From patchwork Tue Nov 16 07:35:24 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Wu Jianhua X-Patchwork-Id: 31438 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a6b:d206:0:0:0:0:0 with SMTP id q6csp6995887iob; Mon, 15 Nov 2021 23:35:50 -0800 (PST) X-Google-Smtp-Source: ABdhPJyjSJri66qAzAMN/gOUtFTl+P8MGoaWJ06MolB3eDpVbKqC9/Sa/sAqdR787khVKmOknpw0 X-Received: by 2002:a17:906:3a9b:: with SMTP id y27mr6991808ejd.563.1637048150605; Mon, 15 Nov 2021 23:35:50 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1637048150; cv=none; d=google.com; s=arc-20160816; b=RAWa/0cOtdpXGzql7nLIGFYHq9K4W0JuzzDRQ0Tfm23QWWJNvt2G4jPM3+eDgGPCOB xVwLwKjlvtWA6KQPrloCunQY2piAi0hxobt2qs/el96MzZSoBEq4YRt5tZLUppz7ZTAa TZbh/cMFjo13GcbHbNY2CwzUhex1OI8OKXT6MVMg1gRiOBbcgy5OtvnIB7ZUwDNSzXCg iKeldTO6/YzpelnxXFSaD6Qh8ovUx1ZIsXxxddJ3mmRxWC/UNZwNz709YjWZNs7HPrbD hU23kW/UqI08H64Z/ZDIdFfCtq03dz0lfk6ei2CXfi6LqxU8ujDp8kCUhwZbkvdu9yud 1WQA== 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=ASWTmcbkb2HyBVRfENAbtUp4hL75zOFdKBaeKLghgho=; b=AIgc5TpvCzgHfpajnFwUWhUM/0iP14nUNv2J9fLNCod30/sWpfPNMamczrCxqXDhSR kkAuUzJzuCD6sZk8ke1uDaSmttKCv8tH1GBYJNV8jNMNAxJzk9xLvYyzRvdbxrFmuxnF vq67JGC54RfwOIr8srnmDlS2QWN6Zerd60FiJY62UTZ10e2zjVGW8J+VGzsvs5NPTxHf BmEW+RrrwsULycnjJVwb4Lk8E7kjVb4GyPPVQJCEGbdsrs3yiaQaRD5pzcaKV+Y6ychh arh4zzM6ZHgqYDK9CKwP2lpGq90mCuvEjwdD+64FE3uDEXqmVwAEBo+StfANsth1ZIN+ 3kdA== 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 o15si39563716edi.471.2021.11.15.23.35.49; Mon, 15 Nov 2021 23:35:50 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; 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 7AE72680157; Tue, 16 Nov 2021 09:35:46 +0200 (EET) 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 62363680157 for ; Tue, 16 Nov 2021 09:35:39 +0200 (EET) X-IronPort-AV: E=McAfee;i="6200,9189,10169"; a="232361810" X-IronPort-AV: E=Sophos;i="5.87,238,1631602800"; d="scan'208";a="232361810" Received: from orsmga005.jf.intel.com ([10.7.209.41]) by fmsmga104.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 15 Nov 2021 23:35:37 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.87,238,1631602800"; d="scan'208";a="671856717" Received: from otc-skl-e5-server.sh.intel.com ([10.239.43.106]) by orsmga005.jf.intel.com with ESMTP; 15 Nov 2021 23:35:35 -0800 From: Wu Jianhua To: ffmpeg-devel@ffmpeg.org Date: Tue, 16 Nov 2021 15:35:24 +0800 Message-Id: <20211116073524.59212-1-jianhua.wu@intel.com> X-Mailer: git-send-email 2.17.1 Subject: [FFmpeg-devel] [PATCH v4] 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: C+fNq/Wyi10B 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 -i input.264 -vf hwupload=extra_hw_frames=16,gblur_vulkan,hwdownload,format=yuv420p output.264 Signed-off-by: Wu Jianhua --- configure | 1 + libavfilter/Makefile | 1 + libavfilter/allfilters.c | 1 + libavfilter/vf_gblur_vulkan.c | 524 ++++++++++++++++++++++++++++++++++ 4 files changed, 527 insertions(+) create mode 100644 libavfilter/vf_gblur_vulkan.c diff --git a/configure b/configure index 891824757b..1b47f6512d 100755 --- a/configure +++ b/configure @@ -3613,6 +3613,7 @@ frei0r_deps_any="libdl LoadLibrary" frei0r_filter_deps="frei0r" frei0r_src_filter_deps="frei0r" fspp_filter_deps="gpl" +gblur_vulkan_filter_deps="vulkan libglslang" histeq_filter_deps="gpl" hqdn3d_filter_deps="gpl" interlace_filter_deps="gpl" diff --git a/libavfilter/Makefile b/libavfilter/Makefile index e2059766b0..08edc92d8c 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -289,6 +289,7 @@ 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 be94249024..f250020159 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -274,6 +274,7 @@ 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..92f4461dd7 --- /dev/null +++ b/libavfilter/vf_gblur_vulkan.c @@ -0,0 +1,524 @@ +/* + * 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 { + FFVulkanContext vkctx; + FFVkQueueFamilyCtx qf; + FFVkExecContext *exec; + FFVulkanPipeline *pl_hor; + FFVulkanPipeline *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 size; + int planes; + int kernel_size; + float sigma; + float sigmaV; + AVFrame *tmpframe; +} 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 odd\n"); + s->size++; + } + if (s->sigmaV <= 0) + s->sigmaV = s->sigma; + + s->kernel_size = (s->size >> 1) + 1; + s->tmpframe = NULL; +} + +static av_cold int init_filter(AVFilterContext *ctx, AVFrame *in) +{ + int err; + char *kernel_def; + uint8_t *kernel_mapped; + FFSPIRVShader *shd; + GBlurVulkanContext *s = ctx->priv; + const int planes = av_pix_fmt_count_planes(s->vkctx.output_format); + + FFVulkanDescriptorSetBinding image_descs[] = { + { + .name = "input_image", + .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, + .dimensions = 2, + .elems = planes, + .stages = VK_SHADER_STAGE_COMPUTE_BIT, + }, + { + .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, + }, + }; + + FFVulkanDescriptorSetBinding 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, + }; + + image_descs[0].sampler = ff_vk_init_sampler(ctx, 1, VK_FILTER_LINEAR); + if (!image_descs[0].sampler) + return AVERROR_EXTERNAL; + + init_gaussian_params(s); + + kernel_def = av_asprintf("float kernel[%i];", s->kernel_size); + if (!kernel_def) + return AVERROR(ENOMEM); + + buf_desc.buf_content = kernel_def; + + ff_vk_qf_init(ctx, &s->qf, VK_QUEUE_COMPUTE_BIT, 0); + + { /* 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, &s->qf); + if (!s->pl_hor) { + err = AVERROR(ENOMEM); + goto fail; + } + + shd = ff_vk_init_shader(ctx, s->pl_hor, "gblur_compute_hor", image_descs[0].stages); + if (!shd) { + err = AVERROR(ENOMEM); + goto fail; + } + + 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, &kernel_mapped, 1, 0)); + + init_gaussian_kernel((float *)kernel_mapped, s->sigma, 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, &s->qf); + if (!s->pl_ver) { + err = AVERROR(ENOMEM); + goto fail; + } + + shd = ff_vk_init_shader(ctx, s->pl_ver, "gblur_compute_ver", image_descs[0].stages); + if (!shd) { + err = AVERROR(ENOMEM); + goto fail; + } + + 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, &kernel_mapped, 1, 0)); + + init_gaussian_kernel((float *)kernel_mapped, s->sigmaV, 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->qf)); + + s->initialized = 1; + + return 0; + +fail: + av_free(kernel_def); + return err; +} + +static av_cold void gblur_vulkan_uninit(AVFilterContext *avctx) +{ + GBlurVulkanContext *s = avctx->priv; + + av_frame_free(&s->tmpframe); + + ff_vk_filter_uninit(avctx); + ff_vk_free_buf(avctx, &s->params_buf_hor); + ff_vk_free_buf(avctx, &s->params_buf_ver); + + s->initialized = 0; +} + +static int process_frames(AVFilterContext *avctx, AVFrame *outframe, AVFrame *inframe) +{ + int err; + VkCommandBuffer cmd_buf; + + const VkFormat *input_formats = NULL; + const VkFormat *output_formats = NULL; + GBlurVulkanContext *s = avctx->priv; + FFVulkanFunctions *vk = &s->vkctx.vkfn; + AVVkFrame *in = (AVVkFrame *)inframe->data[0]; + AVVkFrame *tmp = (AVVkFrame *)s->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, + }, + }; + + vk->CmdPipelineBarrier(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); + + vk->CmdDispatch(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); + + vk->CmdDispatch(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); + ff_vk_add_exec_dep(avctx, s->exec, outframe, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT); + + err = ff_vk_submit_exec_queue(avctx, s->exec); + if (err) + return err; + + ff_vk_qf_rotate(&s->qf); + + return 0; +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; + 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; + } + + if (!s->initialized) { + RET(init_filter(ctx, in)); + s->tmpframe = ff_get_video_buffer(outlink, outlink->w, outlink->h); + if (!s->tmpframe) { + err = AVERROR(ENOMEM); + goto fail; + } + } + + RET(process_frames(ctx, out, in)); + + RET(av_frame_copy_props(out, in)); + + av_frame_free(&in); + + return ff_filter_frame(outlink, out); + +fail: + av_frame_free(&in); + av_frame_free(&out); + av_frame_free(&s->tmpframe); + + 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, + FILTER_INPUTS(gblur_vulkan_inputs), + FILTER_OUTPUTS(gblur_vulkan_outputs), + FILTER_SINGLE_PIXFMT(AV_PIX_FMT_VULKAN), + .priv_class = &gblur_vulkan_class, + .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE, +};