Bump version: 6.56 → 6.57

cpu/encoder: fixed luma plane size calculation

.bumpversion.cfg

@@ -1,7 +1,7 @@
 [bumpversion]
 commit = True
 tag = True
-current_version = 6.56
+current_version = 6.57
 parse = (?P<major>\d+)\.(?P<minor>\d+)
 serialize = 
 	{major}.{minor}

Makefile

@@ -72,7 +72,7 @@ apps:
 
 python:
 	$(MAKE) -C python
-	$(ECHO) ln -sf python/root/usr/lib/python*/site-packages/*.so .
+	$(ECHO) ln -sf python/build/lib.linux-*/*.so -t .
 
 
 janus:

janus/src/rtp.c

@@ -47,9 +47,9 @@ void us_rtp_assign(us_rtp_s *rtp, uint payload, bool video) {
 	rtp->ssrc = us_triple_u32(us_get_now_monotonic_u64());
 }
 
-void us_rtp_write_header(us_rtp_s *rtp, u32 pts, bool marked) {
+void us_rtp_write_header(us_rtp_s *rtp, u32 pts, bool last_header) {
 	u32 word0 = 0x80000000;
-	if (marked) {
+	if (last_header) {
 		word0 |= 1 << 23;
 	}
 	word0 |= (rtp->payload & 0x7F) << 16;

janus/src/rtp.h

@@ -66,4 +66,4 @@ us_rtp_s *us_rtp_init(void);
 void us_rtp_destroy(us_rtp_s *rtp);
 
 void us_rtp_assign(us_rtp_s *rtp, uint payload, bool video);
-void us_rtp_write_header(us_rtp_s *rtp, u32 pts, bool marked);
+void us_rtp_write_header(us_rtp_s *rtp, u32 pts, bool last_header);

janus/src/rtpv.c

@@ -26,8 +26,6 @@
 #include "rtpv.h"
 
 #include <stdlib.h>
-#include <inttypes.h>
-#include <assert.h>
 
 #include <linux/videodev2.h>
 
@@ -38,9 +36,12 @@
 #include "rtp.h"
 
 
-void _rtpv_process_nalu(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool marked);
-
 static sz _find_annexb(const u8 *data, uz size);
+static bool _is_nalu_small(uz size);
+
+static void _rtpv_process_nalu(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool first_nalu, bool last_nalu);
+static void _rtpv_process_nalu_small(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool first_nalu, bool last_nalu);
+static void _rtpv_process_nalu_big(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool first_nalu, bool last_nalu);
 
 
 us_rtpv_s *us_rtpv_init(us_rtp_callback_f callback) {
@@ -60,77 +61,109 @@ void us_rtpv_destroy(us_rtpv_s *rtpv) {
 #define _PRE 3 // Annex B prefix length
 
 void us_rtpv_wrap(us_rtpv_s *rtpv, const us_frame_s *frame, bool zero_playout_delay) {
-	// There is a complicated logic here but everything works as it should:
-	//   - https://github.com/pikvm/ustreamer/issues/115#issuecomment-893071775
+	US_A(frame->format == V4L2_PIX_FMT_H264);
 
-	assert(frame->format == V4L2_PIX_FMT_H264);
+	if (frame->used <= _PRE) {
+		return;
+	}
 
-	rtpv->rtp->first_of_frame = true;
-	rtpv->rtp->last_of_frame = false;
 	rtpv->rtp->zero_playout_delay = zero_playout_delay;
 	rtpv->rtp->grab_ntp_ts = us_get_now_ntp() - us_ld_to_ntp(us_get_now_monotonic() - frame->grab_begin_ts);
 
 	const u32 pts = us_get_now_monotonic_u64() * 9 / 100; // PTS units are in 90 kHz
-	sz last_offset = -_PRE;
 
-	while (true) { // Find and iterate by nalus
-		const uz next_start = last_offset + _PRE;
-		sz offset = _find_annexb(frame->data + next_start, frame->used - next_start);
-		if (offset < 0) {
-			break;
-		}
-		offset += next_start;
+	// Bytestream can be like:
+	//   [00-]00-00-01 NALU [00-]00-00-01 NALU
+	// ... with optional 00 prefix for _PRE.
+	// We need to iterate by NALUs only and throw away separators.
 
-		if (last_offset >= 0) {
-			const u8 *const data = frame->data + last_offset + _PRE;
-			uz size = offset - last_offset - _PRE;
-			if (data[size - 1] == 0) { // Check for extra 00
+	const sz offset_to_first = _find_annexb(frame->data, frame->used);
+	if (offset_to_first < 0) {
+		return;
+	}
+	uz begin = offset_to_first + _PRE;
+
+	bool first_nalu = true;
+	while (begin < frame->used) { // Find and iterate by NALUs
+		const u8 *const data = frame->data + begin;
+		const sz offset_to_next = _find_annexb(data, frame->used - begin);
+
+		if (offset_to_next >= 0) { // Process NALUs between prefixes
+			uz size = offset_to_next;
+			if (size > 1 && data[size - 1] == 0) { // Check for extra trailing zero
 				--size;
 			}
-			_rtpv_process_nalu(rtpv, data, size, pts, false);
+			if (size > 1) { // Skip too short NALUs
+				_rtpv_process_nalu(rtpv, data, size, pts, first_nalu, false);
+				first_nalu = false;
+			}
+			begin += offset_to_next + _PRE;
+
+		} else { // Process the tail
+			const uz size = frame->used - begin;
+			if (size > 1) {
+				_rtpv_process_nalu(rtpv, data, size, pts, first_nalu, true);
+				first_nalu = false;
+			}
+			break;
 		}
-
-		last_offset = offset;
-	}
-
-	if (last_offset >= 0) {
-		const u8 *const data = frame->data + last_offset + _PRE;
-		uz size = frame->used - last_offset - _PRE;
-		_rtpv_process_nalu(rtpv, data, size, pts, true);
 	}
 }
 
-void _rtpv_process_nalu(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool marked) {
+static sz _find_annexb(const u8 *data, uz size) {
+	// Parses buffer for 00 00 01 start codes
+	if (size >= _PRE) {
+		for (uz i = 0; i <= size - _PRE; ++i) {
+			if (
+				data[i] == 0
+				&& data[i + 1] == 0
+				&& data[i + 2] == 1
+			) {
+				return i;
+			}
+		}
+	}
+	return -1;
+}
+
+#undef _PRE
+
+static bool _is_nalu_small(uz size) {
+	return (size + US_RTP_HEADER_SIZE <= US_RTP_TOTAL_SIZE);
+}
+
+static void _rtpv_process_nalu(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool first_nalu, bool last_nalu) {
+	US_A(size > 1);
+	if (_is_nalu_small(size)) {
+		_rtpv_process_nalu_small(rtpv, data, size, pts, first_nalu, last_nalu);
+	} else {
+		_rtpv_process_nalu_big(rtpv, data, size, pts, first_nalu, last_nalu);
+	}
+}
+
+static void _rtpv_process_nalu_small(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool first_nalu, bool last_nalu) {
+	US_A(size > 1);
+	US_A(_is_nalu_small(size));
+
+	us_rtp_s *rtp = rtpv->rtp;
+	u8 *dg = rtp->datagram;
+
+	us_rtp_write_header(rtp, pts, last_nalu);
+	memcpy(dg + US_RTP_HEADER_SIZE, data, size);
+	rtp->used = size + US_RTP_HEADER_SIZE;
+	rtp->first_of_frame = first_nalu;
+	rtp->last_of_frame = last_nalu;
+	rtpv->callback(rtp);
+}
+
+static void _rtpv_process_nalu_big(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool first_nalu, bool last_nalu) {
+	US_A(size > 1);
+	US_A(!_is_nalu_small(size));
+
 	const uint ref_idc = (data[0] >> 5) & 3;
 	const uint type = data[0] & 0x1F;
-	u8 *dg = rtpv->rtp->datagram;
-
-	// Set *_of_frame flags only for non-SPS/PPS packages
-/*
-#	define CALL_FOR_SERVICE { \
-			const bool m_fof = rtpv->rtp->first_of_frame; \
-			const bool m_lof = rtpv->rtp->last_of_frame; \
-			rtpv->rtp->first_of_frame = false; \
-			rtpv->rtp->last_of_frame = false; \
-			rtpv->callback(rtpv->rtp); \
-			rtpv->rtp->first_of_frame = m_fof; \
-			rtpv->rtp->last_of_frame = m_lof; \
-		}
-*/
-
-	if (size + US_RTP_HEADER_SIZE <= US_RTP_TOTAL_SIZE) {
-		us_rtp_write_header(rtpv->rtp, pts, marked);
-		memcpy(dg + US_RTP_HEADER_SIZE, data, size);
-		rtpv->rtp->used = size + US_RTP_HEADER_SIZE;
-//		if (type == 7 || type == 8) {
-//			CALL_FOR_SERVICE;
-//		} else {*/
-			rtpv->rtp->last_of_frame = true;
-			rtpv->callback(rtpv->rtp);
-			rtpv->rtp->first_of_frame = false;
-//		}
-		return;
-	}
+	us_rtp_s *rtp = rtpv->rtp;
+	u8 *dg = rtp->datagram;
 
 	const uz fu_overhead = US_RTP_HEADER_SIZE + 2; // FU-A overhead
 
@@ -145,7 +178,7 @@ void _rtpv_process_nalu(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool
 			frag_size = remaining;
 		}
 
-		us_rtp_write_header(rtpv->rtp, pts, (marked && last));
+		us_rtp_write_header(rtp, pts, (last_nalu && last));
 
 		dg[US_RTP_HEADER_SIZE] = 28 | (ref_idc << 5);
 
@@ -159,33 +192,13 @@ void _rtpv_process_nalu(us_rtpv_s *rtpv, const u8 *data, uz size, u32 pts, bool
 		dg[US_RTP_HEADER_SIZE + 1] = fu;
 
 		memcpy(dg + fu_overhead, src, frag_size);
-		rtpv->rtp->used = fu_overhead + frag_size;
-//		if (type == 7 || type == 8) {
-//			CALL_FOR_SERVICE;
-//		} else {
-			rtpv->rtp->last_of_frame = last;
-			rtpv->callback(rtpv->rtp);
-			rtpv->rtp->first_of_frame = false;
-//		}
+		rtp->used = fu_overhead + frag_size;
+		rtp->first_of_frame = (first_nalu && first);
+		rtp->last_of_frame = (last_nalu && last);
+		rtpv->callback(rtp);
 
 		src += frag_size;
 		remaining -= frag_size;
 		first = false;
 	}
-
-#	undef CALL_FOR_SERVICE
 }
-
-static sz _find_annexb(const u8 *data, uz size) {
-	// Parses buffer for 00 00 01 start codes
-	if (size >= _PRE) {
-		for (uz index = 0; index <= size - _PRE; ++index) {
-			if (data[index] == 0 && data[index + 1] == 0 && data[index + 2] == 1) {
-				return index;
-			}
-		}
-	}
-	return -1;
-}
-
-#undef _PRE

man/ustreamer-dump.1

@@ -1,6 +1,6 @@
 .\" Manpage for ustreamer-dump.
 .\" Open an issue or pull request to https://github.com/pikvm/ustreamer to correct errors or typos
-.TH USTREAMER-DUMP 1 "version 6.56" "January 2021"
+.TH USTREAMER-DUMP 1 "version 6.57" "January 2021"
 
 .SH NAME
 ustreamer-dump \- Dump uStreamer's memory sink to file

man/ustreamer.1

@@ -1,6 +1,6 @@
 .\" Manpage for ustreamer.
 .\" Open an issue or pull request to https://github.com/pikvm/ustreamer to correct errors or typos
-.TH USTREAMER 1 "version 6.56" "November 2020"
+.TH USTREAMER 1 "version 6.57" "November 2020"
 
 .SH NAME
 ustreamer \- stream MJPEG video from any V4L2 device to the network

pkg/arch/PKGBUILD

@@ -3,7 +3,7 @@
 
 
 pkgname=ustreamer
-pkgver=6.56
+pkgver=6.57
 pkgrel=1
 pkgdesc="Lightweight and fast MJPEG-HTTP streamer"
 url="https://github.com/pikvm/ustreamer"

pkg/openwrt/Makefile

@@ -9,7 +9,7 @@
 include $(TOPDIR)/rules.mk
 
 PKG_NAME:=ustreamer
-PKG_VERSION:=6.56
+PKG_VERSION:=6.57
 PKG_RELEASE:=1
 PKG_MAINTAINER:=Maxim Devaev <mdevaev@gmail.com>
 

python/Makefile

@@ -9,12 +9,11 @@ all: root
 root: $(shell find src -type f,l) setup.py
 	$(info == PY_BUILD ustreamer-*.so)
 	rm -rf root
-	$(ECHO) $(PY) -m build --skip-dependency-check --no-isolation
-	$(ECHO) $(PY) -m pip install dist/*.whl --ignore-installed --root=./root
+	$(ECHO) $(PY) -m build --wheel --skip-dependency-check --no-isolation
 
 
 install:
-	$(PY) -m pip install dist/*.whl --ignore-installed --prefix=$(PREFIX) --root=$(if $(R_DESTDIR),$(R_DESTDIR),/)
+	$(PY) -m installer --prefix=$(PREFIX) $(if $(R_DESTDIR),--destdir=$(R_DESTDIR),) dist/*.whl
 
 
 clean:

python/setup.py

@@ -34,7 +34,7 @@ def main() -> None:
     flags = _find_flags()
     setup(
         name="ustreamer",
-        version="6.56",
+        version="6.57",
         description="uStreamer tools",
         author="Maxim Devaev",
         author_email="mdevaev@gmail.com",

src/libs/const.h

@@ -26,7 +26,7 @@
 
 
 #define US_VERSION_MAJOR 6
-#define US_VERSION_MINOR 56
+#define US_VERSION_MINOR 57
 
 #define US_MAKE_VERSION2(_major, _minor) #_major "." #_minor
 #define US_MAKE_VERSION1(_major, _minor) US_MAKE_VERSION2(_major, _minor)

src/ustreamer/encoders/cpu/encoder.c

@@ -216,15 +216,11 @@ static void _jpeg_write_scanlines_yuv_planar(struct jpeg_compress_struct *jpeg,
 	US_CALLOC(line_buf, frame->width * 3);
 
 	const uint padding = us_frame_get_padding(frame);
-	const uint image_size = frame->width * frame->height;
-	const uint chroma_array_size = (frame->used - image_size) / 2;
-	const uint chroma_matrix_order = (image_size / chroma_array_size) == 16 ? 4 : 2;
+	const uint luma_array_size = (frame->width + padding) * frame->height;
+	const uint chroma_array_size = (frame->used - luma_array_size) >> 1;
 	const u8 *data = frame->data;
-	const u8 *chroma1_data = frame->data + image_size;
-	const u8 *chroma2_data = frame->data + image_size + chroma_array_size;
-
-	//US_LOG_DEBUG("Planar data: Image Size %u, Chroma Array Size %u, Chroma Matrix Order %u",
-	//	image_size, chroma_array_size, chroma_matrix_order);
+	const u8 *chroma1_data = frame->data + luma_array_size;
+	const u8 *chroma2_data = frame->data + luma_array_size + chroma_array_size;
 
 	while (jpeg->next_scanline < frame->height) {
 		u8 *ptr = line_buf;
@@ -234,7 +230,7 @@ static void _jpeg_write_scanlines_yuv_planar(struct jpeg_compress_struct *jpeg,
 			u8 y = data[x];
 			u8 u;
 			u8 v;
-			uint chroma_position = x / chroma_matrix_order;
+			uint chroma_position = x >> 1;
 
 			switch (frame->format) {
 				case V4L2_PIX_FMT_YUV420:
@@ -257,9 +253,9 @@ static void _jpeg_write_scanlines_yuv_planar(struct jpeg_compress_struct *jpeg,
 
 		data += frame->width + padding;
 
-		if (jpeg->next_scanline > 0 && jpeg->next_scanline % chroma_matrix_order == 0) {
-			chroma1_data += (frame->width + padding) / chroma_matrix_order;
-			chroma2_data += (frame->width + padding) / chroma_matrix_order;
+		if (jpeg->next_scanline > 0 && jpeg->next_scanline % 2 == 0) {
+			chroma1_data += (frame->width + padding) >> 1;
+			chroma2_data += (frame->width + padding) >> 1;
 		}
 
 		JSAMPROW scanlines[1] = {line_buf};