From patchwork Tue Mar 13 23:32:18 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Danil Iashchenko X-Patchwork-Id: 7963 Delivered-To: ffmpegpatchwork@gmail.com Received: by 10.2.1.70 with SMTP id c67csp1361945jad; Tue, 13 Mar 2018 16:39:18 -0700 (PDT) X-Google-Smtp-Source: AG47ELtbWpqF0Mf+zGugDDjIaabGCsPkZTbnC0YL+Q1EcWMjnicQyRMRPiFpGtl93Q3NYHTAXGoC X-Received: by 10.28.147.73 with SMTP id v70mr2227613wmd.128.1520984358004; Tue, 13 Mar 2018 16:39:18 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1520984357; cv=none; d=google.com; s=arc-20160816; b=aB6sUvqO6CJ7S/YVz6rxXflqnKjuCCs9fKIl9Ihh0o15oZhAf/eRtEVcz3yAXSQ0d2 orQflktFoFOyzI/Fa+FTOYxu+0yMnnyqkVbQTGPYlt0OFhumEb2bRvkbvlNxjoTPe7FA z7qRXeVM9GRB0auHsf+F9CF+WWundmZpsPyzWm6UJ2O4KQ0kJz+3YvYbAIiZ6DBjN3D5 35/akJtULyjngU2VRSeJc1wF1pwxjXoiv6+ulBVymfpNhhvImefqXtwod81iWlTmamO3 4p0najCJXo4hfXAj+R58GU3SsRnBbfIR7AVnfJgKATzAsCVCsKta/7XoL2JnNZMP3J+n 5Rjw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:mime-version:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:references:in-reply-to:message-id:date :to:from:dkim-signature:delivered-to:arc-authentication-results; bh=kjXsQ0ucZMVxnZN14pyyXCsTbNaQjmYAI6em0j1FZmA=; b=A5Qr06Ddcc4fsE57Wk/w2vdLwt2qI81XO/WHelCaVjD3hKKVnKXNcV1/omBobg25H9 OtCCeEBqORN6UHCkWsfZinTsCictfpJ5HQtR7cEDw1kBQZtyaQVru1FlC/g2CcDKBW0Z sWK1uIk7fPsX9sKIFtIyloCDpGVturmG8vxpYI+hImwPwN4dKh8AW0MribeH9LNqomt6 OymFNYWmE14bt2+xgs1dz7Kq2FlG6gOti9bvOOz6fAT+/V4oRyHJyiV6y6dbCJWHvWUc yqo1skz2e1JrSbYGSksG5qZpRXNGKnnE+ZNbbGzhqv/JjBaUwwntGwKwbinCPXqVh6AC PjOA== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com header.s=20161025 header.b=D7NfM+AB; 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=QUARANTINE dis=NONE) header.from=gmail.com Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id 31si864456wri.115.2018.03.13.16.39.17; Tue, 13 Mar 2018 16:39:17 -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=@gmail.com header.s=20161025 header.b=D7NfM+AB; 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=QUARANTINE dis=NONE) header.from=gmail.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 76D46689DC2; Wed, 14 Mar 2018 01:39:04 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-lf0-f50.google.com (mail-lf0-f50.google.com [209.85.215.50]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 6C7B6689C29 for ; Wed, 14 Mar 2018 01:38:58 +0200 (EET) Received: by mail-lf0-f50.google.com with SMTP id t132-v6so2025726lfe.2 for ; Tue, 13 Mar 2018 16:39:09 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=mfFmXPHHW52/VE5Vayyhh2I/98HtKsyG3KmzXzwMYFE=; b=D7NfM+ABAtn2MyBmysfT/DL9GBn4adbPJl/AoHC+GTFNyPHwHfxj1Z+63WV7eoonZD xa7OO5Jb+BjDZNF5gZCsJlvdppG/NZ/593dnAod0iZ0stNufBIDO3Z1jAnrVawd1bkPs LWFmF37tGw6we18YMlGsBFZoy/X62NIwKnRHYbIkMf9oitcjb/8Q8Ff5NZn/tW21sovh wvK6fDNp/RFx6GYw4gzBmHJS0UYB9Hfxl1O6bJUgKJoyan5O4JyvbARQ5k39O07iB1k3 oXYpVuONgivM6aY8F+FroD0jLBUxLJXEhCy1wtOa3ETxlKudIJ+fJKjNultWhKDuNbmJ byPg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=mfFmXPHHW52/VE5Vayyhh2I/98HtKsyG3KmzXzwMYFE=; b=TWPqWXXUyndrwIsvdyE7SyyGzLg8fR6Re3Ugbb1RwKl2LdDs8LhxaMFvH4yEix7ebl Mk28OesVkc3MAf6dWjaf1Ui69PwguqB4vyDctexNOaQfKqaLG9SQCRW3xZOA5X9PZ1ps 7hr97COfXW5gyEC1hpfn2Ob9T0O/Z1Qal5dOfhf4F+KmMlC/8XEvPkZWEQ1c0ROA7xZ4 ybcrKwbO4zsL18nhi3Hgs8ZTrmobL7D1T0WHpHoANPEUf+cJ+y0QuIMcFUPRuc7khBLl MXT3aOrkOP1ycFPpCRXA1WV/XYU9vNWdrW6jyW/AShWZWGxH+WafQSOedNoL00k0QHVo Kt7w== X-Gm-Message-State: AElRT7HH43Euy+I/uLe0tfoW1ZUyhqVNsCWZUYvoU5zXq5qPxZnUwmBr FM01RR8IaZHuQLO5qVsAHm67UJE= X-Received: by 10.46.33.156 with SMTP id h28mr1677216lji.126.1520983955131; Tue, 13 Mar 2018 16:32:35 -0700 (PDT) Received: from dan-acer.lan (campus.ifmo.ru. [194.85.161.2]) by smtp.gmail.com with ESMTPSA id n19sm283519ljc.71.2018.03.13.16.32.34 (version=TLS1_2 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Tue, 13 Mar 2018 16:32:34 -0700 (PDT) From: Danil Iashchenko To: ffmpeg-devel@ffmpeg.org Date: Wed, 14 Mar 2018 02:32:18 +0300 Message-Id: <1520983938-26228-1-git-send-email-danyaschenko@gmail.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: References: Subject: [FFmpeg-devel] [PATCH] libavfilter: Add OpenCL convolution filter. X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 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: Danil Iashchenko MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Behaves like the existing convolution filter, except working on OpenCL hardware frames. Takes exactly the same options: 4 convolution matrices, 4 rdiv values, 4 bias values. If not specified, default parameters are applied. Matrices can be different sizes. filter applies: matrix0, rdiv0, bias0 to image plane0. matrix1, rdiv1, bias1 to image plane1. matrix2, rdiv2, bias2 to image plane2. matrix3, rdiv3, bias3 to image plane3. About Kernel parameters: dst - destination image src - source image coef_matrices_sizes - stores sizes of matrix{0..3} consecutively one after the other coef_matrices - stores matrices{0..3} consecutively one after the other rdivs - stores rdiv{0..3} parameters consecutively one after the other biases - stores bias{0..3} parameters consecutively one after the other About sscanf. I had (!err_code) condition, because I would never get empty line as option(if not specified, I always have default matrix), but changed to (err_code != 1) due to read-ability. Also, before sscanf I split matrix with spaces, so I process each single value of matrix seperately from others and check if they are ok. about rdiv_buffer, bias_buffer, size_buffer objects: they should be buffer objects, because they store sequence of values, not a single value --- configure | 1 + libavfilter/Makefile | 1 + libavfilter/allfilters.c | 1 + libavfilter/opencl/convolution.cl | 43 ++++ libavfilter/opencl_source.h | 1 + libavfilter/vf_convolution_opencl.c | 449 ++++++++++++++++++++++++++++++++++++ 6 files changed, 496 insertions(+) create mode 100644 libavfilter/opencl/convolution.cl create mode 100644 libavfilter/vf_convolution_opencl.c diff --git a/configure b/configure index 6916b45..bf5c312 100755 --- a/configure +++ b/configure @@ -3210,6 +3210,7 @@ blackframe_filter_deps="gpl" boxblur_filter_deps="gpl" bs2b_filter_deps="libbs2b" colormatrix_filter_deps="gpl" +convolution_opencl_filter_deps="opencl" convolve_filter_deps="avcodec" convolve_filter_select="fft" coreimage_filter_deps="coreimage appkit" diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 6a60836..d005934 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -156,6 +156,7 @@ OBJS-$(CONFIG_COLORLEVELS_FILTER) += vf_colorlevels.o OBJS-$(CONFIG_COLORMATRIX_FILTER) += vf_colormatrix.o OBJS-$(CONFIG_COLORSPACE_FILTER) += vf_colorspace.o colorspacedsp.o OBJS-$(CONFIG_CONVOLUTION_FILTER) += vf_convolution.o +OBJS-$(CONFIG_CONVOLUTION_OPENCL_FILTER) += vf_convolution_opencl.o opencl.o opencl/convolution.o OBJS-$(CONFIG_CONVOLVE_FILTER) += vf_convolve.o framesync.o OBJS-$(CONFIG_COPY_FILTER) += vf_copy.o OBJS-$(CONFIG_COREIMAGE_FILTER) += vf_coreimage.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index 9adb109..f2dc55e 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -166,6 +166,7 @@ static void register_all(void) REGISTER_FILTER(COLORMATRIX, colormatrix, vf); REGISTER_FILTER(COLORSPACE, colorspace, vf); REGISTER_FILTER(CONVOLUTION, convolution, vf); + REGISTER_FILTER(CONVOLUTION_OPENCL, convolution_opencl, vf); REGISTER_FILTER(CONVOLVE, convolve, vf); REGISTER_FILTER(COPY, copy, vf); REGISTER_FILTER(COREIMAGE, coreimage, vf); diff --git a/libavfilter/opencl/convolution.cl b/libavfilter/opencl/convolution.cl new file mode 100644 index 0000000..b142aa8 --- /dev/null +++ b/libavfilter/opencl/convolution.cl @@ -0,0 +1,43 @@ +/* + * 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 + */ + +__kernel void convolution_global(__write_only image2d_t dst, + __read_only image2d_t src, + __constant int *coef_matrices_sizes, + __constant float *coef_matrices, + __constant float *rdivs, + __constant float *biases) +{ + const sampler_t sampler = (CLK_NORMALIZED_COORDS_FALSE | CLK_FILTER_NEAREST); + + int2 loc = (int2)(get_global_id(0), get_global_id(1)); + + float4 convPix = (float4)(0.0f, 0.0f, 0.0f, 0.0f); + for (int i = 0; i < 4; i++) { + int half_matrix_size = coef_matrices_sizes[i] / 2; + for (int conv_i = -half_matrix_size; conv_i <= half_matrix_size; conv_i++) { + for (int conv_j = -half_matrix_size; conv_j <= half_matrix_size; conv_j++) { + float4 px = read_imagef(src, sampler, loc + (int2)(conv_j, conv_i)); + convPix[i] += px[i] * coef_matrices[(coef_matrices_sizes[i]*coef_matrices_sizes[i]*i) + + (conv_i+half_matrix_size) * coef_matrices_sizes[i] + (conv_j+half_matrix_size)]; + } + } + convPix[i] = convPix[i] * rdivs[i] + biases[i]; + } + write_imagef(dst, loc, convPix); +} diff --git a/libavfilter/opencl_source.h b/libavfilter/opencl_source.h index 23cdfc6..158fb3e 100644 --- a/libavfilter/opencl_source.h +++ b/libavfilter/opencl_source.h @@ -19,6 +19,7 @@ #ifndef AVFILTER_OPENCL_SOURCE_H #define AVFILTER_OPENCL_SOURCE_H +extern const char *ff_opencl_source_convolution; extern const char *ff_opencl_source_overlay; extern const char *ff_opencl_source_unsharp; diff --git a/libavfilter/vf_convolution_opencl.c b/libavfilter/vf_convolution_opencl.c new file mode 100644 index 0000000..e78c8ef --- /dev/null +++ b/libavfilter/vf_convolution_opencl.c @@ -0,0 +1,449 @@ +/* + * 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 "libavutil/common.h" +#include "libavutil/imgutils.h" +#include "libavutil/mem.h" +#include "libavutil/opt.h" +#include "libavutil/pixdesc.h" +#include "libavutil/avstring.h" + + +#include "avfilter.h" +#include "internal.h" +#include "opencl.h" +#include "opencl_source.h" +#include "video.h" + +typedef struct ConvolutionOpenCLContext { + OpenCLFilterContext ocf; + + int initialised; + cl_kernel kernel; + cl_command_queue command_queue; + + char *matrix_str[4]; + float rdiv[4]; + float bias[4]; + cl_int size[4]; + + cl_int total_matrix_len; + cl_int matrix_length[4]; + cl_mem matrix; + cl_mem rdiv_buffer; + cl_mem bias_buffer; + cl_mem size_buffer; + +} ConvolutionOpenCLContext; + + +static int convolution_opencl_init(AVFilterContext *avctx) +{ + ConvolutionOpenCLContext *ctx = avctx->priv; + cl_int cle; + int err; + + err = ff_opencl_filter_load_program(avctx, &ff_opencl_source_convolution, 1); + if (err < 0) + goto fail; + + ctx->command_queue = clCreateCommandQueue(ctx->ocf.hwctx->context, + ctx->ocf.hwctx->device_id, + 0, &cle); + if (!ctx->command_queue) { + av_log(avctx, AV_LOG_ERROR, "Failed to create OpenCL " + "command queue: %d.\n", cle); + err = AVERROR(EIO); + goto fail; + } + + ctx->kernel = clCreateKernel(ctx->ocf.program, "convolution_global", &cle); + if (!ctx->kernel) { + av_log(avctx, AV_LOG_ERROR, "Failed to create kernel: %d.\n", cle); + err = AVERROR(EIO); + goto fail; + } + + ctx->initialised = 1; + return 0; + +fail: + if (ctx->command_queue) + clReleaseCommandQueue(ctx->command_queue); + if (ctx->kernel) + clReleaseKernel(ctx->kernel); + return err; +} + + +static int convolution_opencl_make_filter_params(AVFilterContext *avctx) +{ + ConvolutionOpenCLContext *ctx = avctx->priv; + float *matrix = NULL; + float *matrix_rdiv = NULL; + float *matrix_bias = NULL; + int *matrix_size = NULL; + size_t matrix_bytes; + cl_mem buffer, buffer_rdiv, buffer_bias, buffer_size; + cl_int cle; + int err, sscanf_err; + int i, j, cnt; + char *p, *arg, *saveptr = NULL; + float input_matrix[4][49]; + + ctx->total_matrix_len = 0; + for (i = 0; i < 4; i++) { + p = ctx->matrix_str[i]; + while (ctx->matrix_length[i] < 49) { + if (!(arg = av_strtok(p, " ", &saveptr))) + break; + p = NULL; + sscanf_err = sscanf(arg, "%f", &input_matrix[i][ctx->matrix_length[i]]); + if (sscanf_err != 1) { + av_log(ctx, AV_LOG_ERROR, "Matrix is sequence of 9, 25 or 49 signed numbers\n"); + return AVERROR(EINVAL); + } + ctx->matrix_length[i]++; + } + if (ctx->matrix_length[i] == 9) { + ctx->size[i] = 3; + } else if (ctx->matrix_length[i] == 25) { + ctx->size[i] = 5; + } else if (ctx->matrix_length[i] == 49) { + ctx->size[i] = 7; + } else { + av_log(ctx, AV_LOG_ERROR, "Invalid matrix size:%d\n", ctx->matrix_length[i]); + return AVERROR(EINVAL); + } + ctx->total_matrix_len += ctx->matrix_length[i]; + + } + matrix_bytes = sizeof(float)*ctx->total_matrix_len; + matrix = av_malloc(matrix_bytes); + if (!matrix) { + err = AVERROR(ENOMEM); + goto fail; + } + cnt = 0; + for (i = 0; i < 4; i++) { + for (j = 0; j < ctx->matrix_length[i]; j++) { + matrix[cnt++] = input_matrix[i][j]; + } + } + + buffer = clCreateBuffer(ctx->ocf.hwctx->context, + CL_MEM_READ_ONLY | + CL_MEM_COPY_HOST_PTR | + CL_MEM_HOST_NO_ACCESS, + matrix_bytes, matrix, &cle); + if (!buffer) { + av_log(avctx, AV_LOG_ERROR, "Failed to create matrix buffer: " + "%d.\n", cle); + err = AVERROR(EIO); + goto fail; + } + ctx->matrix = buffer; + + + matrix_bytes = sizeof(float)*4; + matrix_rdiv = av_malloc(matrix_bytes); + if (!matrix_rdiv) { + err = AVERROR(ENOMEM); + goto fail; + } + + for (j = 0; j < 4; j++) { + matrix_rdiv[j] = ctx->rdiv[j]; + } + buffer_rdiv = clCreateBuffer(ctx->ocf.hwctx->context, + CL_MEM_READ_ONLY | + CL_MEM_COPY_HOST_PTR | + CL_MEM_HOST_NO_ACCESS, + matrix_bytes, matrix_rdiv, &cle); + if (!buffer_rdiv) { + av_log(avctx, AV_LOG_ERROR, "Failed to create rdiv buffer: " + "%d.\n", cle); + err = AVERROR(EIO); + goto fail; + } + ctx->rdiv_buffer = buffer_rdiv; + + + matrix_bytes = sizeof(float)*4; + matrix_bias = av_malloc(matrix_bytes); + if (!matrix_bias) { + err = AVERROR(ENOMEM); + goto fail; + } + + for (j = 0; j < 4; j++) { + matrix_bias[j] = ctx->bias[j]; + } + buffer_bias = clCreateBuffer(ctx->ocf.hwctx->context, + CL_MEM_READ_ONLY | + CL_MEM_COPY_HOST_PTR | + CL_MEM_HOST_NO_ACCESS, + matrix_bytes, matrix_bias, &cle); + if (!buffer_bias) { + av_log(avctx, AV_LOG_ERROR, "Failed to create bias buffer: " + "%d.\n", cle); + err = AVERROR(EIO); + goto fail; + } + ctx->bias_buffer = buffer_bias; + + matrix_bytes = sizeof(int)*4; + matrix_size = av_malloc(matrix_bytes); + if (!matrix_size) { + err = AVERROR(ENOMEM); + goto fail; + } + + for (j = 0; j < 4; j++) { + matrix_size[j] = ctx->size[j]; + } + buffer_size = clCreateBuffer(ctx->ocf.hwctx->context, + CL_MEM_READ_ONLY | + CL_MEM_COPY_HOST_PTR | + CL_MEM_HOST_NO_ACCESS, + matrix_bytes, matrix_size, &cle); + if (!buffer_size) { + av_log(avctx, AV_LOG_ERROR, "Failed to create bias buffer: " + "%d.\n", cle); + err = AVERROR(EIO); + goto fail; + } + ctx->size_buffer = buffer_size; + + err = 0; +fail: + av_freep(&matrix_size); + av_freep(&matrix_bias); + av_freep(&matrix_rdiv); + av_freep(&matrix); + return err; +} + +static int convolution_opencl_filter_frame(AVFilterLink *inlink, AVFrame *input) +{ + AVFilterContext *avctx = inlink->dst; + AVFilterLink *outlink = avctx->outputs[0]; + ConvolutionOpenCLContext *ctx = avctx->priv; + AVFrame *output = NULL; + cl_int cle; + size_t global_work[2]; + cl_mem src, dst; + int err, p; + + av_log(ctx, AV_LOG_DEBUG, "Filter input: %s, %ux%u (%"PRId64").\n", + av_get_pix_fmt_name(input->format), + input->width, input->height, input->pts); + + if (!input->hw_frames_ctx) + return AVERROR(EINVAL); + + if (!ctx->initialised) { + err = convolution_opencl_init(avctx); + if (err < 0) + goto fail; + + err = convolution_opencl_make_filter_params(avctx); + if (err < 0) + goto fail; + } + + output = ff_get_video_buffer(outlink, outlink->w, outlink->h); + if (!output) { + err = AVERROR(ENOMEM); + goto fail; + } + + for (p = 0; p < FF_ARRAY_ELEMS(output->data); p++) { + src = (cl_mem) input->data[p]; + dst = (cl_mem)output->data[p]; + + + if (!dst) + break; + + cle = clSetKernelArg(ctx->kernel, 0, sizeof(cl_mem), &dst); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to set kernel " + "destination image argument: %d.\n", cle); + goto fail; + } + cle = clSetKernelArg(ctx->kernel, 1, sizeof(cl_mem), &src); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to set kernel " + "source image argument: %d.\n", cle); + goto fail; + } + cle = clSetKernelArg(ctx->kernel, 2, sizeof(cl_mem), &ctx->size_buffer); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to set kernel " + "matrix size argument: %d.\n", cle); + goto fail; + } + cle = clSetKernelArg(ctx->kernel, 3, sizeof(cl_mem), &ctx->matrix); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to set kernel " + "matrix argument: %d.\n", cle); + goto fail; + } + cle = clSetKernelArg(ctx->kernel, 4, sizeof(cl_mem), &ctx->rdiv_buffer); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to set kernel " + "div argument: %d.\n", cle); + goto fail; + } + cle = clSetKernelArg(ctx->kernel, 5, sizeof(cl_mem), &ctx->bias_buffer); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to set kernel " + "bias argument: %d.\n", cle); + goto fail; + } + + + + global_work[0] = output->width; + global_work[1] = output->height; + + + av_log(avctx, AV_LOG_DEBUG, "Run kernel on plane %d " + "(%"SIZE_SPECIFIER"x%"SIZE_SPECIFIER").\n", + p, global_work[0], global_work[1]); + + cle = clEnqueueNDRangeKernel(ctx->command_queue, ctx->kernel, 2, NULL, + global_work, NULL, + 0, NULL, NULL); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to enqueue kernel: %d.\n", + cle); + err = AVERROR(EIO); + goto fail; + } + } + + cle = clFinish(ctx->command_queue); + if (cle != CL_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to finish command queue: %d.\n", + cle); + err = AVERROR(EIO); + goto fail; + } + + err = av_frame_copy_props(output, input); + if (err < 0) + goto fail; + + av_frame_free(&input); + + av_log(ctx, AV_LOG_DEBUG, "Filter output: %s, %ux%u (%"PRId64").\n", + av_get_pix_fmt_name(output->format), + output->width, output->height, output->pts); + + return ff_filter_frame(outlink, output); + +fail: + clFinish(ctx->command_queue); + av_frame_free(&input); + av_frame_free(&output); + return err; +} + +static av_cold void convolution_opencl_uninit(AVFilterContext *avctx) +{ + ConvolutionOpenCLContext *ctx = avctx->priv; + cl_int cle; + + if (ctx->matrix) { + clReleaseMemObject(ctx->matrix); + clReleaseMemObject(ctx->rdiv_buffer); + clReleaseMemObject(ctx->bias_buffer); + clReleaseMemObject(ctx->size_buffer); + } + + if (ctx->kernel) { + cle = clReleaseKernel(ctx->kernel); + if (cle != CL_SUCCESS) + av_log(avctx, AV_LOG_ERROR, "Failed to release " + "kernel: %d.\n", cle); + } + + if (ctx->command_queue) { + cle = clReleaseCommandQueue(ctx->command_queue); + if (cle != CL_SUCCESS) + av_log(avctx, AV_LOG_ERROR, "Failed to release " + "command queue: %d.\n", cle); + } + + ff_opencl_filter_uninit(avctx); +} + +#define OFFSET(x) offsetof(ConvolutionOpenCLContext, x) +#define FLAGS (AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM) +static const AVOption convolution_opencl_options[] = { + { "0m", "set matrix for 1st plane", OFFSET(matrix_str[0]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS }, + { "1m", "set matrix for 2nd plane", OFFSET(matrix_str[1]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS }, + { "2m", "set matrix for 3rd plane", OFFSET(matrix_str[2]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS }, + { "3m", "set matrix for 4th plane", OFFSET(matrix_str[3]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS }, + { "0rdiv", "set rdiv for 1st plane", OFFSET(rdiv[0]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS}, + { "1rdiv", "set rdiv for 2nd plane", OFFSET(rdiv[1]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS}, + { "2rdiv", "set rdiv for 3rd plane", OFFSET(rdiv[2]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS}, + { "3rdiv", "set rdiv for 4th plane", OFFSET(rdiv[3]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS}, + { "0bias", "set bias for 1st plane", OFFSET(bias[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS}, + { "1bias", "set bias for 2nd plane", OFFSET(bias[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS}, + { "2bias", "set bias for 3rd plane", OFFSET(bias[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS}, + { "3bias", "set bias for 4th plane", OFFSET(bias[3]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS}, + { NULL } +}; + +AVFILTER_DEFINE_CLASS(convolution_opencl); + + +static const AVFilterPad convolution_opencl_inputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .filter_frame = &convolution_opencl_filter_frame, + .config_props = &ff_opencl_filter_config_input, + }, + { NULL } +}; + +static const AVFilterPad convolution_opencl_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .config_props = &ff_opencl_filter_config_output, + }, + { NULL } +}; + +AVFilter ff_vf_convolution_opencl = { + .name = "convolution_opencl", + .description = NULL_IF_CONFIG_SMALL("Apply convolution mask to input video"), + .priv_size = sizeof(ConvolutionOpenCLContext), + .priv_class = &convolution_opencl_class, + .init = &ff_opencl_filter_init, + .uninit = &convolution_opencl_uninit, + .query_formats = &ff_opencl_filter_query_formats, + .inputs = convolution_opencl_inputs, + .outputs = convolution_opencl_outputs, + .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE, +};