The following command is on how to apply transpose_vulkan filter:
ffmpeg -init_hw_device vulkan -i input.264 -vf \
hwupload=extra_hw_frames=16,transpose_vulkan,hwdownload,format=yuv420p
output.264
Signed-off-by: Wu Jianhua <jianhua...@intel.com>
---
configure | 1 +
libavfilter/Makefile | 1 +
libavfilter/allfilters.c | 1 +
libavfilter/vf_transpose_vulkan.c | 316 ++++++++++++++++++++++++++++++
4 files changed, 319 insertions(+)
create mode 100644 libavfilter/vf_transpose_vulkan.c
diff --git a/configure b/configure
index a98a18abaa..12cb49e877 100755
--- a/configure
+++ b/configure
@@ -3718,6 +3718,7 @@ tonemap_vaapi_filter_deps="vaapi
VAProcFilterParameterBufferHDRToneMapping"
tonemap_opencl_filter_deps="opencl const_nan"
transpose_opencl_filter_deps="opencl"
transpose_vaapi_filter_deps="vaapi VAProcPipelineCaps_rotation_flags"
+transpose_vulkan_filter_deps="vulkan spirv_compiler"
unsharp_opencl_filter_deps="opencl"
uspp_filter_deps="gpl avcodec"
vaguedenoiser_filter_deps="gpl"
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index c8082c4a2f..8744cc3c63 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -483,6 +483,7 @@ OBJS-$(CONFIG_TRANSPOSE_FILTER) +=
vf_transpose.o
OBJS-$(CONFIG_TRANSPOSE_NPP_FILTER) += vf_transpose_npp.o
OBJS-$(CONFIG_TRANSPOSE_OPENCL_FILTER) += vf_transpose_opencl.o opencl.o
opencl/transpose.o
OBJS-$(CONFIG_TRANSPOSE_VAAPI_FILTER) += vf_transpose_vaapi.o
vaapi_vpp.o
+OBJS-$(CONFIG_TRANSPOSE_VULKAN_FILTER) += vf_transpose_vulkan.o vulkan.o
vulkan_filter.o
OBJS-$(CONFIG_TRIM_FILTER) += trim.o
OBJS-$(CONFIG_UNPREMULTIPLY_FILTER) += vf_premultiply.o framesync.o
OBJS-$(CONFIG_UNSHARP_FILTER) += vf_unsharp.o
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index b1af2cbcc8..9e16b4e71e 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -462,6 +462,7 @@ extern const AVFilter ff_vf_transpose;
extern const AVFilter ff_vf_transpose_npp;
extern const AVFilter ff_vf_transpose_opencl;
extern const AVFilter ff_vf_transpose_vaapi;
+extern const AVFilter ff_vf_transpose_vulkan;
extern const AVFilter ff_vf_trim;
extern const AVFilter ff_vf_unpremultiply;
extern const AVFilter ff_vf_unsharp;
diff --git a/libavfilter/vf_transpose_vulkan.c
b/libavfilter/vf_transpose_vulkan.c
new file mode 100644
index 0000000000..c9bae413c3
--- /dev/null
+++ b/libavfilter/vf_transpose_vulkan.c
@@ -0,0 +1,316 @@
+/*
+ * copyright (c) 2021 Wu Jianhua <jianhua...@intel.com>
+ * 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_filter.h"
+#include "internal.h"
+
+#define CGS 32
+
+typedef struct TransposeVulkanContext {
+ FFVulkanContext vkctx;
+ FFVkQueueFamilyCtx qf;
+ FFVkExecContext *exec;
+ FFVulkanPipeline *pl;
+
+ VkDescriptorImageInfo input_images[3];
+ VkDescriptorImageInfo output_images[3];
+
+ int initialized;
+} TransposeVulkanContext;
+
+static av_cold int init_filter(AVFilterContext *ctx, AVFrame *in)
+{
+ int err = 0;
+ FFVkSPIRVShader *shd;
+ TransposeVulkanContext *s = ctx->priv;
+ FFVulkanContext *vkctx = &s->vkctx;
+ const int planes = av_pix_fmt_count_planes(s->vkctx.output_format);
+
+ FFVulkanDescriptorSetBinding image_descs[] = {
+ {
+ .name = "input_images",
+ .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+ .dimensions = 2,
+ .elems = planes,
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ .updater = s->input_images,
+ },
+ {
+ .name = "output_images",
+ .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,
+ .updater = s->output_images,
+ },
+ };
+
+ image_descs[0].sampler = ff_vk_init_sampler(vkctx, 1, VK_FILTER_LINEAR);
+ if (!image_descs[0].sampler)
+ return AVERROR_EXTERNAL;
+
+ ff_vk_qf_init(vkctx, &s->qf, VK_QUEUE_COMPUTE_BIT, 0);
+
+ {
+ s->pl = ff_vk_create_pipeline(vkctx, &s->qf);
+ if (!s->pl)
+ return AVERROR(ENOMEM);
+
+ shd = ff_vk_init_shader(s->pl, "transpose_compute",
image_descs[0].stages);
+ if (!shd)
+ return AVERROR(ENOMEM);
+
+ ff_vk_set_compute_shader_sizes(shd, (int [3]){ CGS, 1, 1 });
+ RET(ff_vk_add_descriptor_set(vkctx, s->pl, shd, image_descs,
FF_ARRAY_ELEMS(image_descs), 0));
+
+ 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_images[%i]); ,i);
+ GLSLC(1, if (IS_WITHIN(pos, size)) { );
+ GLSLF(2, vec4 res = texture(input_images[%i], pos.yx); ,i);
+ GLSLF(2, imageStore(output_images[%i], pos, res); ,i);
+ GLSLC(1, } );
+ }
+ GLSLC(0, } );
+
+ RET(ff_vk_compile_shader(vkctx, shd, "main"));
+ RET(ff_vk_init_pipeline_layout(vkctx, s->pl));
+ RET(ff_vk_init_compute_pipeline(vkctx, s->pl));
+ }
+
+ RET(ff_vk_create_exec_ctx(vkctx, &s->exec, &s->qf));
+ s->initialized = 1;
+
+fail:
+ return err;
+}
+
+static int process_frames(AVFilterContext *avctx, AVFrame *outframe, AVFrame
*inframe)
+{
+ int err = 0;
+ VkCommandBuffer cmd_buf;
+ TransposeVulkanContext *s = avctx->priv;
+ FFVulkanContext *vkctx = &s->vkctx;
+ FFVulkanFunctions *vk = &s->vkctx.vkfn;
+ const int planes = av_pix_fmt_count_planes(s->vkctx.output_format);
+
+ AVVkFrame *in = (AVVkFrame *)inframe->data[0];
+ AVVkFrame *out = (AVVkFrame *)outframe->data[0];
+
+ const VkFormat *input_formats =
av_vkfmt_from_pixfmt(s->vkctx.input_format);
+ const VkFormat *output_formats =
av_vkfmt_from_pixfmt(s->vkctx.output_format);
+
+ ff_vk_start_exec_recording(vkctx, s->exec);
+ cmd_buf = ff_vk_get_exec_buf(s->exec);
+
+ for (int i = 0; i < planes; i++) {
+ RET(ff_vk_create_imageview(vkctx, s->exec,
+ &s->input_images[i].imageView, in->img[i],
+ input_formats[i],
+ ff_comp_identity_map));
+
+ RET(ff_vk_create_imageview(vkctx, 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->output_images[i].imageLayout = VK_IMAGE_LAYOUT_GENERAL;
+ }
+
+ ff_vk_update_descriptor_set(vkctx, s->pl, 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,
+ .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;
+
+ out->layout[i] = barriers[1].newLayout;
+ out->access[i] = barriers[1].dstAccessMask;
+ }
+
+ ff_vk_bind_pipeline_exec(vkctx, s->exec, s->pl);
+ vk->CmdDispatch(cmd_buf, FFALIGN(s->vkctx.output_width, CGS)/CGS,
+ s->vkctx.output_height, 1);
+
+ ff_vk_add_exec_dep(vkctx, s->exec, inframe,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
+ ff_vk_add_exec_dep(vkctx, s->exec, outframe,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT);
+
+ err = ff_vk_submit_exec_queue(vkctx, s->exec);
+ if (err)
+ return err;
+
+ ff_vk_qf_rotate(&s->qf);
+
+ return 0;
+
+fail:
+ ff_vk_discard_exec_deps(s->exec);
+ return err;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *in)
+{
+ int err;
+ AVFrame *out = NULL;
+ AVFilterContext *ctx = inlink->dst;
+ TransposeVulkanContext *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));
+
+ RET(process_frames(ctx, out, in));
+
+ RET(av_frame_copy_props(out, in));
+
+ if (in->sample_aspect_ratio.num)
+ out->sample_aspect_ratio = in->sample_aspect_ratio;
+ else {
+ out->sample_aspect_ratio.num = in->sample_aspect_ratio.den;
+ out->sample_aspect_ratio.den = in->sample_aspect_ratio.num;
+ }
+
+ av_frame_free(&in);
+
+ return ff_filter_frame(outlink, out);
+
+fail:
+ av_frame_free(&in);
+ av_frame_free(&out);
+ return err;
+}
+
+static av_cold void transpose_vulkan_uninit(AVFilterContext *avctx)
+{
+ TransposeVulkanContext *s = avctx->priv;
+ ff_vk_uninit(&s->vkctx);
+
+ s->initialized = 0;
+}
+
+static int config_props_output(AVFilterLink *outlink)
+{
+ int err = 0;
+ AVFilterContext *avctx = outlink->src;
+ TransposeVulkanContext *s = avctx->priv;
+ FFVulkanContext *vkctx = &s->vkctx;
+ AVFilterLink *inlink = avctx->inputs[0];
+
+ vkctx->output_width = inlink->h;
+ vkctx->output_height = inlink->w;
+
+ RET(ff_vk_filter_config_output(outlink));
+
+ outlink->w = inlink->h;
+ outlink->h = inlink->w;
+
+ if (inlink->sample_aspect_ratio.num)
+ outlink->sample_aspect_ratio = av_div_q((AVRational) { 1, 1 },
+ inlink->sample_aspect_ratio);
+ else
+ outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
+
+ err = 0;
+
+fail:
+ return err;
+}
+
+static const AVOption transpose_vulkan_options[] = {
+ { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(transpose_vulkan);
+
+static const AVFilterPad transpose_vulkan_inputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .filter_frame = &filter_frame,
+ .config_props = &ff_vk_filter_config_input,
+ }
+};
+
+static const AVFilterPad transpose_vulkan_outputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .config_props = &config_props_output,
+ }
+};
+
+const AVFilter ff_vf_transpose_vulkan = {
+ .name = "transpose_vulkan",
+ .description = NULL_IF_CONFIG_SMALL("Transpose Vulkan Filter"),
+ .priv_size = sizeof(TransposeVulkanContext),
+ .init = &ff_vk_filter_init,
+ .uninit = &transpose_vulkan_uninit,
+ FILTER_INPUTS(transpose_vulkan_inputs),
+ FILTER_OUTPUTS(transpose_vulkan_outputs),
+ FILTER_SINGLE_PIXFMT(AV_PIX_FMT_VULKAN),
+ .priv_class = &transpose_vulkan_class,
+ .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
+};
--
2.25.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
