diff --git a/configure b/configure index d7fba1620b..073c3ad9bd 100755 --- a/configure +++ b/configure @@ -4258,7 +4258,7 @@ frc_amf_filter_deps="amf windows_h" scale_qsv_filter_deps="libmfx" scale_qsv_filter_select="qsvvpp" scdet_filter_select="scene_sad" -scdet_vulkan_filter_deps="vulkan spirv_library" +scdet_vulkan_filter_deps="vulkan spirv_compiler" select_filter_select="scene_sad" sharpness_vaapi_filter_deps="vaapi" showcqt_filter_deps="avformat swscale" diff --git a/libavfilter/vf_scdet_vulkan.c b/libavfilter/vf_scdet_vulkan.c index ee2bf248a2..552904a2bb 100644 --- a/libavfilter/vf_scdet_vulkan.c +++ b/libavfilter/vf_scdet_vulkan.c @@ -19,13 +19,15 @@ */ #include "libavutil/avassert.h" -#include "libavutil/vulkan_spirv.h" #include "libavutil/opt.h" #include "libavutil/timestamp.h" #include "vulkan_filter.h" #include "filters.h" +extern const unsigned char ff_scdet_comp_spv_data[]; +extern const unsigned int ff_scdet_comp_spv_len; + typedef struct SceneDetectVulkanContext { FFVulkanContext vkctx; @@ -52,26 +54,14 @@ typedef struct SceneDetectBuf { static av_cold int init_filter(AVFilterContext *ctx) { int err; - uint8_t *spv_data; - size_t spv_len; - void *spv_opaque = NULL; SceneDetectVulkanContext *s = ctx->priv; FFVulkanContext *vkctx = &s->vkctx; - FFVulkanShader *shd; - FFVkSPIRVCompiler *spv; - FFVulkanDescriptorSetBinding *desc; const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(s->vkctx.input_format); const int lumaonly = !(pixdesc->flags & AV_PIX_FMT_FLAG_RGB) && (pixdesc->flags & AV_PIX_FMT_FLAG_PLANAR); s->nb_planes = lumaonly ? 1 : av_pix_fmt_count_planes(s->vkctx.input_format); - spv = ff_vk_spirv_init(); - if (!spv) { - av_log(ctx, AV_LOG_ERROR, "Unable to initialize SPIR-V compiler!\n"); - return AVERROR_EXTERNAL; - } - s->qf = ff_vk_qf_find(vkctx, VK_QUEUE_COMPUTE_BIT, 0); if (!s->qf) { av_log(ctx, AV_LOG_ERROR, "Device has no compute queues\n"); @@ -80,76 +70,41 @@ static av_cold int init_filter(AVFilterContext *ctx) } RET(ff_vk_exec_pool_init(vkctx, s->qf, &s->e, s->qf->num*4, 0, 0, 0, NULL)); - RET(ff_vk_shader_init(vkctx, &s->shd, "scdet", - VK_SHADER_STAGE_COMPUTE_BIT, - (const char *[]) { "GL_KHR_shader_subgroup_arithmetic" }, 1, - 32, 32, 1, - 0)); - shd = &s->shd; - desc = (FFVulkanDescriptorSetBinding []) { - { - .name = "prev_img", + SPEC_LIST_CREATE(sl, 2, 2*sizeof(uint32_t)) + SPEC_LIST_ADD(sl, 0, 32, s->nb_planes); + SPEC_LIST_ADD(sl, 1, 32, SLICES); + + ff_vk_shader_load(&s->shd, VK_SHADER_STAGE_COMPUTE_BIT, sl, + (int []) { 32, 32, 1 }, 0); + + const FFVulkanDescriptorSetBinding desc[] = { + { /* prev_img */ + .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, + .stages = VK_SHADER_STAGE_COMPUTE_BIT, + .elems = av_pix_fmt_count_planes(s->vkctx.input_format), + }, + { /* cur_img */ .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, - .mem_layout = ff_vk_shader_rep_fmt(s->vkctx.input_format, FF_VK_REP_UINT), - .mem_quali = "readonly", - .dimensions = 2, - .elems = av_pix_fmt_count_planes(s->vkctx.input_format), .stages = VK_SHADER_STAGE_COMPUTE_BIT, - }, { - .name = "cur_img", - .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, - .mem_layout = ff_vk_shader_rep_fmt(s->vkctx.input_format, FF_VK_REP_UINT), - .mem_quali = "readonly", - .dimensions = 2, .elems = av_pix_fmt_count_planes(s->vkctx.input_format), - .stages = VK_SHADER_STAGE_COMPUTE_BIT, - }, { - .name = "sad_buffer", + }, + { /* sad_buffer */ .type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, .stages = VK_SHADER_STAGE_COMPUTE_BIT, - .buf_content = "uint frame_sad[];", } }; + ff_vk_shader_add_descriptor_set(vkctx, &s->shd, desc, 3, 0, 0); - RET(ff_vk_shader_add_descriptor_set(vkctx, &s->shd, desc, 3, 0, 0)); - - GLSLC(0, shared uint wg_sum; ); - GLSLC(0, void main() ); - GLSLC(0, { ); - GLSLF(1, const uint slice = gl_WorkGroupID.x %% %u; ,SLICES); - GLSLC(1, const ivec2 pos = ivec2(gl_GlobalInvocationID.xy); ); - GLSLC(1, wg_sum = 0; ); - GLSLC(1, barrier(); ); - for (int i = 0; i < s->nb_planes; i++) { - GLSLF(1, if (IS_WITHIN(pos, imageSize(cur_img[%d]))) { ,i); - GLSLF(2, uvec4 prev = imageLoad(prev_img[%d], pos); ,i); - GLSLF(2, uvec4 cur = imageLoad(cur_img[%d], pos); ,i); - GLSLC(2, uvec4 sad = abs(ivec4(cur) - ivec4(prev)); ); - GLSLC(2, uint sum = subgroupAdd(sad.x + sad.y + sad.z); ); - GLSLC(2, if (subgroupElect()) ); - GLSLC(3, atomicAdd(wg_sum, sum); ); - GLSLC(1, } ); - } - GLSLC(1, barrier(); ); - GLSLC(1, if (gl_LocalInvocationIndex == 0) ); - GLSLC(2, atomicAdd(frame_sad[slice], wg_sum); ); - GLSLC(0, } ); - - RET(spv->compile_shader(vkctx, spv, &s->shd, &spv_data, &spv_len, "main", - &spv_opaque)); - RET(ff_vk_shader_link(vkctx, &s->shd, spv_data, spv_len, "main")); + RET(ff_vk_shader_link(vkctx, &s->shd, + ff_scdet_comp_spv_data, + ff_scdet_comp_spv_len, "main")); RET(ff_vk_shader_register_exec(vkctx, &s->e, &s->shd)); s->initialized = 1; fail: - if (spv_opaque) - spv->free_shader(spv, &spv_opaque); - if (spv) - spv->uninit(&spv); - return err; } diff --git a/libavfilter/vulkan/Makefile b/libavfilter/vulkan/Makefile index 59c7de46ce..1f9e68c166 100644 --- a/libavfilter/vulkan/Makefile +++ b/libavfilter/vulkan/Makefile @@ -7,6 +7,7 @@ OBJS-$(CONFIG_CHROMABER_VULKAN_FILTER) += vulkan/chromaber.comp.spv.o OBJS-$(CONFIG_COLOR_VULKAN_FILTER) += vulkan/color.comp.spv.o OBJS-$(CONFIG_GBLUR_VULKAN_FILTER) += vulkan/gblur.comp.spv.o OBJS-$(CONFIG_SCALE_VULKAN_FILTER) += vulkan/debayer.comp.spv.o +OBJS-$(CONFIG_SCDET_VULKAN_FILTER) += vulkan/scdet.comp.spv.o OBJS-$(CONFIG_FLIP_VULKAN_FILTER) += vulkan/flip.comp.spv.o OBJS-$(CONFIG_TRANSPOSE_VULKAN_FILTER) += vulkan/transpose.comp.spv.o OBJS-$(CONFIG_V360_VULKAN_FILTER) += vulkan/v360.comp.spv.o diff --git a/libavfilter/vulkan/scdet.comp.glsl b/libavfilter/vulkan/scdet.comp.glsl new file mode 100644 index 0000000000..a38189a759 --- /dev/null +++ b/libavfilter/vulkan/scdet.comp.glsl @@ -0,0 +1,62 @@ +/* + * Copyright 2025 (c) Niklas Haas + * + * 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 + */ + +#pragma shader_stage(compute) + +#extension GL_EXT_shader_image_load_formatted : require +#extension GL_EXT_scalar_block_layout : require +#extension GL_EXT_nonuniform_qualifier : require +#extension GL_KHR_shader_subgroup_arithmetic : require +#extension GL_EXT_null_initializer : require + +layout (constant_id = 0) const uint planes = 0; +layout (constant_id = 1) const uint slices = 0; + +layout (local_size_x_id = 253, local_size_y_id = 254, local_size_z_id = 255) in; + +layout (set = 0, binding = 0) uniform readonly uimage2D prev_img[]; +layout (set = 0, binding = 1) uniform readonly uimage2D cur_img[]; +layout (set = 0, binding = 2, scalar) buffer sad_buffer { + uint frame_sad[]; +}; + +shared uint wg_sum = { }; + +void main() +{ + const uint slice = gl_WorkGroupID.x % slices; + const ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + + for (uint i = 0; i < planes; i++) { + const ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + if (all(lessThan(pos, imageSize(cur_img[i])))) { + uvec4 prev = imageLoad(prev_img[i], pos); + uvec4 cur = imageLoad(cur_img[i], pos); + uvec4 sad = abs(ivec4(cur) - ivec4(prev)); + uint sum = subgroupAdd(sad.x + sad.y + sad.z); + if (subgroupElect()) + atomicAdd(wg_sum, sum); + } + } + + barrier(); + if (gl_LocalInvocationIndex == 0) + atomicAdd(frame_sad[slice], wg_sum); +}