From patchwork Wed Sep 8 06:12:24 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Wu Jianhua X-Patchwork-Id: 30067 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6602:2a4a:0:0:0:0 with SMTP id k10csp109295iov; Tue, 7 Sep 2021 23:12:56 -0700 (PDT) X-Google-Smtp-Source: ABdhPJyaq/d9Tnq5eNzay/V25sT1G4eM7W3R5SDKzw1Kx/oW6aKMHvn+rk94JmHM23PeJ0W6Y47a X-Received: by 2002:aa7:cb0f:: with SMTP id s15mr2161126edt.190.1631081576259; Tue, 07 Sep 2021 23:12:56 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1631081576; cv=none; d=google.com; s=arc-20160816; b=mbSYMCqbdVvMJcYgA4Nk2iJokC4tlAzzD5MdpS5dGYI4F0CzNNDipNNtpBeJfoZFjt N3cALZQruKR8KI7UG/lebtHQhOc3ZxiXBG4Q2So5WDHAQaZhFymKidb5BZRaIuJohO20 2kKZ43cAUCSRUdXschCI+rXFSBASUt4ySSl6Xm70kSawU+WuBYraI7QzKF9kTEEDN+cR hPd4D3pZqVq5NeqtAyHht25uPQ/sQRMXY5Xf4OBaYjWqiDJjhHm9o0le/pPj7hbhgBO2 nwugVUGkCHTkllcN9V+xuMfnql+ylUlAwFQRGmV2kJqi1sHcCI6B6Uctwmvizk4YM4f2 JGgQ== 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:message-id:date:to:from:delivered-to; bh=XK1DOLmlE7cWlfTe2+BNR8HHh2uXct6Pr5Ft8J+IFEc=; b=iBweePyo1atO26V6tUSvM9TdV1qhmjw+wPTBisK4IKasgoDLldakd+cmK50p3D+8D6 38HbcJJOTToxO5rR3czQ6fMgf2OMhy2h45i2dEVZSpxxWZyGKTi7D5gg86e6aHHVOpqZ 9p4v580xG5zOEZqGwg/btC6zO/3/MwSRKnTqKO0pYZSZazgVEwu5R905xEH6TJ8pI1o2 jrqX0lIif6E6qhfSQnkOJjyYIKSu5OJX9qXCcZAHu71cspLKIa4IJVmssxLeu0wPm8lL /JDw3IRpnk8RTioTNkG8FbcfxNmuvTxYFkkJrWXrIAf9KnASSlpy5SdUjb/6vBjr4rIE asJw== ARC-Authentication-Results: i=1; mx.google.com; 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=intel.com Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id ee35si1199393edb.219.2021.09.07.23.12.55; Tue, 07 Sep 2021 23:12:56 -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; 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=intel.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 801C768A77C; Wed, 8 Sep 2021 09:12:51 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mga04.intel.com (mga04.intel.com [192.55.52.120]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 0286A68A5E7 for ; Wed, 8 Sep 2021 09:12:43 +0300 (EEST) X-IronPort-AV: E=McAfee;i="6200,9189,10100"; a="218530948" X-IronPort-AV: E=Sophos;i="5.85,277,1624345200"; d="scan'208";a="218530948" Received: from fmsmga003.fm.intel.com ([10.253.24.29]) by fmsmga104.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 07 Sep 2021 23:12:40 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.85,277,1624345200"; d="scan'208";a="538402926" Received: from otc-skl-e5-server.sh.intel.com ([10.239.43.106]) by FMSMGA003.fm.intel.com with ESMTP; 07 Sep 2021 23:12:39 -0700 From: Wu Jianhua To: ffmpeg-devel@ffmpeg.org Date: Wed, 8 Sep 2021 14:12:24 +0800 Message-Id: <20210908061224.18500-1-jianhua.wu@intel.com> X-Mailer: git-send-email 2.17.1 Subject: [FFmpeg-devel] [PATCH] libavfilter: add a gblur_vulkan filter 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: Wu Jianhua MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: dWGzRC4viyLl This commit adds a powerful and customizable gblur Vulkan filter, which provides a maximum 127x127 kernel size of Gaussian Filter. The size could be adjusted by requirements on quality or performance. The following command is on how to apply gblur_vulkan filter: ffmpeg -init_hw_device vulkan=vul:0 -filter_hw_device vul -i input.264 -vf hwupload=extra_hw_frames=16,gblur_vulkan,hwdownload,format=yuv420p output.264 Signed-off-by: Wu Jianhua --- Changelog | 2 +- configure | 1 + libavfilter/Makefile | 1 + libavfilter/allfilters.c | 1 + libavfilter/vf_gblur_vulkan.c | 509 ++++++++++++++++++++++++++++++++++ 5 files changed, 513 insertions(+), 1 deletion(-) create mode 100644 libavfilter/vf_gblur_vulkan.c diff --git a/Changelog b/Changelog index 5b58401183..bc36fa09e4 100644 --- a/Changelog +++ b/Changelog @@ -148,7 +148,7 @@ version 4.3: - LEGO Racers ALP (.tun & .pcm) demuxer - AMQP 0-9-1 protocol (RabbitMQ) - Vulkan support -- avgblur_vulkan, overlay_vulkan, scale_vulkan and chromaber_vulkan filters +- avgblur_vulkan, overlay_vulkan, scale_vulkan, chromaber_vulkan and gblur_vulkan filters - ADPCM IMA MTF decoder - FWSE demuxer - DERF DPCM decoder diff --git a/configure b/configure index af410a9d11..4b9a0d8e07 100755 --- a/configure +++ b/configure @@ -3601,6 +3601,7 @@ freezedetect_filter_select="scene_sad" frei0r_filter_deps="frei0r libdl" frei0r_src_filter_deps="frei0r libdl" fspp_filter_deps="gpl" +gblur_vulkan_filter_deps="vulkan_lib libglslang" histeq_filter_deps="gpl" hqdn3d_filter_deps="gpl" interlace_filter_deps="gpl" diff --git a/libavfilter/Makefile b/libavfilter/Makefile index af957a5ac0..5f74e33599 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -286,6 +286,7 @@ OBJS-$(CONFIG_FREEZEFRAMES_FILTER) += vf_freezeframes.o OBJS-$(CONFIG_FREI0R_FILTER) += vf_frei0r.o OBJS-$(CONFIG_FSPP_FILTER) += vf_fspp.o qp_table.o OBJS-$(CONFIG_GBLUR_FILTER) += vf_gblur.o +OBJS-$(CONFIG_GBLUR_VULKAN_FILTER) += vf_gblur_vulkan.o vulkan.o OBJS-$(CONFIG_GEQ_FILTER) += vf_geq.o OBJS-$(CONFIG_GRADFUN_FILTER) += vf_gradfun.o OBJS-$(CONFIG_GRAPHMONITOR_FILTER) += f_graphmonitor.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index 0c6b2347c8..b5576de3af 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -271,6 +271,7 @@ extern const AVFilter ff_vf_freezeframes; extern const AVFilter ff_vf_frei0r; extern const AVFilter ff_vf_fspp; extern const AVFilter ff_vf_gblur; +extern const AVFilter ff_vf_gblur_vulkan; extern const AVFilter ff_vf_geq; extern const AVFilter ff_vf_gradfun; extern const AVFilter ff_vf_graphmonitor; diff --git a/libavfilter/vf_gblur_vulkan.c b/libavfilter/vf_gblur_vulkan.c new file mode 100644 index 0000000000..c1693f5d66 --- /dev/null +++ b/libavfilter/vf_gblur_vulkan.c @@ -0,0 +1,509 @@ +/* + * copyright (c) 2021 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 "libavutil/random_seed.h" +#include "libavutil/opt.h" +#include "vulkan.h" +#include "internal.h" + +#define CGS 32 +#define GBLUR_MAX_KERNEL_SIZE 127 + +typedef struct GBlurVulkanContext { + VulkanFilterContext vkctx; + FFVkExecContext *exec; + VulkanPipeline *pl_hor; + VulkanPipeline *pl_ver; + FFVkBuffer params_buf_hor; + FFVkBuffer params_buf_ver; + + VkDescriptorImageInfo input_images[3]; + VkDescriptorImageInfo tmp_images[3]; + VkDescriptorImageInfo output_images[3]; + VkDescriptorBufferInfo params_desc_hor; + VkDescriptorBufferInfo params_desc_ver; + + int initialized; + int planes; + int kernel_size; + int size; + float sigma; + float sigmaV; + float *kernel; +} GBlurVulkanContext; + +static const char gblur_horizontal[] = { + C(0, void gblur(const ivec2 pos, const int index) ) + C(0, { ) + C(1, vec4 sum = texture(input_image[index], pos) * kernel[0]; ) + C(0, ) + C(1, for(int i = 1; i < kernel.length(); i++) { ) + C(2, sum += texture(input_image[index], pos + vec2(i, 0.0)) * kernel[i]; ) + C(2, sum += texture(input_image[index], pos - vec2(i, 0.0)) * kernel[i]; ) + C(1, } ) + C(0, ) + C(1, imageStore(output_image[index], pos, sum); ) + C(0, } ) +}; + +static const char gblur_vertical[] = { + C(0, void gblur(const ivec2 pos, const int index) ) + C(0, { ) + C(1, vec4 sum = texture(input_image[index], pos) * kernel[0]; ) + C(0, ) + C(1, for(int i = 1; i < kernel.length(); i++) { ) + C(2, sum += texture(input_image[index], pos + vec2(0.0, i)) * kernel[i]; ) + C(2, sum += texture(input_image[index], pos - vec2(0.0, i)) * kernel[i]; ) + C(1, } ) + C(0, ) + C(1, imageStore(output_image[index], pos, sum); ) + C(0, } ) +}; + +static inline float gaussian(float sigma, float x) +{ + return 1.0 / (sqrt(2.0 * M_PI) * sigma) * + exp(-(x * x) / (2.0 * sigma * sigma)); +} + +static inline float gaussian_simpson_integration(float sigma, float a, float b) +{ + return (b - a) * (1.0 / 6.0) * ((gaussian(sigma, a) + + 4.0 * gaussian(sigma, (a + b) * 0.5) + gaussian(sigma, b))); +} + +static void init_gaussian_kernel(float *kernel, float sigma, float kernel_size) +{ + int x; + float sum; + + sum = 0; + for (x = 0; x < kernel_size; x++) { + kernel[x] = gaussian_simpson_integration(sigma, x - 0.5f, x + 0.5f); + if (!x) + sum += kernel[x]; + else + sum += kernel[x] * 2.0; + } + /* Normalized */ + sum = 1.0 / sum; + for (x = 0; x < kernel_size; x++) { + kernel[x] *= sum; + } +} + +static av_cold void init_gaussian_params(GBlurVulkanContext *s) +{ + if (!(s->size & 1)) { + av_log(s, AV_LOG_WARNING, "kernel size should be even\n"); + s->size++; + } + if (s->sigmaV <= 0) + s->sigmaV = s->sigma; + + s->kernel_size = (s->size >> 1) + 1; + s->kernel = av_mallocz(sizeof(float) * s->kernel_size); +} + +static av_cold int init_filter(AVFilterContext *ctx, AVFrame *in) +{ + int err; + uint8_t *data_mapped; + char *kernel_def; + SPIRVShader *shd; + GBlurVulkanContext *s = ctx->priv; + + int planes = av_pix_fmt_count_planes(s->vkctx.output_format); + VkSampler *sampler = ff_vk_init_sampler(ctx, 1, VK_FILTER_LINEAR); + + VulkanDescriptorSetBinding image_descs[] = { + { + .name = "input_image", + .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .dimensions = 2, + .elems = planes, + .stages = VK_SHADER_STAGE_COMPUTE_BIT, + .samplers = DUP_SAMPLER_ARRAY4(*sampler), + }, + { + .name = "output_image", + .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .mem_layout = ff_vk_shader_rep_fmt(s->vkctx.output_format), + .mem_quali = "writeonly", + .dimensions = 2, + .elems = planes, + .stages = VK_SHADER_STAGE_COMPUTE_BIT, + }, + }; + + VulkanDescriptorSetBinding buf_desc = { + .name = "data", + .type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, + .mem_quali = "readonly", + .mem_layout = "std430", + .stages = VK_SHADER_STAGE_COMPUTE_BIT, + .updater = NULL, + .buf_content = NULL, + }; + + init_gaussian_params(s); + + kernel_def = av_asprintf("float kernel[%i];", s->kernel_size); + buf_desc.buf_content = kernel_def; + + if (!sampler) + return AVERROR_EXTERNAL; + + s->vkctx.queue_family_idx = s->vkctx.hwctx->queue_family_comp_index; + s->vkctx.queue_count = GET_QUEUE_COUNT(s->vkctx.hwctx, 0, 1, 0); + s->vkctx.cur_queue_idx = av_get_random_seed() % s->vkctx.queue_count; + + { /* Create shader for the horizontal pass */ + image_descs[0].updater = s->input_images; + image_descs[1].updater = s->tmp_images; + buf_desc.updater = &s->params_desc_hor; + + s->pl_hor = ff_vk_create_pipeline(ctx); + + if (!s->pl_hor) + return AVERROR(ENOMEM); + + shd = ff_vk_init_shader(ctx, s->pl_hor, "gblur_compute_hor", image_descs[0].stages); + ff_vk_set_compute_shader_sizes(ctx, shd, (int [3]){ CGS, CGS, 1 }); + + RET(ff_vk_add_descriptor_set(ctx, s->pl_hor, shd, image_descs, FF_ARRAY_ELEMS(image_descs), 0)); + RET(ff_vk_add_descriptor_set(ctx, s->pl_hor, shd, &buf_desc, 1, 0)); + + GLSLD( gblur_horizontal ); + GLSLC(0, void main() ); + GLSLC(0, { ); + GLSLC(1, ivec2 size; ); + GLSLC(1, const ivec2 pos = ivec2(gl_GlobalInvocationID.xy); ); + for (int i = 0; i < planes; i++) { + GLSLC(0, ); + GLSLF(1, size = imageSize(output_image[%i]); ,i); + GLSLC(1, if (IS_WITHIN(pos, size)) { ); + if (s->planes & (1 << i)) { + GLSLF(2, gblur(pos, %i); ,i); + } else { + GLSLF(2, vec4 res = texture(input_image[%i], pos); ,i); + GLSLF(2, imageStore(output_image[%i], pos, res); ,i); + } + GLSLC(1, } ); + } + GLSLC(0, } ); + + RET(ff_vk_compile_shader(ctx, shd, "main")); + + RET(ff_vk_init_pipeline_layout(ctx, s->pl_hor)); + RET(ff_vk_init_compute_pipeline(ctx, s->pl_hor)); + + RET(ff_vk_create_buf(ctx, &s->params_buf_hor, sizeof(float) * s->kernel_size, + VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)); + RET(ff_vk_map_buffers(ctx, &s->params_buf_hor, &data_mapped, 1, 0)); + + init_gaussian_kernel(s->kernel, s->sigma, s->kernel_size); + memcpy(data_mapped, s->kernel, sizeof(float) * s->kernel_size); + + RET(ff_vk_unmap_buffers(ctx, &s->params_buf_hor, 1, 1)); + + s->params_desc_hor.buffer = s->params_buf_hor.buf; + s->params_desc_hor.range = VK_WHOLE_SIZE; + + ff_vk_update_descriptor_set(ctx, s->pl_hor, 1); + } + + { /* Create shader for the vertical pass */ + image_descs[0].updater = s->tmp_images; + image_descs[1].updater = s->output_images; + buf_desc.updater = &s->params_desc_ver; + + s->pl_ver = ff_vk_create_pipeline(ctx); + if (!s->pl_ver) + return AVERROR(ENOMEM); + + shd = ff_vk_init_shader(ctx, s->pl_ver, "gblur_compute_ver", image_descs[0].stages); + ff_vk_set_compute_shader_sizes(ctx, shd, (int [3]){ CGS, CGS, 1 }); + + RET(ff_vk_add_descriptor_set(ctx, s->pl_ver, shd, image_descs, FF_ARRAY_ELEMS(image_descs), 0)); + RET(ff_vk_add_descriptor_set(ctx, s->pl_ver, shd, &buf_desc, 1, 0)); + + GLSLD( gblur_vertical ); + GLSLC(0, void main() ); + GLSLC(0, { ); + GLSLC(1, ivec2 size; ); + GLSLC(1, const ivec2 pos = ivec2(gl_GlobalInvocationID.xy); ); + for (int i = 0; i < planes; i++) { + GLSLC(0, ); + GLSLF(1, size = imageSize(output_image[%i]); ,i); + GLSLC(1, if (IS_WITHIN(pos, size)) { ); + if (s->planes & (1 << i)) { + GLSLF(2, gblur(pos, %i); ,i); + } else { + GLSLF(2, vec4 res = texture(input_image[%i], pos); ,i); + GLSLF(2, imageStore(output_image[%i], pos, res); ,i); + } + GLSLC(1, } ); + } + GLSLC(0, } ); + + RET(ff_vk_compile_shader(ctx, shd, "main")); + + RET(ff_vk_init_pipeline_layout(ctx, s->pl_ver)); + RET(ff_vk_init_compute_pipeline(ctx, s->pl_ver)); + + RET(ff_vk_create_buf(ctx, &s->params_buf_ver, sizeof(float) * s->kernel_size, + VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)); + RET(ff_vk_map_buffers(ctx, &s->params_buf_ver, &data_mapped, 1, 0)); + + init_gaussian_kernel(s->kernel, s->sigmaV, s->kernel_size); + memcpy(data_mapped, s->kernel, sizeof(float) * s->kernel_size); + + RET(ff_vk_unmap_buffers(ctx, &s->params_buf_ver, 1, 1)); + + s->params_desc_ver.buffer = s->params_buf_ver.buf; + s->params_desc_ver.range = VK_WHOLE_SIZE; + + ff_vk_update_descriptor_set(ctx, s->pl_ver, 1); + } + + RET(ff_vk_create_exec_ctx(ctx, &s->exec)); + + s->initialized = 1; + err = 0; + +fail: + av_free(kernel_def); + return err; +} + +static av_cold void gblur_vulkan_uninit(AVFilterContext *avctx) +{ + GBlurVulkanContext *s = avctx->priv; + + av_free(s->kernel); + ff_vk_free_buf(avctx, &s->params_buf_hor); + ff_vk_free_buf(avctx, &s->params_buf_ver); + ff_vk_filter_uninit(avctx); + + s->initialized = 0; +} + +static int process_frames(AVFilterContext *avctx, AVFrame *outframe, AVFrame *tmpframe, AVFrame *inframe) +{ + int err; + const VkFormat *input_formats = NULL; + const VkFormat *output_formats = NULL; + VkCommandBuffer cmd_buf; + GBlurVulkanContext *s = avctx->priv; + + AVVkFrame *in = (AVVkFrame *)inframe->data[0]; + AVVkFrame *tmp = (AVVkFrame *)tmpframe->data[0]; + AVVkFrame *out = (AVVkFrame *)outframe->data[0]; + + int planes = av_pix_fmt_count_planes(s->vkctx.output_format); + + ff_vk_start_exec_recording(avctx, s->exec); + cmd_buf = ff_vk_get_exec_buf(avctx, s->exec); + + input_formats = av_vkfmt_from_pixfmt(s->vkctx.input_format); + output_formats = av_vkfmt_from_pixfmt(s->vkctx.output_format); + for (int i = 0; i < planes; i++) { + RET(ff_vk_create_imageview(avctx, s->exec, &s->input_images[i].imageView, + in->img[i], + input_formats[i], + ff_comp_identity_map)); + + RET(ff_vk_create_imageview(avctx, s->exec, &s->tmp_images[i].imageView, + tmp->img[i], + output_formats[i], + ff_comp_identity_map)); + + RET(ff_vk_create_imageview(avctx, s->exec, &s->output_images[i].imageView, + out->img[i], + output_formats[i], + ff_comp_identity_map)); + + s->input_images[i].imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; + s->tmp_images[i].imageLayout = VK_IMAGE_LAYOUT_GENERAL; + s->output_images[i].imageLayout = VK_IMAGE_LAYOUT_GENERAL; + } + + ff_vk_update_descriptor_set(avctx, s->pl_hor, 0); + ff_vk_update_descriptor_set(avctx, s->pl_ver, 0); + + for (int i = 0; i < planes; i++) { + VkImageMemoryBarrier barriers[] = { + { + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .srcAccessMask = 0, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT, + .oldLayout = in->layout[i], + .newLayout = s->input_images[i].imageLayout, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = in->img[i], + .subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .subresourceRange.levelCount = 1, + .subresourceRange.layerCount = 1, + }, + { + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .srcAccessMask = 0, + .dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_SHADER_READ_BIT, + .oldLayout = tmp->layout[i], + .newLayout = s->tmp_images[i].imageLayout, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = tmp->img[i], + .subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .subresourceRange.levelCount = 1, + .subresourceRange.layerCount = 1, + }, + { + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .srcAccessMask = 0, + .dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT, + .oldLayout = out->layout[i], + .newLayout = s->output_images[i].imageLayout, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = out->img[i], + .subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT, + .subresourceRange.levelCount = 1, + .subresourceRange.layerCount = 1, + }, + }; + + vkCmdPipelineBarrier(cmd_buf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, + VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, + 0, NULL, 0, NULL, FF_ARRAY_ELEMS(barriers), barriers); + + in->layout[i] = barriers[0].newLayout; + in->access[i] = barriers[0].dstAccessMask; + + tmp->layout[i] = barriers[1].newLayout; + tmp->access[i] = barriers[1].dstAccessMask; + + out->layout[i] = barriers[2].newLayout; + out->access[i] = barriers[2].dstAccessMask; + } + + ff_vk_bind_pipeline_exec(avctx, s->exec, s->pl_hor); + + vkCmdDispatch(cmd_buf, FFALIGN(s->vkctx.output_width, CGS)/CGS, + FFALIGN(s->vkctx.output_height, CGS)/CGS, 1); + + ff_vk_bind_pipeline_exec(avctx, s->exec, s->pl_ver); + + vkCmdDispatch(cmd_buf, FFALIGN(s->vkctx.output_width, CGS)/CGS, + FFALIGN(s->vkctx.output_height, CGS)/CGS, 1); + + ff_vk_add_exec_dep(avctx, s->exec, inframe, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 1); + ff_vk_add_exec_dep(avctx, s->exec, outframe, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0); + + return ff_vk_submit_exec_queue(avctx, s->exec); + +fail: + ff_vk_discard_exec_deps(avctx, s->exec); + return err; +} + +static int gblur_vulkan_filter_frame(AVFilterLink *link, AVFrame *in) +{ + int err; + AVFrame *out = NULL; + AVFrame *tmp = NULL; + AVFilterContext *ctx = link->dst; + GBlurVulkanContext *s = ctx->priv; + AVFilterLink *outlink = ctx->outputs[0]; + + out = ff_get_video_buffer(outlink, outlink->w, outlink->h); + if (!out) { + err = AVERROR(ENOMEM); + goto fail; + } + + tmp = ff_get_video_buffer(outlink, outlink->w, outlink->h); + if (!tmp) { + err = AVERROR(ENOMEM); + goto fail; + } + + if (!s->initialized) + RET(init_filter(ctx, in)); + + RET(process_frames(ctx, out, tmp, in)); + + RET(av_frame_copy_props(out, in)); + + av_frame_free(&in); + av_frame_free(&tmp); + + return ff_filter_frame(outlink, out); + +fail: + av_frame_free(&in); + av_frame_free(&tmp); + av_frame_free(&out); + + return err; +} + +#define OFFSET(x) offsetof(GBlurVulkanContext, x) +#define FLAGS (AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM) +static const AVOption gblur_vulkan_options[] = { + { "sigma", "set sigma", OFFSET(sigma), AV_OPT_TYPE_FLOAT, {.dbl = 0.5}, 0.01, 1024.0, FLAGS }, + { "sigmaV", "set vertical sigma", OFFSET(sigmaV), AV_OPT_TYPE_FLOAT, {.dbl = 0}, 0.0, 1024.0, FLAGS }, + { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64 = 0xF}, 0, 0xF, FLAGS }, + { "size", "set kernel Size", OFFSET(size), AV_OPT_TYPE_INT, {.i64 = 19}, 1, GBLUR_MAX_KERNEL_SIZE, FLAGS }, + { NULL }, +}; + +AVFILTER_DEFINE_CLASS(gblur_vulkan); + +static const AVFilterPad gblur_vulkan_inputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .filter_frame = &gblur_vulkan_filter_frame, + .config_props = &ff_vk_filter_config_input, + } +}; + +static const AVFilterPad gblur_vulkan_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .config_props = &ff_vk_filter_config_output, + } +}; + +const AVFilter ff_vf_gblur_vulkan = { + .name = "gblur_vulkan", + .description = NULL_IF_CONFIG_SMALL("Gaussian Blur in Vulkan"), + .priv_size = sizeof(GBlurVulkanContext), + .init = &ff_vk_filter_init, + .uninit = &gblur_vulkan_uninit, + .query_formats = &ff_vk_filter_query_formats, + .priv_class = &gblur_vulkan_class, + FILTER_INPUTS(gblur_vulkan_inputs), + FILTER_OUTPUTS(gblur_vulkan_outputs), + .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE, +};