[FFmpeg-devel] [PATCH v3 1/4] libavfilter: add vf_xfade_vulkan
Marvin Scholz
epirat07 at gmail.com
Wed Jun 7 01:22:59 EEST 2023
This is an initial version of vf_xfade_vulkan based
on vf_xfade_opencl, for now only a subset of transitions
are supported.
---
Changes to v2:
- Fixed activate handling, same as in my patch for the xfade filter
- Added all remaining transitions the OpenCL filter supports
configure | 1 +
libavfilter/Makefile | 1 +
libavfilter/allfilters.c | 1 +
libavfilter/vf_xfade_vulkan.c | 499 ++++++++++++++++++++++++++++++++++
4 files changed, 502 insertions(+)
create mode 100644 libavfilter/vf_xfade_vulkan.c
diff --git a/configure b/configure
index 2992dae283..378cab1f3d 100755
--- a/configure
+++ b/configure
@@ -3824,6 +3824,7 @@ scale_vulkan_filter_deps="vulkan spirv_compiler"
vpp_qsv_filter_deps="libmfx"
vpp_qsv_filter_select="qsvvpp"
xfade_opencl_filter_deps="opencl"
+xfade_vulkan_filter_deps="vulkan spirv_compiler"
yadif_cuda_filter_deps="ffnvcodec"
yadif_cuda_filter_deps_any="cuda_nvcc cuda_llvm"
yadif_videotoolbox_filter_deps="metal corevideo videotoolbox"
diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 18935b1616..ff149a3733 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -552,6 +552,7 @@ OBJS-$(CONFIG_XBR_FILTER) += vf_xbr.o
OBJS-$(CONFIG_XCORRELATE_FILTER) += vf_convolve.o framesync.o
OBJS-$(CONFIG_XFADE_FILTER) += vf_xfade.o
OBJS-$(CONFIG_XFADE_OPENCL_FILTER) += vf_xfade_opencl.o opencl.o opencl/xfade.o
+OBJS-$(CONFIG_XFADE_VULKAN_FILTER) += vf_xfade_vulkan.o vulkan.o vulkan_filter.o
OBJS-$(CONFIG_XMEDIAN_FILTER) += vf_xmedian.o framesync.o
OBJS-$(CONFIG_XSTACK_FILTER) += vf_stack.o framesync.o
OBJS-$(CONFIG_YADIF_FILTER) += vf_yadif.o yadif_common.o
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index f1f781101b..6593e4eb83 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -519,6 +519,7 @@ extern const AVFilter ff_vf_xbr;
extern const AVFilter ff_vf_xcorrelate;
extern const AVFilter ff_vf_xfade;
extern const AVFilter ff_vf_xfade_opencl;
+extern const AVFilter ff_vf_xfade_vulkan;
extern const AVFilter ff_vf_xmedian;
extern const AVFilter ff_vf_xstack;
extern const AVFilter ff_vf_yadif;
diff --git a/libavfilter/vf_xfade_vulkan.c b/libavfilter/vf_xfade_vulkan.c
new file mode 100644
index 0000000000..f18a7b33aa
--- /dev/null
+++ b/libavfilter/vf_xfade_vulkan.c
@@ -0,0 +1,499 @@
+/*
+ * 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/avassert.h"
+#include "libavutil/random_seed.h"
+#include "libavutil/opt.h"
+#include "vulkan_filter.h"
+#include "vulkan_spirv.h"
+#include "filters.h"
+#include "internal.h"
+
+#define IN_A 0
+#define IN_B 1
+#define IN_NB 2
+
+typedef struct XFadeParameters {
+ float progress;
+} XFadeParameters;
+
+typedef struct XFadeVulkanContext {
+ FFVulkanContext vkctx;
+
+ int transition;
+ int64_t duration;
+ int64_t offset;
+
+ int initialized;
+ FFVulkanPipeline pl;
+ FFVkExecPool e;
+ FFVkQueueFamilyCtx qf;
+ FFVkSPIRVShader shd;
+ VkSampler sampler;
+
+ // PTS when the fade should start (in IN_A timebase)
+ int64_t start_pts;
+
+ // PTS offset between IN_A and IN_B
+ int64_t inputs_offset_pts;
+
+ // Duration of the transition
+ int64_t duration_pts;
+
+ // Current PTS of the first input (IN_A)
+ int64_t pts;
+
+ // If frames are currently just passed through
+ // unmodified, like before and after the actual
+ // transition.
+ int passthrough;
+
+ int status[IN_NB];
+} XFadeVulkanContext;
+
+enum XFadeTransitions {
+ FADE,
+ WIPELEFT,
+ WIPERIGHT,
+ NB_TRANSITIONS,
+};
+
+static const char transition_fade[] = {
+ C(0, void transition(int idx, ivec2 pos, float progress) )
+ C(0, { )
+ C(1, vec4 a = texture(a_images[idx], pos); )
+ C(1, vec4 b = texture(b_images[idx], pos); )
+ C(1, imageStore(output_images[idx], pos, mix(a, b, progress)); )
+ C(0, } )
+};
+
+static const char transition_wipeleft[] = {
+ C(0, void transition(int idx, ivec2 pos, float progress) )
+ C(0, { )
+ C(1, ivec2 size = imageSize(output_images[idx]); )
+ C(1, int s = int(size.x * (1.0 - progress)); )
+ C(1, vec4 a = texture(a_images[idx], pos); )
+ C(1, vec4 b = texture(b_images[idx], pos); )
+ C(1, imageStore(output_images[idx], pos, pos.x > s ? b : a); )
+ C(0, } )
+};
+
+static const char transition_wiperight[] = {
+ C(0, void transition(int idx, ivec2 pos, float progress) )
+ C(0, { )
+ C(1, ivec2 size = imageSize(output_images[idx]); )
+ C(1, int s = int(size.x * progress); )
+ C(1, vec4 a = texture(a_images[idx], pos); )
+ C(1, vec4 b = texture(b_images[idx], pos); )
+ C(1, imageStore(output_images[idx], pos, pos.x > s ? a : b); )
+ C(0, } )
+};
+
+static const char* transitions_map[NB_TRANSITIONS] = {
+ [FADE] = transition_fade,
+ [WIPELEFT] = transition_wipeleft,
+ [WIPERIGHT] = transition_wiperight,
+};
+
+static av_cold int init_vulkan(AVFilterContext *avctx)
+{
+ int err = 0;
+ uint8_t *spv_data;
+ size_t spv_len;
+ void *spv_opaque = NULL;
+ XFadeVulkanContext *s = avctx->priv;
+ FFVulkanContext *vkctx = &s->vkctx;
+ const int planes = av_pix_fmt_count_planes(s->vkctx.output_format);
+ FFVkSPIRVShader *shd = &s->shd;
+ FFVkSPIRVCompiler *spv;
+ FFVulkanDescriptorSetBinding *desc;
+
+ spv = ff_vk_spirv_init();
+ if (!spv) {
+ av_log(avctx, AV_LOG_ERROR, "Unable to initialize SPIR-V compiler!\n");
+ return AVERROR_EXTERNAL;
+ }
+
+ ff_vk_qf_init(vkctx, &s->qf, VK_QUEUE_COMPUTE_BIT);
+ RET(ff_vk_exec_pool_init(vkctx, &s->qf, &s->e, s->qf.nb_queues*4, 0, 0, 0, NULL));
+ RET(ff_vk_init_sampler(vkctx, &s->sampler, 1, VK_FILTER_NEAREST));
+ RET(ff_vk_shader_init(&s->pl, &s->shd, "xfade_compute",
+ VK_SHADER_STAGE_COMPUTE_BIT, 0));
+
+ ff_vk_shader_set_compute_sizes(&s->shd, 32, 32, 1);
+
+ desc = (FFVulkanDescriptorSetBinding []) {
+ {
+ .name = "a_images",
+ .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+ .dimensions = 2,
+ .elems = planes,
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ .samplers = DUP_SAMPLER(s->sampler),
+ },
+ {
+ .name = "b_images",
+ .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
+ .dimensions = 2,
+ .elems = planes,
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ .samplers = DUP_SAMPLER(s->sampler),
+ },
+ {
+ .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,
+ },
+ };
+
+ RET(ff_vk_pipeline_descriptor_set_add(vkctx, &s->pl, shd, desc, 3, 0, 0));
+
+ GLSLC(0, layout(push_constant, std430) uniform pushConstants { );
+ GLSLC(1, float progress; );
+ GLSLC(0, }; );
+
+ ff_vk_add_push_constant(&s->pl, 0, sizeof(XFadeParameters),
+ VK_SHADER_STAGE_COMPUTE_BIT);
+
+ // Add the right transition type function to the shader
+ GLSLD(transitions_map[s->transition]);
+
+ GLSLC(0, void main() );
+ GLSLC(0, { );
+ GLSLC(1, ivec2 pos = ivec2(gl_GlobalInvocationID.xy); );
+ GLSLF(1, int planes = %i; ,planes);
+ GLSLC(1, for (int i = 0; i < planes; i++) { );
+ GLSLC(2, transition(i, pos, progress); );
+ GLSLC(1, } );
+ GLSLC(0, } );
+
+ RET(spv->compile_shader(spv, avctx, shd, &spv_data, &spv_len, "main",
+ &spv_opaque));
+ RET(ff_vk_shader_create(vkctx, shd, spv_data, spv_len, "main"));
+
+ RET(ff_vk_init_compute_pipeline(vkctx, &s->pl, shd));
+ RET(ff_vk_exec_pipeline_register(vkctx, &s->e, &s->pl));
+
+ s->initialized = 1;
+
+fail:
+ if (spv_opaque)
+ spv->free_shader(spv, &spv_opaque);
+ if (spv)
+ spv->uninit(&spv);
+
+ return err;
+}
+
+static int xfade_frame(AVFilterContext *avctx, AVFrame *frame_a, AVFrame *frame_b)
+{
+ int err;
+ AVFilterLink *outlink = avctx->outputs[0];
+ XFadeVulkanContext *s = avctx->priv;
+ float progress;
+
+ AVFrame *output = ff_get_video_buffer(outlink, outlink->w, outlink->h);
+ if (!output) {
+ err = AVERROR(ENOMEM);
+ goto fail;
+ }
+
+ if (!s->initialized) {
+ AVHWFramesContext *a_fc = (AVHWFramesContext*)frame_a->hw_frames_ctx->data;
+ AVHWFramesContext *b_fc = (AVHWFramesContext*)frame_b->hw_frames_ctx->data;
+ if (a_fc->sw_format != b_fc->sw_format) {
+ av_log(avctx, AV_LOG_ERROR,
+ "Currently the sw format of the first input needs to match the second!\n");
+ return AVERROR(EINVAL);
+ }
+ RET(init_vulkan(avctx));
+ }
+
+ RET(av_frame_copy_props(output, frame_a));
+ output->pts = s->pts;
+
+ progress = av_clipf(
+ (float)(s->pts - s->start_pts) / s->duration_pts,
+ 0.f, 1.f);
+
+ RET(ff_vk_filter_process_Nin(&s->vkctx, &s->e, &s->pl, output,
+ (AVFrame *[]){ frame_a, frame_b }, 2, s->sampler,
+ &(XFadeParameters){ progress }, sizeof(XFadeParameters)));
+
+ return ff_filter_frame(outlink, output);
+
+fail:
+ av_frame_free(&output);
+ return err;
+}
+
+static int config_props_output(AVFilterLink *outlink)
+{
+ int err;
+ AVFilterContext *avctx = outlink->src;
+ XFadeVulkanContext *s = avctx->priv;
+ AVFilterLink *inlink_a = avctx->inputs[IN_A];
+ AVFilterLink *inlink_b = avctx->inputs[IN_B];
+
+ if (inlink_a->w != inlink_b->w || inlink_a->h != inlink_b->h) {
+ av_log(avctx, AV_LOG_ERROR, "First input link %s parameters "
+ "(size %dx%d) do not match the corresponding "
+ "second input link %s parameters (size %dx%d)\n",
+ avctx->input_pads[IN_A].name, inlink_a->w, inlink_a->h,
+ avctx->input_pads[IN_B].name, inlink_b->w, inlink_b->h);
+ return AVERROR(EINVAL);
+ }
+
+ if (inlink_a->time_base.num != inlink_b->time_base.num ||
+ inlink_a->time_base.den != inlink_b->time_base.den) {
+ av_log(avctx, AV_LOG_ERROR, "First input link %s timebase "
+ "(%d/%d) does not match the corresponding "
+ "second input link %s timebase (%d/%d)\n",
+ avctx->input_pads[IN_A].name, inlink_a->time_base.num, inlink_a->time_base.den,
+ avctx->input_pads[IN_B].name, inlink_b->time_base.num, inlink_b->time_base.den);
+ return AVERROR(EINVAL);
+ }
+
+ s->start_pts = s->inputs_offset_pts = AV_NOPTS_VALUE;
+
+ outlink->time_base = inlink_a->time_base;
+ outlink->frame_rate = inlink_a->frame_rate;
+ outlink->sample_aspect_ratio = inlink_a->sample_aspect_ratio;
+
+ if (s->duration)
+ s->duration_pts = av_rescale_q(s->duration, AV_TIME_BASE_Q, inlink_a->time_base);
+ RET(ff_vk_filter_config_output(outlink));
+
+fail:
+ return err;
+}
+
+static int forward_frame(XFadeVulkanContext *s,
+ AVFilterLink *inlink, AVFilterLink *outlink)
+{
+ int64_t status_pts;
+ int ret = 0, status;
+ AVFrame *frame = NULL;
+
+ ret = ff_inlink_consume_frame(inlink, &frame);
+ if (ret < 0)
+ return ret;
+
+ if (ret > 0) {
+ // If we do not have an offset yet, it's because we
+ // never got a first input. Just offset to 0
+ if (s->inputs_offset_pts == AV_NOPTS_VALUE)
+ s->inputs_offset_pts = -frame->pts;
+
+ // We got a frame, nothing to do other than adjusting the timestamp
+ frame->pts += s->inputs_offset_pts;
+ return ff_filter_frame(outlink, frame);
+ }
+
+ // Forward status with our timestamp
+ if (ff_inlink_acknowledge_status(inlink, &status, &status_pts)) {
+ if (s->inputs_offset_pts == AV_NOPTS_VALUE)
+ s->inputs_offset_pts = -status_pts;
+
+ ff_outlink_set_status(outlink, status, status_pts + s->inputs_offset_pts);
+ return 0;
+ }
+
+ // No frame available, request one if needed
+ if (ff_outlink_frame_wanted(outlink))
+ ff_inlink_request_frame(inlink);
+
+ return 0;
+}
+
+static int activate(AVFilterContext *avctx)
+{
+ XFadeVulkanContext *s = avctx->priv;
+ AVFilterLink *in_a = avctx->inputs[IN_A];
+ AVFilterLink *in_b = avctx->inputs[IN_B];
+ AVFilterLink *outlink = avctx->outputs[0];
+ int64_t status_pts;
+
+ FF_FILTER_FORWARD_STATUS_BACK_ALL(outlink, avctx);
+
+ // Check if we already transitioned or IN_A ended prematurely,
+ // in which case just forward the frames from IN_B with adjusted
+ // timestamps until EOF.
+ if (s->status[IN_A] && !s->status[IN_B])
+ return forward_frame(s, in_b, outlink);
+
+ // We did not finish transitioning yet and the first stream
+ // did not end either, so check if there are more frames to consume.
+ if (ff_inlink_check_available_frame(in_a)) {
+ AVFrame *peeked_frame = ff_inlink_peek_frame(in_a, 0);
+ s->pts = peeked_frame->pts;
+
+ if (s->start_pts == AV_NOPTS_VALUE)
+ s->start_pts =
+ s->pts + av_rescale_q(s->offset, AV_TIME_BASE_Q, in_a->time_base);
+
+ // Check if we are not yet transitioning, in which case
+ // just request and forward the input frame.
+ if (s->start_pts > s->pts) {
+ AVFrame *frame_a = NULL;
+ s->passthrough = 1;
+ ff_inlink_consume_frame(in_a, &frame_a);
+ return ff_filter_frame(outlink, frame_a);
+ }
+ s->passthrough = 0;
+
+ // We are transitioning, so we need a frame from IN_B
+ if (ff_inlink_check_available_frame(in_b)) {
+ int ret;
+ AVFrame *frame_a = NULL, *frame_b = NULL;
+ ff_inlink_consume_frame(avctx->inputs[IN_A], &frame_a);
+ ff_inlink_consume_frame(avctx->inputs[IN_B], &frame_b);
+
+ // Calculate PTS offset to first input
+ if (s->inputs_offset_pts == AV_NOPTS_VALUE)
+ s->inputs_offset_pts = s->pts - frame_b->pts;
+
+ // Check if we finished transitioning, in which case we
+ // report back EOF to IN_A as it is no longer needed.
+ if (s->pts - s->start_pts > s->duration_pts) {
+ s->status[IN_A] = AVERROR_EOF;
+ ff_inlink_set_status(in_a, AVERROR_EOF);
+ s->passthrough = 1;
+ }
+ ret = xfade_frame(avctx, frame_a, frame_b);
+ av_frame_free(&frame_a);
+ av_frame_free(&frame_b);
+ return ret;
+ }
+
+ // We did not get a frame from IN_B, check its status.
+ if (ff_inlink_acknowledge_status(in_b, &s->status[IN_B], &status_pts)) {
+ // We should transition, but IN_B is EOF so just report EOF output now.
+ ff_outlink_set_status(outlink, s->status[IN_B], s->pts);
+ return 0;
+ }
+
+ // We did not get a frame for IN_B but no EOF either, so just request more.
+ if (ff_outlink_frame_wanted(outlink)) {
+ ff_inlink_request_frame(in_b);
+ return 0;
+ }
+ }
+
+ // We did not get a frame from IN_A, check its status.
+ if (ff_inlink_acknowledge_status(in_a, &s->status[IN_A], &status_pts)) {
+ // No more frames from IN_A, do not report EOF though, we will just
+ // forward the IN_B frames in the next activate calls.
+ s->passthrough = 1;
+ ff_filter_set_ready(avctx, 100);
+ return 0;
+ }
+
+ // We have no frames yet from IN_A and no EOF, so request some.
+ if (ff_outlink_frame_wanted(outlink)) {
+ ff_inlink_request_frame(in_a);
+ return 0;
+ }
+
+ return FFERROR_NOT_READY;
+}
+
+static av_cold void uninit(AVFilterContext *avctx)
+{
+ XFadeVulkanContext *s = avctx->priv;
+ FFVulkanContext *vkctx = &s->vkctx;
+ FFVulkanFunctions *vk = &vkctx->vkfn;
+
+ ff_vk_exec_pool_free(vkctx, &s->e);
+ ff_vk_pipeline_free(vkctx, &s->pl);
+ ff_vk_shader_free(vkctx, &s->shd);
+
+ if (s->sampler)
+ vk->DestroySampler(vkctx->hwctx->act_dev, s->sampler,
+ vkctx->hwctx->alloc);
+
+ ff_vk_uninit(&s->vkctx);
+
+ s->initialized = 0;
+}
+
+static AVFrame *get_video_buffer(AVFilterLink *inlink, int w, int h)
+{
+ XFadeVulkanContext *s = inlink->dst->priv;
+
+ return s->passthrough ?
+ ff_null_get_video_buffer (inlink, w, h) :
+ ff_default_get_video_buffer(inlink, w, h);
+}
+
+#define OFFSET(x) offsetof(XFadeVulkanContext, x)
+#define FLAGS (AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM)
+
+static const AVOption xfade_vulkan_options[] = {
+ { "transition", "set cross fade transition", OFFSET(transition), AV_OPT_TYPE_INT, {.i64=FADE}, 0, NB_TRANSITIONS-1, FLAGS, "transition" },
+ { "fade", "fade transition", 0, AV_OPT_TYPE_CONST, {.i64=FADE}, 0, 0, FLAGS, "transition" },
+ { "wipeleft", "wipe left transition", 0, AV_OPT_TYPE_CONST, {.i64=WIPELEFT}, 0, 0, FLAGS, "transition" },
+ { "wiperight", "wipe right transition", 0, AV_OPT_TYPE_CONST, {.i64=WIPERIGHT}, 0, 0, FLAGS, "transition" },
+ { "duration", "set cross fade duration", OFFSET(duration), AV_OPT_TYPE_DURATION, {.i64=1000000}, 0, 60000000, FLAGS },
+ { "offset", "set cross fade start relative to first input stream", OFFSET(offset), AV_OPT_TYPE_DURATION, {.i64=0}, INT64_MIN, INT64_MAX, FLAGS },
+ { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(xfade_vulkan);
+
+static const AVFilterPad xfade_vulkan_inputs[] = {
+ {
+ .name = "main",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .get_buffer.video = &get_video_buffer,
+ .config_props = &ff_vk_filter_config_input,
+ },
+ {
+ .name = "xfade",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .get_buffer.video = &get_video_buffer,
+ .config_props = &ff_vk_filter_config_input,
+ },
+};
+
+static const AVFilterPad xfade_vulkan_outputs[] = {
+ {
+ .name = "default",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .config_props = &config_props_output,
+ },
+};
+
+const AVFilter ff_vf_xfade_vulkan = {
+ .name = "xfade_vulkan",
+ .description = NULL_IF_CONFIG_SMALL("Cross fade one video with another video."),
+ .priv_size = sizeof(XFadeVulkanContext),
+ .init = &ff_vk_filter_init,
+ .uninit = &uninit,
+ .activate = &activate,
+ FILTER_INPUTS(xfade_vulkan_inputs),
+ FILTER_OUTPUTS(xfade_vulkan_outputs),
+ FILTER_SINGLE_PIXFMT(AV_PIX_FMT_VULKAN),
+ .priv_class = &xfade_vulkan_class,
+ .flags_internal = FF_FILTER_FLAG_HWFRAME_AWARE,
+ .flags = AVFILTER_FLAG_HWDEVICE,
+};
--
2.37.0 (Apple Git-136)
More information about the ffmpeg-devel
mailing list