From patchwork Tue Jul 5 15:16:51 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: mohamed Elhadidy X-Patchwork-Id: 36673 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:8b27:b0:88:1bbf:7fd2 with SMTP id l39csp3557798pzh; Tue, 5 Jul 2022 08:17:33 -0700 (PDT) X-Google-Smtp-Source: AGRyM1v3B8vFzGzgV89A5tEUfVZgLfIvoA90BYaxodatFQUZM7L8HG7FSYJj8HG3zo/mXpY3g+YY X-Received: by 2002:a05:6402:194f:b0:43a:298f:f39c with SMTP id f15-20020a056402194f00b0043a298ff39cmr18444884edz.106.1657034253414; Tue, 05 Jul 2022 08:17:33 -0700 (PDT) Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id q10-20020a1709066aca00b007156fe1aa78si22763895ejs.403.2022.07.05.08.17.32; Tue, 05 Jul 2022 08:17:33 -0700 (PDT) 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=@hotmail.com header.s=selector1 header.b=F3O84YYj; arc=fail (body hash mismatch); 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; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=hotmail.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 45C1E68B9F2; Tue, 5 Jul 2022 18:17:29 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from EUR05-AM6-obe.outbound.protection.outlook.com (mail-am6eur05olkn2029.outbound.protection.outlook.com [40.92.91.29]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 0325E68B898 for ; Tue, 5 Jul 2022 18:17:22 +0300 (EEST) ARC-Seal: i=1; a=rsa-sha256; s=arcselector9901; d=microsoft.com; cv=none; b=SCa9boOmZv6E304DDR7MLTfXjA++YXiybriqUQZYYYMpNau4wtVhpIf9Mnfn5kxvXNqcIvK+X5bGeEqYTY1d/LwvW/06pQpGrIjuyVVZMZx/sgnvUWzBiWTmi5+jA4YvlvAqVoJZ8RFfGywHDvznyq4Jzm5N7mr+TzpIP92nkUHFKH5n3F+i9DlI2vYH5FAyf8eu/vermBImZlBnJZVB5Uw5Bbl69APoz/IZT3NPkXlsBTsx6NdRgXvZG1NQf5q8casXIVZbSxqmEY6mZATLcilSV8iHBioR2mYiQ3st1m1hDB0c0VFuf6f5m2JD6NCjpagV9UpwGh1QxUiwkEUQAw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=microsoft.com; s=arcselector9901; h=From:Date:Subject:Message-ID:Content-Type:MIME-Version:X-MS-Exchange-AntiSpam-MessageData-ChunkCount:X-MS-Exchange-AntiSpam-MessageData-0:X-MS-Exchange-AntiSpam-MessageData-1; bh=3Xn2U2vQJwM6qnxD+1vDTGW5SLe7k33JhqOuzcwLW9M=; b=RNM6elogqtoALGL5/ETQGPI7HG6JOaldJqgCyvF+Oos44wLkmvuryoZDsh0kJOFshjKlvAoESgUDOt39edv7WJI0pYE8AIEfHoDZECc7SGAXOm0Jo3x1IxWACQz+UinWfEleRPnGl68I/LAnWAn91ydKSr3IETZGlwHCqh1Y4m/Rw5OBcf6vL3e6xFoRS3JH4MqcDzyKh6xMbwsH+2thxPTaQY/cL0566yPsJaBdSo+yyVYEOhN0BOMuxl9eSaI40lRUzoRdOUJzYpgo+olIEtLBy3Ng/B2KxB8+9+IMa6GCBrHsnry/AGxWNtB5/T7HVrXhrra6tFQviX2r5e9CQg== ARC-Authentication-Results: i=1; mx.microsoft.com 1; spf=none; dmarc=none; dkim=none; arc=none DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=hotmail.com; s=selector1; h=From:Date:Subject:Message-ID:Content-Type:MIME-Version:X-MS-Exchange-SenderADCheck; bh=3Xn2U2vQJwM6qnxD+1vDTGW5SLe7k33JhqOuzcwLW9M=; b=F3O84YYj6sWh67omaWrP5nqXKwDzXba0VA5lzfx19EMXP9UfMKm8zHCt//vjaImOdLrVZKyDpKVJiWEZz/VfOmKv+IcETvyP5IJQ+stmYdvIsTJHrLhwfvcpGk81SrQ/Z9/rH+jutiK+aZ+iFkpyq1wxiXbVWCb2NPvTnv6kFHR+BrfiD0k0Is/Rnt9PlTCeBfeXFenGOsbLX1CZtQvGjrLfgEzk31hwRgfgKPbPCsnims/1+zN1B8L406m0nAMEjrxDRRtrBggj8jf8GMkFIRWDsVtnujyODkzBVj/jE+tsstNpd0amGowOjU3MecBLo1DdrvT6YfpC9pLXM7H5qg== Received: from AM9PR04MB8940.eurprd04.prod.outlook.com (2603:10a6:20b:40b::6) by DB9PR04MB9401.eurprd04.prod.outlook.com (2603:10a6:10:36b::20) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.20.5395.21; Tue, 5 Jul 2022 15:17:20 +0000 Received: from AM9PR04MB8940.eurprd04.prod.outlook.com ([fe80::d976:d179:896b:6a2a]) by AM9PR04MB8940.eurprd04.prod.outlook.com ([fe80::d976:d179:896b:6a2a%4]) with mapi id 15.20.5395.021; Tue, 5 Jul 2022 15:17:20 +0000 From: mohamed Elhadidy To: ffmpeg-devel@ffmpeg.org Date: Tue, 5 Jul 2022 17:16:51 +0200 Message-ID: X-Mailer: git-send-email 2.36.1 In-Reply-To: References: X-TMN: [z+E3An5+/RKbtlsSO3grjM5u+mQhOKoM] X-ClientProxiedBy: MR2P264CA0073.FRAP264.PROD.OUTLOOK.COM (2603:10a6:500:32::13) To AM9PR04MB8940.eurprd04.prod.outlook.com (2603:10a6:20b:40b::6) X-Microsoft-Original-Message-ID: <20220705151650.48289-1-mohamedpower_50@hotmail.com> MIME-Version: 1.0 X-MS-Exchange-MessageSentRepresentingType: 1 X-MS-PublicTrafficType: Email X-MS-Office365-Filtering-Correlation-Id: cce64e1b-3474-4d81-a39e-08da5e997456 X-MS-TrafficTypeDiagnostic: DB9PR04MB9401:EE_ X-Microsoft-Antispam: BCL:0; X-Microsoft-Antispam-Message-Info: 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 X-MS-Exchange-AntiSpam-MessageData-ChunkCount: 1 X-MS-Exchange-AntiSpam-MessageData-0: 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 X-OriginatorOrg: sct-15-20-4755-11-msonline-outlook-03a34.templateTenant X-MS-Exchange-CrossTenant-Network-Message-Id: cce64e1b-3474-4d81-a39e-08da5e997456 X-MS-Exchange-CrossTenant-AuthSource: AM9PR04MB8940.eurprd04.prod.outlook.com X-MS-Exchange-CrossTenant-AuthAs: Internal X-MS-Exchange-CrossTenant-OriginalArrivalTime: 05 Jul 2022 15:17:20.4780 (UTC) X-MS-Exchange-CrossTenant-FromEntityHeader: Hosted X-MS-Exchange-CrossTenant-Id: 84df9e7f-e9f6-40af-b435-aaaaaaaaaaaa X-MS-Exchange-CrossTenant-RMS-PersistedConsumerOrg: 00000000-0000-0000-0000-000000000000 X-MS-Exchange-Transport-CrossTenantHeadersStamped: DB9PR04MB9401 Subject: [FFmpeg-devel] [PATCH v2] Chromakey CUDA 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: Mohamed Khaled Mohamed <56936494+MohamedElhadidy0019@users.noreply.github.com> Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: d6nRxxdUZX3X From: Mohamed Khaled Mohamed <56936494+MohamedElhadidy0019@users.noreply.github.com> V2 ChromeKey CUDA Added Chromakey CUDA filter v2 some changes for coding conventions GSoC'22 Added CUDA chromakeyfilter libavfilter/vf_chromakey_cuda.cu:the CUDA kernel for the filter libavfilter/vf_chromakey_cuda.c: the C side that calls the kernel and gets user input libavfilter/allfilters.c: added the filter to it libavfilter/Makefile: added the filter to it cuda/cuda_runtime.h: added two math CUDA functions that are used in the filter --- compat/cuda/cuda_runtime.h | 2 + libavfilter/Makefile | 2 + libavfilter/allfilters.c | 1 + libavfilter/vf_chromakey_cuda.c | 450 +++++++++++++++++++++++++++++++ libavfilter/vf_chromakey_cuda.cu | 164 +++++++++++ 5 files changed, 619 insertions(+) create mode 100644 libavfilter/vf_chromakey_cuda.c create mode 100644 libavfilter/vf_chromakey_cuda.cu diff --git a/compat/cuda/cuda_runtime.h b/compat/cuda/cuda_runtime.h index 30cd085e48..5837c1ad37 100644 --- a/compat/cuda/cuda_runtime.h +++ b/compat/cuda/cuda_runtime.h @@ -181,7 +181,9 @@ static inline __device__ double trunc(double a) { return __builtin_trunc(a); } static inline __device__ float fabsf(float a) { return __builtin_fabsf(a); } static inline __device__ float fabs(float a) { return __builtin_fabsf(a); } static inline __device__ double fabs(double a) { return __builtin_fabs(a); } +static inline __device__ float sqrtf(float a) { return __builtin_sqrtf(a); } +static inline __device__ float __saturatef(float a) { return __nvvm_saturate_f(a); } static inline __device__ float __sinf(float a) { return __nvvm_sin_approx_f(a); } static inline __device__ float __cosf(float a) { return __nvvm_cos_approx_f(a); } static inline __device__ float __expf(float a) { return __nvvm_ex2_approx_f(a * (float)__builtin_log2(__builtin_exp(1))); } diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 22b0a0ca15..f02571c710 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -210,6 +210,8 @@ OBJS-$(CONFIG_CAS_FILTER) += vf_cas.o OBJS-$(CONFIG_CHROMABER_VULKAN_FILTER) += vf_chromaber_vulkan.o vulkan.o vulkan_filter.o OBJS-$(CONFIG_CHROMAHOLD_FILTER) += vf_chromakey.o OBJS-$(CONFIG_CHROMAKEY_FILTER) += vf_chromakey.o +OBJS-$(CONFIG_CHROMAKEY_CUDA_FILTER) += vf_chromakey_cuda.o vf_chromakey_cuda.ptx.o + OBJS-$(CONFIG_CHROMANR_FILTER) += vf_chromanr.o OBJS-$(CONFIG_CHROMASHIFT_FILTER) += vf_chromashift.o OBJS-$(CONFIG_CIESCOPE_FILTER) += vf_ciescope.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index ec70feef11..da1a96b23c 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -195,6 +195,7 @@ extern const AVFilter ff_vf_cas; extern const AVFilter ff_vf_chromaber_vulkan; extern const AVFilter ff_vf_chromahold; extern const AVFilter ff_vf_chromakey; +extern const AVFilter ff_vf_chromakey_cuda; extern const AVFilter ff_vf_chromanr; extern const AVFilter ff_vf_chromashift; extern const AVFilter ff_vf_ciescope; diff --git a/libavfilter/vf_chromakey_cuda.c b/libavfilter/vf_chromakey_cuda.c new file mode 100644 index 0000000000..b3b037d43a --- /dev/null +++ b/libavfilter/vf_chromakey_cuda.c @@ -0,0 +1,450 @@ +/* + * Copyright (c) 2022 Mohamed Khaled + * + * 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/avstring.h" +#include "libavutil/common.h" +#include "libavutil/hwcontext.h" +#include "libavutil/hwcontext_cuda_internal.h" +#include "libavutil/cuda_check.h" +#include "libavutil/internal.h" +#include "libavutil/opt.h" +#include "libavutil/pixdesc.h" + +#include "avfilter.h" +#include "formats.h" +#include "internal.h" +#include "video.h" +#include "cuda/load_helper.h" + +#define FIXNUM(x) lrint((x) * (1 << 10)) +#define RGB_TO_U(rgb) (((-FIXNUM(0.16874) * rgb[0] - FIXNUM(0.33126) * rgb[1] + FIXNUM(0.50000) * rgb[2] + (1 << 9) - 1) >> 10) + 128) +#define RGB_TO_V(rgb) (((FIXNUM(0.50000) * rgb[0] - FIXNUM(0.41869) * rgb[1] - FIXNUM(0.08131) * rgb[2] + (1 << 9) - 1) >> 10) + 128) + +static const enum AVPixelFormat supported_formats[] = { + AV_PIX_FMT_YUV420P, + AV_PIX_FMT_NV12, + AV_PIX_FMT_YUVA420P +}; + +#define DIV_UP(a, b) (((a) + (b)-1) / (b)) +#define BLOCKX 32 +#define BLOCKY 16 +#define CHECK_CU(x) FF_CUDA_CHECK_DL(ctx, s->hwctx->internal->cuda_dl, x) + +typedef struct ChromakeyCUDAContext +{ + const AVClass *class; + AVCUDADeviceContext *hwctx; + enum AVPixelFormat in_fmt, out_fmt; + const AVPixFmtDescriptor *in_desc, *out_desc; + int in_planes, out_planes; + int in_plane_depths[4]; + int in_plane_channels[4]; + uint8_t chromakey_rgba[4]; + uint16_t chromakey_uv[2]; + int is_yuv; + float similarity; + float blend; + AVBufferRef *frames_ctx; + AVFrame *frame; + AVFrame *tmp_frame; + char *w_expr; + char *h_expr; + CUcontext cu_ctx; + CUmodule cu_module; + CUfunction cu_func; + CUfunction cu_func_uv; + CUstream cu_stream; +} ChromakeyCUDAContext; + +static av_cold int cudachromakey_init(AVFilterContext *ctx) +{ + ChromakeyCUDAContext *s = ctx->priv; + s->frame = av_frame_alloc(); + if (!s->frame) + return AVERROR(ENOMEM); + s->tmp_frame = av_frame_alloc(); + if (!s->tmp_frame) + return AVERROR(ENOMEM); + return 0; +} + +static av_cold void cudachromakey_uninit(AVFilterContext *ctx) +{ + ChromakeyCUDAContext *s = ctx->priv; + if (s->hwctx && s->cu_module) + { + CudaFunctions *cu = s->hwctx->internal->cuda_dl; + CUcontext context; + CHECK_CU(cu->cuCtxPushCurrent(s->hwctx->cuda_ctx)); + CHECK_CU(cu->cuModuleUnload(s->cu_module)); + s->cu_module = NULL; + CHECK_CU(cu->cuCtxPopCurrent(&context)); + } + av_frame_free(&s->frame); + av_buffer_unref(&s->frames_ctx); + av_frame_free(&s->tmp_frame); +} + +static av_cold int init_hwframe_ctx(ChromakeyCUDAContext *s, AVBufferRef *device_ctx, int width, int height) +{ + AVBufferRef *out_ref = NULL; + AVHWFramesContext *out_ctx; + int ret; + out_ref = av_hwframe_ctx_alloc(device_ctx); + if (!out_ref) + return AVERROR(ENOMEM); + out_ctx = (AVHWFramesContext *)out_ref->data; + out_ctx->format = AV_PIX_FMT_CUDA; + out_ctx->sw_format = s->out_fmt; + out_ctx->width = width; + out_ctx->height = height; + ret = av_hwframe_ctx_init(out_ref); + if (ret < 0) + goto fail; + av_frame_unref(s->frame); + ret = av_hwframe_get_buffer(out_ref, s->frame, 0); + if (ret < 0) + goto fail; + av_buffer_unref(&s->frames_ctx); + s->frames_ctx = out_ref; + return 0; +fail: + av_buffer_unref(&out_ref); + return ret; +} + +static int format_is_supported(enum AVPixelFormat fmt) +{ + int i; + for (i = 0; i < FF_ARRAY_ELEMS(supported_formats); i++) + if (supported_formats[i] == fmt) + return 1; + return 0; +} + +static av_cold void set_format_info(AVFilterContext *ctx, enum AVPixelFormat in_format, enum AVPixelFormat out_format) +{ + ChromakeyCUDAContext *s = ctx->priv; + int i, p, d; + s->in_fmt = in_format; + s->out_fmt = out_format; + s->in_desc = av_pix_fmt_desc_get(s->in_fmt); + s->out_desc = av_pix_fmt_desc_get(s->out_fmt); + s->in_planes = av_pix_fmt_count_planes(s->in_fmt); + s->out_planes = av_pix_fmt_count_planes(s->out_fmt); + // find maximum step of each component of each plane + // For our subset of formats, this should accurately tell us how many channels CUDA needs + // i.e. 1 for Y plane, 2 for UV plane of NV12, 4 for single plane of RGB0 formats + for (i = 0; i < s->in_desc->nb_components; i++) + { + d = (s->in_desc->comp[i].depth + 7) / 8; + p = s->in_desc->comp[i].plane; + s->in_plane_channels[p] = FFMAX(s->in_plane_channels[p], s->in_desc->comp[i].step / d); + s->in_plane_depths[p] = s->in_desc->comp[i].depth; + } +} + +static av_cold int init_processing_chain(AVFilterContext *ctx, int width, int height) +{ + ChromakeyCUDAContext *s = ctx->priv; + AVHWFramesContext *in_frames_ctx; + int ret; + /* check that we have a hw context */ + if (!ctx->inputs[0]->hw_frames_ctx) + { + av_log(ctx, AV_LOG_ERROR, "No hw context provided on input\n"); + return AVERROR(EINVAL); + } + in_frames_ctx = (AVHWFramesContext *)ctx->inputs[0]->hw_frames_ctx->data; + if (!format_is_supported(in_frames_ctx->sw_format)) + { + av_log(ctx, AV_LOG_ERROR, "Unsupported format: %s\n", av_get_pix_fmt_name(in_frames_ctx->sw_format)); + return AVERROR(ENOSYS); + } + set_format_info(ctx, in_frames_ctx->sw_format, AV_PIX_FMT_YUVA420P); + ret = init_hwframe_ctx(s, in_frames_ctx->device_ref, width, height); + if (ret < 0) + return ret; + ctx->outputs[0]->hw_frames_ctx = av_buffer_ref(s->frames_ctx); + if (!ctx->outputs[0]->hw_frames_ctx) + return AVERROR(ENOMEM); + + return 0; +} + +static av_cold int cudachromakey_load_functions(AVFilterContext *ctx) +{ + ChromakeyCUDAContext *s = ctx->priv; + CUcontext context, cuda_ctx = s->hwctx->cuda_ctx; + CudaFunctions *cu = s->hwctx->internal->cuda_dl; + int ret; + extern const unsigned char ff_vf_chromakey_cuda_ptx_data[]; + extern const unsigned int ff_vf_chromakey_cuda_ptx_len; + ret = CHECK_CU(cu->cuCtxPushCurrent(cuda_ctx)); + if (ret < 0) + return ret; + ret = ff_cuda_load_module(ctx, s->hwctx, &s->cu_module, + ff_vf_chromakey_cuda_ptx_data, ff_vf_chromakey_cuda_ptx_len); + if (ret < 0) + goto fail; + ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func, s->cu_module, "Process_uchar")); + if (ret < 0) + { + av_log(ctx, AV_LOG_FATAL, "Failed loading Process_uchar\n"); + goto fail; + } + ret = CHECK_CU(cu->cuModuleGetFunction(&s->cu_func_uv, s->cu_module, "Process_uchar2")); + if (ret < 0) + { + av_log(ctx, AV_LOG_FATAL, "Failed loading Process_uchar2\n"); + goto fail; + } +fail: + CHECK_CU(cu->cuCtxPopCurrent(&context)); + + return ret; +} + +static av_cold int cudachromakey_config_props(AVFilterLink *outlink) +{ + AVFilterContext *ctx = outlink->src; + AVFilterLink *inlink = outlink->src->inputs[0]; + ChromakeyCUDAContext *s = ctx->priv; + AVHWFramesContext *frames_ctx = (AVHWFramesContext *)inlink->hw_frames_ctx->data; + AVCUDADeviceContext *device_hwctx = frames_ctx->device_ctx->hwctx; + int ret; + s->hwctx = device_hwctx; + s->cu_stream = s->hwctx->stream; + ret = init_processing_chain(ctx, inlink->w, inlink->h); + if (ret < 0) + return ret; + outlink->sample_aspect_ratio = inlink->sample_aspect_ratio; + ret = cudachromakey_load_functions(ctx); + if (ret < 0) + return ret; + return 0; +} + +static int call_cuda_kernel(AVFilterContext *ctx, CUfunction func, + CUtexObject src_tex[3], AVFrame *out_frame, + int width, int height, int pitch, + int width_uv, int height_uv, int pitch_uv, + float u_key, float v_key, float similarity, + float blend) +{ + ChromakeyCUDAContext *s = ctx->priv; + CudaFunctions *cu = s->hwctx->internal->cuda_dl; + int ret; + CUdeviceptr dst_devptr[4] = { + (CUdeviceptr)out_frame->data[0], (CUdeviceptr)out_frame->data[1], + (CUdeviceptr)out_frame->data[2], (CUdeviceptr)out_frame->data[3] + }; + + void *args_uchar[] = { + &src_tex[0], &src_tex[1], &src_tex[2], + &dst_devptr[0], &dst_devptr[1], &dst_devptr[2], &dst_devptr[3], + &width, &height, &pitch, + &width_uv, &height_uv, &pitch_uv, &u_key, &v_key, &similarity, &blend}; + + ret = CHECK_CU(cu->cuLaunchKernel(func, + DIV_UP(width, BLOCKX), DIV_UP(height, BLOCKY), 1, + BLOCKX, BLOCKY, 1, 0, s->cu_stream, args_uchar, NULL)); + if (ret < 0) + return ret; + // This ruins performance, but catches errors that happened in the Kernel immediately + // Remove for performance/production test + ret = CHECK_CU(cu->cuStreamSynchronize(s->cu_stream)); + + return ret; +} + +static int cudachromakey_process_internal(AVFilterContext *ctx, + AVFrame *out, AVFrame *in) +{ + ChromakeyCUDAContext *s = ctx->priv; + CudaFunctions *cu = s->hwctx->internal->cuda_dl; + CUcontext context, cuda_ctx = s->hwctx->cuda_ctx; + float u_key, v_key; + int i, ret; + CUtexObject tex[3] = {0, 0, 0}; + if (s->is_yuv) + { + u_key = s->chromakey_rgba[1]; + v_key = s->chromakey_rgba[2]; + } + else + { + u_key = RGB_TO_U(s->chromakey_rgba); + v_key = RGB_TO_V(s->chromakey_rgba); + } + ret = CHECK_CU(cu->cuCtxPushCurrent(cuda_ctx)); + if (ret < 0) + return ret; + for (i = 0; i < s->in_planes; i++) + { + CUDA_TEXTURE_DESC tex_desc = { + .filterMode = CU_TR_FILTER_MODE_LINEAR, + .flags = 0, // CU_TRSF_READ_AS_INTEGER to get raw ints instead of normalized floats from tex2D + }; + CUDA_RESOURCE_DESC res_desc = { + .resType = CU_RESOURCE_TYPE_PITCH2D, + .res.pitch2D.format = CU_AD_FORMAT_UNSIGNED_INT8, + .res.pitch2D.numChannels = s->in_plane_channels[i], + .res.pitch2D.pitchInBytes = in->linesize[i], + .res.pitch2D.devPtr = (CUdeviceptr)in->data[i], + }; + if (i == 1 || i == 2) + { + res_desc.res.pitch2D.width = AV_CEIL_RSHIFT(in->width, s->in_desc->log2_chroma_w); + res_desc.res.pitch2D.height = AV_CEIL_RSHIFT(in->height, s->in_desc->log2_chroma_h); + } + else + { + res_desc.res.pitch2D.width = in->width; + res_desc.res.pitch2D.height = in->height; + } + ret = CHECK_CU(cu->cuTexObjectCreate(&tex[i], &res_desc, &tex_desc, NULL)); + if (ret < 0) + goto exit; + } + ret = call_cuda_kernel(ctx, (s->in_plane_channels[1] > 1) ? s->cu_func_uv : s->cu_func, + tex, out, + out->width, out->height, out->linesize[0], + AV_CEIL_RSHIFT(out->width, s->out_desc->log2_chroma_w), + AV_CEIL_RSHIFT(out->height, s->out_desc->log2_chroma_h), + out->linesize[1], + u_key, v_key, s->similarity, s->blend); + if (ret < 0) + goto exit; +exit: + for (i = 0; i < s->in_planes; i++) + if (tex[i]) + CHECK_CU(cu->cuTexObjectDestroy(tex[i])); + CHECK_CU(cu->cuCtxPopCurrent(&context)); + return ret; +} + +static int cudachromakey_process(AVFilterContext *ctx, AVFrame *out, AVFrame *in) +{ + ChromakeyCUDAContext *s = ctx->priv; + AVFrame *src = in; + int ret; + ret = cudachromakey_process_internal(ctx, s->frame, src); + if (ret < 0) + return ret; + src = s->frame; ret = av_hwframe_get_buffer(src->hw_frames_ctx, s->tmp_frame, 0); + if (ret < 0) + return ret; + av_frame_move_ref(out, s->frame); + av_frame_move_ref(s->frame, s->tmp_frame); + ret = av_frame_copy_props(out, in); + if (ret < 0) + return ret; + return 0; +} + +static int cudachromakey_filter_frame(AVFilterLink *link, AVFrame *in) +{ + AVFilterContext *ctx = link->dst; + ChromakeyCUDAContext *s = ctx->priv; + AVFilterLink *outlink = ctx->outputs[0]; + CudaFunctions *cu = s->hwctx->internal->cuda_dl; + AVFrame *out = NULL; + CUcontext context; + int ret = 0; + out = av_frame_alloc(); + if (!out) + { + ret = AVERROR(ENOMEM); + goto fail; + } + ret = CHECK_CU(cu->cuCtxPushCurrent(s->hwctx->cuda_ctx)); + if (ret < 0) + goto fail; + ret = cudachromakey_process(ctx, out, in); + CHECK_CU(cu->cuCtxPopCurrent(&context)); + if (ret < 0) + goto fail; + av_frame_free(&in); + return ff_filter_frame(outlink, out); +fail: + av_frame_free(&in); + av_frame_free(&out); + return ret; +} + +static AVFrame *cudachromakey_get_video_buffer(AVFilterLink *inlink, int w, int h) +{ + return ff_default_get_video_buffer(inlink, w, h); +} + +#define OFFSET(x) offsetof(ChromakeyCUDAContext, x) +#define FLAGS (AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM) +static const AVOption options[] = { + {"color", "set the chromakey key color", OFFSET(chromakey_rgba), AV_OPT_TYPE_COLOR, {.str = "black"}, 0, 0, FLAGS}, + {"similarity", "set the chromakey similarity value", OFFSET(similarity), AV_OPT_TYPE_FLOAT, {.dbl = 0.01}, 0.01, 1.0, FLAGS}, + {"blend", "set the chromakey key blend value", OFFSET(blend), AV_OPT_TYPE_FLOAT, {.dbl = 0.0}, 0.0, 1.0, FLAGS}, + {"yuv", "color parameter is in yuv instead of rgb", OFFSET(is_yuv), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, FLAGS}, + {"w", "Output video width", OFFSET(w_expr), AV_OPT_TYPE_STRING, {.str = "iw"}, .flags = FLAGS}, + {"h", "Output video height", OFFSET(h_expr), AV_OPT_TYPE_STRING, {.str = "ih"}, .flags = FLAGS}, + {NULL}, +}; + +static const AVClass cudachromakey_class = { + .class_name = "cudachromakey", + .item_name = av_default_item_name, + .option = options, + .version = LIBAVUTIL_VERSION_INT, +}; + +static const AVFilterPad cudachromakey_inputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .filter_frame = cudachromakey_filter_frame, + .get_buffer.video = cudachromakey_get_video_buffer, + }, +}; + +static const AVFilterPad cudachromakey_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .config_props = cudachromakey_config_props, + }, +}; + +const AVFilter ff_vf_chromakey_cuda = { + .name = "chromakey_cuda", + .description = NULL_IF_CONFIG_SMALL("GPU accelerated chromakey filter"), + .init = cudachromakey_init, + .uninit = cudachromakey_uninit, + .priv_size = sizeof(ChromakeyCUDAContext), + .priv_class = &cudachromakey_class, + FILTER_INPUTS(cudachromakey_inputs), + FILTER_OUTPUTS(cudachromakey_outputs), + FILTER_SINGLE_PIXFMT(AV_PIX_FMT_CUDA), + .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE, +}; diff --git a/libavfilter/vf_chromakey_cuda.cu b/libavfilter/vf_chromakey_cuda.cu new file mode 100644 index 0000000000..f6e7a80c42 --- /dev/null +++ b/libavfilter/vf_chromakey_cuda.cu @@ -0,0 +1,164 @@ +/* + * Copyright (c) 2022 Mohamed Khaled + * + * 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 "cuda/vector_helpers.cuh" + +extern "C" +{ +/** + * @brief function contains the main logic of chroma keying, and changes the alpahc channel with the suitable value + * + * @param src_tex texture U or texture UV , decided based on the passed is_uchar2 flag + * @param src_tex_V texture V , used only if is_uchar2 flag is false + * @param dst_A alpha channel destination + * @param width_uv width of uv channels + * @param height_uv height of uv channels + * @param width width of alpha channel + * @param height height of alpha channel + * @param pitch pitch of alpha channel + * @param x current x coordinate of pixel + * @param y current y coordinate of pixel + * @param chromakey_uv uv values for chroma keying + * @param similarity similarity of keying + * @param blend blend of keying + * @return void + */ +__device__ static inline void change_alpha_channel( + cudaTextureObject_t src_tex, cudaTextureObject_t src_tex_V, uchar *dst_A, + int width_uv, int height_uv, + int width, int height, int pitch, + int x, int y, + float2 chromakey_uv, float similarity,float blend) +{ + uchar window_size = 3; + int start_r = x - window_size / 2; + int start_c = y - window_size / 2; + int resize_ratio = width / width_uv; + uchar counter = 0; + float diff = 0.0f; + float du, dv; + + // loop over the eight neighbourhood of the current pixel(x,y) + for (uchar i = 0; i < window_size; i++) + { + for (uchar j = 0; j < window_size; j++) + { + int r = start_r + i; + int c = start_c + j; + bool check_flag = (r >= 0 && r < width_uv && c >= 0 && c < height_uv); + if (!check_flag) + continue; + float u_value, v_value; + if (!src_tex_V){ + float2 temp_uv = tex2D(src_tex, r, c); + u_value = temp_uv.x; + v_value = temp_uv.y; + }else{ + u_value = tex2D(src_tex, r, c); + v_value = tex2D(src_tex_V, r, c); + } + du = (u_value * 255.0f) - chromakey_uv.x; + dv = (v_value * 255.0f) - chromakey_uv.y; + diff += sqrtf((du * du + dv * dv) / (255.0f * 255.0f * 2.f)); + counter++; + } + } + + if (counter > 0) + diff = diff / counter; + else + diff /= 9.0f; + + uchar alpha_value; + if (blend>0.0001f) + alpha_value=__saturatef((diff - similarity) / blend)*255; + else + alpha_value=(diff < similarity ? 0 : 1)*255; + + //write the value in the alpha channel with regarding the ratio of (alpha_size : uv_size) + for (uchar k = 0; k < resize_ratio; k++) + { + for (uchar l = 0; l < resize_ratio; l++) + { + int x_resize = x * resize_ratio + k; + int y_resize = y * resize_ratio + l; + int a_channel_resize = y_resize * pitch + x_resize; + if (y_resize >= height || x_resize >= width) + continue; + dst_A[a_channel_resize] = alpha_value; + } + } +} + +__global__ void Process_uchar( + cudaTextureObject_t src_tex_Y, cudaTextureObject_t src_tex_U, cudaTextureObject_t src_tex_V, + uchar *dst_Y, uchar *dst_U, uchar *dst_V, uchar *dst_A, + int width, int height, int pitch, + int width_uv, int height_uv, int pitch_uv, + float u_key, float v_key, float similarity,float blend) +{ + int x = blockIdx.x * blockDim.x + threadIdx.x; + int y = blockIdx.y * blockDim.y + threadIdx.y; + + if (y >= height || x >= width) + return; + dst_Y[y * pitch + x] = tex2D(src_tex_Y, x, y)*255; + if (y >= height_uv || x >= width_uv) + return; + + int uv_index = y * pitch_uv + x; + dst_U[uv_index]=tex2D(src_tex_U,x,y)*255; + dst_V[uv_index]=tex2D(src_tex_V,x,y)*255; + + change_alpha_channel(src_tex_U, src_tex_V, dst_A, + width_uv, height_uv, + width, height, pitch, + x, y, + make_float2(u_key,v_key), similarity, blend); +} + +__global__ void Process_uchar2( + cudaTextureObject_t src_tex_Y, cudaTextureObject_t src_tex_UV, cudaTextureObject_t unused1, + uchar *dst_Y, uchar *dst_U, uchar *dst_V, uchar *dst_A, + int width, int height, int pitch, + int width_uv, int height_uv,int pitch_uv, + float u_key,float v_key, float similarity, float blend) +{ + int x = blockIdx.x * blockDim.x + threadIdx.x; // x coordinate of current pixel + int y = blockIdx.y * blockDim.y + threadIdx.y; // y coordinate of current pixel + + if (y >= height || x >= width) + return; + dst_Y[y * pitch + x] = tex2D(src_tex_Y, x, y)*255; + + if (y >= height_uv || x >= width_uv) + return; + int uv_index = y * pitch_uv + x; + float2 uv_temp =tex2D(src_tex_UV,x,y); + dst_U[uv_index]=uv_temp.x*255; + dst_V[uv_index]=uv_temp.y*255; + + change_alpha_channel(src_tex_UV, (cudaTextureObject_t)nullptr, + dst_A, width_uv, height_uv, + width,height, pitch, + x, y, + make_float2(u_key,v_key), similarity,blend); +} +}