]> git.sesse.net Git - nageru/blobdiff - x264_encoder.cpp
Fix an issue where the mixer lagging too much behind CEF would cause us to display...
[nageru] / x264_encoder.cpp
index fcd20e7aa519b33345a86b08728177bddf515b8b..ccf6942e1813ef7aa4748b45b531d108dd06d64f 100644 (file)
@@ -1,18 +1,24 @@
 #include "x264_encoder.h"
 
 #include <assert.h>
+#include <dlfcn.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 #include <x264.h>
+#include <atomic>
 #include <cstdint>
+#include <functional>
+#include <mutex>
 
 #include "defs.h"
 #include "flags.h"
+#include "metrics.h"
 #include "mux.h"
 #include "print_latency.h"
 #include "timebase.h"
+#include "x264_dynamic.h"
 #include "x264_speed_control.h"
 
 extern "C" {
@@ -20,13 +26,30 @@ extern "C" {
 #include <libavformat/avformat.h>
 }
 
+using namespace movit;
 using namespace std;
 using namespace std::chrono;
+using namespace std::placeholders;
 
 namespace {
 
+// X264Encoder can be restarted if --record-x264-video is set, so make these
+// metrics global.
+atomic<int64_t> metric_x264_queued_frames{0};
+atomic<int64_t> metric_x264_max_queued_frames{X264_QUEUE_LENGTH};
+atomic<int64_t> metric_x264_dropped_frames{0};
+atomic<int64_t> metric_x264_output_frames_i{0};
+atomic<int64_t> metric_x264_output_frames_p{0};
+atomic<int64_t> metric_x264_output_frames_b{0};
+Histogram metric_x264_crf;
+LatencyHistogram x264_latency_histogram;
+once_flag x264_metrics_inited;
+
 void update_vbv_settings(x264_param_t *param)
 {
+       if (global_flags.x264_bitrate == -1) {
+               return;
+       }
        if (global_flags.x264_vbv_buffer_size < 0) {
                param->rc.i_vbv_buffer_size = param->rc.i_bitrate;  // One-second VBV.
        } else {
@@ -42,11 +65,26 @@ void update_vbv_settings(x264_param_t *param)
 }  // namespace
 
 X264Encoder::X264Encoder(AVOutputFormat *oformat)
-       : wants_global_headers(oformat->flags & AVFMT_GLOBALHEADER)
+       : wants_global_headers(oformat->flags & AVFMT_GLOBALHEADER),
+         dyn(load_x264_for_bit_depth(global_flags.x264_bit_depth))
 {
-       frame_pool.reset(new uint8_t[global_flags.width * global_flags.height * 2 * X264_QUEUE_LENGTH]);
+       call_once(x264_metrics_inited, [](){
+               global_metrics.add("x264_queued_frames", &metric_x264_queued_frames, Metrics::TYPE_GAUGE);
+               global_metrics.add("x264_max_queued_frames", &metric_x264_max_queued_frames, Metrics::TYPE_GAUGE);
+               global_metrics.add("x264_dropped_frames", &metric_x264_dropped_frames);
+               global_metrics.add("x264_output_frames", {{ "type", "i" }}, &metric_x264_output_frames_i);
+               global_metrics.add("x264_output_frames", {{ "type", "p" }}, &metric_x264_output_frames_p);
+               global_metrics.add("x264_output_frames", {{ "type", "b" }}, &metric_x264_output_frames_b);
+
+               metric_x264_crf.init_uniform(50);
+               global_metrics.add("x264_crf", &metric_x264_crf);
+               x264_latency_histogram.init("x264");
+       });
+
+       size_t bytes_per_pixel = global_flags.x264_bit_depth > 8 ? 2 : 1;
+       frame_pool.reset(new uint8_t[global_flags.width * global_flags.height * 2 * bytes_per_pixel * X264_QUEUE_LENGTH]);
        for (unsigned i = 0; i < X264_QUEUE_LENGTH; ++i) {
-               free_frames.push(frame_pool.get() + i * (global_flags.width * global_flags.height * 2));
+               free_frames.push(frame_pool.get() + i * (global_flags.width * global_flags.height * 2 * bytes_per_pixel));
        }
        encoder_thread = thread(&X264Encoder::encoder_thread_func, this);
 }
@@ -56,19 +94,26 @@ X264Encoder::~X264Encoder()
        should_quit = true;
        queued_frames_nonempty.notify_all();
        encoder_thread.join();
+       if (dyn.handle) {
+               dlclose(dyn.handle);
+       }
 }
 
-void X264Encoder::add_frame(int64_t pts, int64_t duration, const uint8_t *data, const ReceivedTimestamps &received_ts)
+void X264Encoder::add_frame(int64_t pts, int64_t duration, YCbCrLumaCoefficients ycbcr_coefficients, const uint8_t *data, const ReceivedTimestamps &received_ts)
 {
+       assert(!should_quit);
+
        QueuedFrame qf;
        qf.pts = pts;
        qf.duration = duration;
+       qf.ycbcr_coefficients = ycbcr_coefficients;
        qf.received_ts = received_ts;
 
        {
                lock_guard<mutex> lock(mu);
                if (free_frames.empty()) {
                        fprintf(stderr, "WARNING: x264 queue full, dropping frame with pts %ld\n", pts);
+                       ++metric_x264_dropped_frames;
                        return;
                }
 
@@ -76,23 +121,28 @@ void X264Encoder::add_frame(int64_t pts, int64_t duration, const uint8_t *data,
                free_frames.pop();
        }
 
-       memcpy(qf.data, data, global_flags.width * global_flags.height * 2);
+       size_t bytes_per_pixel = global_flags.x264_bit_depth > 8 ? 2 : 1;
+       memcpy(qf.data, data, global_flags.width * global_flags.height * 2 * bytes_per_pixel);
 
        {
                lock_guard<mutex> lock(mu);
                queued_frames.push(qf);
                queued_frames_nonempty.notify_all();
+               metric_x264_queued_frames = queued_frames.size();
        }
 }
        
 void X264Encoder::init_x264()
 {
        x264_param_t param;
-       x264_param_default_preset(&param, global_flags.x264_preset.c_str(), global_flags.x264_tune.c_str());
+       dyn.x264_param_default_preset(&param, global_flags.x264_preset.c_str(), global_flags.x264_tune.c_str());
 
        param.i_width = global_flags.width;
        param.i_height = global_flags.height;
        param.i_csp = X264_CSP_NV12;
+       if (global_flags.x264_bit_depth > 8) {
+               param.i_csp |= X264_CSP_HIGH_DEPTH;
+       }
        param.b_vfr_input = 1;
        param.i_timebase_num = 1;
        param.i_timebase_den = TIMEBASE;
@@ -101,16 +151,24 @@ void X264Encoder::init_x264()
                param.i_frame_reference = 16;  // Because speedcontrol is never allowed to change this above what we set at start.
        }
 
-       // NOTE: These should be in sync with the ones in h264encode.cpp (sbs_rbsp()).
+       // NOTE: These should be in sync with the ones in quicksync_encoder.cpp (sps_rbsp()).
        param.vui.i_vidformat = 5;  // Unspecified.
        param.vui.b_fullrange = 0;
        param.vui.i_colorprim = 1;  // BT.709.
-       param.vui.i_transfer = 2;  // Unspecified (since we use sRGB).
-       param.vui.i_colmatrix = 6;  // BT.601/SMPTE 170M.
-
+       param.vui.i_transfer = 13;  // sRGB.
+       if (global_flags.ycbcr_rec709_coefficients) {
+               param.vui.i_colmatrix = 1;  // BT.709.
+       } else {
+               param.vui.i_colmatrix = 6;  // BT.601/SMPTE 170M.
+       }
 
-       param.rc.i_rc_method = X264_RC_ABR;
-       param.rc.i_bitrate = global_flags.x264_bitrate;
+       if (!isinf(global_flags.x264_crf)) {
+               param.rc.i_rc_method = X264_RC_CRF;
+               param.rc.f_rf_constant = global_flags.x264_crf;
+       } else {
+               param.rc.i_rc_method = X264_RC_ABR;
+               param.rc.i_bitrate = global_flags.x264_bitrate;
+       }
        update_vbv_settings(&param);
        if (param.rc.i_vbv_max_bitrate > 0) {
                // If the user wants VBV control to cap the max rate, it is
@@ -147,24 +205,28 @@ void X264Encoder::init_x264()
        for (const string &str : global_flags.x264_extra_param) {
                const size_t pos = str.find(',');
                if (pos == string::npos) {
-                       if (x264_param_parse(&param, str.c_str(), nullptr) != 0) {
+                       if (dyn.x264_param_parse(&param, str.c_str(), nullptr) != 0) {
                                fprintf(stderr, "ERROR: x264 rejected parameter '%s'\n", str.c_str());
                        }
                } else {
                        const string key = str.substr(0, pos);
                        const string value = str.substr(pos + 1);
-                       if (x264_param_parse(&param, key.c_str(), value.c_str()) != 0) {
+                       if (dyn.x264_param_parse(&param, key.c_str(), value.c_str()) != 0) {
                                fprintf(stderr, "ERROR: x264 rejected parameter '%s' set to '%s'\n",
                                        key.c_str(), value.c_str());
                        }
                }
        }
 
-       x264_param_apply_profile(&param, "high");
+       if (global_flags.x264_bit_depth > 8) {
+               dyn.x264_param_apply_profile(&param, "high10");
+       } else {
+               dyn.x264_param_apply_profile(&param, "high");
+       }
 
        param.b_repeat_headers = !wants_global_headers;
 
-       x264 = x264_encoder_open(&param);
+       x264 = dyn.x264_encoder_open(&param);
        if (x264 == nullptr) {
                fprintf(stderr, "ERROR: x264 initialization failed.\n");
                exit(1);
@@ -178,7 +240,7 @@ void X264Encoder::init_x264()
                x264_nal_t *nal;
                int num_nal;
 
-               x264_encoder_headers(x264, &nal, &num_nal);
+               dyn.x264_encoder_headers(x264, &nal, &num_nal);
 
                for (int i = 0; i < num_nal; ++i) {
                        if (nal[i].i_type == NAL_SEI) {
@@ -197,6 +259,7 @@ void X264Encoder::encoder_thread_func()
                perror("nice()");
                // No exit; it's not fatal.
        }
+       pthread_setname_np(pthread_self(), "x264_encode");
        init_x264();
        x264_init_done = true;
 
@@ -218,6 +281,7 @@ void X264Encoder::encoder_thread_func()
                                qf.data = nullptr;
                        }
 
+                       metric_x264_queued_frames = queued_frames.size();
                        frames_left = !queued_frames.empty();
                }
 
@@ -230,9 +294,9 @@ void X264Encoder::encoder_thread_func()
 
                // We should quit only if the should_quit flag is set _and_ we have nothing
                // in either queue.
-       } while (!should_quit || frames_left || x264_encoder_delayed_frames(x264) > 0);
+       } while (!should_quit || frames_left || dyn.x264_encoder_delayed_frames(x264) > 0);
 
-       x264_encoder_close(x264);
+       dyn.x264_encoder_close(x264);
 }
 
 void X264Encoder::encode_frame(X264Encoder::QueuedFrame qf)
@@ -243,15 +307,24 @@ void X264Encoder::encode_frame(X264Encoder::QueuedFrame qf)
        x264_picture_t *input_pic = nullptr;
 
        if (qf.data) {
-               x264_picture_init(&pic);
+               dyn.x264_picture_init(&pic);
 
                pic.i_pts = qf.pts;
-               pic.img.i_csp = X264_CSP_NV12;
-               pic.img.i_plane = 2;
-               pic.img.plane[0] = qf.data;
-               pic.img.i_stride[0] = global_flags.width;
-               pic.img.plane[1] = qf.data + global_flags.width * global_flags.height;
-               pic.img.i_stride[1] = global_flags.width / 2 * sizeof(uint16_t);
+               if (global_flags.x264_bit_depth > 8) {
+                       pic.img.i_csp = X264_CSP_NV12 | X264_CSP_HIGH_DEPTH;
+                       pic.img.i_plane = 2;
+                       pic.img.plane[0] = qf.data;
+                       pic.img.i_stride[0] = global_flags.width * sizeof(uint16_t);
+                       pic.img.plane[1] = qf.data + global_flags.width * global_flags.height * sizeof(uint16_t);
+                       pic.img.i_stride[1] = global_flags.width / 2 * sizeof(uint32_t);
+               } else {
+                       pic.img.i_csp = X264_CSP_NV12;
+                       pic.img.i_plane = 2;
+                       pic.img.plane[0] = qf.data;
+                       pic.img.i_stride[0] = global_flags.width;
+                       pic.img.plane[1] = qf.data + global_flags.width * global_flags.height;
+                       pic.img.i_stride[1] = global_flags.width / 2 * sizeof(uint16_t);
+               }
                pic.opaque = reinterpret_cast<void *>(intptr_t(qf.duration));
 
                input_pic = &pic;
@@ -259,33 +332,36 @@ void X264Encoder::encode_frame(X264Encoder::QueuedFrame qf)
                frames_being_encoded[qf.pts] = qf.received_ts;
        }
 
-       // See if we have a new bitrate to change to.
-       unsigned new_rate = new_bitrate_kbit.exchange(0);  // Read and clear.
-       if (new_rate != 0) {
-               if (speed_control) {
-                       speed_control->set_config_override_function([new_rate](x264_param_t *param) {
-                               param->rc.i_bitrate = new_rate;
-                               update_vbv_settings(param);
-                       });
-               } else {
-                       x264_param_t param;
-                       x264_encoder_parameters(x264, &param);
-                       param.rc.i_bitrate = new_rate;
-                       update_vbv_settings(&param);
-                       x264_encoder_reconfig(x264, &param);
-               }
+       unsigned new_rate = new_bitrate_kbit.load();  // Can be 0 for no change.
+       if (speed_control) {
+               speed_control->set_config_override_function(bind(&speed_control_override_func, new_rate, qf.ycbcr_coefficients, _1));
+       } else {
+               x264_param_t param;
+               dyn.x264_encoder_parameters(x264, &param);
+               speed_control_override_func(new_rate, qf.ycbcr_coefficients, &param);
+               dyn.x264_encoder_reconfig(x264, &param);
        }
 
        if (speed_control) {
                speed_control->before_frame(float(free_frames.size()) / X264_QUEUE_LENGTH, X264_QUEUE_LENGTH, 1e6 * qf.duration / TIMEBASE);
        }
-       x264_encoder_encode(x264, &nal, &num_nal, input_pic, &pic);
+       dyn.x264_encoder_encode(x264, &nal, &num_nal, input_pic, &pic);
        if (speed_control) {
                speed_control->after_frame();
        }
 
        if (num_nal == 0) return;
 
+       if (IS_X264_TYPE_I(pic.i_type)) {
+               ++metric_x264_output_frames_i;
+       } else if (IS_X264_TYPE_B(pic.i_type)) {
+               ++metric_x264_output_frames_b;
+       } else {
+               ++metric_x264_output_frames_p;
+       }
+
+       metric_x264_crf.count_event(pic.prop.f_crf_avg);
+
        if (frames_being_encoded.count(pic.i_pts)) {
                ReceivedTimestamps received_ts = frames_being_encoded[pic.i_pts];
                frames_being_encoded.erase(pic.i_pts);
@@ -293,7 +369,7 @@ void X264Encoder::encode_frame(X264Encoder::QueuedFrame qf)
                static int frameno = 0;
                print_latency("Current x264 latency (video inputs → network mux):",
                        received_ts, (pic.i_type == X264_TYPE_B || pic.i_type == X264_TYPE_BREF),
-                       &frameno);
+                       &frameno, &x264_latency_histogram);
        } else {
                assert(false);
        }
@@ -331,5 +407,22 @@ void X264Encoder::encode_frame(X264Encoder::QueuedFrame qf)
        }
        pkt.duration = reinterpret_cast<intptr_t>(pic.opaque);
 
-       mux->add_packet(pkt, pic.i_pts, pic.i_dts);
+       for (Mux *mux : muxes) {
+               mux->add_packet(pkt, pic.i_pts, pic.i_dts);
+       }
+}
+
+void X264Encoder::speed_control_override_func(unsigned bitrate_kbit, movit::YCbCrLumaCoefficients ycbcr_coefficients, x264_param_t *param)
+{
+       if (bitrate_kbit != 0) {
+               param->rc.i_bitrate = bitrate_kbit;
+               update_vbv_settings(param);
+       }
+
+       if (ycbcr_coefficients == YCBCR_REC_709) {
+               param->vui.i_colmatrix = 1;  // BT.709.
+       } else {
+               assert(ycbcr_coefficients == YCBCR_REC_601);
+               param->vui.i_colmatrix = 6;  // BT.601/SMPTE 170M.
+       }
 }