]> git.sesse.net Git - nageru/blobdiff - nageru/ffmpeg_capture.cpp
Suppress repeated chroma location warnings from FFmpeg.
[nageru] / nageru / ffmpeg_capture.cpp
index b9eb982008f166fbc7cab97b0e3acc990ee19340..c6824069aedc8786954531e3fb1f8f62deb52391 100644 (file)
@@ -27,6 +27,10 @@ extern "C" {
 #include <utility>
 #include <vector>
 
+#include <Eigen/Core>
+#include <Eigen/LU>
+#include <movit/colorspace_conversion_effect.h>
+
 #include "bmusb/bmusb.h"
 #include "shared/ffmpeg_raii.h"
 #include "ffmpeg_util.h"
@@ -35,12 +39,17 @@ extern "C" {
 #include "ref_counted_frame.h"
 #include "shared/timebase.h"
 
+#ifdef HAVE_SRT
+#include <srt/srt.h>
+#endif
+
 #define FRAME_SIZE (8 << 20)  // 8 MB.
 
 using namespace std;
 using namespace std::chrono;
 using namespace bmusb;
 using namespace movit;
+using namespace Eigen;
 
 namespace {
 
@@ -134,7 +143,7 @@ AVPixelFormat decide_dst_format(AVPixelFormat src_format, bmusb::PixelFormat dst
        return av_get_pix_fmt(best_format);
 }
 
-YCbCrFormat decode_ycbcr_format(const AVPixFmtDescriptor *desc, const AVFrame *frame, bool is_mjpeg)
+YCbCrFormat decode_ycbcr_format(const AVPixFmtDescriptor *desc, const AVFrame *frame, bool is_mjpeg, AVColorSpace *last_colorspace, AVChromaLocation *last_chroma_location)
 {
        YCbCrFormat format;
        AVColorSpace colorspace = frame->colorspace;
@@ -154,11 +163,14 @@ YCbCrFormat decode_ycbcr_format(const AVPixFmtDescriptor *desc, const AVFrame *f
                format.luma_coefficients = (frame->height >= 720 ? YCBCR_REC_709 : YCBCR_REC_601);
                break;
        default:
-               fprintf(stderr, "Unknown Y'CbCr coefficient enum %d from FFmpeg; choosing Rec. 709.\n",
-                       colorspace);
+               if (colorspace != *last_colorspace) {
+                       fprintf(stderr, "Unknown Y'CbCr coefficient enum %d from FFmpeg; choosing Rec. 709.\n",
+                               colorspace);
+               }
                format.luma_coefficients = YCBCR_REC_709;
                break;
        }
+       *last_colorspace = colorspace;
 
        format.full_range = is_full_range(desc);
        format.num_levels = 1 << desc->comp[0].depth;
@@ -191,12 +203,15 @@ YCbCrFormat decode_ycbcr_format(const AVPixFmtDescriptor *desc, const AVFrame *f
                format.cb_y_position = 1.0;
                break;
        default:
-               fprintf(stderr, "Unknown chroma location coefficient enum %d from FFmpeg; choosing center.\n",
-                       frame->chroma_location);
+               if (frame->chroma_location != *last_chroma_location) {
+                       fprintf(stderr, "Unknown chroma location coefficient enum %d from FFmpeg; choosing center.\n",
+                               frame->chroma_location);
+               }
                format.cb_x_position = 0.5;
                format.cb_y_position = 0.5;
                break;
        }
+       *last_chroma_location = frame->chroma_location;
 
        if (is_mjpeg && !format.full_range) {
                // Limited-range MJPEG is only detected by FFmpeg whenever a special
@@ -214,6 +229,32 @@ YCbCrFormat decode_ycbcr_format(const AVPixFmtDescriptor *desc, const AVFrame *f
        return format;
 }
 
+RGBTriplet get_neutral_color(AVDictionary *metadata)
+{
+       if (metadata == nullptr) {
+               return RGBTriplet(1.0f, 1.0f, 1.0f);
+       }
+       AVDictionaryEntry *entry = av_dict_get(metadata, "WhitePoint", nullptr, 0);
+       if (entry == nullptr) {
+               return RGBTriplet(1.0f, 1.0f, 1.0f);
+       }
+
+       unsigned x_nom, x_den, y_nom, y_den;
+       if (sscanf(entry->value, " %u:%u , %u:%u", &x_nom, &x_den, &y_nom, &y_den) != 4) {
+               fprintf(stderr, "WARNING: Unable to parse white point '%s', using default white point\n", entry->value);
+               return RGBTriplet(1.0f, 1.0f, 1.0f);
+       }
+
+       double x = double(x_nom) / x_den;
+       double y = double(y_nom) / y_den;
+       double z = 1.0 - x - y;
+
+       Matrix3d rgb_to_xyz_matrix = movit::ColorspaceConversionEffect::get_xyz_matrix(COLORSPACE_sRGB);
+       Vector3d rgb = rgb_to_xyz_matrix.inverse() * Vector3d(x, y, z);
+
+       return RGBTriplet(rgb[0], rgb[1], rgb[2]);
+}
+
 }  // namespace
 
 FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned height)
@@ -226,12 +267,36 @@ FFmpegCapture::FFmpegCapture(const string &filename, unsigned width, unsigned he
        avformat_network_init();  // In case someone wants this.
 }
 
+#ifdef HAVE_SRT
+FFmpegCapture::FFmpegCapture(int srt_sock, const string &stream_id)
+       : srt_sock(srt_sock),
+         width(0),  // Don't resize; SRT streams typically have stable resolution, and should behave much like regular cards in general.
+         height(0),
+         pixel_format(bmusb::PixelFormat_8BitYCbCrPlanar),
+         video_timebase{1, 1}
+{
+       if (stream_id.empty()) {
+               description = "SRT stream";
+       } else {
+               description = stream_id;
+       }
+       play_as_fast_as_possible = true;
+       play_once = true;
+       last_frame = steady_clock::now();
+}
+#endif
+
 FFmpegCapture::~FFmpegCapture()
 {
        if (has_dequeue_callbacks) {
                dequeue_cleanup_callback();
        }
        swr_free(&resampler);
+#ifdef HAVE_SRT
+       if (srt_sock != -1) {
+               srt_close(srt_sock);
+       }
+#endif
 }
 
 void FFmpegCapture::configure_card()
@@ -273,12 +338,12 @@ std::map<uint32_t, VideoMode> FFmpegCapture::get_available_video_modes() const
        VideoMode mode;
 
        char buf[256];
-       snprintf(buf, sizeof(buf), "%ux%u", width, height);
+       snprintf(buf, sizeof(buf), "%ux%u", sws_last_width, sws_last_height);
        mode.name = buf;
        
        mode.autodetect = false;
-       mode.width = width;
-       mode.height = height;
+       mode.width = sws_last_width;
+       mode.height = sws_last_height;
        mode.frame_rate_num = 60;
        mode.frame_rate_den = 1;
        mode.interlaced = false;
@@ -299,22 +364,38 @@ void FFmpegCapture::producer_thread_func()
                        filename_copy = filename;
                }
 
-               string pathname = search_for_file(filename_copy);
+               string pathname;
+               if (srt_sock == -1) {
+                       pathname = search_for_file(filename_copy);
+               } else {
+                       pathname = description;
+               }
                if (pathname.empty()) {
-                       fprintf(stderr, "%s not found, sleeping one second and trying again...\n", filename_copy.c_str());
                        send_disconnected_frame();
+                       if (play_once) {
+                               break;
+                       }
                        producer_thread_should_quit.sleep_for(seconds(1));
+                       fprintf(stderr, "%s not found, sleeping one second and trying again...\n", filename_copy.c_str());
                        continue;
                }
                should_interrupt = false;
                if (!play_video(pathname)) {
                        // Error.
-                       fprintf(stderr, "Error when playing %s, sleeping one second and trying again...\n", pathname.c_str());
                        send_disconnected_frame();
+                       if (play_once) {
+                               break;
+                       }
+                       fprintf(stderr, "Error when playing %s, sleeping one second and trying again...\n", pathname.c_str());
                        producer_thread_should_quit.sleep_for(seconds(1));
                        continue;
                }
 
+               if (play_once) {
+                       send_disconnected_frame();
+                       break;
+               }
+
                // Probably just EOF, will exit the loop above on next test.
        }
 
@@ -328,19 +409,21 @@ void FFmpegCapture::send_disconnected_frame()
 {
        // Send an empty frame to signal that we have no signal anymore.
        FrameAllocator::Frame video_frame = video_frame_allocator->alloc_frame();
+       size_t frame_width = width == 0 ? global_flags.width : width;
+       size_t frame_height = height == 0 ? global_flags.height : height;
        if (video_frame.data) {
                VideoFormat video_format;
-               video_format.width = width;
-               video_format.height = height;
+               video_format.width = frame_width;
+               video_format.height = frame_height;
                video_format.frame_rate_nom = 60;
                video_format.frame_rate_den = 1;
                video_format.is_connected = false;
                if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
-                       video_format.stride = width * 4;
-                       video_frame.len = width * height * 4;
+                       video_format.stride = frame_width * 4;
+                       video_frame.len = frame_width * frame_height * 4;
                        memset(video_frame.data, 0, video_frame.len);
                } else {
-                       video_format.stride = width;
+                       video_format.stride = frame_width;
                        current_frame_ycbcr_format.luma_coefficients = YCBCR_REC_709;
                        current_frame_ycbcr_format.full_range = true;
                        current_frame_ycbcr_format.num_levels = 256;
@@ -350,9 +433,9 @@ void FFmpegCapture::send_disconnected_frame()
                        current_frame_ycbcr_format.cb_y_position = 0.0f;
                        current_frame_ycbcr_format.cr_x_position = 0.0f;
                        current_frame_ycbcr_format.cr_y_position = 0.0f;
-                       video_frame.len = width * height * 2;
-                       memset(video_frame.data, 0, width * height);
-                       memset(video_frame.data + width * height, 128, width * height);  // Valid for both NV12 and planar.
+                       video_frame.len = frame_width * frame_height * 2;
+                       memset(video_frame.data, 0, frame_width * frame_height);
+                       memset(video_frame.data + frame_width * frame_height, 128, frame_width * frame_height);  // Valid for both NV12 and planar.
                }
 
                frame_callback(-1, AVRational{1, TIMEBASE}, -1, AVRational{1, TIMEBASE}, timecode++,
@@ -360,6 +443,13 @@ void FFmpegCapture::send_disconnected_frame()
                        FrameAllocator::Frame(), /*audio_offset=*/0, AudioFormat());
                last_frame_was_connected = false;
        }
+
+       if (play_once) {
+               disconnected = true;
+               if (card_disconnected_callback != nullptr) {
+                       card_disconnected_callback();
+               }
+       }
 }
 
 bool FFmpegCapture::play_video(const string &pathname)
@@ -375,8 +465,26 @@ bool FFmpegCapture::play_video(const string &pathname)
        } else {
                last_modified = buf.st_mtim;
        }
-
-       auto format_ctx = avformat_open_input_unique(pathname.c_str(), nullptr, nullptr, AVIOInterruptCB{ &FFmpegCapture::interrupt_cb_thunk, this });
+       last_colorspace = static_cast<AVColorSpace>(-1);
+       last_chroma_location = static_cast<AVChromaLocation>(-1);
+
+       AVFormatContextWithCloser format_ctx;
+       if (srt_sock == -1) {
+               // Regular file.
+               format_ctx = avformat_open_input_unique(pathname.c_str(), /*fmt=*/nullptr,
+                       /*options=*/nullptr,
+                       AVIOInterruptCB{ &FFmpegCapture::interrupt_cb_thunk, this });
+       } else {
+#ifdef HAVE_SRT
+               // SRT socket, already opened.
+               AVInputFormat *mpegts_fmt = av_find_input_format("mpegts");
+               format_ctx = avformat_open_input_unique(&FFmpegCapture::read_srt_thunk, this,
+                       mpegts_fmt, /*options=*/nullptr,
+                       AVIOInterruptCB{ &FFmpegCapture::interrupt_cb_thunk, this });
+#else
+               assert(false);
+#endif
+       }
        if (format_ctx == nullptr) {
                fprintf(stderr, "%s: Error opening file\n", pathname.c_str());
                return false;
@@ -494,6 +602,22 @@ bool FFmpegCapture::play_video(const string &pathname)
                }
 
                VideoFormat video_format = construct_video_format(frame.get(), video_timebase);
+               if (video_format.frame_rate_nom == 0 || video_format.frame_rate_den == 0) {
+                       // Invalid frame rate; try constructing it from the previous frame length.
+                       // (This is especially important if we are the master card, for SRT,
+                       // since it affects audio. Not all senders have good timebases
+                       // (e.g., Larix rounds first to timebase 1000 and then multiplies by
+                       // 90 from there, it seems), but it's much better to have an oscillating
+                       // value than just locking at 60.
+                       if (last_pts != 0 && frame->pts > last_pts) {
+                               int64_t pts_diff = frame->pts - last_pts;
+                               video_format.frame_rate_nom = video_timebase.den;
+                               video_format.frame_rate_den = video_timebase.num * pts_diff;
+                       } else {
+                               video_format.frame_rate_nom = 60;
+                               video_format.frame_rate_den = 1;
+                       }
+               }
                UniqueFrame video_frame = make_video_frame(frame.get(), pathname, &error);
                if (error) {
                        return false;
@@ -575,6 +699,7 @@ bool FFmpegCapture::play_video(const string &pathname)
                                        // audio discontinuity.)
                                        timecode += MAX_FPS * 2 + 1;
                                }
+                               last_neutral_color = get_neutral_color(frame->metadata);
                                frame_callback(frame->pts, video_timebase, audio_pts, audio_timebase, timecode++,
                                        video_frame.get_and_release(), 0, video_format,
                                        audio_frame.get_and_release(), 0, audio_format);
@@ -824,23 +949,18 @@ void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::
 VideoFormat FFmpegCapture::construct_video_format(const AVFrame *frame, AVRational video_timebase)
 {
        VideoFormat video_format;
-       video_format.width = width;
-       video_format.height = height;
+       video_format.width = frame_width(frame);
+       video_format.height = frame_height(frame);
        if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
-               video_format.stride = width * 4;
+               video_format.stride = frame_width(frame) * 4;
        } else if (pixel_format == FFmpegCapture::PixelFormat_NV12) {
-               video_format.stride = width;
+               video_format.stride = frame_width(frame);
        } else {
                assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar);
-               video_format.stride = width;
+               video_format.stride = frame_width(frame);
        }
        video_format.frame_rate_nom = video_timebase.den;
        video_format.frame_rate_den = frame->pkt_duration * video_timebase.num;
-       if (video_format.frame_rate_nom == 0 || video_format.frame_rate_den == 0) {
-               // Invalid frame rate.
-               video_format.frame_rate_nom = 60;
-               video_format.frame_rate_den = 1;
-       }
        video_format.has_signal = true;
        video_format.is_connected = true;
        return video_format;
@@ -862,7 +982,7 @@ UniqueFrame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &
                sws_dst_format = decide_dst_format(AVPixelFormat(frame->format), pixel_format);
                sws_ctx.reset(
                        sws_getContext(frame->width, frame->height, AVPixelFormat(frame->format),
-                               width, height, sws_dst_format,
+                               frame_width(frame), frame_height(frame), sws_dst_format,
                                SWS_BICUBIC, nullptr, nullptr, nullptr));
                sws_last_width = frame->width;
                sws_last_height = frame->height;
@@ -878,50 +998,81 @@ UniqueFrame FFmpegCapture::make_video_frame(const AVFrame *frame, const string &
        int linesizes[4] = { 0, 0, 0, 0 };
        if (pixel_format == bmusb::PixelFormat_8BitBGRA) {
                pic_data[0] = video_frame->data;
-               linesizes[0] = width * 4;
-               video_frame->len = (width * 4) * height;
+               linesizes[0] = frame_width(frame) * 4;
+               video_frame->len = (frame_width(frame) * 4) * frame_height(frame);
        } else if (pixel_format == PixelFormat_NV12) {
                pic_data[0] = video_frame->data;
-               linesizes[0] = width;
+               linesizes[0] = frame_width(frame);
 
-               pic_data[1] = pic_data[0] + width * height;
-               linesizes[1] = width;
+               pic_data[1] = pic_data[0] + frame_width(frame) * frame_height(frame);
+               linesizes[1] = frame_width(frame);
 
-               video_frame->len = (width * 2) * height;
+               video_frame->len = (frame_width(frame) * 2) * frame_height(frame);
 
                const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format);
-               current_frame_ycbcr_format = decode_ycbcr_format(desc, frame, is_mjpeg);
+               current_frame_ycbcr_format = decode_ycbcr_format(desc, frame, is_mjpeg, &last_colorspace, &last_chroma_location);
        } else {
                assert(pixel_format == bmusb::PixelFormat_8BitYCbCrPlanar);
                const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(sws_dst_format);
 
-               int chroma_width = AV_CEIL_RSHIFT(int(width), desc->log2_chroma_w);
-               int chroma_height = AV_CEIL_RSHIFT(int(height), desc->log2_chroma_h);
+               int chroma_width = AV_CEIL_RSHIFT(int(frame_width(frame)), desc->log2_chroma_w);
+               int chroma_height = AV_CEIL_RSHIFT(int(frame_height(frame)), desc->log2_chroma_h);
 
                pic_data[0] = video_frame->data;
-               linesizes[0] = width;
+               linesizes[0] = frame_width(frame);
 
-               pic_data[1] = pic_data[0] + width * height;
+               pic_data[1] = pic_data[0] + frame_width(frame) * frame_height(frame);
                linesizes[1] = chroma_width;
 
                pic_data[2] = pic_data[1] + chroma_width * chroma_height;
                linesizes[2] = chroma_width;
 
-               video_frame->len = width * height + 2 * chroma_width * chroma_height;
+               video_frame->len = frame_width(frame) * frame_height(frame) + 2 * chroma_width * chroma_height;
 
-               current_frame_ycbcr_format = decode_ycbcr_format(desc, frame, is_mjpeg);
+               current_frame_ycbcr_format = decode_ycbcr_format(desc, frame, is_mjpeg, &last_colorspace, &last_chroma_location);
        }
        sws_scale(sws_ctx.get(), frame->data, frame->linesize, 0, frame->height, pic_data, linesizes);
 
        return video_frame;
 }
 
-int FFmpegCapture::interrupt_cb_thunk(void *unique)
+int FFmpegCapture::interrupt_cb_thunk(void *opaque)
 {
-       return reinterpret_cast<FFmpegCapture *>(unique)->interrupt_cb();
+       return reinterpret_cast<FFmpegCapture *>(opaque)->interrupt_cb();
 }
 
 int FFmpegCapture::interrupt_cb()
 {
        return should_interrupt.load();
 }
+
+unsigned FFmpegCapture::frame_width(const AVFrame *frame) const
+{
+       if (width == 0) {
+               return frame->width;
+       } else {
+               return width;
+       }
+}
+
+unsigned FFmpegCapture::frame_height(const AVFrame *frame) const
+{
+       if (height == 0) {
+               return frame->height;
+       } else {
+               return width;
+       }
+}
+
+#ifdef HAVE_SRT
+int FFmpegCapture::read_srt_thunk(void *opaque, uint8_t *buf, int buf_size)
+{
+       return reinterpret_cast<FFmpegCapture *>(opaque)->read_srt(buf, buf_size);
+}
+
+int FFmpegCapture::read_srt(uint8_t *buf, int buf_size)
+{
+       SRT_MSGCTRL mc = srt_msgctrl_default;
+       return srt_recvmsg2(srt_sock, reinterpret_cast<char *>(buf), buf_size, &mc);
+}
+#endif