#include "flow.h"
#include "jpeg_frame_view.h"
#include "movit/util.h"
+#include "pbo_pool.h"
#include "player.h"
#include "shared/context.h"
#include "shared/httpd.h"
+#include "shared/metrics.h"
#include "shared/shared_defs.h"
#include "shared/mux.h"
#include "util.h"
using namespace std;
using namespace std::chrono;
+namespace {
+
+once_flag video_metrics_inited;
+Summary metric_jpeg_encode_time_seconds;
+Summary metric_fade_latency_seconds;
+Summary metric_interpolation_latency_seconds;
+Summary metric_fade_fence_wait_time_seconds;
+Summary metric_interpolation_fence_wait_time_seconds;
+
+void wait_for_upload(shared_ptr<Frame> &frame)
+{
+ if (frame->uploaded_interpolation != nullptr) {
+ glWaitSync(frame->uploaded_interpolation.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
+ frame->uploaded_interpolation.reset();
+ }
+}
+
+} // namespace
+
extern HTTPD *global_httpd;
struct VectorDestinationManager {
string encode_jpeg(const uint8_t *y_data, const uint8_t *cb_data, const uint8_t *cr_data, unsigned width, unsigned height, const string exif_data)
{
+ steady_clock::time_point start = steady_clock::now();
VectorDestinationManager dest;
jpeg_compress_struct cinfo;
jpeg_finish_compress(&cinfo);
jpeg_destroy_compress(&cinfo);
+ steady_clock::time_point stop = steady_clock::now();
+ metric_jpeg_encode_time_seconds.count_event(duration<double>(stop - start).count());
+
return move(dest.dest);
}
+string encode_jpeg_from_pbo(void *contents, unsigned width, unsigned height, const string exif_data)
+{
+ unsigned chroma_width = width / 2;
+
+ const uint8_t *y = (const uint8_t *)contents;
+ const uint8_t *cb = (const uint8_t *)contents + width * height;
+ const uint8_t *cr = (const uint8_t *)contents + width * height + chroma_width * height;
+ return encode_jpeg(y, cb, cr, width, height, move(exif_data));
+}
+
VideoStream::VideoStream(AVFormatContext *file_avctx)
: avctx(file_avctx), output_fast_forward(file_avctx != nullptr)
{
+ call_once(video_metrics_inited, [] {
+ vector<double> quantiles{ 0.01, 0.1, 0.25, 0.5, 0.75, 0.9, 0.99 };
+ metric_jpeg_encode_time_seconds.init(quantiles, 60.0);
+ global_metrics.add("jpeg_encode_time_seconds", &metric_jpeg_encode_time_seconds);
+ metric_fade_fence_wait_time_seconds.init(quantiles, 60.0);
+ global_metrics.add("fade_fence_wait_time_seconds", &metric_fade_fence_wait_time_seconds);
+ metric_interpolation_fence_wait_time_seconds.init(quantiles, 60.0);
+ global_metrics.add("interpolation_fence_wait_time_seconds", &metric_interpolation_fence_wait_time_seconds);
+ metric_fade_latency_seconds.init(quantiles, 60.0);
+ global_metrics.add("fade_latency_seconds", &metric_fade_latency_seconds);
+ metric_interpolation_latency_seconds.init(quantiles, 60.0);
+ global_metrics.add("interpolation_latency_seconds", &metric_interpolation_latency_seconds);
+ });
+
ycbcr_converter.reset(new YCbCrConverter(YCbCrConverter::OUTPUT_TO_DUAL_YCBCR, /*resource_pool=*/nullptr));
ycbcr_semiplanar_converter.reset(new YCbCrConverter(YCbCrConverter::OUTPUT_TO_SEMIPLANAR, /*resource_pool=*/nullptr));
shared_ptr<Frame> frame1 = decode_jpeg_with_cache(frame1_spec, DECODE_IF_NOT_IN_CACHE, &frame_reader, &did_decode);
shared_ptr<Frame> frame2 = decode_jpeg_with_cache(frame2_spec, DECODE_IF_NOT_IN_CACHE, &frame_reader, &did_decode);
+ wait_for_upload(frame1);
+ wait_for_upload(frame2);
ycbcr_semiplanar_converter->prepare_chain_for_fade(frame1, frame2, fade_alpha)->render_to_fbo(resources->fade_fbo, global_flags.width, global_flags.height);
// Set a fence we can wait for to make sure the CPU sees the read.
glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
check_error();
+ qf.fence_created = steady_clock::now();
qf.fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
check_error();
qf.resources = move(resources);
FrameOnDisk frame_spec = frame_no == 1 ? frame2 : frame1;
bool did_decode;
shared_ptr<Frame> frame = decode_jpeg_with_cache(frame_spec, DECODE_IF_NOT_IN_CACHE, &frame_reader, &did_decode);
+ wait_for_upload(frame);
ycbcr_converter->prepare_chain_for_conversion(frame)->render_to_fbo(resources->input_fbos[frame_no], global_flags.width, global_flags.height);
if (frame_no == 1) {
qf.exif_data = frame->exif_data; // Use the white point from the last frame.
// Now decode the image we are fading against.
bool did_decode;
shared_ptr<Frame> frame2 = decode_jpeg_with_cache(secondary_frame, DECODE_IF_NOT_IN_CACHE, &frame_reader, &did_decode);
+ wait_for_upload(frame2);
// Then fade against it, putting it into the fade Y' and CbCr textures.
RGBTriplet neutral_color = get_neutral_color(qf.exif_data);
// Set a fence we can wait for to make sure the CPU sees the read.
glMemoryBarrier(GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT);
check_error();
+ qf.fence_created = steady_clock::now();
qf.fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
check_error();
qf.resources = move(resources);
namespace {
-shared_ptr<Frame> frame_from_pbo(void *contents, size_t width, size_t height)
+RefCountedTexture clone_r8_texture(GLuint src_tex, unsigned width, unsigned height)
{
- size_t chroma_width = width / 2;
-
- const uint8_t *y = (const uint8_t *)contents;
- const uint8_t *cb = (const uint8_t *)contents + width * height;
- const uint8_t *cr = (const uint8_t *)contents + width * height + chroma_width * height;
+ GLuint tex;
+ glCreateTextures(GL_TEXTURE_2D, 1, &tex);
+ check_error();
+ glTextureStorage2D(tex, 1, GL_R8, width, height);
+ check_error();
+ glCopyImageSubData(src_tex, GL_TEXTURE_2D, 0, 0, 0, 0,
+ tex, GL_TEXTURE_2D, 0, 0, 0, 0,
+ width, height, 1);
+ check_error();
+ glTextureParameteri(tex, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+ check_error();
+ glTextureParameteri(tex, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
+ check_error();
+ glTextureParameteri(tex, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+ check_error();
+ glTextureParameteri(tex, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+ check_error();
- shared_ptr<Frame> frame(new Frame);
- frame->y.reset(new uint8_t[width * height]);
- frame->cb.reset(new uint8_t[chroma_width * height]);
- frame->cr.reset(new uint8_t[chroma_width * height]);
- for (unsigned yy = 0; yy < height; ++yy) {
- memcpy(frame->y.get() + width * yy, y + width * yy, width);
- memcpy(frame->cb.get() + chroma_width * yy, cb + chroma_width * yy, chroma_width);
- memcpy(frame->cr.get() + chroma_width * yy, cr + chroma_width * yy, chroma_width);
- }
- frame->is_semiplanar = false;
- frame->width = width;
- frame->height = height;
- frame->chroma_subsampling_x = 2;
- frame->chroma_subsampling_y = 1;
- frame->pitch_y = width;
- frame->pitch_chroma = chroma_width;
- return frame;
+ return RefCountedTexture(new GLuint(tex), TextureDeleter());
}
} // namespace
abort();
}
+ init_pbo_pool();
+
while (!should_quit) {
QueuedFrame qf;
{
add_audio_or_silence(qf);
} else if (qf.type == QueuedFrame::FADED) {
+ steady_clock::time_point start = steady_clock::now();
glClientWaitSync(qf.fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
-
- shared_ptr<Frame> frame = frame_from_pbo(qf.resources->pbo_contents, global_flags.width, global_flags.height);
- assert(frame->exif_data.empty());
+ steady_clock::time_point stop = steady_clock::now();
+ metric_fade_fence_wait_time_seconds.count_event(duration<double>(stop - start).count());
+ metric_fade_latency_seconds.count_event(duration<double>(stop - qf.fence_created).count());
// Now JPEG encode it, and send it on to the stream.
- string jpeg = encode_jpeg(frame->y.get(), frame->cb.get(), frame->cr.get(), global_flags.width, global_flags.height, /*exif_data=*/"");
+ string jpeg = encode_jpeg_from_pbo(qf.resources->pbo_contents, global_flags.width, global_flags.height, /*exif_data=*/"");
AVPacket pkt;
av_init_packet(&pkt);
add_audio_or_silence(qf);
} else if (qf.type == QueuedFrame::INTERPOLATED || qf.type == QueuedFrame::FADED_INTERPOLATED) {
+ steady_clock::time_point start = steady_clock::now();
glClientWaitSync(qf.fence.get(), /*flags=*/0, GL_TIMEOUT_IGNORED);
+ steady_clock::time_point stop = steady_clock::now();
+ metric_interpolation_fence_wait_time_seconds.count_event(duration<double>(stop - start).count());
+ metric_interpolation_latency_seconds.count_event(duration<double>(stop - qf.fence_created).count());
// Send it on to display.
- shared_ptr<Frame> frame = frame_from_pbo(qf.resources->pbo_contents, global_flags.width, global_flags.height);
if (qf.display_decoded_func != nullptr) {
- qf.display_decoded_func(frame);
+ shared_ptr<Frame> frame(new Frame);
+ if (qf.type == QueuedFrame::FADED_INTERPOLATED) {
+ frame->y = clone_r8_texture(qf.resources->fade_y_output_tex, global_flags.width, global_flags.height);
+ } else {
+ frame->y = clone_r8_texture(qf.output_tex, global_flags.width, global_flags.height);
+ }
+ frame->cb = clone_r8_texture(qf.resources->cb_tex, global_flags.width / 2, global_flags.height);
+ frame->cr = clone_r8_texture(qf.resources->cr_tex, global_flags.width / 2, global_flags.height);
+ frame->width = global_flags.width;
+ frame->height = global_flags.height;
+ frame->chroma_subsampling_x = 2;
+ frame->chroma_subsampling_y = 1;
+ frame->uploaded_ui_thread = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
+ qf.display_decoded_func(move(frame));
}
// Now JPEG encode it, and send it on to the stream.
- string jpeg = encode_jpeg(frame->y.get(), frame->cb.get(), frame->cr.get(), global_flags.width, global_flags.height, move(qf.exif_data));
+ string jpeg = encode_jpeg_from_pbo(qf.resources->pbo_contents, global_flags.width, global_flags.height, move(qf.exif_data));
if (qf.flow_tex != 0) {
compute_flow->release_texture(qf.flow_tex);
}