqDPRDZr8nsVxpWvPrV9ryIyQGl4cXN ODzX1OHtnkGc56ClKUtyhA9kQNefsGXVCWRq9SdEQwoD7xRumrYygS+g5Dv96mIZvmKC H3A+PW/yLbeoJZ/7vL4aEBUMO6Bd06v4IdPAaJZUw6cDSiThCcA3QMMY8HyR6txftQRQ 6WPQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=7gNw/4CXN1oLZhy2+kIU5BSjQcCzXw2PsX8C0wSKbTY=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=QKW7EZFDd1d7z9WVWTFuWQnYDu6YJL2SiQOf/ft2XxOW3yfbS60e1SJmYJuFXFhK3g PktoBk0rktopfAFYc/+akslcR4u62XpaC6WZwTlKwZALTZoLJZXivev3V6sukXq/ApsF NyyUHrr59QItl7mF7nfBDBwsll7JSt/IlFQ6Q4f5XgGosKB1iQapAk3yR+vIBAE4XpoB qPOPSlgI/RtoBBdF+byziVAnYpo2+hrmomDRm+2MEUCdM6kj6rSbKyolvCJM8mfuFpdL 03hY5YYRQCiYB9+Gi1iL8Ia2u98BcCGWt+ZdKm/t4uHzUI7rYXYRqTK+fi4uMs1AKNkD 81ew== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=XeBhRNH4; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id nc8-20020a1709071c0800b00a1c68c96a1dsi274127ejc.743.2023.12.04.22.52.28; Mon, 04 Dec 2023 22:52:29 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=XeBhRNH4; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 65C3D68CF19; Tue, 5 Dec 2023 08:51:43 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id DE5E768CEF5 for ; Tue, 5 Dec 2023 08:51:34 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759100; x=1733295100; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=iSPx+eaa/FF3jdcGUYq0iMH3H1U/C3XJ1GNgGcS1B1c=; b=XeBhRNH41tw16vR4sJufKZU33PGLx3aqnIeFhjRhonCAjScGZGYerdFJ tVaLFOHFz+Uv7BWuZcpU2QaRjN7+doRSTIO7hSLES6XRZNNwJtY3iJrb3 37I3ruLPzMLPb0vLB2RVKfqmvp44s/GzHgaT5/HcCwkfcMOj3nJJBBD8g Y+NePuGYW1sKczVeeKSvMAwLaVyEcPxPlcZzGOxjdsxR9QKbTbJFUjI6k zF3tgQEmrzBrbDsHMs+q7jd6fDV0E1JfKC+pPcaX6TbGzEoKWTJnOEVhV 453jVQ5M5tSRjCRh3r2W/jMGNJs7FoyMDgZcugy/J+Q67YxBZ1YfUCaJZ w==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012392" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012392" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:28 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853633" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853633" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:26 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:48 +0800 Message-ID: <20231205065034.1621-6-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 6/9] avcodec: add D3D12VA hardware accelerated MPEG-2 decoding X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: XuSGk0LnUIID From: Wu Jianhua The command below is how to enable d3d12va: ffmpeg -hwaccel d3d12va -i input.mp4 output.mp4 Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- configure | 2 + libavcodec/Makefile | 1 + libavcodec/d3d12va_mpeg2.c | 191 ++++++++++++++++++++++++++++++++++++ libavcodec/dxva2_internal.h | 6 ++ libavcodec/dxva2_mpeg2.c | 20 ++-- libavcodec/hwaccels.h | 1 + libavcodec/mpeg12dec.c | 6 ++ 7 files changed, 217 insertions(+), 10 deletions(-) create mode 100644 libavcodec/d3d12va_mpeg2.c diff --git a/configure b/configure index b5a096de7d..7981833ec3 100755 --- a/configure +++ b/configure @@ -3130,6 +3130,8 @@ mpeg2_d3d11va_hwaccel_deps="d3d11va" mpeg2_d3d11va_hwaccel_select="mpeg2video_decoder" mpeg2_d3d11va2_hwaccel_deps="d3d11va" mpeg2_d3d11va2_hwaccel_select="mpeg2video_decoder" +mpeg2_d3d12va_hwaccel_deps="d3d12va" +mpeg2_d3d12va_hwaccel_select="mpeg2video_decoder" mpeg2_dxva2_hwaccel_deps="dxva2" mpeg2_dxva2_hwaccel_select="mpeg2video_decoder" mpeg2_nvdec_hwaccel_deps="nvdec" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 730dcf373f..889f844031 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1027,6 +1027,7 @@ OBJS-$(CONFIG_MPEG1_VDPAU_HWACCEL) += vdpau_mpeg12.o OBJS-$(CONFIG_MPEG1_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o OBJS-$(CONFIG_MPEG2_D3D11VA_HWACCEL) += dxva2_mpeg2.o OBJS-$(CONFIG_MPEG2_DXVA2_HWACCEL) += dxva2_mpeg2.o +OBJS-$(CONFIG_MPEG2_D3D12VA_HWACCEL) += dxva2_mpeg2.o d3d12va_mpeg2.o OBJS-$(CONFIG_MPEG2_NVDEC_HWACCEL) += nvdec_mpeg12.o OBJS-$(CONFIG_MPEG2_QSV_HWACCEL) += qsvdec.o OBJS-$(CONFIG_MPEG2_VAAPI_HWACCEL) += vaapi_mpeg2.o diff --git a/libavcodec/d3d12va_mpeg2.c b/libavcodec/d3d12va_mpeg2.c new file mode 100644 index 0000000000..91bf3f8b75 --- /dev/null +++ b/libavcodec/d3d12va_mpeg2.c @@ -0,0 +1,191 @@ +/* + * Direct3D12 MPEG-2 HW acceleration + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config_components.h" +#include "libavutil/avassert.h" +#include "libavutil/hwcontext_d3d12va_internal.h" +#include "mpegutils.h" +#include "mpegvideodec.h" +#include "d3d12va_decode.h" +#include "dxva2_internal.h" + +#define MAX_SLICES 1024 +#define INVALID_REF 0xffff + +#define REF_RESOURCE(index) if (index != INVALID_REF) { \ + ctx->ref_resources[index] = frames_hwctx->texture_infos[index].texture; \ +} + +typedef struct D3D12DecodePictureContext { + DXVA_PictureParameters pp; + DXVA_QmatrixData qm; + unsigned slice_count; + DXVA_SliceInfo slices[MAX_SLICES]; + const uint8_t *bitstream; + unsigned bitstream_size; +} D3D12DecodePictureContext; + +static int d3d12va_mpeg2_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size) +{ + const MpegEncContext *s = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + D3D12DecodePictureContext *ctx_pic = s->current_picture_ptr->hwaccel_picture_private; + DXVA_QmatrixData *qm = &ctx_pic->qm; + + if (!ctx) + return -1; + + av_assert0(ctx_pic); + + ctx->used_mask = 0; + + ff_dxva2_mpeg2_fill_picture_parameters(avctx, (AVDXVAContext *)ctx, &ctx_pic->pp); + ff_dxva2_mpeg2_fill_quantization_matrices(avctx, (AVDXVAContext *)ctx, &ctx_pic->qm); + + // Post processing operations are not supported in D3D12 Video + ctx_pic->pp.wDeblockedPictureIndex = INVALID_REF; + + ctx_pic->bitstream = NULL; + ctx_pic->bitstream_size = 0; + ctx_pic->slice_count = 0; + + return 0; +} + +static int d3d12va_mpeg2_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const MpegEncContext *s = avctx->priv_data; + D3D12DecodePictureContext *ctx_pic = s->current_picture_ptr->hwaccel_picture_private; + + int is_field = s->picture_structure != PICT_FRAME; + + if (ctx_pic->slice_count >= MAX_SLICES) { + return AVERROR(ERANGE); + } + + if (!ctx_pic->bitstream) + ctx_pic->bitstream = buffer; + ctx_pic->bitstream_size += size; + + ff_dxva2_mpeg2_fill_slice(avctx, &ctx_pic->slices[ctx_pic->slice_count++], + buffer - ctx_pic->bitstream, buffer, size); + + return 0; +} + +static int update_input_arguments(AVCodecContext *avctx, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *input_args, ID3D12Resource *buffer) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + const MpegEncContext *s = avctx->priv_data; + D3D12DecodePictureContext *ctx_pic = s->current_picture_ptr->hwaccel_picture_private; + + const int is_field = s->picture_structure != PICT_FRAME; + const unsigned mb_count = s->mb_width * (s->mb_height >> is_field); + + int i; + uint8_t *mapped_data = NULL; + D3D12_VIDEO_DECODE_FRAME_ARGUMENT *args = &input_args->FrameArguments[input_args->NumFrameArguments++]; + + D3D12_RANGE range = { + .Begin = 0, + .End = ctx_pic->bitstream_size, + }; + + if (FAILED(ID3D12Resource_Map(buffer, 0, &range, &mapped_data))) { + av_log(avctx, AV_LOG_ERROR, "Failed to map D3D12 Buffer resource!\n"); + return AVERROR(EINVAL); + } + + for (i = 0; i < ctx_pic->slice_count; i++) { + DXVA_SliceInfo *slice = &ctx_pic->slices[i]; + + if (i < ctx_pic->slice_count - 1) + slice->wNumberMBsInSlice = slice[1].wNumberMBsInSlice - slice[0].wNumberMBsInSlice; + else + slice->wNumberMBsInSlice = mb_count - slice[0].wNumberMBsInSlice; + } + + memcpy(mapped_data, ctx_pic->bitstream, ctx_pic->bitstream_size); + + ID3D12Resource_Unmap(buffer, 0, &range); + + args->Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_SLICE_CONTROL; + args->Size = sizeof(DXVA_SliceInfo) * ctx_pic->slice_count; + args->pData = ctx_pic->slices; + + input_args->CompressedBitstream = (D3D12_VIDEO_DECODE_COMPRESSED_BITSTREAM){ + .pBuffer = buffer, + .Offset = 0, + .Size = ctx_pic->bitstream_size, + }; + + return 0; +} + +static int d3d12va_mpeg2_end_frame(AVCodecContext *avctx) +{ + int ret; + MpegEncContext *s = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + D3D12DecodePictureContext *ctx_pic = s->current_picture_ptr->hwaccel_picture_private; + + if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0) + return -1; + + ret = ff_d3d12va_common_end_frame(avctx, s->current_picture_ptr->f, &ctx_pic->pp, sizeof(ctx_pic->pp), + &ctx_pic->qm, sizeof(ctx_pic->qm), update_input_arguments); + if (!ret) + ff_mpeg_draw_horiz_band(s, 0, avctx->height); + + return ret; +} + +static int d3d12va_mpeg2_decode_init(AVCodecContext *avctx) +{ + const MpegEncContext *s = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_MPEG2; + + ctx->max_num_ref = 3; + + return ff_d3d12va_decode_init(avctx); +} + +#if CONFIG_MPEG2_D3D12VA_HWACCEL +const FFHWAccel ff_mpeg2_d3d12va_hwaccel = { + .p.name = "mpeg2_d3d12va", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_MPEG2VIDEO, + .p.pix_fmt = AV_PIX_FMT_D3D12, + .init = d3d12va_mpeg2_decode_init, + .uninit = ff_d3d12va_decode_uninit, + .start_frame = d3d12va_mpeg2_start_frame, + .decode_slice = d3d12va_mpeg2_decode_slice, + .end_frame = d3d12va_mpeg2_end_frame, + .frame_params = ff_d3d12va_common_frame_params, + .frame_priv_data_size = sizeof(D3D12DecodePictureContext), + .priv_data_size = sizeof(D3D12VADecodeContext), +}; +#endif diff --git a/libavcodec/dxva2_internal.h b/libavcodec/dxva2_internal.h index 7a30672ecd..46a359aec6 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -181,4 +181,10 @@ int ff_dxva2_vp9_fill_picture_parameters(const AVCodecContext *avctx, AVDXVACont int ff_dxva2_av1_fill_picture_parameters(const AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PicParams_AV1 *pp); #endif +void ff_dxva2_mpeg2_fill_picture_parameters(AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PictureParameters *pp); + +void ff_dxva2_mpeg2_fill_quantization_matrices(AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_QmatrixData *qm); + +void ff_dxva2_mpeg2_fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice, unsigned position, const uint8_t *buffer, unsigned size); + #endif /* AVCODEC_DXVA2_INTERNAL_H */ diff --git a/libavcodec/dxva2_mpeg2.c b/libavcodec/dxva2_mpeg2.c index b739dc1e54..4c9ed1d761 100644 --- a/libavcodec/dxva2_mpeg2.c +++ b/libavcodec/dxva2_mpeg2.c @@ -40,16 +40,15 @@ struct dxva2_picture_context { unsigned bitstream_size; }; -static void fill_picture_parameters(AVCodecContext *avctx, +void ff_dxva2_mpeg2_fill_picture_parameters(AVCodecContext *avctx, AVDXVAContext *ctx, - const struct MpegEncContext *s, DXVA_PictureParameters *pp) { + const struct MpegEncContext *s = avctx->priv_data; const Picture *current_picture = s->current_picture_ptr; int is_field = s->picture_structure != PICT_FRAME; memset(pp, 0, sizeof(*pp)); - pp->wDecodedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f, 1); pp->wDeblockedPictureIndex = 0; if (s->pict_type != AV_PICTURE_TYPE_I) pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f, 0); @@ -59,6 +58,7 @@ static void fill_picture_parameters(AVCodecContext *avctx, pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f, 0); else pp->wBackwardRefPictureIndex = 0xffff; + pp->wDecodedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f, 1); pp->wPicWidthInMBminus1 = s->mb_width - 1; pp->wPicHeightInMBminus1 = (s->mb_height >> is_field) - 1; pp->bMacroblockWidthMinus1 = 15; @@ -106,11 +106,11 @@ static void fill_picture_parameters(AVCodecContext *avctx, pp->bBitstreamConcealmentMethod = 0; } -static void fill_quantization_matrices(AVCodecContext *avctx, +void ff_dxva2_mpeg2_fill_quantization_matrices(AVCodecContext *avctx, AVDXVAContext *ctx, - const struct MpegEncContext *s, DXVA_QmatrixData *qm) { + const struct MpegEncContext *s = avctx->priv_data; int i; for (i = 0; i < 4; i++) qm->bNewQmatrix[i] = 1; @@ -123,12 +123,12 @@ static void fill_quantization_matrices(AVCodecContext *avctx, } } -static void fill_slice(AVCodecContext *avctx, - const struct MpegEncContext *s, +void ff_dxva2_mpeg2_fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice, unsigned position, const uint8_t *buffer, unsigned size) { + const struct MpegEncContext *s = avctx->priv_data; int is_field = s->picture_structure != PICT_FRAME; GetBitContext gb; @@ -266,8 +266,8 @@ static int dxva2_mpeg2_start_frame(AVCodecContext *avctx, return -1; assert(ctx_pic); - fill_picture_parameters(avctx, ctx, s, &ctx_pic->pp); - fill_quantization_matrices(avctx, ctx, s, &ctx_pic->qm); + ff_dxva2_mpeg2_fill_picture_parameters(avctx, ctx, &ctx_pic->pp); + ff_dxva2_mpeg2_fill_quantization_matrices(avctx, ctx, &ctx_pic->qm); ctx_pic->slice_count = 0; ctx_pic->bitstream_size = 0; @@ -293,7 +293,7 @@ static int dxva2_mpeg2_decode_slice(AVCodecContext *avctx, ctx_pic->bitstream_size += size; position = buffer - ctx_pic->bitstream; - fill_slice(avctx, s, &ctx_pic->slice[ctx_pic->slice_count++], position, + ff_dxva2_mpeg2_fill_slice(avctx, &ctx_pic->slice[ctx_pic->slice_count++], position, buffer, size); return 0; } diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index 1c4825f2fb..927314c5e3 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -54,6 +54,7 @@ extern const struct FFHWAccel ff_mpeg1_vdpau_hwaccel; extern const struct FFHWAccel ff_mpeg1_videotoolbox_hwaccel; extern const struct FFHWAccel ff_mpeg2_d3d11va_hwaccel; extern const struct FFHWAccel ff_mpeg2_d3d11va2_hwaccel; +extern const struct FFHWAccel ff_mpeg2_d3d12va_hwaccel; extern const struct FFHWAccel ff_mpeg2_dxva2_hwaccel; extern const struct FFHWAccel ff_mpeg2_nvdec_hwaccel; extern const struct FFHWAccel ff_mpeg2_vaapi_hwaccel; diff --git a/libavcodec/mpeg12dec.c b/libavcodec/mpeg12dec.c index 020b7b7564..d07eed8744 100644 --- a/libavcodec/mpeg12dec.c +++ b/libavcodec/mpeg12dec.c @@ -850,6 +850,9 @@ static const enum AVPixelFormat mpeg2_hwaccel_pixfmt_list_420[] = { AV_PIX_FMT_D3D11VA_VLD, AV_PIX_FMT_D3D11, #endif +#if CONFIG_MPEG2_D3D12VA_HWACCEL + AV_PIX_FMT_D3D12, +#endif #if CONFIG_MPEG2_VAAPI_HWACCEL AV_PIX_FMT_VAAPI, #endif @@ -2620,6 +2623,9 @@ const FFCodec ff_mpeg2video_decoder = { #if CONFIG_MPEG2_D3D11VA2_HWACCEL HWACCEL_D3D11VA2(mpeg2), #endif +#if CONFIG_MPEG2_D3D12VA_HWACCEL + HWACCEL_D3D12VA(mpeg2), +#endif #if CONFIG_MPEG2_NVDEC_HWACCEL HWACCEL_NVDEC(mpeg2), #endif From patchwork Tue Dec 5 06:46:49 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44912 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145972pzc; Mon, 4 Dec 2023 22:52:57 -0800 (PST) X-Google-Smtp-Source: AGHT+IG/wVmra34h4t1lZiqg0Q9NJrj8aiazDaS9DbDX47TFrQqph1xyVBOoVFADKg/sxJlVtS9q X-Received: by 2002:a17:906:2c3:b0:a19:a1ba:da6f with SMTP id 3-20020a17090602c300b00a19a1bada6fmr3483421ejk.150.1701759177252; Mon, 04 Dec 2023 22:52:57 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759177; cv=none; d=google.com; s=arc-20160816; b=lpDti6h51ECfQd8MnbY+az2njidxfVM1a8D/JhPyYgg3TblbSDpdHh7izVhlBMmGqw lUv3BaE2pts2NsvX0oRhDbfJqK9Bgtz8k6EO6sIfI3UyMwZONYXdsLMEuHMYvmYDWGii TIJ8aM7vi0mADeyYt4oGrN8gJ3xXqfhBU6lS+ktAhlhQw8+WB1J1SxdyX0ro4eaycW3W vBFb4ZyWCHZ3IwH/Hn1fV8BT7xcHJUpLibw1fi5uYFFZVFTmj+UCPMf6fYZyMZTnIKVI my325cx3mIDmbaVk10+KSv1+ckyIRosnXLl6ZbiR8x/x4kPmnYRoojI7fznBe3ylCfeu 3qKQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=Q8WPzFbo3KotIj9RVnRkNjpdultR+5Zg/Hb1M65Cf94=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=U/Oq4tIn1ALHLLN2mDOQ4b0oQj88dHamZD+kokE1NfZZ1w/k4WzPABhSNgF4wgZ/gd wB9ul/MDMntoBBLF/G0ovn+hb+wfV/5zKyFBdfBquZfMknsTj1lcfM0Gz6uQt0SMgGq/ 0XurA0c7WeY1ZKeJke4xbPLcsD/DeEyJJS11g+20OWtw4KOmjsMNcGCGfqX17eDjGeZa MfmjgFbOaSb2kLqo+H6Mnd5s+I7JGy1e2/M++Z1eJ4sq32yVUgGA1rnRlFIxspYuKDE1 dubYI4yK2LmMYjYjQ7BwbNmMFveP4fWc22ZUdVUct9L3tnjbKRnEzBoJuPmKFaWFGOTt GB8Q== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=XawlR31B; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id bz2-20020a1709070aa200b00a1bbb2fc09dsi857798ejc.541.2023.12.04.22.52.56; Mon, 04 Dec 2023 22:52:57 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=XawlR31B; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 470DB68CF34; Tue, 5 Dec 2023 08:51:46 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 061C268CEB1 for ; Tue, 5 Dec 2023 08:51:37 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759103; x=1733295103; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=/w+upb28vv87x6wFi9aVqyrVoRfke55jA5rEsVKo6UQ=; b=XawlR31B/SFi09sge/+zKtKhIXkEF3VzgGjfd1H0tNeneQ9A+/45xOwO t3t85gvklstELgKPIWXaGx7EDkKXGM6cWPsFGjK6gw9vjFTIuJzPG4hPp lgvfZ5oxLxC4ibSSuReGZmf6sjxYqFkamwQETYOIQOKclnTABieO+er74 vSo0VeIwLFzlJwwgzze7Xhr6vcjdRjwEY1Ff6YPt/cCkDfZDvMbyPvNqu zjn+2oLrefXxGPxABMr2R3B9hSAHE1R5EhK6pF5Cq/stUkJ2R0WYhwc5g WCXA0Yp1wJW3GPFo+Jv1WVi5yCOZqZHpYLBYnGCO3qZsWI53w8t+CkjUy g==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012394" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012394" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:29 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853646" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853646" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:28 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:49 +0800 Message-ID: <20231205065034.1621-7-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 7/9] avcodec: add D3D12VA hardware accelerated VC1 decoding X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: vQAA0r8Vqmna From: Wu Jianhua The command below is how to enable d3d12va: ffmpeg -hwaccel d3d12va -i input.mp4 output.mp4 Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- configure | 3 + libavcodec/Makefile | 1 + libavcodec/d3d12va_vc1.c | 214 ++++++++++++++++++++++++++++++++++++ libavcodec/dxva2_internal.h | 4 + libavcodec/dxva2_vc1.c | 15 +-- libavcodec/hwaccels.h | 2 + libavcodec/vc1dec.c | 9 ++ 7 files changed, 241 insertions(+), 7 deletions(-) create mode 100644 libavcodec/d3d12va_vc1.c diff --git a/configure b/configure index 7981833ec3..4b8545ace2 100755 --- a/configure +++ b/configure @@ -3156,6 +3156,8 @@ vc1_d3d11va_hwaccel_deps="d3d11va" vc1_d3d11va_hwaccel_select="vc1_decoder" vc1_d3d11va2_hwaccel_deps="d3d11va" vc1_d3d11va2_hwaccel_select="vc1_decoder" +vc1_d3d12va_hwaccel_deps="d3d12va" +vc1_d3d12va_hwaccel_select="vc1_decoder" vc1_dxva2_hwaccel_deps="dxva2" vc1_dxva2_hwaccel_select="vc1_decoder" vc1_nvdec_hwaccel_deps="nvdec" @@ -3186,6 +3188,7 @@ vp9_videotoolbox_hwaccel_deps="videotoolbox" vp9_videotoolbox_hwaccel_select="vp9_decoder" wmv3_d3d11va_hwaccel_select="vc1_d3d11va_hwaccel" wmv3_d3d11va2_hwaccel_select="vc1_d3d11va2_hwaccel" +wmv3_d3d12va_hwaccel_select="vc1_d3d12va_hwaccel" wmv3_dxva2_hwaccel_select="vc1_dxva2_hwaccel" wmv3_nvdec_hwaccel_select="vc1_nvdec_hwaccel" wmv3_vaapi_hwaccel_select="vc1_vaapi_hwaccel" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 889f844031..5dd5ea68a6 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -1039,6 +1039,7 @@ OBJS-$(CONFIG_MPEG4_VDPAU_HWACCEL) += vdpau_mpeg4.o OBJS-$(CONFIG_MPEG4_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o OBJS-$(CONFIG_VC1_D3D11VA_HWACCEL) += dxva2_vc1.o OBJS-$(CONFIG_VC1_DXVA2_HWACCEL) += dxva2_vc1.o +OBJS-$(CONFIG_VC1_D3D12VA_HWACCEL) += dxva2_vc1.o d3d12va_vc1.o OBJS-$(CONFIG_VC1_NVDEC_HWACCEL) += nvdec_vc1.o OBJS-$(CONFIG_VC1_QSV_HWACCEL) += qsvdec.o OBJS-$(CONFIG_VC1_VAAPI_HWACCEL) += vaapi_vc1.o diff --git a/libavcodec/d3d12va_vc1.c b/libavcodec/d3d12va_vc1.c new file mode 100644 index 0000000000..3d15abd1f1 --- /dev/null +++ b/libavcodec/d3d12va_vc1.c @@ -0,0 +1,214 @@ +/* + * Direct3D12 WMV3/VC-1 HW acceleration + * + * copyright (c) 2022-2023 Wu Jianhua + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config_components.h" +#include "libavutil/avassert.h" +#include "libavutil/hwcontext_d3d12va_internal.h" +#include "mpegutils.h" +#include "mpegvideodec.h" +#include "vc1.h" +#include "vc1data.h" +#include "d3d12va_decode.h" +#include "dxva2_internal.h" + +#define MAX_SLICES 1024 +#define INVALID_REF 0xffff + +#define REF_RESOURCE(index) if (index != INVALID_REF) { \ + ctx->ref_resources[index] = frames_hwctx->texture_infos[index].texture; \ +} + +typedef struct D3D12DecodePictureContext { + DXVA_PictureParameters pp; + unsigned slice_count; + DXVA_SliceInfo slices[MAX_SLICES]; + const uint8_t *bitstream; + unsigned bitstream_size; +} D3D12DecodePictureContext; + +static int d3d12va_vc1_start_frame(AVCodecContext *avctx, av_unused const uint8_t *buffer, av_unused uint32_t size) +{ + const VC1Context *v = avctx->priv_data; + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + D3D12DecodePictureContext *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private; + + if (!ctx) + return -1; + + av_assert0(ctx_pic); + + ctx->used_mask = 0; + + ff_dxva2_vc1_fill_picture_parameters(avctx, (AVDXVAContext *)ctx, &ctx_pic->pp); + ctx_pic->pp.wDeblockedPictureIndex = INVALID_REF; + + ctx_pic->bitstream = NULL; + ctx_pic->bitstream_size = 0; + ctx_pic->slice_count = 0; + + return 0; +} + +static int d3d12va_vc1_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size) +{ + const VC1Context *v = avctx->priv_data; + D3D12DecodePictureContext *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private; + + if (ctx_pic->slice_count >= MAX_SLICES) { + return AVERROR(ERANGE); + } + + if (avctx->codec_id == AV_CODEC_ID_VC1 && + size >= 4 && IS_MARKER(AV_RB32(buffer))) { + buffer += 4; + size -= 4; + } + + if (!ctx_pic->bitstream) + ctx_pic->bitstream = buffer; + ctx_pic->bitstream_size += size; + + ff_dxva2_vc1_fill_slice(avctx, &ctx_pic->slices[ctx_pic->slice_count++], + buffer - ctx_pic->bitstream, size); + + return 0; +} + +static int update_input_arguments(AVCodecContext *avctx, D3D12_VIDEO_DECODE_INPUT_STREAM_ARGUMENTS *input_args, ID3D12Resource *buffer) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + AVHWFramesContext *frames_ctx = D3D12VA_FRAMES_CONTEXT(avctx); + AVD3D12VAFramesContext *frames_hwctx = frames_ctx->hwctx; + const VC1Context *v = avctx->priv_data; + const MpegEncContext *s = &v->s; + D3D12DecodePictureContext *ctx_pic = s->current_picture_ptr->hwaccel_picture_private; + D3D12_VIDEO_DECODE_FRAME_ARGUMENT *args = &input_args->FrameArguments[input_args->NumFrameArguments++]; + + const unsigned mb_count = s->mb_width * (s->mb_height >> v->field_mode); + uint8_t *mapped_data, *mapped_ptr; + + static const uint8_t start_code[] = { 0, 0, 1, 0x0d }; + + if (FAILED(ID3D12Resource_Map(buffer, 0, NULL, &mapped_data))) { + av_log(avctx, AV_LOG_ERROR, "Failed to map D3D12 Buffer resource!\n"); + return AVERROR(EINVAL); + } + + mapped_ptr = mapped_data; + for (int i = 0; i < ctx_pic->slice_count; i++) { + DXVA_SliceInfo *slice = &ctx_pic->slices[i]; + unsigned position = slice->dwSliceDataLocation; + unsigned size = slice->dwSliceBitsInBuffer / 8; + + slice->dwSliceDataLocation = mapped_ptr - mapped_data; + if (i < ctx_pic->slice_count - 1) + slice->wNumberMBsInSlice = slice[1].wNumberMBsInSlice - slice[0].wNumberMBsInSlice; + else + slice->wNumberMBsInSlice = mb_count - slice[0].wNumberMBsInSlice; + + if (avctx->codec_id == AV_CODEC_ID_VC1) { + memcpy(mapped_ptr, start_code, sizeof(start_code)); + if (i == 0 && v->second_field) + mapped_ptr[3] = 0x0c; + else if (i > 0) + mapped_ptr[3] = 0x0b; + + mapped_ptr += sizeof(start_code); + slice->dwSliceBitsInBuffer += sizeof(start_code) * 8; + } + + memcpy(mapped_ptr, &ctx_pic->bitstream[position], size); + mapped_ptr += size; + } + + ID3D12Resource_Unmap(buffer, 0, NULL); + + args->Type = D3D12_VIDEO_DECODE_ARGUMENT_TYPE_SLICE_CONTROL; + args->Size = sizeof(DXVA_SliceInfo) * ctx_pic->slice_count; + args->pData = ctx_pic->slices; + + input_args->CompressedBitstream = (D3D12_VIDEO_DECODE_COMPRESSED_BITSTREAM){ + .pBuffer = buffer, + .Offset = 0, + .Size = mapped_ptr - mapped_data, + }; + + return 0; +} + +static int d3d12va_vc1_end_frame(AVCodecContext *avctx) +{ + const VC1Context *v = avctx->priv_data; + D3D12DecodePictureContext *ctx_pic = v->s.current_picture_ptr->hwaccel_picture_private; + + if (ctx_pic->slice_count <= 0 || ctx_pic->bitstream_size <= 0) + return -1; + + return ff_d3d12va_common_end_frame(avctx, v->s.current_picture_ptr->f, + &ctx_pic->pp, sizeof(ctx_pic->pp), + NULL, 0, + update_input_arguments); +} + +static int d3d12va_vc1_decode_init(AVCodecContext *avctx) +{ + D3D12VADecodeContext *ctx = D3D12VA_DECODE_CONTEXT(avctx); + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_VC1; + + ctx->max_num_ref = 3; + + return ff_d3d12va_decode_init(avctx); +} + +#if CONFIG_WMV3_D3D12VA_HWACCEL +const FFHWAccel ff_wmv3_d3d12va_hwaccel = { + .p.name = "wmv3_d3d12va", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_WMV3, + .p.pix_fmt = AV_PIX_FMT_D3D12, + .init = d3d12va_vc1_decode_init, + .uninit = ff_d3d12va_decode_uninit, + .start_frame = d3d12va_vc1_start_frame, + .decode_slice = d3d12va_vc1_decode_slice, + .end_frame = d3d12va_vc1_end_frame, + .frame_params = ff_d3d12va_common_frame_params, + .frame_priv_data_size = sizeof(D3D12DecodePictureContext), + .priv_data_size = sizeof(D3D12VADecodeContext), +}; +#endif + +#if CONFIG_VC1_D3D12VA_HWACCEL +const FFHWAccel ff_vc1_d3d12va_hwaccel = { + .p.name = "vc1_d3d12va", + .p.type = AVMEDIA_TYPE_VIDEO, + .p.id = AV_CODEC_ID_VC1, + .p.pix_fmt = AV_PIX_FMT_D3D12, + .init = d3d12va_vc1_decode_init, + .uninit = ff_d3d12va_decode_uninit, + .start_frame = d3d12va_vc1_start_frame, + .decode_slice = d3d12va_vc1_decode_slice, + .end_frame = d3d12va_vc1_end_frame, + .frame_params = ff_d3d12va_common_frame_params, + .frame_priv_data_size = sizeof(D3D12DecodePictureContext), + .priv_data_size = sizeof(D3D12VADecodeContext), +}; +#endif diff --git a/libavcodec/dxva2_internal.h b/libavcodec/dxva2_internal.h index 46a359aec6..224a867ebc 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -187,4 +187,8 @@ void ff_dxva2_mpeg2_fill_quantization_matrices(AVCodecContext *avctx, AVDXVACont void ff_dxva2_mpeg2_fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice, unsigned position, const uint8_t *buffer, unsigned size); +void ff_dxva2_vc1_fill_picture_parameters(AVCodecContext *avctx, AVDXVAContext *ctx, DXVA_PictureParameters *pp); + +void ff_dxva2_vc1_fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice, unsigned position, unsigned size); + #endif /* AVCODEC_DXVA2_INTERNAL_H */ diff --git a/libavcodec/dxva2_vc1.c b/libavcodec/dxva2_vc1.c index 8ee23feabf..2b1b3f97ea 100644 --- a/libavcodec/dxva2_vc1.c +++ b/libavcodec/dxva2_vc1.c @@ -40,10 +40,11 @@ struct dxva2_picture_context { unsigned bitstream_size; }; -static void fill_picture_parameters(AVCodecContext *avctx, - AVDXVAContext *ctx, const VC1Context *v, +void ff_dxva2_vc1_fill_picture_parameters(AVCodecContext *avctx, + AVDXVAContext *ctx, DXVA_PictureParameters *pp) { + const VC1Context *v = avctx->priv_data; const MpegEncContext *s = &v->s; const Picture *current_picture = s->current_picture_ptr; int intcomp = 0; @@ -57,8 +58,6 @@ static void fill_picture_parameters(AVCodecContext *avctx, } memset(pp, 0, sizeof(*pp)); - pp->wDecodedPictureIndex = - pp->wDeblockedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f, 1); if (s->pict_type != AV_PICTURE_TYPE_I && !v->bi_type) pp->wForwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->last_picture.f, 0); else @@ -67,6 +66,8 @@ static void fill_picture_parameters(AVCodecContext *avctx, pp->wBackwardRefPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, s->next_picture.f, 0); else pp->wBackwardRefPictureIndex = 0xffff; + pp->wDecodedPictureIndex = + pp->wDeblockedPictureIndex = ff_dxva2_get_surface_index(avctx, ctx, current_picture->f, 1); if (v->profile == PROFILE_ADVANCED) { /* It is the cropped width/height -1 of the frame */ pp->wPicWidthInMBminus1 = avctx->width - 1; @@ -163,7 +164,7 @@ static void fill_picture_parameters(AVCodecContext *avctx, pp->bBitstreamConcealmentMethod = 0; } -static void fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice, +void ff_dxva2_vc1_fill_slice(AVCodecContext *avctx, DXVA_SliceInfo *slice, unsigned position, unsigned size) { const VC1Context *v = avctx->priv_data; @@ -322,7 +323,7 @@ static int dxva2_vc1_start_frame(AVCodecContext *avctx, return -1; assert(ctx_pic); - fill_picture_parameters(avctx, ctx, v, &ctx_pic->pp); + ff_dxva2_vc1_fill_picture_parameters(avctx, ctx, &ctx_pic->pp); ctx_pic->slice_count = 0; ctx_pic->bitstream_size = 0; @@ -356,7 +357,7 @@ static int dxva2_vc1_decode_slice(AVCodecContext *avctx, ctx_pic->bitstream_size += size; position = buffer - ctx_pic->bitstream; - fill_slice(avctx, &ctx_pic->slice[ctx_pic->slice_count++], position, size); + ff_dxva2_vc1_fill_slice(avctx, &ctx_pic->slice[ctx_pic->slice_count++], position, size); return 0; } diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h index 927314c5e3..5171e4c7d7 100644 --- a/libavcodec/hwaccels.h +++ b/libavcodec/hwaccels.h @@ -67,6 +67,7 @@ extern const struct FFHWAccel ff_mpeg4_videotoolbox_hwaccel; extern const struct FFHWAccel ff_prores_videotoolbox_hwaccel; extern const struct FFHWAccel ff_vc1_d3d11va_hwaccel; extern const struct FFHWAccel ff_vc1_d3d11va2_hwaccel; +extern const struct FFHWAccel ff_vc1_d3d12va_hwaccel; extern const struct FFHWAccel ff_vc1_dxva2_hwaccel; extern const struct FFHWAccel ff_vc1_nvdec_hwaccel; extern const struct FFHWAccel ff_vc1_vaapi_hwaccel; @@ -83,6 +84,7 @@ extern const struct FFHWAccel ff_vp9_vdpau_hwaccel; extern const struct FFHWAccel ff_vp9_videotoolbox_hwaccel; extern const struct FFHWAccel ff_wmv3_d3d11va_hwaccel; extern const struct FFHWAccel ff_wmv3_d3d11va2_hwaccel; +extern const struct FFHWAccel ff_wmv3_d3d12va_hwaccel; extern const struct FFHWAccel ff_wmv3_dxva2_hwaccel; extern const struct FFHWAccel ff_wmv3_nvdec_hwaccel; extern const struct FFHWAccel ff_wmv3_vaapi_hwaccel; diff --git a/libavcodec/vc1dec.c b/libavcodec/vc1dec.c index 871944d781..6eff2ec04c 100644 --- a/libavcodec/vc1dec.c +++ b/libavcodec/vc1dec.c @@ -1373,6 +1373,9 @@ static const enum AVPixelFormat vc1_hwaccel_pixfmt_list_420[] = { AV_PIX_FMT_D3D11VA_VLD, AV_PIX_FMT_D3D11, #endif +#if CONFIG_VC1_D3D12VA_HWACCEL + AV_PIX_FMT_D3D12, +#endif #if CONFIG_VC1_NVDEC_HWACCEL AV_PIX_FMT_CUDA, #endif @@ -1408,6 +1411,9 @@ const FFCodec ff_vc1_decoder = { #if CONFIG_VC1_D3D11VA2_HWACCEL HWACCEL_D3D11VA2(vc1), #endif +#if CONFIG_VC1_D3D12VA_HWACCEL + HWACCEL_D3D12VA(vc1), +#endif #if CONFIG_VC1_NVDEC_HWACCEL HWACCEL_NVDEC(vc1), #endif @@ -1445,6 +1451,9 @@ const FFCodec ff_wmv3_decoder = { #if CONFIG_WMV3_D3D11VA2_HWACCEL HWACCEL_D3D11VA2(wmv3), #endif +#if CONFIG_WMV3_D3D12VA_HWACCEL + HWACCEL_D3D12VA(wmv3), +#endif #if CONFIG_WMV3_NVDEC_HWACCEL HWACCEL_NVDEC(wmv3), #endif From patchwork Tue Dec 5 06:46:50 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44910 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145877pzc; Mon, 4 Dec 2023 22:52:38 -0800 (PST) X-Google-Smtp-Source: AGHT+IFADExPtvHSbbajozy0aTTbzLrpXNUMJhxcZjplOJCh3CBxyUQTQc25YZXFx0dwOC2KUwH8 X-Received: by 2002:a50:cc97:0:b0:54c:4837:93f7 with SMTP id q23-20020a50cc97000000b0054c483793f7mr3264323edi.62.1701759158640; Mon, 04 Dec 2023 22:52:38 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759158; cv=none; d=google.com; s=arc-20160816; b=fqBy6OCFCXNWdon3MLroSwtGwJd1X1ogQGnurdVOOww565vkOGVupY14iukRSLsx2b aYEOaLhcaozBISHOfYRIgtGFhF4BRJSAfyPccv01Oduj7LVALd4hqWCqF+uRqCTDLteP eIIB2D33J6STG3RE601X7imKId0cdvcJvJOZnTsv06hpSZLFGiMgdJ1vEVaYFTp9TYNF WqI1CWR9yUWH4LbNI5Po8jGlByBqWZDMG5mHH5sHXSBOUSFrA6zE+YyrAKv2fBpsm4qL lAq+XfD1CuTKNnucClGyHcFSFY+70KIhwhhTa6j+3WNR0gzvzky+EThnLGDY7Wfl7dLw j6QA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=s/Lx4zar+GjwnZdxbg+MFHYww/5ieYE4qR1I21phr+Y=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=On4IjkVh3yAiJF5oUi2ClXe+gxQSfEOfH7mY7lZvUqyPqFiAmo/jtP6PXX4wcaGTxh Yc+BCpqR6Q9gJjjHW9ABztYqp2LwP7HIkRCvfD5ae4NPzqclzBuA38Pzf7bCuq3+Df6S ispLUcTLsTaf55v3cxroil70gInhF40wMYttJv2TRHroCGtzk/5XQkTAZ6B990Oalprk KebaGn92z7IZuP8zui+JaNV7/3o0JpodHMEEajOmbxybshUgP3qYmpGnsVhx/Gkz1U4H UlivvBg3KfX1oQQYCGl0epa/akbHEfet0S8tNxVMe1qL0+7U78NRYrveWVGGHeXcKy/P 3rZg== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=PvYZLrum; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id g8-20020a056402090800b0054b5f8bfc9dsi586408edz.615.2023.12.04.22.52.38; Mon, 04 Dec 2023 22:52:38 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=PvYZLrum; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 7390A68CF22; Tue, 5 Dec 2023 08:51:44 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 08D0F68CEF7 for ; Tue, 5 Dec 2023 08:51:39 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759100; x=1733295100; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=vWfp0LoQUI3O+qtmeDZcoiwpkLWgnGf2S5kzL7BknGI=; b=PvYZLrumxSgGPSBjNeZ+MXUGF+RpThDdwRp5H239l/yGcCqJARqO4ZK+ xovA8rfybLajqNOyhZlr8RzmrbLOQ07lcX1suV0sD8U7gy6APYtjd1Uza ttPF9Z/Uva1Dgw/CrqWA2HdmLhj3G56OBd+S0VXSFEW4HrhlyJU4asfXR VcG/X9mkq3Nh2YARwdnvzJU/ApsrqMX46fkUHMQU1YPMHv54WWOFCwrAV iCe7y9YchrPvKASG7qYkJYTrTQ42oAxz3GIngVEAypT++pqD3zapRyvxW OqRqLAKwVR04l1s/DHhSHIqYGGgLAaE/kn7AAPXCQzMTDeDrErJ681a7L A==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012396" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012396" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:30 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853655" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853655" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:29 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:50 +0800 Message-ID: <20231205065034.1621-8-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 8/9] Changelog: D3D12VA hardware accelerated H264, HEVC, VP9, AV1, MPEG-2 and VC1 decoding X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu , Wu Jianhua Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: Takjr6LyREyk From: Wu Jianhua Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- Changelog | 1 + 1 file changed, 1 insertion(+) diff --git a/Changelog b/Changelog index f00bc27ca4..d4f8bd42b7 100644 --- a/Changelog +++ b/Changelog @@ -7,6 +7,7 @@ version : - EVC encoding using external library libxeve - QOA decoder and demuxer - aap filter +- D3D12VA hardware accelerated H264, HEVC, VP9, AV1, MPEG-2 and VC1 decoding version 6.1: - libaribcaption decoder From patchwork Tue Dec 5 06:46:51 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44911 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:9153:b0:181:818d:5e7f with SMTP id x19csp145919pzc; Mon, 4 Dec 2023 22:52:48 -0800 (PST) X-Google-Smtp-Source: AGHT+IFYGvrHlYLrhme5MdT5yR3OQiKF71fZpOUjy0ndONk7RWfOHPeTA0rPOnv6s6VyDkk2h4qA X-Received: by 2002:a2e:8006:0:b0:2c9:f6af:300a with SMTP id j6-20020a2e8006000000b002c9f6af300amr2391708ljg.39.1701759168108; Mon, 04 Dec 2023 22:52:48 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701759168; cv=none; d=google.com; s=arc-20160816; b=LwGl+qs3RHuC2JxQhK08vnkKC+WT9eckCuP02fFwgI3lg1gsBkfWel6MZ9uY7p89oF lAjkjY1Gi7w5BrL3kW0Dlcwa5HQlXYvH84iw1vT9IZg9fwniIns66u8+F+IRcQf/400T LCbrA6ca7vQ2779g8YH+mkTWVifat+8eP9fbkBEV7F7xUKkkaVRLPYC/SuE6lIlZWMeY 99ClqocTwZPlVItV2UzRUvu4PVTueEcWAbOEj42sqkHgsTjY4x+MsROp3OAYAEjCEHYg BjPbAQd/xl/AAVmpfbcfTh/ZzOga4jz2Wc8Fa58xGZohE3KxPnjcmXw+G5E93xUhjF4i q0Cg== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:dkim-signature:delivered-to; bh=0rX9Chw19YTl2js+BniWw8lSM3w8MMPLa5N3htb/Hi8=; fh=CX/m9qTuMDwrotqtm4RkAOJT6yXlKL2vcfWDitFPXZs=; b=q7H1bQN8zhAUhtOU1JsJntUEZb52ZG7+G1+ejGAMzf4iltWh2DyCuHw6JCqjupmRbD ODFztQtGzQYVxWPw3JTA5TA1r8PRoeX5gP7RCDHzo+cTj9+QTOroKR1fWkEpDqRb8Tsa TisiBjXYXd5bp0muTs5il5TBOpY5UiAMhNxlHV9Vj65gyxWGWAL5TlbwFoQCNXVnvic4 TJu7Ed3hd58RIznJqE7ePfF1jcO7VdtpV+Dg0O7e7tWvFN8gz7utZT+TKOhAYcreRfAK Y7OgNlhwyr1a0Qhu2UD1xfmN7Zi9+Mv4q5dC3slo4FA4n+No6DFUU4R1D3TvAAz2hWM+ RwFw== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=QczYh9z7; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id rh12-20020a17090720ec00b00a19e50282fcsi4112515ejb.217.2023.12.04.22.52.47; Mon, 04 Dec 2023 22:52:48 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=QczYh9z7; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 677A668CF13; Tue, 5 Dec 2023 08:51:45 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.93]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id D19A568CF09 for ; Tue, 5 Dec 2023 08:51:40 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701759101; x=1733295101; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=Z/MowF5pFwPy0D0kYdFZ/H4vcxYKHcE+5fr8NSF+Ljc=; b=QczYh9z7Pe4QhFTdQ/o267flFkK0g5ibz6cYE0Pw+om02NNykFw1qLEi HSQbC9WU3Rm9UTxaUo4+ojH3u1eG7W77N0VEaWqd0iifCCd3kxlF+AI+I Auqzs8ne7yde/MkPEXtlMY/2gaATn7erkU/vzGbgxI7YZUhkSwDOfYc0t ln4zuWpE+vTx/379stO50dmp5YJbEpp6+2Eyn1UgQ+IP6eIaA9C0m8MlB +Pd0JCW2WsPuH0CV6G/lrTyWFpjU4E2NWWaSBXqX32F40jtQqsAs/drdS c+emEeNTCfjHhxTwKbm4qF09yOJVc/mq0PqJk8qJqmhppe3rDIDZk7Low A==; X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="391012398" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="391012398" Received: from orsmga008.jf.intel.com ([10.7.209.65]) by fmsmga102.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 04 Dec 2023 22:51:31 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10914"; a="799853661" X-IronPort-AV: E=Sophos;i="6.04,251,1695711600"; d="scan'208";a="799853661" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by orsmga008.jf.intel.com with ESMTP; 04 Dec 2023 22:51:30 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Dec 2023 14:46:51 +0800 Message-ID: <20231205065034.1621-9-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231205065034.1621-1-tong1.wu@intel.com> References: <20231205065034.1621-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v12 9/9] avcodec/d3d12va_hevc: enable allow_profile_mismatch flag for d3d12va msp profile X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Tong Wu Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: RgtxYzpbTmAn Same as d3d11va, this flag enables main still picture profile for d3d12va. User should add this flag when decoding main still picture profile. Signed-off-by: Tong Wu --- libavcodec/d3d12va_hevc.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/libavcodec/d3d12va_hevc.c b/libavcodec/d3d12va_hevc.c index 6537e46065..7e0c295a2d 100644 --- a/libavcodec/d3d12va_hevc.c +++ b/libavcodec/d3d12va_hevc.c @@ -176,8 +176,13 @@ static int d3d12va_hevc_decode_init(AVCodecContext *avctx) break; case FF_PROFILE_HEVC_MAIN_STILL_PICTURE: - av_log(avctx, AV_LOG_ERROR, "D3D12 doesn't support PROFILE_HEVC_MAIN_STILL_PICTURE!\n"); - return AVERROR(EINVAL); + if (avctx->hwaccel_flags & AV_HWACCEL_FLAG_ALLOW_PROFILE_MISMATCH) { + ctx->cfg.DecodeProfile = D3D12_VIDEO_DECODE_PROFILE_HEVC_MAIN; + break; + } else { + av_log(avctx, AV_LOG_ERROR, "D3D12 doesn't support PROFILE_HEVC_MAIN_STILL_PICTURE!\n"); + return AVERROR(EINVAL); + } case FF_PROFILE_HEVC_MAIN: default: