From patchwork Sat Dec 2 10:12:41 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44855 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1655985pzk; Sat, 2 Dec 2023 02:13:59 -0800 (PST) X-Google-Smtp-Source: AGHT+IEf/WMkY3G7+OD22I5nQWJuLBk4n+8riISIy6pg5MldqX7gqeq377Fj4vW+RvlGAW7DkyX2 X-Received: by 2002:a50:aacb:0:b0:54c:4837:903b with SMTP id r11-20020a50aacb000000b0054c4837903bmr1843919edc.51.1701512039288; Sat, 02 Dec 2023 02:13:59 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512039; cv=none; d=google.com; s=arc-20160816; b=BAmF3X1cFVlwxct3b+7UybtTHDP6n0XjLR/q/IuvckNFepQZQUNbTrtzK3CpNmDooL VtCb3chwxhxUHliyXKuk8NNSl6l58+xHieUWO6LnKVeP5ewcUVkn4r0o/XKlyoj7l+fz efgAqCPRlFmgrq9JhVmvZZn1FkRkkPhc4B40cY0V9p+fdj3Er89m7gzjDEzqLDE7CBLk iiHmnn0556vTBsFrFXA4YF9/vjAhMYtcO5IJERfo5u/7Cj21QEn3X/KkcWcfp99AUTCy phLaXkOZ/c74AKs2XsWceksxliyKLRJaz1VOUbVLygg1ZUjpPjaSV4ICuV2XCvTUirO6 0Tvw== 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=gLK4QN8JoRCwSEajbRSaq3adEHmsoxaK/TLv+Ay1iyQ=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=kw0wGSdGgZWko9E1dlj5Ci0pnUGkhBrDBzwcEcAOt3dCWToOQ8AoIMTWN+3MG7TJUB GauDsjkZ/GwGTv8IPQJBc5xml8WOGzKGQhE0kgPMxK+f++WBmXbYBQwC2MWtlHX4W8x8 59St8Aqr9m7yBUALzMaZ7IazmfI5crVgozDCrqAkq4yqhFhg3Tp4vUOhWHndgB0KFbun bCVwwL2lpxfsr91V4s+RbX5LqnFLh02ScDmghBvJ+48qYe2tAPK3AEvLidO6mmx4vI0T flErx9r2iKfoZS4+ZNvmlC20Vuqn7kq7BfZR3wyQklyRWqwGM2MmB7ncN88tbHxtCeAK P4lA== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=M3+oJSzx; 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 m25-20020a509319000000b0054c7023973esi865430eda.141.2023.12.02.02.13.48; Sat, 02 Dec 2023 02:13:59 -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=M3+oJSzx; 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 264B468CEB9; Sat, 2 Dec 2023 12:13:43 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 76F4D68BFFF for ; Sat, 2 Dec 2023 12:13:35 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512020; x=1733048020; h=from:to:cc:subject:date:message-id:mime-version: content-transfer-encoding; bh=hwI/BVHlTzw4n5v1ljd32Rs0nzsRHJ8lbSL51/y95R4=; b=M3+oJSzxf/6+v25a1LU6TiTFocTDAvxUZCDE6k55VT4PALe4FyizPqPI oE4XcvUxg3tLeYsNc/RWUSJy2pFq+Lk0ixhbxCaqn4g7Y/tOb25B1HFqx b/JIluxYaT/sZpL77Ku+qJmCJHLd5NJ4SVEbWxQx0JVEPt+QNeoKoQFTk RTid2T7V41nlbKIJL0PtvGcPOYBrGUQj+wd33ecZfC0kxy94GV9HWlUUG g/Y+5ahaN2zhvNa4Vyo2StR4e6nlZ2k8s8biwWB9Tcl8sj6aPbEgYDjoQ 9js4v6BXtoLUSmLhAnGVhl6JqpRjemboXtGptS+b1iqKQS6z4SUga147g w==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794820" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794820" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:32 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711594" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711594" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:30 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:41 +0800 Message-ID: <20231202101250.1410-1-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: PHKop9jWq4Ck From: Wu Jianhua Signed-off-by: Wu Jianhua Signed-off-by: Tong Wu --- configure | 5 + doc/APIchanges | 5 + libavutil/Makefile | 3 + libavutil/hwcontext.c | 4 + libavutil/hwcontext.h | 1 + libavutil/hwcontext_d3d12va.c | 704 +++++++++++++++++++++++++ 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, 930 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 838e627084..cdeed9bab1 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..a21206377b --- /dev/null +++ b/libavutil/hwcontext_d3d12va.c @@ -0,0 +1,704 @@ +/* + * 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 for downloading\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, s->sync_ctx.fence_value, 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; + D3D12VAFramesContext *s = ctx->internal->priv; + 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, s->sync_ctx.fence_value, 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 Sat Dec 2 10:12:42 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44856 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656012pzk; Sat, 2 Dec 2023 02:14:04 -0800 (PST) X-Google-Smtp-Source: AGHT+IHA/Ofc7075BQRBPhNIHNW+HVhMUQsFK94lEcAH96iJ36Rl3pkuQrrpcHCehdxtfbeujPzk X-Received: by 2002:a7b:c848:0:b0:40b:5e59:ccaa with SMTP id c8-20020a7bc848000000b0040b5e59ccaamr1017585wml.139.1701512043901; Sat, 02 Dec 2023 02:14:03 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512043; cv=none; d=google.com; s=arc-20160816; b=rE0nZ3EmGOxuBIdEknEHCpqXsRN6lH14UnGxALOMulzAFVZnkfhJfOMcPX7H0xVIUJ +/ieUOIVV43/5r/5ARj3ausc3QAPPeFsN0hnoG0jhZJHu80CdOS0mmekDlMTzbWRpWI2 +jAUU3avlNXUeq36qFDHosom/GsMuv03kCH4fQYuo2e3CYfmkU+Tt0Cbvt/DGK3eWUJH rgz4KvO5TNd2lLhO5GqRN/cueJo3zpdQcNiKgwbaiWCZ0sLyQ4kXdNigu5ZL+3P99e/o R0MeOB+IDnWAsrECaxqZXsEJtO+2oPNra905RNWBi5Il1S8bt1qhUpRKpGzZ4Ezr7L8f 96qg== 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=ZvS2rUmToI4S2RelJr3lrgoiioh5bcncHeUdnB8u+9Y=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=Uw2/Nfaq1MNpZB1VBysvuo0U72WiLNIscsJHtsYmPOQCqOZwZGb2YvdDccAnR+B7HV t23PjU1ugKCm4xBTJ2QxzWAz18dsyOzrTmabOg4GHRg+ZD1OPQ87JJhrmjk1GV1WMZ+S uJu/0ZBJezHZd1QLIgvTaPNRPYMlRnPXnu4tqcNexhzEvIC3qOIFJV0qe/iCEFO3vwgq BipMyy/Xe0zKZAjUBIT7Z72jNFzaxLHAc0IItEAvN4a9B5YMc4Ledb4V5ed9WBVOuwzj cUWr4TUA/Wv2SS9y23BbvywsJ4kC/1uQES1FghDCIut1wH7/xLIGkMgCTbQ7U9VJtdo/ SiLg== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=NNN503uN; 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 l15-20020a170906a40f00b009ffd688dc76si2641274ejz.318.2023.12.02.02.14.03; Sat, 02 Dec 2023 02:14:03 -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=NNN503uN; 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 5B24768CEDE; Sat, 2 Dec 2023 12:13:44 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id E41E268C6D8 for ; Sat, 2 Dec 2023 12:13:36 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512022; x=1733048022; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=C+8qZnYZl0Nhap52xHaFu1YcIAj5Vdb106wWl2k8+oc=; b=NNN503uNd5a3gKxvCnNTY/QauihjokkyM36RVezlgAqwtiVHSMMsokU1 liPuZpQY+iZipASLF7zsmY0/1ljSXVbN2R3tPoccZUTv3eIAUHm2ogFcQ rjlKtIgcu7K3WgyKpGLz6z0K2YFdbkzt5pu18JQjkXhoWwDzstOLNgngo hxNTFwdmPkoDOvUz5Ur85qaP7dQRQk6C+X9Ew59U3+1K3jq83ovMplt7m 4v3Xc0LSAS0wRKUqhI1G+xkIuN9712O3ZUG3lXIK+QwoF6rAQFOM2CWgi e4rS4werx/UPzCF1FpzMi6xG4n7fJYICBY3WTNZs2BKOxpFVVtTkF2/ML A==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794823" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794823" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:33 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711608" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711608" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:32 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:42 +0800 Message-ID: <20231202101250.1410-2-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: 3BymSqa+itdn 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 | 564 ++++++++++++++++++++++++++++++++++++ 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, 1061 insertions(+), 67 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 cdeed9bab1..8709ec2620 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 748806e702..80a8cf5f57 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..c8228fdaef --- /dev/null +++ b/libavcodec/d3d12va_decode.c @@ -0,0 +1,564 @@ +/* + * 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); +} + +static inline int d3d12va_get_num_surfaces(enum AVCodecID codec_id) +{ + int num_surfaces = 1; + switch (codec_id) { + case AV_CODEC_ID_H264: + case AV_CODEC_ID_HEVC: + num_surfaces += 16; + break; + + case AV_CODEC_ID_AV1: + num_surfaces += 12; + break; + + case AV_CODEC_ID_VP9: + num_surfaces += 8; + break; + + default: + num_surfaces += 2; + } + + return num_surfaces; +} + +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; + + frames_ctx->initial_pool_size = d3d12va_get_num_surfaces(avctx->codec_id); + + 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(sizeof(ID3D12Resource *), ctx->max_num_ref); + if (!ctx->ref_resources) + return AVERROR(ENOMEM); + + ctx->ref_subresources = av_calloc(sizeof(UINT), ctx->max_num_ref); + 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, ctx->sync_ctx.fence_value, 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..645549c20c 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 +120,26 @@ 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) -#endif + (ff_dxva2_is_d3d11(avctx) || DXVA2_VAR(ctx, surface_count))) unsigned ff_dxva2_get_surface_index(const AVCodecContext *avctx, - const AVDXVAContext *, - const AVFrame *frame); + AVDXVAContext *, const AVFrame *frame, int curr); + +unsigned ff_d3d12va_get_surface_index(const AVCodecContext *avctx, + D3D12VADecodeContext *ctx, const AVFrame *frame, + int curr); int ff_dxva2_commit_buffer(AVCodecContext *, AVDXVAContext *, DECODER_BUFFER_DESC *, @@ -161,4 +163,10 @@ 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 Sat Dec 2 10:12:43 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44857 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656072pzk; Sat, 2 Dec 2023 02:14:14 -0800 (PST) X-Google-Smtp-Source: AGHT+IEQkA0PEwAzKBM5p08966O4d5KyMR91tuBosdhxmfe43cNx/8WjMdxNdXrS6AY1mO4+JKBE X-Received: by 2002:a17:906:2e8f:b0:a19:a19a:eacc with SMTP id o15-20020a1709062e8f00b00a19a19aeaccmr2007402eji.133.1701512054352; Sat, 02 Dec 2023 02:14:14 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512054; cv=none; d=google.com; s=arc-20160816; b=qIb4PGYC8ZJSmSt30Uq9LjlI/nyByqGIb5PeJOYHuX6eS0Bxxbb0QSDiUkAgdk+FFh VoLL573P0Q8Hn03jlVsLXF2bgJINnBCyr6Fj+hCahhmqVxczRoAhXXZmxUDCZgrCy0aH jjveNfxkcQLfNuzdVj/+JSTShFEcOtvqSqJ0VNdCjDPGhN0LqjCfGEZNtc9gbnZ+vasS OOLWSjpHdvfiW7pFp59seFcYct+DfQAFkxDobxRXWvWnnHiS5s9NQMAK7L5PuZ3nBtSI XkQw+3uMj7q1BvxnUlzXyGIBIYoW2dUjM14LDQX4Iz6VmLA9SfgzsBvKZdErGZFb43CF 7DlQ== 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=CWM6kW9EnC7gQ/bcBN9SRHndJq8xp0fWA6OjUhYR0YU=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=wPfgwu+n42WrWN/mvGjoMHjkJj0FBXhKG4dJvvoApq/XFeMtL9bMiM/kfdjzy6MJrd S8+FhLY9jDMW/G76X35fu3lf3R7aEx/XJfyVc7htEXKGtI9ny91/pt0HvlSKb1+Xqa+K RKdOZYNhTVre+V/OaWi7nF5G4hQ/qW064+q3NL4jZSgbvees/E9TGfJGTn8dY3W5tVCp fD0UXWNJ+kboMCzOBmwjNOXQLhjkWtm47GP4poSfpeSFVZbutE1bCZu/4+RPC61cSIeX 18Me4nRoO3toh28eyOALBv0Fk7JwxOOKSJLv8GjZunqkMxbQBxqvMYbne7EJPaVTlOGM 7Zqg== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=SP0jPvo5; 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 lt26-20020a170906fa9a00b009e5e19050a1si2617050ejb.425.2023.12.02.02.14.13; Sat, 02 Dec 2023 02:14:14 -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=SP0jPvo5; 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 7DEC468CEE4; Sat, 2 Dec 2023 12:13:46 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 2A0C268CECE for ; Sat, 2 Dec 2023 12:13:37 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512023; x=1733048023; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=1jXbvW3W2AC+Yyw+N7lGmdlzvjLUHsw3NtFOrjyE4RM=; b=SP0jPvo5ZdJxEjiHypUg+nxapu3CC8qwyndfqO3wqMOCcShk1k8KHT1o ECuFjs7flS/HV883xtvBR4XsAUV4vQ6H0vePp/KIra9iu8MoBt5KjeL7Q 5LG7BSUfNbh58VEZf2htMxOamylVIadP77aJ9BIBl2BSkL1GiN36YnuOz pQTHhFnv3mKo4hMxWjjtNY5Z9IA+DyxZi000T5rfQVxw1LdmFlo5KLqmN hB9G/+ayHlv6JI85blQFTqA9BLRBJlDmNdpvYjpOroHgOFEqtbBgaPApa sL98AtWTRm68yE18FQ4oum+RjvQk9h15Q9IxPDYZq1DhaIbamOgGw48Ma w==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794826" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794826" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:35 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711615" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711615" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:33 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:43 +0800 Message-ID: <20231202101250.1410-3-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: xHr6i1fs2XG2 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 8709ec2620..26bd31266c 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 80a8cf5f57..18108ebf74 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 645549c20c..a3b751bbf4 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -169,4 +169,8 @@ 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 Sat Dec 2 10:12:44 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44858 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656141pzk; Sat, 2 Dec 2023 02:14:24 -0800 (PST) X-Google-Smtp-Source: AGHT+IFX/tBERy3el7jiqMj7aABIH5xBefFQXBelPzBw5l8KSlHJi+1c2i2yH09wtLOAlqHNZpMU X-Received: by 2002:a50:9e8d:0:b0:54c:4837:75a2 with SMTP id a13-20020a509e8d000000b0054c483775a2mr1816172edf.78.1701512063733; Sat, 02 Dec 2023 02:14:23 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512063; cv=none; d=google.com; s=arc-20160816; b=xCRjqznBXrawClz2X3yAesT1vVIjuqxn+BmTeltqN3hlV4+hOxIKgA66ZCpHk8lAwM g8tiuTr92R9wTHEvTxkel2oajM1pYdRW77AnE1ZF0znlxRGxjXtxR1Vo97ddYuIBesWW dCmYlU3bV4y/kEi/Jh4XpEV2O3Cdz8m18J2q3nl2mTcH8zjyXzf6C85o2HPAmf0AwCH5 SpYr+KX+AIY8TmvxedC8O+ktvpDy/W/XHldkvaxk89xMYzmAnl0MNpi3fleVrGvJhKPa qsaSnOYqPb87kN+NCVCQa/P9Ha7b4g0ezy87NO2vrdcUqmzfpFA0/qfU+ajrF3faI2k/ GH2w== 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=5ZDI6gj49sz1IxnD8vywo2/aJ3cMkv1+kKZVS3BxzyE=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=xRUzZbe2YeXKdi+ZMIn9I/8kL9RQoZt3XJ0OBubJZO8C3Gml/UwhM7/a7Cn3IcQf+/ zLmsIh7rwwvFxCJYkxxgnqn+tuw/miSaFQk1TbJmwede8WogVx4r/GFxaVfw10JVXET9 lPP/xS9yXsrE9lnlJY1W67Qe6Mqeg/qPpbleJxnFRiHBdG4D9hLOFIdd8N9z68H+0RRS RkBma/sw53oyPtWT1k9lKZAgsD46wQmOGFgHtCnb1A/aaR5WFwzmt/JJJZkPBOKpNDJ1 kQiUEvV3q+g1yY47wVtILFOXkLqXMJjZr4k1+tC4a1Km81KLF9aPCfHwU/hv59qaOlwv qqFQ== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=OLqdPuo5; 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 z29-20020a50cd1d000000b0054c5ef5de9asi1345648edi.27.2023.12.02.02.14.23; Sat, 02 Dec 2023 02:14:23 -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=OLqdPuo5; 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 9437C68CA16; Sat, 2 Dec 2023 12:13:49 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 3704668CEED for ; Sat, 2 Dec 2023 12:13:42 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512027; x=1733048027; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=Jf1u/tlXTp1YnJw8JZJ0mUUTSm4Bj5E0T4yWipZN+FI=; b=OLqdPuo5zc0Hk0xJKGAB+QfRpddW0TiZdC7wPcJh7tb7vB4W5MwxAtLx mBb+Jz/5gnTwGIaIcZp1aISNckSiZWGmJwitExodheW0eqx+3E7acfc+A TehhO76B09QLX4n7+QNADQQhG/O2lia9AkBVra8VyaebMq37RxjbYZHi1 LfVrDkkaCMmqlz7hAXJk6OSKmZsqPDT8qwfEcpy/u2Mv+XWzizHrXswvi 5YEapDVrQWqFAJLw1pRUxByv8KaPCP0j9BCk5Dk7FgRspGPFJE0hq86hW BUJZr1qXNQP4yu6UOflXNh6sbX8ibwcjXSFaTfJO4kabDbM321387ocEr A==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794828" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794828" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:36 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711631" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711631" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:35 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:44 +0800 Message-ID: <20231202101250.1410-4-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: F7bK1n+cjWfz 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 26bd31266c..f62ab51c45 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 18108ebf74..673f45d3b8 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 a3b751bbf4..4b317f4c8c 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -173,4 +173,6 @@ 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 Sat Dec 2 10:12:45 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44859 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656206pzk; Sat, 2 Dec 2023 02:14:34 -0800 (PST) X-Google-Smtp-Source: AGHT+IH+u+cfcq5Njn457WsOMyKc418KM9e7wmeHuXqN89Vp/gmPVAbOKeKnf7yfwb3Q27Q6shyo X-Received: by 2002:a50:c29a:0:b0:54c:48aa:cd15 with SMTP id o26-20020a50c29a000000b0054c48aacd15mr3135160edf.37.1701512074463; Sat, 02 Dec 2023 02:14:34 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512074; cv=none; d=google.com; s=arc-20160816; b=sAxHgadg00msXCOL/jmIFyNE5FbRzxxBOGBRB7JKAruaSTLIHTG2YAUQJKz5AqCn/i Um7D7wY/xYb8kEfvn/Ilqhh83akx6fdYpYpAMDUbosoMBCPM1WO1u8/z4J76/nH5qhtl dbhrdKu0EUIG1VR9fn0qDVWQoGGxNfo8zTXNxgDWzSgpLP2uPlT7MmFTuE4sJY9Zw+Zu L7FaC5SJC9Om5R9DfpfCoAmauEyMfR3OefFPnfS/dqHVVyYVJn7F8q/P6wS9FG8oe7GD /m3I6wztu1BAERB0JFfxiC1akma+1vD7ur02DBIOTRCtRZBZxXiqyrFv1HaBAUI4UOf3 imiw== 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=TqPnipW7lSyh3KM9wxFyJj8X0EvQCxFn8OuGlks9oxg=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=HAiFg9Ew4SPUQ/CDtZYdJql1Kfedj1xWluRuLDcXmx1mB7UlFtDYFl3T3JV7hJZaK6 2YK3iSujP5S2hYENzzMfnQaZaLEHGff2Ory37A5mbRc5yW60mRJ4+ZsGtyMpi5Vpj8zj eAii6iwuQDLy+SrK1nyC05oswaQP6CcCTkhJ1AR9ooAlkZvcKNXCrpGeaUi0TO7ujgKr GZkL+tKuO7s1NZYyjYMMHo6lrG2Pk4GK07z3/ivJLZ1o5SJh0HsqHHpdFAH8bI2RcHBe dglLTEDOGMicN1JVKQ/EwQsSddwXLoXMm3kQvy7BRMj+X0hq4UWgVBatfQLWhDLyp83B pAlg== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=BuY+xRZM; 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 f25-20020a50d559000000b0054c54fb37ddsi1445393edj.188.2023.12.02.02.14.33; Sat, 02 Dec 2023 02:14:34 -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=BuY+xRZM; 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 A8C2868CEF7; Sat, 2 Dec 2023 12:13:50 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 6D84F68CEE1 for ; Sat, 2 Dec 2023 12:13:43 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512028; x=1733048028; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=+FBd8sfVeQd1rK6r/ZDNmpUhC9Paqu8v4u7/8NIvwSk=; b=BuY+xRZM0bqZEYfn7F/PFsxyPHcqCMIzM5EqWj86RzF5KrYLgFB6pI3f v69u+pK/sNPXKDKtVV+FBc9JW13bAco+YaEivg6cRdoH3UITtW4U0Q541 6CGH2HehDx4TmMsxRjkcXivmW1z9sZPBb9UWuWgiRQyg4ykmTxhXWTAj9 Ax8x3cWZ0r2C5DJqt+SqhiuirsDGv7FpKCDAWt0nrUWxsqP+w2VqMET7s Yzvu3YTai7hoOoUi7ulm55RrEd67bvXhgTeUIpwLhYJ+ysS6v+1ONpcPK P1zqVYZx45S2bIFAVDus6R9dPgJl0Wf5lSrta+E/SnJKPZzdJSp//Co7P w==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794830" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794830" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:37 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711642" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711642" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:36 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:45 +0800 Message-ID: <20231202101250.1410-5-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: 1953VHPbVKxe 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 f62ab51c45..7feeeddaf5 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 673f45d3b8..0f152df17c 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 4b317f4c8c..1a7ebd6e8e 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -175,4 +175,8 @@ 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 Sat Dec 2 10:12:46 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44860 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656267pzk; Sat, 2 Dec 2023 02:14:43 -0800 (PST) X-Google-Smtp-Source: AGHT+IGwfiCMJ4uE4xu91OhZ5Xxijsvb4BK4wvtk8IVoo0uxjHIxYdtK4EvpYNpzCgcrWpkQFDvg X-Received: by 2002:a17:907:1109:b0:9e3:fbab:e091 with SMTP id qu9-20020a170907110900b009e3fbabe091mr1759463ejb.15.1701512083334; Sat, 02 Dec 2023 02:14:43 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512083; cv=none; d=google.com; s=arc-20160816; b=jxAWq5L2UAs3IuipzTdxTAk0o8Ixm/nn7MtVXlpFOGkL+0Zy/XkoOKkjlpRKlXgemn yyHIhxKshkcUVgPyoKZyB272vbk4ew9VLALXzcqBSfKai3/ciKKICSHJSaNfhYBUKZXQ nBnrxlDF/gHfrAP6iwJhL0hRbSXWcC3B9s1e0pCF+jw9MW0MKr4fHBNuW4BrnVf8NmbT wxu/S7YtI4RbQYBBiXV7uwi+RPXOsRR5tgMieoOdg9nUEaUqgWhYMxJJNdDsXRwuf4n1 LHyBADMj3kyGHjllswVKGyl4J2CVjOheT5teVRlx+CPjNyIb7yI7K4W6bq/yCVJIxhW/ 24rg== 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=LjAMIIhdHo/I+yadO48T9PcYkLRSeunBG/XAxf52GSA=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=gQ9K+N1UVyB2bPa5WDFm63mnZ+VDaxypa6zOfSoqhegp2NG27SqOFSHRd0UyNpcCAK P4KIBs7O3UeNf6dcZWcJfCl/JX8KClOzNhaD18+c9ZiNl9YNT5ZI8JjMFILpu7qHRXM8 9JD1J0tSjQCZdSGejnxoY4MX+yof2A+IR9TsYAKySYtPa1dz49I4usHMdmIs6I+Umq2b pUGtufDM4axiqbC5mBq+xYJFb9FkCxjVvwEo7r7tCRI2j6sGj4zq6+tS8bf2+cd+29He 0lLlwZxOMldYvobPtVhNigahMm13f5DBCUjUsnYVbNPqt327BfGJFweN6pShu8fxVbMi 3vrA== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=E0vtXeUW; 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 hp28-20020a1709073e1c00b009bf138f9d95si2777109ejc.48.2023.12.02.02.14.42; Sat, 02 Dec 2023 02:14:43 -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=E0vtXeUW; 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 9861168CF01; Sat, 2 Dec 2023 12:13:51 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 47E6B68CEF7 for ; Sat, 2 Dec 2023 12:13:44 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512029; x=1733048029; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=XCzYHTSz2fqrNaZmJ8PcBYtiiqW02luEn5MQ7+SRb/M=; b=E0vtXeUWGmc2WTZBVO4SkmkFD5hVkR8zVEEFODuRQxPIu1CvE+9C+PaQ WiMoS9Wj7FlS4TQSqO5Axc1yMcilGRNTJtIB86spshFcRVeGJpvT9QSFv 5evCmCXtg1/2jx7rY4xrKvqO6PuauMtZztM1P3Msb1lo4IeG+hnN9KDFB 2pgYeHB6Mp/53LGnrkCGPL/ou5yK6ryGo5U6j7huBRWxYkZbUEYNk3Kuw 6NdSu6L6eA0xKgoYX+j0+Uzz/AMq4WZA3+Dye5EBYeUBZvRMMKEJanyQH XlngmmbBwbLcTSnIu5Es8OUK0iAFhd2BpXYWRVSzRHNvPv+9G5+iIaI5w w==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794832" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794832" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:38 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711650" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711650" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:37 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:46 +0800 Message-ID: <20231202101250.1410-6-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: yLy4i1OIYj2o 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 7feeeddaf5..000ea76c34 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 0f152df17c..00c1c01cbd 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 1a7ebd6e8e..7f84b0d2fa 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -179,4 +179,10 @@ 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 Sat Dec 2 10:12:47 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44863 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656428pzk; Sat, 2 Dec 2023 02:15:13 -0800 (PST) X-Google-Smtp-Source: AGHT+IF/sryxRr9gzXmDclJgmtPvxTvIrvnp7bapsymWYvx8vrd6a+pqX//WO/gEnXYYP4DToyyN X-Received: by 2002:a17:906:3bc7:b0:a19:a1ba:8cee with SMTP id v7-20020a1709063bc700b00a19a1ba8ceemr1548086ejf.140.1701512112778; Sat, 02 Dec 2023 02:15:12 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512112; cv=none; d=google.com; s=arc-20160816; b=cD/lMRwAiFIWDCpbWTwOFFaSA2KP2ysMJIqWneNdPvX02u7F7Gq9XMeONnRUI9/ONp 75GjHkIZXsXtONzYhmf5HuufMBIKVdsIq0PQU4Tnt++p3jyRi3fdEerJSA7RvAnSzDJt +8CDSjEkocMFLIUH3j4lUZ0hqBL4Q+20YIbYM0ScVXHl5C4qLe6jzs6m33mux4SUKYnh zDZn7n1IU9hvAhUGpRqiVCAZzPnux7dC3ac3+pYXip/A/NrmOvut98kX+QG+WSsTB+gz YmOQbBEWECiRQxzGj2VMWlVfq/uQjn78GrOvFD6F1J+Plw3ML10kylvkOMLji+YN+Ai4 ViVg== 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=WsEu87UxYPf6f/fklICUOA489dhkiN+/0ADeAzstF4U=; fh=9GN8hc/WTnabikg2VxPLeDN+i3R37IqYi/LhxNANmV0=; b=YOdBmQ5qjuqNN7NOfS2MUD2DVTPbZRRdb9mU/x0NiM/Kt+ZEL2xQ7RJQPDMAmBOCco k9tSpfDVYnVcJ75BrvekKcuMox8K2BXNN/UozwCK77L59aszZXZFWtYODWEEQnGCLwnm 8xeWZkK8sg0wSP4oZ1gCiEs0Vh7j+asfbo8MP3l9dwkGWG9ftqqRriCLGTrAlSjHhkCQ NYdk3rI8w0MA9TwHKaBaXprs1+Pq4YU+85D7qfY1hslRwbuUm0P7glzfETG+mKATYwDZ K7QXctj6pYbEc7SEjwIs/mP7R7O/8rEcApfKjnJEpN5vHrVDRZlCrKwrvexBtwwqpCez Zg8g== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=TRIwarH+; 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 y5-20020a17090629c500b009a9fbfccdc2si2569940eje.369.2023.12.02.02.15.11; Sat, 02 Dec 2023 02:15:12 -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=TRIwarH+; 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 9F90768CF47; Sat, 2 Dec 2023 12:13:55 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 279DD68CF2A for ; Sat, 2 Dec 2023 12:13:47 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512033; x=1733048033; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=OuLwLRm4UVudLMpdzhSIfRL7Wc1Ja5IcsHKgVfayegU=; b=TRIwarH+AIhblk6r5cK5HtihaAKDnoROv32aeXX/JPoMNA9PIGuDGWdf fubib++ONNDzfLwm5ScNzzt92fsjuPlOifaSBKIQMAN97FqSTs6TlPfvY NmpyRpcg1q0G4Qwc2ZBr2OuXorVq64u1KTHsrumeeav+obkSKuJ6Fzk6d slvkn+gB7jeJUjd5LiNSnexcDn0tTozt5eV87KjjKMxYPm1xQnhjXqHBz PvubHKyIvp62w7L8T7nd+DH5wp22DmLRIJfqTJI1NNH4lSkj94rhQPx/T LnBznHOgGR+SuvRsN4CikyDUmd8WBW+rZozGjlSSYpgNf7dzosCYgxDDU g==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794836" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794836" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:40 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711661" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711661" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:39 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:47 +0800 Message-ID: <20231202101250.1410-7-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: KvdMjPPwZhxG 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 000ea76c34..72f6edce91 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 00c1c01cbd..cc2f922b8e 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 7f84b0d2fa..022a1bbea5 100644 --- a/libavcodec/dxva2_internal.h +++ b/libavcodec/dxva2_internal.h @@ -185,4 +185,8 @@ 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 Sat Dec 2 10:12:48 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44861 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656312pzk; Sat, 2 Dec 2023 02:14:53 -0800 (PST) X-Google-Smtp-Source: AGHT+IHmzkjYiqYIvmt7/rJ9l8/M2jfLu2dqOLn4N1B64wvEGiO14jRrrNT1DNX/hxawpPg/VW2O X-Received: by 2002:a1c:4b0e:0:b0:40b:5e21:e286 with SMTP id y14-20020a1c4b0e000000b0040b5e21e286mr915688wma.115.1701512093582; Sat, 02 Dec 2023 02:14:53 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512093; cv=none; d=google.com; s=arc-20160816; b=aWRvmLJATBM1je81Aw5Tn+lXPPwK3mc9mqqPgxtIqN3IYJHEiw+AATKQbkLz4ZjZDl JYumfVfU3GujggNrAqLRQUd1GY2IWu4gL2Nji/mRs/HBcNjlvvxjZpKaHQiSt9zxiFlP dCnA6/BkPfk7MqFCE0pIln3ArfqBF53fDCKBh4M7hTyJtSW0v8LDsjErCY/hpC3BL0th b5iCxBD8+VH9AqkO33aRGnf8BNjZsQUAhvWjL1EOY9Ffkx79n7FyF/iaYZsi0ClcftQG xVoTnnELidQB3cCM5heQGFpGcyNLt9LheIg2UsU5WXWVbTxvA3IT8a8aJdp+m9dk3/ys E3jA== 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=fdzR6P5jcuIidyUUYsvjZ/ACF/WJOBO+2BOBhKyk+iRiT4eTDJZIzyNtNiwDsjAuOd +DAqXakc1lJ8I+9cfeLLKX5rVM4tkEAP/MhQDiI3A6t2fGhEquGprZv2THYflWHJtxka t/akdtGWHBojDrk8vAuQOgenJ3hvGS+vUDYKO6GJRV1JOsrlGt8ieV3//9KqAW5x+Ggd DKSWBSrRnDGw9FdziPbxuGvc+V6jxwqpf1COiy4F1fOME/YvcOJdO3qf5OtAL4MQNPBV 3S71KqIaxiJnHSDMN9XU+nfbUy9UVdGlKvGmYpmvqm8DpWSXS2+9AtxflfnbOkDjgZTm Ihow== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=OcqyKf0L; 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 g8-20020a50d0c8000000b0054c470cb514si1827975edf.232.2023.12.02.02.14.52; Sat, 02 Dec 2023 02:14:53 -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=OcqyKf0L; 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 9052668CF23; Sat, 2 Dec 2023 12:13:52 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 7472168CEF6 for ; Sat, 2 Dec 2023 12:13:49 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512029; x=1733048029; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=vWfp0LoQUI3O+qtmeDZcoiwpkLWgnGf2S5kzL7BknGI=; b=OcqyKf0LXt6/0Lip5GO3G4f0vL8a5TAiO3IseoyTBcfSnSrCoac6zN39 Bhs8gOqnlm+XpdD1JVl9JJ9WZkxK7UmoftwvIIoKBQ1TUGAyFBVI6AmeW Xru9nfCe/+W8jSutsTzFujD8ve1/UlcfJs9bdQzehLqtBnd0lIAHVLSTG Y8UZx6/A6U3kbyHcfLZAJabfEJNFelGCSfz7H3xvzXjBsc6SoIBpDzvA7 DPVYI5B/XHbjB4ZKl9ktQSeJiN7elGbNrzaGzrxiSPWO2V0gk5gZYVo10 Ksv0x7+9L045GOQXK+fpTiv3YzrzBVDy/bwYXiT3Vc/8MGTSYlY8AfA5b w==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794841" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794841" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:41 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711666" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711666" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:40 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:48 +0800 Message-ID: <20231202101250.1410-8-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: Eb0pQh6tqHHq 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 Sat Dec 2 10:12:49 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Wu, Tong1" X-Patchwork-Id: 44862 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:a301:b0:181:818d:5e7f with SMTP id x1csp1656363pzk; Sat, 2 Dec 2023 02:15:02 -0800 (PST) X-Google-Smtp-Source: AGHT+IE0JZY6GSnGmU6pwicp1Ijo9ZZkJRXjIkQj9Rg9jMbAGMw2nCCeKV21MlwI60vtNEWDuOvn X-Received: by 2002:a17:906:4b:b0:a1a:c370:221f with SMTP id 11-20020a170906004b00b00a1ac370221fmr182041ejg.28.1701512102194; Sat, 02 Dec 2023 02:15:02 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1701512102; cv=none; d=google.com; s=arc-20160816; b=VPEAzpEFtKM6+j6+1ap6J5s8DypQ9EODdDO5ruSWT83YNFlYEqKt5Nn6GZS8ojlmKb EYGSyZgsQX6Qg5rniUNmqo7+JnvNs3rcNrPPcR9UVmgL27+W4dkcxP95GA7aVThOKZG2 xiPjLtM9pu6YNmN7SDFcZrGCWBFIj5VIkocnvUo536b8EQZT8Mcxhd1UAZK9Y35SVrJg 62w/07M1m+zKtOE6I6MNQ8mIuifu2o5CK2Z8k1zo6EbiB2M1QP9FpIzz+hvEtBX8oQod TvZdXlYXLY19iitORY9u7V8euUZ24Pr9Qkx1+j5fBcIOn1YlTDdDhCMhCxG+Ga3e1ARz Wpzg== 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=PkJupL0J54i/DBo9MCBBK3ubXTIr0G8iQEgoVPzoUlERI6+412SsDLiGKkQ0wygAFR OYH5zwqFAQqLGCKHrjj6qiHHQ0w/n06rCDLIZoU0ynkIgOD8oVcTavRON+CqnZHAQ0g2 aDCP+nKID/pBXFOl3QsrPcReQPCknboV+cjiJhZZ5ArxXsrLIy2L+ADrw5jx2oYGS8IY ozres/UuihQu7mylg/JoEF+Ipg0uutXAvAs3AIOSE5blLmnviQbb6N4xhe+twO0LdEeM YrcfJKu1vV7v9vs6T+3M1Ivz3K5dqDKJyAkJQYBRdPvU6sJ0OUfypiEzaTwSSnyQhzsL CdiA== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@intel.com header.s=Intel header.b=BK34uRol; 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 rh8-20020a17090720e800b00a182c81c1c7si2485535ejb.707.2023.12.02.02.15.01; Sat, 02 Dec 2023 02:15:02 -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=BK34uRol; 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 9722F68CF28; Sat, 2 Dec 2023 12:13:54 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mgamail.intel.com (mgamail.intel.com [192.55.52.43]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 4985768CEEE for ; Sat, 2 Dec 2023 12:13:50 +0200 (EET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1701512030; x=1733048030; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=Z/MowF5pFwPy0D0kYdFZ/H4vcxYKHcE+5fr8NSF+Ljc=; b=BK34uRolrVDBn5DBTua+Hmnvvlcce98BVm9HgvIh+6oBCI0erdbcmlVj IFBNIYI7CmymSYTE3WwzcasBV8RSo04ePtkIP4onS3ea4YvzSwXqakyuA M7z8pdHBLvLs0iGBUH4ZUBAMScWnOt9543Wo7dZTZBLjSzOMjT2HAL7KX P0ZPRYzSHuqCTS4qHTAZZujlhdi46MC+SHnoa59tsSjW5Dw/7p+/f0+ns RFT5OObwvOO6ia4C+/udBpUwfNYGbB75VwRMRRpg43NaO9UV6GnbAED4g gVep0bKFbBJ5cpP9uuYzVMpYFWXQBaEVHcjpwfy/RBTwla4BHE8zu9J9t A==; X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="479794843" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="479794843" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2023 02:13:42 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10911"; a="773711678" X-IronPort-AV: E=Sophos;i="6.04,245,1695711600"; d="scan'208";a="773711678" Received: from desktop-qn7n0nf.sh.intel.com (HELO localhost.localdomain) ([10.239.160.66]) by fmsmga007.fm.intel.com with ESMTP; 02 Dec 2023 02:13:41 -0800 From: Tong Wu To: ffmpeg-devel@ffmpeg.org Date: Sat, 2 Dec 2023 18:12:49 +0800 Message-ID: <20231202101250.1410-9-tong1.wu@intel.com> X-Mailer: git-send-email 2.41.0.windows.1 In-Reply-To: <20231202101250.1410-1-tong1.wu@intel.com> References: <20231202101250.1410-1-tong1.wu@intel.com> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH v10 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: VAlh6BVeotKo 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: