X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=h264encode.cpp;h=a478c2a1a2459a2fa24d3ed06c34a598ea1bf8d1;hb=c957e3daa14e615a3aa5d6bd4d15ddf409812fd3;hp=d4fc0ad859498dcfba828f51d91ff0ec4d4991e2;hpb=e18d9bad93d5bf766d52d0cb66db4c3d3f8a711b;p=nageru diff --git a/h264encode.cpp b/h264encode.cpp index d4fc0ad..a478c2a 100644 --- a/h264encode.cpp +++ b/h264encode.cpp @@ -1,28 +1,50 @@ //#include "sysdeps.h" +#include "h264encode.h" + +#include +#include +#include +#include +#include +#include +extern "C" { +#include +#include +#include +#include +#include +#include +} +#include #include -#include #include -#include -#include -#include -#include -#include -#include +#include #include -#include -#include -#include -#include #include -#include -#include +#include #include -#include -#include +#include +#include +#include +#include +#include +#include +#include #include #include -#include -#include "h264encode.h" +#include +#include +#include + +#include "context.h" +#include "defs.h" +#include "httpd.h" +#include "timebase.h" + +using namespace std; + +class QOpenGLContext; +class QSurface; #define CHECK_VASTATUS(va_status, func) \ if (va_status != VA_STATUS_SUCCESS) { \ @@ -30,6 +52,8 @@ exit(1); \ } +#define BUFFER_OFFSET(i) ((char *)NULL + (i)) + //#include "loadsurface.h" #define NAL_REF_IDC_NONE 0 @@ -59,62 +83,13 @@ #define PROFILE_IDC_HIGH 100 #define BITSTREAM_ALLOCATE_STEPPING 4096 - #define SURFACE_NUM 16 /* 16 surfaces for source YUV */ -static VADisplay va_dpy; -static VAProfile h264_profile = (VAProfile)~0; -static VAConfigAttrib config_attrib[VAConfigAttribTypeMax]; -static int config_attrib_num = 0, enc_packed_header_idx; - -struct GLSurface { - VASurfaceID src_surface, ref_surface; - VABufferID coded_buf; - - VAImage surface_image; - GLuint y_tex, cbcr_tex; - EGLImage y_egl_image, cbcr_egl_image; -}; -GLSurface gl_surfaces[SURFACE_NUM]; - -static VAConfigID config_id; -static VAContextID context_id; -static VAEncSequenceParameterBufferH264 seq_param; -static VAEncPictureParameterBufferH264 pic_param; -static VAEncSliceParameterBufferH264 slice_param; -static VAPictureH264 CurrentCurrPic; -static VAPictureH264 ReferenceFrames[16], RefPicList0_P[32], RefPicList0_B[32], RefPicList1_B[32]; - -static unsigned int MaxFrameNum = (2<<16); -static unsigned int MaxPicOrderCntLsb = (2<<8); -static unsigned int Log2MaxFrameNum = 16; -static unsigned int Log2MaxPicOrderCntLsb = 8; - -static unsigned int num_ref_frames = 2; -static unsigned int numShortTerm = 0; -static int constraint_set_flag = 0; -static int h264_packedheader = 0; /* support pack header? */ -static int h264_maxref = (1<<16|1); -static int h264_entropy_mode = 1; /* cabac */ - -static char *coded_fn = NULL; -static FILE *coded_fp = NULL; - -static int frame_width = 176; -static int frame_height = 144; -static int frame_width_mbaligned; -static int frame_height_mbaligned; -static int frame_rate = 60; -static unsigned int frame_bitrate = 0; -static unsigned int frame_slices = 1; -static double frame_size = 0; -static int initial_qp = 15; -//static int initial_qp = 28; -static int minimal_qp = 0; -static int intra_period = 30; -static int intra_idr_period = 60; -static int ip_period = 1; -static int rc_mode = -1; -static int rc_default_modes[] = { + +static constexpr unsigned int MaxFrameNum = (2<<16); +static constexpr unsigned int MaxPicOrderCntLsb = (2<<8); +static constexpr unsigned int Log2MaxFrameNum = 16; +static constexpr unsigned int Log2MaxPicOrderCntLsb = 8; +static constexpr int rc_default_modes[] = { // Priority list of modes. VA_RC_VBR, VA_RC_CQP, VA_RC_VBR_CONSTRAINED, @@ -122,14 +97,6 @@ static int rc_default_modes[] = { VA_RC_VCM, VA_RC_NONE, }; -static unsigned long long current_frame_encoding = 0; -static unsigned long long current_frame_display = 0; -static unsigned long long current_IDR_display = 0; -static unsigned int current_frame_num = 0; -static int current_frame_type; - -static int misc_priv_type = 0; -static int misc_priv_value = 0; /* thread to save coded data */ #define SRC_SURFACE_FREE 0 @@ -144,6 +111,154 @@ typedef struct __bitstream bitstream; using namespace std; +class H264EncoderImpl { +public: + H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd); + ~H264EncoderImpl(); + void add_audio(int64_t pts, vector audio); + bool begin_frame(GLuint *y_tex, GLuint *cbcr_tex); + void end_frame(RefCountedGLsync fence, int64_t pts, const vector &input_frames); + void shutdown(); + +private: + struct storage_task { + unsigned long long display_order; + int frame_type; + vector audio; + int64_t pts, dts; + }; + struct PendingFrame { + RefCountedGLsync fence; + vector input_frames; + int64_t pts; + }; + + void encode_thread_func(); + void encode_remaining_frames_as_p(int encoding_frame_num, int gop_start_display_frame_num, int64_t last_dts); + void encode_frame(PendingFrame frame, int encoding_frame_num, int display_frame_num, int gop_start_display_frame_num, + int frame_type, int64_t pts, int64_t dts); + void storage_task_thread(); + void storage_task_enqueue(storage_task task); + void save_codeddata(storage_task task); + int render_packedsequence(); + int render_packedpicture(); + void render_packedslice(); + int render_sequence(); + int render_picture(int frame_type, int display_frame_num, int gop_start_display_frame_num); + void sps_rbsp(bitstream *bs); + void pps_rbsp(bitstream *bs); + int build_packed_pic_buffer(unsigned char **header_buffer); + int render_slice(int encoding_frame_num, int display_frame_num, int gop_start_display_frame_num, int frame_type); + void slice_header(bitstream *bs); + int build_packed_seq_buffer(unsigned char **header_buffer); + int build_packed_slice_buffer(unsigned char **header_buffer); + int init_va(const string &va_display); + int deinit_va(); + VADisplay va_open_display(const string &va_display); + void va_close_display(VADisplay va_dpy); + int setup_encode(); + int release_encode(); + void update_ReferenceFrames(int frame_type); + int update_RefPicList(int frame_type); + + bool is_shutdown = false; + bool use_zerocopy; + int drm_fd = -1; + + thread encode_thread, storage_thread; + + mutex storage_task_queue_mutex; + condition_variable storage_task_queue_changed; + int srcsurface_status[SURFACE_NUM]; // protected by storage_task_queue_mutex + queue storage_task_queue; // protected by storage_task_queue_mutex + bool storage_thread_should_quit = false; // protected by storage_task_queue_mutex + + mutex frame_queue_mutex; + condition_variable frame_queue_nonempty; + bool encode_thread_should_quit = false; // under frame_queue_mutex + + int current_storage_frame; + + map pending_video_frames; // under frame_queue_mutex + map> pending_audio_frames; // under frame_queue_mutex + QSurface *surface; + + AVCodecContext *context_audio; + HTTPD *httpd; + + Display *x11_display = nullptr; + + // Encoder parameters + VADisplay va_dpy; + VAProfile h264_profile = (VAProfile)~0; + VAConfigAttrib config_attrib[VAConfigAttribTypeMax]; + int config_attrib_num = 0, enc_packed_header_idx; + + struct GLSurface { + VASurfaceID src_surface, ref_surface; + VABufferID coded_buf; + + VAImage surface_image; + GLuint y_tex, cbcr_tex; + + // Only if use_zerocopy == true. + EGLImage y_egl_image, cbcr_egl_image; + + // Only if use_zerocopy == false. + GLuint pbo; + uint8_t *y_ptr, *cbcr_ptr; + size_t y_offset, cbcr_offset; + }; + GLSurface gl_surfaces[SURFACE_NUM]; + + VAConfigID config_id; + VAContextID context_id; + VAEncSequenceParameterBufferH264 seq_param; + VAEncPictureParameterBufferH264 pic_param; + VAEncSliceParameterBufferH264 slice_param; + VAPictureH264 CurrentCurrPic; + VAPictureH264 ReferenceFrames[16], RefPicList0_P[32], RefPicList0_B[32], RefPicList1_B[32]; + + // Static quality settings. + static constexpr unsigned int frame_bitrate = 15000000 / 60; // Doesn't really matter; only initial_qp does. + static constexpr unsigned int num_ref_frames = 2; + static constexpr int initial_qp = 15; + static constexpr int minimal_qp = 0; + static constexpr int intra_period = 30; + static constexpr int intra_idr_period = MAX_FPS; // About a second; more at lower frame rates. Not ideal. + + // Quality settings that are meant to be static, but might be overridden + // by the profile. + int constraint_set_flag = 0; + int h264_packedheader = 0; /* support pack header? */ + int h264_maxref = (1<<16|1); + int h264_entropy_mode = 1; /* cabac */ + int ip_period = 3; + + int rc_mode = -1; + unsigned int current_frame_num = 0; + unsigned int numShortTerm = 0; + + int frame_width; + int frame_height; + int frame_width_mbaligned; + int frame_height_mbaligned; +}; + +// Supposedly vaRenderPicture() is supposed to destroy the buffer implicitly, +// but if we don't delete it here, we get leaks. The GStreamer implementation +// does the same. +static void render_picture_and_delete(VADisplay dpy, VAContextID context, VABufferID *buffers, int num_buffers) +{ + VAStatus va_status = vaRenderPicture(dpy, context, buffers, num_buffers); + CHECK_VASTATUS(va_status, "vaRenderPicture"); + + for (int i = 0; i < num_buffers; ++i) { + va_status = vaDestroyBuffer(dpy, buffers[i]); + CHECK_VASTATUS(va_status, "vaDestroyBuffer"); + } +} + static unsigned int va_swap32(unsigned int val) { @@ -191,7 +306,11 @@ bitstream_put_ui(bitstream *bs, unsigned int val, int size_in_bits) bs->buffer[pos] = (bs->buffer[pos] << size_in_bits | val); } else { size_in_bits -= bit_left; - bs->buffer[pos] = (bs->buffer[pos] << bit_left) | (val >> size_in_bits); + if (bit_left >= 32) { + bs->buffer[pos] = (val >> size_in_bits); + } else { + bs->buffer[pos] = (bs->buffer[pos] << bit_left) | (val >> size_in_bits); + } bs->buffer[pos] = va_swap32(bs->buffer[pos]); if (pos + 1 == bs->max_size_in_dword) { @@ -270,7 +389,7 @@ static void nal_header(bitstream *bs, int nal_ref_idc, int nal_unit_type) bitstream_put_ui(bs, nal_unit_type, 5); } -static void sps_rbsp(bitstream *bs) +void H264EncoderImpl::sps_rbsp(bitstream *bs) { int profile_idc = PROFILE_IDC_BASELINE; @@ -333,12 +452,22 @@ static void sps_rbsp(bitstream *bs) bitstream_put_ui(bs, 1, 1); /* vui_parameters_present_flag */ bitstream_put_ui(bs, 0, 1); /* aspect_ratio_info_present_flag */ bitstream_put_ui(bs, 0, 1); /* overscan_info_present_flag */ - bitstream_put_ui(bs, 0, 1); /* video_signal_type_present_flag */ + bitstream_put_ui(bs, 1, 1); /* video_signal_type_present_flag */ + { + bitstream_put_ui(bs, 5, 3); /* video_format (5 = Unspecified) */ + bitstream_put_ui(bs, 0, 1); /* video_full_range_flag */ + bitstream_put_ui(bs, 1, 1); /* colour_description_present_flag */ + { + bitstream_put_ui(bs, 1, 8); /* colour_primaries (1 = BT.709) */ + bitstream_put_ui(bs, 2, 8); /* transfer_characteristics (2 = unspecified, since we use sRGB) */ + bitstream_put_ui(bs, 6, 8); /* matrix_coefficients (6 = BT.601/SMPTE 170M) */ + } + } bitstream_put_ui(bs, 0, 1); /* chroma_loc_info_present_flag */ bitstream_put_ui(bs, 1, 1); /* timing_info_present_flag */ { bitstream_put_ui(bs, 1, 32); // FPS - bitstream_put_ui(bs, frame_rate * 2, 32); // FPS + bitstream_put_ui(bs, TIMEBASE * 2, 32); // FPS bitstream_put_ui(bs, 1, 1); } bitstream_put_ui(bs, 1, 1); /* nal_hrd_parameters_present_flag */ @@ -368,7 +497,7 @@ static void sps_rbsp(bitstream *bs) } -static void pps_rbsp(bitstream *bs) +void H264EncoderImpl::pps_rbsp(bitstream *bs) { bitstream_put_ue(bs, pic_param.pic_parameter_set_id); /* pic_parameter_set_id */ bitstream_put_ue(bs, pic_param.seq_parameter_set_id); /* seq_parameter_set_id */ @@ -401,7 +530,7 @@ static void pps_rbsp(bitstream *bs) rbsp_trailing_bits(bs); } -static void slice_header(bitstream *bs) +void H264EncoderImpl::slice_header(bitstream *bs) { int first_mb_in_slice = slice_param.macroblock_address; @@ -496,8 +625,7 @@ static void slice_header(bitstream *bs) } } -static int -build_packed_pic_buffer(unsigned char **header_buffer) +int H264EncoderImpl::build_packed_pic_buffer(unsigned char **header_buffer) { bitstream bs; @@ -511,8 +639,8 @@ build_packed_pic_buffer(unsigned char **header_buffer) return bs.bit_offset; } -static int -build_packed_seq_buffer(unsigned char **header_buffer) +int +H264EncoderImpl::build_packed_seq_buffer(unsigned char **header_buffer) { bitstream bs; @@ -526,7 +654,7 @@ build_packed_seq_buffer(unsigned char **header_buffer) return bs.bit_offset; } -static int build_packed_slice_buffer(unsigned char **header_buffer) +int H264EncoderImpl::build_packed_slice_buffer(unsigned char **header_buffer) { bitstream bs; int is_idr = !!pic_param.pic_fields.bits.idr_pic_flag; @@ -580,23 +708,65 @@ static int build_packed_slice_buffer(unsigned char **header_buffer) {IDR(PBB)(PBB)}. */ -/* - * Return displaying order with specified periods and encoding order - * displaying_order: displaying order - * frame_type: frame type - */ +// General pts/dts strategy: +// +// Getting pts and dts right with variable frame rate (VFR) and B-frames can be a +// bit tricky. We assume first of all that the frame rate never goes _above_ +// MAX_FPS, which gives us a frame period N. The decoder can always decode +// in at least this speed, as long at dts <= pts (the frame is not attempted +// presented before it is decoded). Furthermore, we never have longer chains of +// B-frames than a fixed constant C. (In a B-frame chain, we say that the base +// I/P-frame has order O=0, the B-frame depending on it directly has order O=1, +// etc. The last frame in the chain, which no B-frames depend on, is the “tip” +// frame, with an order O <= C.) +// +// Many strategies are possible, but we establish these rules: +// +// - Tip frames have dts = pts - (C-O)*N. +// - Non-tip frames have dts = dts_last + N. +// +// An example, with C=2 and N=10 and the data flow showed with arrows: +// +// I B P B B P +// pts: 30 40 50 60 70 80 +// ↓ ↓ ↓ +// dts: 10 30 20 60 50←40 +// | | ↑ ↑ +// `--|--' | +// `----------' +// +// To show that this works fine also with irregular spacings, let's say that +// the third frame is delayed a bit (something earlier was dropped). Now the +// situation looks like this: +// +// I B P B B P +// pts: 30 40 80 90 100 110 +// ↓ ↓ ↓ +// dts: 10 30 20 90 50←40 +// | | ↑ ↑ +// `--|--' | +// `----------' +// +// The resetting on every tip frame makes sure dts never ends up lagging a lot +// behind pts, and the subtraction of (C-O)*N makes sure pts <= dts. +// +// In the output of this function, if is >= 0, it means to reset the +// dts from the current pts minus , while if it's -1, the frame is not +// a tip frame and should be given a dts based on the previous one. #define FRAME_P 0 #define FRAME_B 1 #define FRAME_I 2 #define FRAME_IDR 7 void encoding2display_order( - unsigned long long encoding_order, int intra_period, + int encoding_order, int intra_period, int intra_idr_period, int ip_period, - unsigned long long *displaying_order, - int *frame_type) + int *displaying_order, + int *frame_type, int *pts_lag) { int encoding_order_gop = 0; + *pts_lag = 0; + if (intra_period == 1) { /* all are I/IDR frames */ *displaying_order = encoding_order; if (intra_idr_period == 0) @@ -609,34 +779,52 @@ void encoding2display_order( if (intra_period == 0) intra_idr_period = 0; - /* new sequence like - * IDR PPPPP IPPPPP - * IDR (PBB)(PBB)(IBB)(PBB) - */ - encoding_order_gop = (intra_idr_period == 0)? encoding_order: - (encoding_order % (intra_idr_period + ((ip_period == 1)?0:1))); + if (ip_period == 1) { + // No B-frames, sequence is like IDR PPPPP IPPPPP. + encoding_order_gop = (intra_idr_period == 0) ? encoding_order : (encoding_order % intra_idr_period); + *displaying_order = encoding_order; + + if (encoding_order_gop == 0) { /* the first frame */ + *frame_type = FRAME_IDR; + } else if (intra_period != 0 && /* have I frames */ + encoding_order_gop >= 2 && + (encoding_order_gop % intra_period == 0)) { + *frame_type = FRAME_I; + } else { + *frame_type = FRAME_P; + } + return; + } + + // We have B-frames. Sequence is like IDR (PBB)(PBB)(IBB)(PBB). + encoding_order_gop = (intra_idr_period == 0) ? encoding_order : (encoding_order % (intra_idr_period + 1)); + *pts_lag = -1; // Most frames are not tip frames. if (encoding_order_gop == 0) { /* the first frame */ *frame_type = FRAME_IDR; *displaying_order = encoding_order; + // IDR frames are a special case; I honestly can't find the logic behind + // why this is the right thing, but it seems to line up nicely in practice :-) + *pts_lag = TIMEBASE / MAX_FPS; } else if (((encoding_order_gop - 1) % ip_period) != 0) { /* B frames */ - *frame_type = FRAME_B; + *frame_type = FRAME_B; *displaying_order = encoding_order - 1; - } else if ((intra_period != 0) && /* have I frames */ - (encoding_order_gop >= 2) && - ((ip_period == 1 && encoding_order_gop % intra_period == 0) || /* for IDR PPPPP IPPPP */ - /* for IDR (PBB)(PBB)(IBB) */ - (ip_period >= 2 && ((encoding_order_gop - 1) / ip_period % (intra_period / ip_period)) == 0))) { - *frame_type = FRAME_I; - *displaying_order = encoding_order + ip_period - 1; + if ((encoding_order_gop % ip_period) == 0) { + *pts_lag = 0; // Last B-frame. + } + } else if (intra_period != 0 && /* have I frames */ + encoding_order_gop >= 2 && + ((encoding_order_gop - 1) / ip_period % (intra_period / ip_period)) == 0) { + *frame_type = FRAME_I; + *displaying_order = encoding_order + ip_period - 1; } else { - *frame_type = FRAME_P; - *displaying_order = encoding_order + ip_period - 1; + *frame_type = FRAME_P; + *displaying_order = encoding_order + ip_period - 1; } } -static const char *rc_to_string(int rcmode) +static const char *rc_to_string(int rc_mode) { switch (rc_mode) { case VA_RC_NONE: @@ -656,191 +844,47 @@ static const char *rc_to_string(int rcmode) } } -#if 0 -static int process_cmdline(int argc, char *argv[]) +VADisplay H264EncoderImpl::va_open_display(const string &va_display) { - char c; - const struct option long_opts[] = { - {"help", no_argument, NULL, 0 }, - {"bitrate", required_argument, NULL, 1 }, - {"minqp", required_argument, NULL, 2 }, - {"initialqp", required_argument, NULL, 3 }, - {"intra_period", required_argument, NULL, 4 }, - {"idr_period", required_argument, NULL, 5 }, - {"ip_period", required_argument, NULL, 6 }, - {"rcmode", required_argument, NULL, 7 }, - {"srcyuv", required_argument, NULL, 9 }, - {"recyuv", required_argument, NULL, 10 }, - {"fourcc", required_argument, NULL, 11 }, - {"syncmode", no_argument, NULL, 12 }, - {"enablePSNR", no_argument, NULL, 13 }, - {"prit", required_argument, NULL, 14 }, - {"priv", required_argument, NULL, 15 }, - {"framecount", required_argument, NULL, 16 }, - {"entropy", required_argument, NULL, 17 }, - {"profile", required_argument, NULL, 18 }, - {NULL, no_argument, NULL, 0 }}; - int long_index; - - while ((c =getopt_long_only(argc, argv, "w:h:n:f:o:?", long_opts, &long_index)) != EOF) { - switch (c) { - case 'w': - frame_width = atoi(optarg); - break; - case 'h': - frame_height = atoi(optarg); - break; - case 'n': - case 'f': - frame_rate = atoi(optarg); - break; - case 'o': - coded_fn = strdup(optarg); - break; - case 0: - print_help(); - exit(0); - case 1: - frame_bitrate = atoi(optarg); - break; - case 2: - minimal_qp = atoi(optarg); - break; - case 3: - initial_qp = atoi(optarg); - break; - case 4: - intra_period = atoi(optarg); - break; - case 5: - intra_idr_period = atoi(optarg); - break; - case 6: - ip_period = atoi(optarg); - break; - case 7: - rc_mode = string_to_rc(optarg); - if (rc_mode < 0) { - print_help(); - exit(1); - } - break; - case 9: - srcyuv_fn = strdup(optarg); - break; - case 11: - srcyuv_fourcc = string_to_fourcc(optarg); - if (srcyuv_fourcc <= 0) { - print_help(); - exit(1); - } - break; - case 13: - calc_psnr = 1; - break; - case 14: - misc_priv_type = strtol(optarg, NULL, 0); - break; - case 15: - misc_priv_value = strtol(optarg, NULL, 0); - break; - case 17: - h264_entropy_mode = atoi(optarg) ? 1: 0; - break; - case 18: - if (strncmp(optarg, "BP", 2) == 0) - h264_profile = VAProfileH264Baseline; - else if (strncmp(optarg, "MP", 2) == 0) - h264_profile = VAProfileH264Main; - else if (strncmp(optarg, "HP", 2) == 0) - h264_profile = VAProfileH264High; - else - h264_profile = (VAProfile)0; - break; - case ':': - case '?': - print_help(); - exit(0); - } - } - - if (ip_period < 1) { - printf(" ip_period must be greater than 0\n"); - exit(0); - } - if (intra_period != 1 && intra_period % ip_period != 0) { - printf(" intra_period must be a multiplier of ip_period\n"); - exit(0); - } - if (intra_period != 0 && intra_idr_period % intra_period != 0) { - printf(" intra_idr_period must be a multiplier of intra_period\n"); - exit(0); - } - - if (frame_bitrate == 0) - frame_bitrate = frame_width * frame_height * 12 * frame_rate / 50; - - if (coded_fn == NULL) { - struct stat buf; - if (stat("/tmp", &buf) == 0) - coded_fn = strdup("/tmp/test.264"); - else if (stat("/sdcard", &buf) == 0) - coded_fn = strdup("/sdcard/test.264"); - else - coded_fn = strdup("./test.264"); - } - - /* store coded data into a file */ - coded_fp = fopen(coded_fn, "w+"); - if (coded_fp == NULL) { - printf("Open file %s failed, exit\n", coded_fn); - exit(1); - } - - frame_width_mbaligned = (frame_width + 15) & (~15); - frame_height_mbaligned = (frame_height + 15) & (~15); - if (frame_width != frame_width_mbaligned || - frame_height != frame_height_mbaligned) { - printf("Source frame is %dx%d and will code clip to %dx%d with crop\n", - frame_width, frame_height, - frame_width_mbaligned, frame_height_mbaligned - ); - } - - return 0; -} -#endif - -static Display *x11_display; -static Window x11_window; - -VADisplay -va_open_display(void) -{ - x11_display = XOpenDisplay(NULL); - if (!x11_display) { - fprintf(stderr, "error: can't connect to X server!\n"); - return NULL; - } - return vaGetDisplay(x11_display); + if (va_display.empty()) { + x11_display = XOpenDisplay(NULL); + if (!x11_display) { + fprintf(stderr, "error: can't connect to X server!\n"); + return NULL; + } + use_zerocopy = true; + return vaGetDisplay(x11_display); + } else if (va_display[0] != '/') { + x11_display = XOpenDisplay(va_display.c_str()); + if (!x11_display) { + fprintf(stderr, "error: can't connect to X server!\n"); + return NULL; + } + use_zerocopy = true; + return vaGetDisplay(x11_display); + } else { + drm_fd = open(va_display.c_str(), O_RDWR); + if (drm_fd == -1) { + perror(va_display.c_str()); + return NULL; + } + use_zerocopy = false; + return vaGetDisplayDRM(drm_fd); + } } -void -va_close_display(VADisplay va_dpy) +void H264EncoderImpl::va_close_display(VADisplay va_dpy) { - if (!x11_display) - return; - - if (x11_window) { - XUnmapWindow(x11_display, x11_window); - XDestroyWindow(x11_display, x11_window); - x11_window = None; - } - XCloseDisplay(x11_display); - x11_display = NULL; + if (x11_display) { + XCloseDisplay(x11_display); + x11_display = nullptr; + } + if (drm_fd != -1) { + close(drm_fd); + } } -static int init_va(void) +int H264EncoderImpl::init_va(const string &va_display) { VAProfile profile_list[]={VAProfileH264High, VAProfileH264Main, VAProfileH264Baseline, VAProfileH264ConstrainedBaseline}; VAEntrypoint *entrypoints; @@ -850,7 +894,7 @@ static int init_va(void) VAStatus va_status; unsigned int i; - va_dpy = va_open_display(); + va_dpy = va_open_display(va_display); va_status = vaInitialize(va_dpy, &major_ver, &minor_ver); CHECK_VASTATUS(va_status, "vaInitialize"); @@ -879,33 +923,30 @@ static int init_va(void) } if (support_encode == 0) { - printf("Can't find VAEntrypointEncSlice for H264 profiles\n"); + printf("Can't find VAEntrypointEncSlice for H264 profiles. If you are using a non-Intel GPU\n"); + printf("but have one in your system, try launching Nageru with --va-display /dev/dri/renderD128\n"); + printf("to use VA-API against DRM instead of X11.\n"); exit(1); } else { switch (h264_profile) { case VAProfileH264Baseline: - printf("Use profile VAProfileH264Baseline\n"); ip_period = 1; constraint_set_flag |= (1 << 0); /* Annex A.2.1 */ h264_entropy_mode = 0; break; case VAProfileH264ConstrainedBaseline: - printf("Use profile VAProfileH264ConstrainedBaseline\n"); constraint_set_flag |= (1 << 0 | 1 << 1); /* Annex A.2.2 */ ip_period = 1; break; case VAProfileH264Main: - printf("Use profile VAProfileH264Main\n"); constraint_set_flag |= (1 << 1); /* Annex A.2.2 */ break; case VAProfileH264High: constraint_set_flag |= (1 << 3); /* Annex A.2.4 */ - printf("Use profile VAProfileH264High\n"); break; default: - printf("unknow profile. Set to Baseline"); h264_profile = VAProfileH264Baseline; ip_period = 1; constraint_set_flag |= (1 << 0); /* Annex A.2.1 */ @@ -935,23 +976,6 @@ static int init_va(void) if (attrib[VAConfigAttribRateControl].value != VA_ATTRIB_NOT_SUPPORTED) { int tmp = attrib[VAConfigAttribRateControl].value; - printf("Support rate control mode (0x%x):", tmp); - - if (tmp & VA_RC_NONE) - printf("NONE "); - if (tmp & VA_RC_CBR) - printf("CBR "); - if (tmp & VA_RC_VBR) - printf("VBR "); - if (tmp & VA_RC_VCM) - printf("VCM "); - if (tmp & VA_RC_CQP) - printf("CQP "); - if (tmp & VA_RC_VBR_CONSTRAINED) - printf("VBR_CONSTRAINED "); - - printf("\n"); - if (rc_mode == -1 || !(rc_mode & tmp)) { if (rc_mode != -1) { printf("Warning: Don't support the specified RateControl mode: %s!!!, switch to ", rc_to_string(rc_mode)); @@ -963,8 +987,6 @@ static int init_va(void) break; } } - - printf("RateControl mode: %s\n", rc_to_string(rc_mode)); } config_attrib[config_attrib_num].type = VAConfigAttribRateControl; @@ -976,29 +998,23 @@ static int init_va(void) if (attrib[VAConfigAttribEncPackedHeaders].value != VA_ATTRIB_NOT_SUPPORTED) { int tmp = attrib[VAConfigAttribEncPackedHeaders].value; - printf("Support VAConfigAttribEncPackedHeaders\n"); - h264_packedheader = 1; config_attrib[config_attrib_num].type = VAConfigAttribEncPackedHeaders; config_attrib[config_attrib_num].value = VA_ENC_PACKED_HEADER_NONE; if (tmp & VA_ENC_PACKED_HEADER_SEQUENCE) { - printf("Support packed sequence headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_SEQUENCE; } if (tmp & VA_ENC_PACKED_HEADER_PICTURE) { - printf("Support packed picture headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_PICTURE; } if (tmp & VA_ENC_PACKED_HEADER_SLICE) { - printf("Support packed slice headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_SLICE; } if (tmp & VA_ENC_PACKED_HEADER_MISC) { - printf("Support packed misc headers\n"); config_attrib[config_attrib_num].value |= VA_ENC_PACKED_HEADER_MISC; } @@ -1007,19 +1023,6 @@ static int init_va(void) } if (attrib[VAConfigAttribEncInterlaced].value != VA_ATTRIB_NOT_SUPPORTED) { - int tmp = attrib[VAConfigAttribEncInterlaced].value; - - printf("Support VAConfigAttribEncInterlaced\n"); - - if (tmp & VA_ENC_INTERLACED_FRAME) - printf("support VA_ENC_INTERLACED_FRAME\n"); - if (tmp & VA_ENC_INTERLACED_FIELD) - printf("Support VA_ENC_INTERLACED_FIELD\n"); - if (tmp & VA_ENC_INTERLACED_MBAFF) - printf("Support VA_ENC_INTERLACED_MBAFF\n"); - if (tmp & VA_ENC_INTERLACED_PAFF) - printf("Support VA_ENC_INTERLACED_PAFF\n"); - config_attrib[config_attrib_num].type = VAConfigAttribEncInterlaced; config_attrib[config_attrib_num].value = VA_ENC_PACKED_HEADER_NONE; config_attrib_num++; @@ -1027,35 +1030,13 @@ static int init_va(void) if (attrib[VAConfigAttribEncMaxRefFrames].value != VA_ATTRIB_NOT_SUPPORTED) { h264_maxref = attrib[VAConfigAttribEncMaxRefFrames].value; - - printf("Support %d RefPicList0 and %d RefPicList1\n", - h264_maxref & 0xffff, (h264_maxref >> 16) & 0xffff ); - } - - if (attrib[VAConfigAttribEncMaxSlices].value != VA_ATTRIB_NOT_SUPPORTED) - printf("Support %d slices\n", attrib[VAConfigAttribEncMaxSlices].value); - - if (attrib[VAConfigAttribEncSliceStructure].value != VA_ATTRIB_NOT_SUPPORTED) { - int tmp = attrib[VAConfigAttribEncSliceStructure].value; - - printf("Support VAConfigAttribEncSliceStructure\n"); - - if (tmp & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS) - printf("Support VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS\n"); - if (tmp & VA_ENC_SLICE_STRUCTURE_POWER_OF_TWO_ROWS) - printf("Support VA_ENC_SLICE_STRUCTURE_POWER_OF_TWO_ROWS\n"); - if (tmp & VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS) - printf("Support VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS\n"); - } - if (attrib[VAConfigAttribEncMacroblockInfo].value != VA_ATTRIB_NOT_SUPPORTED) { - printf("Support VAConfigAttribEncMacroblockInfo\n"); } free(entrypoints); return 0; } -static int setup_encode() +int H264EncoderImpl::setup_encode() { VAStatus va_status; VASurfaceID *tmp_surfaceid; @@ -1114,6 +1095,28 @@ static int setup_encode() for (i = 0; i < SURFACE_NUM; i++) { glGenTextures(1, &gl_surfaces[i].y_tex); glGenTextures(1, &gl_surfaces[i].cbcr_tex); + + if (!use_zerocopy) { + // Create Y image. + glBindTexture(GL_TEXTURE_2D, gl_surfaces[i].y_tex); + glTexStorage2D(GL_TEXTURE_2D, 1, GL_R8, frame_width, frame_height); + + // Create CbCr image. + glBindTexture(GL_TEXTURE_2D, gl_surfaces[i].cbcr_tex); + glTexStorage2D(GL_TEXTURE_2D, 1, GL_RG8, frame_width / 2, frame_height / 2); + + // Generate a PBO to read into. It doesn't necessarily fit 1:1 with the VA-API + // buffers, due to potentially differing pitch. + glGenBuffers(1, &gl_surfaces[i].pbo); + glBindBuffer(GL_PIXEL_PACK_BUFFER, gl_surfaces[i].pbo); + glBufferStorage(GL_PIXEL_PACK_BUFFER, frame_width * frame_height * 2, nullptr, GL_MAP_READ_BIT | GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT); + uint8_t *ptr = (uint8_t *)glMapBufferRange(GL_PIXEL_PACK_BUFFER, 0, frame_width * frame_height * 2, GL_MAP_READ_BIT | GL_MAP_PERSISTENT_BIT); + gl_surfaces[i].y_offset = 0; + gl_surfaces[i].cbcr_offset = frame_width * frame_height; + gl_surfaces[i].y_ptr = ptr + gl_surfaces[i].y_offset; + gl_surfaces[i].cbcr_ptr = ptr + gl_surfaces[i].cbcr_offset; + glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); + } } for (i = 0; i < SURFACE_NUM; i++) { @@ -1124,76 +1127,23 @@ static int setup_encode() return 0; } - - -#define partition(ref, field, key, ascending) \ - while (i <= j) { \ - if (ascending) { \ - while (ref[i].field < key) \ - i++; \ - while (ref[j].field > key) \ - j--; \ - } else { \ - while (ref[i].field > key) \ - i++; \ - while (ref[j].field < key) \ - j--; \ - } \ - if (i <= j) { \ - tmp = ref[i]; \ - ref[i] = ref[j]; \ - ref[j] = tmp; \ - i++; \ - j--; \ - } \ - } \ - -static void sort_one(VAPictureH264 ref[], int left, int right, - int ascending, int frame_idx) -{ - int i = left, j = right; - unsigned int key; - VAPictureH264 tmp; - - if (frame_idx) { - key = ref[(left + right) / 2].frame_idx; - partition(ref, frame_idx, key, ascending); - } else { - key = ref[(left + right) / 2].TopFieldOrderCnt; - partition(ref, TopFieldOrderCnt, (signed int)key, ascending); - } - - /* recursion */ - if (left < j) - sort_one(ref, left, j, ascending, frame_idx); - - if (i < right) - sort_one(ref, i, right, ascending, frame_idx); -} - -static void sort_two(VAPictureH264 ref[], int left, int right, unsigned int key, unsigned int frame_idx, - int partition_ascending, int list0_ascending, int list1_ascending) +// Given a list like 1 9 3 0 2 8 4 and a pivot element 3, will produce +// +// 2 1 0 [3] 4 8 9 +template +static void sort_two(T *begin, T *end, const T &pivot, const C &less_than) { - int i = left, j = right; - VAPictureH264 tmp; - - if (frame_idx) { - partition(ref, frame_idx, key, partition_ascending); - } else { - partition(ref, TopFieldOrderCnt, (signed int)key, partition_ascending); - } - - - sort_one(ref, left, i-1, list0_ascending, frame_idx); - sort_one(ref, j+1, right, list1_ascending, frame_idx); + T *middle = partition(begin, end, [&](const T &elem) { return less_than(elem, pivot); }); + sort(begin, middle, [&](const T &a, const T &b) { return less_than(b, a); }); + sort(middle, end, less_than); } -static int update_ReferenceFrames(void) +void H264EncoderImpl::update_ReferenceFrames(int frame_type) { int i; - if (current_frame_type == FRAME_B) - return 0; + if (frame_type == FRAME_B) + return; CurrentCurrPic.flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE; numShortTerm++; @@ -1203,43 +1153,44 @@ static int update_ReferenceFrames(void) ReferenceFrames[i] = ReferenceFrames[i-1]; ReferenceFrames[0] = CurrentCurrPic; - if (current_frame_type != FRAME_B) - current_frame_num++; + current_frame_num++; if (current_frame_num > MaxFrameNum) current_frame_num = 0; - - return 0; } -static int update_RefPicList(void) +int H264EncoderImpl::update_RefPicList(int frame_type) { - unsigned int current_poc = CurrentCurrPic.TopFieldOrderCnt; + const auto descending_by_frame_idx = [](const VAPictureH264 &a, const VAPictureH264 &b) { + return a.frame_idx > b.frame_idx; + }; + const auto ascending_by_top_field_order_cnt = [](const VAPictureH264 &a, const VAPictureH264 &b) { + return a.TopFieldOrderCnt < b.TopFieldOrderCnt; + }; + const auto descending_by_top_field_order_cnt = [](const VAPictureH264 &a, const VAPictureH264 &b) { + return a.TopFieldOrderCnt > b.TopFieldOrderCnt; + }; - if (current_frame_type == FRAME_P) { + if (frame_type == FRAME_P) { memcpy(RefPicList0_P, ReferenceFrames, numShortTerm * sizeof(VAPictureH264)); - sort_one(RefPicList0_P, 0, numShortTerm-1, 0, 1); - } - - if (current_frame_type == FRAME_B) { + sort(&RefPicList0_P[0], &RefPicList0_P[numShortTerm], descending_by_frame_idx); + } else if (frame_type == FRAME_B) { memcpy(RefPicList0_B, ReferenceFrames, numShortTerm * sizeof(VAPictureH264)); - sort_two(RefPicList0_B, 0, numShortTerm-1, current_poc, 0, - 1, 0, 1); + sort_two(&RefPicList0_B[0], &RefPicList0_B[numShortTerm], CurrentCurrPic, ascending_by_top_field_order_cnt); memcpy(RefPicList1_B, ReferenceFrames, numShortTerm * sizeof(VAPictureH264)); - sort_two(RefPicList1_B, 0, numShortTerm-1, current_poc, 0, - 0, 1, 0); + sort_two(&RefPicList1_B[0], &RefPicList1_B[numShortTerm], CurrentCurrPic, descending_by_top_field_order_cnt); } return 0; } -static int render_sequence(void) +int H264EncoderImpl::render_sequence() { - VABufferID seq_param_buf, rc_param_buf, misc_param_tmpbuf, render_id[2]; + VABufferID seq_param_buf, rc_param_buf, render_id[2]; VAStatus va_status; - VAEncMiscParameterBuffer *misc_param, *misc_param_tmp; + VAEncMiscParameterBuffer *misc_param; VAEncMiscParameterRateControl *misc_rate_ctrl; seq_param.level_idc = 41 /*SH_LEVEL_3*/; @@ -1253,7 +1204,7 @@ static int render_sequence(void) seq_param.max_num_ref_frames = num_ref_frames; seq_param.seq_fields.bits.frame_mbs_only_flag = 1; - seq_param.time_scale = frame_rate * 2; + seq_param.time_scale = TIMEBASE * 2; seq_param.num_units_in_tick = 1; /* Tc = num_units_in_tick / scale */ seq_param.seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = Log2MaxPicOrderCntLsb - 4; seq_param.seq_fields.bits.log2_max_frame_num_minus4 = Log2MaxFrameNum - 4;; @@ -1296,33 +1247,18 @@ static int render_sequence(void) render_id[0] = seq_param_buf; render_id[1] = rc_param_buf; - va_status = vaRenderPicture(va_dpy, context_id, &render_id[0], 2); - CHECK_VASTATUS(va_status, "vaRenderPicture");; - - if (misc_priv_type != 0) { - va_status = vaCreateBuffer(va_dpy, context_id, - VAEncMiscParameterBufferType, - sizeof(VAEncMiscParameterBuffer), - 1, NULL, &misc_param_tmpbuf); - CHECK_VASTATUS(va_status, "vaCreateBuffer"); - vaMapBuffer(va_dpy, misc_param_tmpbuf, (void **)&misc_param_tmp); - misc_param_tmp->type = (VAEncMiscParameterType)misc_priv_type; - misc_param_tmp->data[0] = misc_priv_value; - vaUnmapBuffer(va_dpy, misc_param_tmpbuf); - - va_status = vaRenderPicture(va_dpy, context_id, &misc_param_tmpbuf, 1); - } + render_picture_and_delete(va_dpy, context_id, &render_id[0], 2); return 0; } -static int calc_poc(int pic_order_cnt_lsb) +static int calc_poc(int pic_order_cnt_lsb, int frame_type) { static int PicOrderCntMsb_ref = 0, pic_order_cnt_lsb_ref = 0; int prevPicOrderCntMsb, prevPicOrderCntLsb; int PicOrderCntMsb, TopFieldOrderCnt; - if (current_frame_type == FRAME_IDR) + if (frame_type == FRAME_IDR) prevPicOrderCntMsb = prevPicOrderCntLsb = 0; else { prevPicOrderCntMsb = PicOrderCntMsb_ref; @@ -1340,7 +1276,7 @@ static int calc_poc(int pic_order_cnt_lsb) TopFieldOrderCnt = PicOrderCntMsb + pic_order_cnt_lsb; - if (current_frame_type != FRAME_B) { + if (frame_type != FRAME_B) { PicOrderCntMsb_ref = PicOrderCntMsb; pic_order_cnt_lsb_ref = pic_order_cnt_lsb; } @@ -1348,56 +1284,44 @@ static int calc_poc(int pic_order_cnt_lsb) return TopFieldOrderCnt; } -static int render_picture(void) +int H264EncoderImpl::render_picture(int frame_type, int display_frame_num, int gop_start_display_frame_num) { VABufferID pic_param_buf; VAStatus va_status; int i = 0; - pic_param.CurrPic.picture_id = gl_surfaces[current_frame_display % SURFACE_NUM].ref_surface; + pic_param.CurrPic.picture_id = gl_surfaces[display_frame_num % SURFACE_NUM].ref_surface; pic_param.CurrPic.frame_idx = current_frame_num; pic_param.CurrPic.flags = 0; - pic_param.CurrPic.TopFieldOrderCnt = calc_poc((current_frame_display - current_IDR_display) % MaxPicOrderCntLsb); + pic_param.CurrPic.TopFieldOrderCnt = calc_poc((display_frame_num - gop_start_display_frame_num) % MaxPicOrderCntLsb, frame_type); pic_param.CurrPic.BottomFieldOrderCnt = pic_param.CurrPic.TopFieldOrderCnt; CurrentCurrPic = pic_param.CurrPic; - if (getenv("TO_DEL")) { /* set RefPicList into ReferenceFrames */ - update_RefPicList(); /* calc RefPicList */ - memset(pic_param.ReferenceFrames, 0xff, 16 * sizeof(VAPictureH264)); /* invalid all */ - if (current_frame_type == FRAME_P) { - pic_param.ReferenceFrames[0] = RefPicList0_P[0]; - } else if (current_frame_type == FRAME_B) { - pic_param.ReferenceFrames[0] = RefPicList0_B[0]; - pic_param.ReferenceFrames[1] = RefPicList1_B[0]; - } - } else { - memcpy(pic_param.ReferenceFrames, ReferenceFrames, numShortTerm*sizeof(VAPictureH264)); - for (i = numShortTerm; i < SURFACE_NUM; i++) { - pic_param.ReferenceFrames[i].picture_id = VA_INVALID_SURFACE; - pic_param.ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; - } + memcpy(pic_param.ReferenceFrames, ReferenceFrames, numShortTerm*sizeof(VAPictureH264)); + for (i = numShortTerm; i < SURFACE_NUM; i++) { + pic_param.ReferenceFrames[i].picture_id = VA_INVALID_SURFACE; + pic_param.ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; } - pic_param.pic_fields.bits.idr_pic_flag = (current_frame_type == FRAME_IDR); - pic_param.pic_fields.bits.reference_pic_flag = (current_frame_type != FRAME_B); + pic_param.pic_fields.bits.idr_pic_flag = (frame_type == FRAME_IDR); + pic_param.pic_fields.bits.reference_pic_flag = (frame_type != FRAME_B); pic_param.pic_fields.bits.entropy_coding_mode_flag = h264_entropy_mode; pic_param.pic_fields.bits.deblocking_filter_control_present_flag = 1; pic_param.frame_num = current_frame_num; - pic_param.coded_buf = gl_surfaces[current_frame_display % SURFACE_NUM].coded_buf; + pic_param.coded_buf = gl_surfaces[display_frame_num % SURFACE_NUM].coded_buf; pic_param.last_picture = false; // FIXME pic_param.pic_init_qp = initial_qp; va_status = vaCreateBuffer(va_dpy, context_id, VAEncPictureParameterBufferType, sizeof(pic_param), 1, &pic_param, &pic_param_buf); - CHECK_VASTATUS(va_status, "vaCreateBuffer");; + CHECK_VASTATUS(va_status, "vaCreateBuffer"); - va_status = vaRenderPicture(va_dpy, context_id, &pic_param_buf, 1); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, &pic_param_buf, 1); return 0; } -static int render_packedsequence(void) +int H264EncoderImpl::render_packedsequence() { VAEncPackedHeaderParameterBuffer packedheader_param_buffer; VABufferID packedseq_para_bufid, packedseq_data_bufid, render_id[2]; @@ -1427,8 +1351,7 @@ static int render_packedsequence(void) render_id[0] = packedseq_para_bufid; render_id[1] = packedseq_data_bufid; - va_status = vaRenderPicture(va_dpy, context_id, render_id, 2); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, render_id, 2); free(packedseq_buffer); @@ -1436,7 +1359,7 @@ static int render_packedsequence(void) } -static int render_packedpicture(void) +int H264EncoderImpl::render_packedpicture() { VAEncPackedHeaderParameterBuffer packedheader_param_buffer; VABufferID packedpic_para_bufid, packedpic_data_bufid, render_id[2]; @@ -1465,15 +1388,14 @@ static int render_packedpicture(void) render_id[0] = packedpic_para_bufid; render_id[1] = packedpic_data_bufid; - va_status = vaRenderPicture(va_dpy, context_id, render_id, 2); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, render_id, 2); free(packedpic_buffer); return 0; } -static void render_packedslice() +void H264EncoderImpl::render_packedslice() { VAEncPackedHeaderParameterBuffer packedheader_param_buffer; VABufferID packedslice_para_bufid, packedslice_data_bufid, render_id[2]; @@ -1502,28 +1424,27 @@ static void render_packedslice() render_id[0] = packedslice_para_bufid; render_id[1] = packedslice_data_bufid; - va_status = vaRenderPicture(va_dpy, context_id, render_id, 2); - CHECK_VASTATUS(va_status, "vaRenderPicture"); + render_picture_and_delete(va_dpy, context_id, render_id, 2); free(packedslice_buffer); } -static int render_slice(void) +int H264EncoderImpl::render_slice(int encoding_frame_num, int display_frame_num, int gop_start_display_frame_num, int frame_type) { VABufferID slice_param_buf; VAStatus va_status; int i; - update_RefPicList(); + update_RefPicList(frame_type); /* one frame, one slice */ slice_param.macroblock_address = 0; slice_param.num_macroblocks = frame_width_mbaligned * frame_height_mbaligned/(16*16); /* Measured by MB */ - slice_param.slice_type = (current_frame_type == FRAME_IDR)?2:current_frame_type; - if (current_frame_type == FRAME_IDR) { - if (current_frame_encoding != 0) + slice_param.slice_type = (frame_type == FRAME_IDR)?2:frame_type; + if (frame_type == FRAME_IDR) { + if (encoding_frame_num != 0) ++slice_param.idr_pic_id; - } else if (current_frame_type == FRAME_P) { + } else if (frame_type == FRAME_P) { int refpiclist0_max = h264_maxref & 0xffff; memcpy(slice_param.RefPicList0, RefPicList0_P, refpiclist0_max*sizeof(VAPictureH264)); @@ -1531,7 +1452,7 @@ static int render_slice(void) slice_param.RefPicList0[i].picture_id = VA_INVALID_SURFACE; slice_param.RefPicList0[i].flags = VA_PICTURE_H264_INVALID; } - } else if (current_frame_type == FRAME_B) { + } else if (frame_type == FRAME_B) { int refpiclist0_max = h264_maxref & 0xffff; int refpiclist1_max = (h264_maxref >> 16) & 0xffff; @@ -1551,7 +1472,7 @@ static int render_slice(void) slice_param.slice_alpha_c0_offset_div2 = 0; slice_param.slice_beta_offset_div2 = 0; slice_param.direct_spatial_mv_pred_flag = 1; - slice_param.pic_order_cnt_lsb = (current_frame_display - current_IDR_display) % MaxPicOrderCntLsb; + slice_param.pic_order_cnt_lsb = (display_frame_num - gop_start_display_frame_num) % MaxPicOrderCntLsb; if (h264_packedheader && @@ -1560,51 +1481,99 @@ static int render_slice(void) va_status = vaCreateBuffer(va_dpy, context_id, VAEncSliceParameterBufferType, sizeof(slice_param), 1, &slice_param, &slice_param_buf); - CHECK_VASTATUS(va_status, "vaCreateBuffer");; + CHECK_VASTATUS(va_status, "vaCreateBuffer"); + + render_picture_and_delete(va_dpy, context_id, &slice_param_buf, 1); - va_status = vaRenderPicture(va_dpy, context_id, &slice_param_buf, 1); - CHECK_VASTATUS(va_status, "vaRenderPicture"); - return 0; } -int H264Encoder::save_codeddata(unsigned long long display_order, unsigned long long encode_order, int frame_type) +void H264EncoderImpl::save_codeddata(storage_task task) { VACodedBufferSegment *buf_list = NULL; VAStatus va_status; - unsigned int coded_size = 0; string data; - va_status = vaMapBuffer(va_dpy, gl_surfaces[display_order % SURFACE_NUM].coded_buf, (void **)(&buf_list)); + const int64_t global_delay = int64_t(ip_period - 1) * (TIMEBASE / MAX_FPS); // So we never get negative dts. + + va_status = vaMapBuffer(va_dpy, gl_surfaces[task.display_order % SURFACE_NUM].coded_buf, (void **)(&buf_list)); CHECK_VASTATUS(va_status, "vaMapBuffer"); while (buf_list != NULL) { data.append(reinterpret_cast(buf_list->buf), buf_list->size); - if (coded_fp != nullptr) - coded_size += fwrite(buf_list->buf, 1, buf_list->size, coded_fp); buf_list = (VACodedBufferSegment *) buf_list->next; - - frame_size += coded_size; } - vaUnmapBuffer(va_dpy, gl_surfaces[display_order % SURFACE_NUM].coded_buf); - - AVPacket pkt; - memset(&pkt, 0, sizeof(pkt)); - pkt.buf = nullptr; - pkt.pts = av_rescale_q(display_order, AVRational{1, frame_rate}, avstream->time_base); - pkt.dts = av_rescale_q(encode_order, AVRational{1, frame_rate}, avstream->time_base); - pkt.data = reinterpret_cast(&data[0]); - pkt.size = data.size(); - pkt.stream_index = 0; - if (frame_type == FRAME_IDR || frame_type == FRAME_I) { - pkt.flags = AV_PKT_FLAG_KEY; - } else { - pkt.flags = 0; + vaUnmapBuffer(va_dpy, gl_surfaces[task.display_order % SURFACE_NUM].coded_buf); + + { + // Add video. + AVPacket pkt; + memset(&pkt, 0, sizeof(pkt)); + pkt.buf = nullptr; + pkt.data = reinterpret_cast(&data[0]); + pkt.size = data.size(); + pkt.stream_index = 0; + if (task.frame_type == FRAME_IDR || task.frame_type == FRAME_I) { + pkt.flags = AV_PKT_FLAG_KEY; + } else { + pkt.flags = 0; + } + //pkt.duration = 1; + httpd->add_packet(pkt, task.pts + global_delay, task.dts + global_delay); + } + // Encode and add all audio frames up to and including the pts of this video frame. + for ( ;; ) { + int64_t audio_pts; + vector audio; + { + unique_lock lock(frame_queue_mutex); + frame_queue_nonempty.wait(lock, [this]{ return storage_thread_should_quit || !pending_audio_frames.empty(); }); + if (storage_thread_should_quit && pending_audio_frames.empty()) return; + auto it = pending_audio_frames.begin(); + if (it->first > task.pts) break; + audio_pts = it->first; + audio = move(it->second); + pending_audio_frames.erase(it); + } + + AVFrame *frame = avcodec_alloc_frame(); + frame->nb_samples = audio.size() / 2; + frame->format = AV_SAMPLE_FMT_S32; + frame->channel_layout = AV_CH_LAYOUT_STEREO; + + unique_ptr int_samples(new int32_t[audio.size()]); + int ret = avcodec_fill_audio_frame(frame, 2, AV_SAMPLE_FMT_S32, (const uint8_t*)int_samples.get(), audio.size() * sizeof(int32_t), 1); + if (ret < 0) { + fprintf(stderr, "avcodec_fill_audio_frame() failed with %d\n", ret); + exit(1); + } + for (int i = 0; i < frame->nb_samples * 2; ++i) { + if (audio[i] >= 1.0f) { + int_samples[i] = 2147483647; + } else if (audio[i] <= -1.0f) { + int_samples[i] = -2147483647; + } else { + int_samples[i] = lrintf(audio[i] * 2147483647.0f); + } + } + + AVPacket pkt; + av_init_packet(&pkt); + pkt.data = nullptr; + pkt.size = 0; + int got_output; + avcodec_encode_audio2(context_audio, &pkt, frame, &got_output); + if (got_output) { + pkt.stream_index = 1; + httpd->add_packet(pkt, audio_pts + global_delay, audio_pts + global_delay); + } + // TODO: Delayed frames. + avcodec_free_frame(&frame); + av_free_packet(&pkt); + if (audio_pts == task.pts) break; } - pkt.duration = 1; - av_interleaved_write_frame(avctx, &pkt); #if 0 printf("\r "); /* return back to startpoint */ @@ -1623,38 +1592,29 @@ int H264Encoder::save_codeddata(unsigned long long display_order, unsigned long break; } printf("%08lld", encode_order); - printf("(%06d bytes coded)", coded_size); #endif - - return 0; } // this is weird. but it seems to put a new frame onto the queue -void H264Encoder::storage_task_enqueue(unsigned long long display_order, unsigned long long encode_order, int frame_type) +void H264EncoderImpl::storage_task_enqueue(storage_task task) { - std::unique_lock lock(storage_task_queue_mutex); - - storage_task tmp; - tmp.display_order = display_order; - tmp.encode_order = encode_order; - tmp.frame_type = frame_type; - storage_task_queue.push(tmp); - srcsurface_status[display_order % SURFACE_NUM] = SRC_SURFACE_IN_ENCODING; - + unique_lock lock(storage_task_queue_mutex); + storage_task_queue.push(move(task)); + srcsurface_status[task.display_order % SURFACE_NUM] = SRC_SURFACE_IN_ENCODING; storage_task_queue_changed.notify_all(); } -void H264Encoder::storage_task_thread() +void H264EncoderImpl::storage_task_thread() { for ( ;; ) { storage_task current; { // wait until there's an encoded frame - std::unique_lock lock(storage_task_queue_mutex); + unique_lock lock(storage_task_queue_mutex); storage_task_queue_changed.wait(lock, [this]{ return storage_thread_should_quit || !storage_task_queue.empty(); }); - if (storage_thread_should_quit) return; - current = storage_task_queue.front(); + if (storage_thread_should_quit && storage_task_queue.empty()) return; + current = move(storage_task_queue.front()); storage_task_queue.pop(); } @@ -1663,33 +1623,40 @@ void H264Encoder::storage_task_thread() // waits for data, then saves it to disk. va_status = vaSyncSurface(va_dpy, gl_surfaces[current.display_order % SURFACE_NUM].src_surface); CHECK_VASTATUS(va_status, "vaSyncSurface"); - save_codeddata(current.display_order, current.encode_order, current.frame_type); + save_codeddata(move(current)); { - std::unique_lock lock(storage_task_queue_mutex); + unique_lock lock(storage_task_queue_mutex); srcsurface_status[current.display_order % SURFACE_NUM] = SRC_SURFACE_FREE; storage_task_queue_changed.notify_all(); } } } -static int release_encode() +int H264EncoderImpl::release_encode() { - int i; - - for (i = 0; i < SURFACE_NUM; i++) { - vaDestroyBuffer(va_dpy, gl_surfaces[i].coded_buf); - vaDestroySurfaces(va_dpy, &gl_surfaces[i].src_surface, 1); - vaDestroySurfaces(va_dpy, &gl_surfaces[i].ref_surface, 1); - } - - vaDestroyContext(va_dpy, context_id); - vaDestroyConfig(va_dpy, config_id); + for (unsigned i = 0; i < SURFACE_NUM; i++) { + vaDestroyBuffer(va_dpy, gl_surfaces[i].coded_buf); + vaDestroySurfaces(va_dpy, &gl_surfaces[i].src_surface, 1); + vaDestroySurfaces(va_dpy, &gl_surfaces[i].ref_surface, 1); + + if (!use_zerocopy) { + glBindBuffer(GL_PIXEL_PACK_BUFFER, gl_surfaces[i].pbo); + glUnmapBuffer(GL_PIXEL_PACK_BUFFER); + glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); + glDeleteBuffers(1, &gl_surfaces[i].pbo); + } + glDeleteTextures(1, &gl_surfaces[i].y_tex); + glDeleteTextures(1, &gl_surfaces[i].cbcr_tex); + } - return 0; + vaDestroyContext(va_dpy, context_id); + vaDestroyConfig(va_dpy, config_id); + + return 0; } -static int deinit_va() +int H264EncoderImpl::deinit_va() { vaTerminate(va_dpy); @@ -1699,72 +1666,30 @@ static int deinit_va() } -static int print_input() -{ - printf("\n\nINPUT:Try to encode H264...\n"); - if (rc_mode != -1) - printf("INPUT: RateControl : %s\n", rc_to_string(rc_mode)); - printf("INPUT: Resolution : %dx%dframes\n", frame_width, frame_height); - printf("INPUT: FrameRate : %d\n", frame_rate); - printf("INPUT: Bitrate : %d\n", frame_bitrate); - printf("INPUT: Slieces : %d\n", frame_slices); - printf("INPUT: IntraPeriod : %d\n", intra_period); - printf("INPUT: IDRPeriod : %d\n", intra_idr_period); - printf("INPUT: IpPeriod : %d\n", ip_period); - printf("INPUT: Initial QP : %d\n", initial_qp); - printf("INPUT: Min QP : %d\n", minimal_qp); - printf("INPUT: Coded Clip : %s\n", coded_fn); - - printf("\n\n"); /* return back to startpoint */ - - return 0; -} - - -//H264Encoder::H264Encoder(SDL_Window *window, SDL_GLContext context, int width, int height, const char *output_filename) -H264Encoder::H264Encoder(QSurface *surface, int width, int height, const char *output_filename) - : current_storage_frame(0), surface(surface) - //: width(width), height(height), current_encoding_frame(0) +H264EncoderImpl::H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd) + : current_storage_frame(0), surface(surface), httpd(httpd) { - av_register_all(); - avctx = avformat_alloc_context(); - avctx->oformat = av_guess_format(NULL, output_filename, NULL); - strcpy(avctx->filename, output_filename); - if (avio_open2(&avctx->pb, output_filename, AVIO_FLAG_WRITE, &avctx->interrupt_callback, NULL) < 0) { - fprintf(stderr, "%s: avio_open2() failed\n", output_filename); - exit(1); - } - AVCodec *codec = avcodec_find_encoder(AV_CODEC_ID_H264); - avstream = avformat_new_stream(avctx, codec); - if (avstream == nullptr) { - fprintf(stderr, "%s: avformat_new_stream() failed\n", output_filename); - exit(1); - } - avstream->time_base = AVRational{1, frame_rate}; // TODO - avstream->codec->width = width; - avstream->codec->height = height; - //avstream->codec->time_base = AVRational{1, 60}; // TODO - avstream->codec->time_base = AVRational{1, frame_rate}; - avstream->codec->ticks_per_frame = 1; // or 2? - - if (avformat_write_header(avctx, NULL) < 0) { - fprintf(stderr, "%s: avformat_write_header() failed\n", output_filename); + AVCodec *codec_audio = avcodec_find_encoder(AUDIO_OUTPUT_CODEC); + context_audio = avcodec_alloc_context3(codec_audio); + context_audio->bit_rate = AUDIO_OUTPUT_BIT_RATE; + context_audio->sample_rate = OUTPUT_FREQUENCY; + context_audio->sample_fmt = AUDIO_OUTPUT_SAMPLE_FMT; + context_audio->channels = 2; + context_audio->channel_layout = AV_CH_LAYOUT_STEREO; + context_audio->time_base = AVRational{1, TIMEBASE}; + if (avcodec_open2(context_audio, codec_audio, NULL) < 0) { + fprintf(stderr, "Could not open codec\n"); exit(1); } - coded_fp = fopen("dump.h264", "wb"); - assert(coded_fp != NULL); - frame_width = width; frame_height = height; frame_width_mbaligned = (frame_width + 15) & (~15); frame_height_mbaligned = (frame_height + 15) & (~15); - frame_bitrate = 15000000; // / 60; - current_frame_encoding = 0; - print_input(); + //print_input(); - init_va(); + init_va(va_display); setup_encode(); // No frames are ready yet. @@ -1774,48 +1699,32 @@ H264Encoder::H264Encoder(QSurface *surface, int width, int height, const char *o memset(&pic_param, 0, sizeof(pic_param)); memset(&slice_param, 0, sizeof(slice_param)); - storage_thread = std::thread(&H264Encoder::storage_task_thread, this); + storage_thread = thread(&H264EncoderImpl::storage_task_thread, this); - copy_thread = std::thread([this]{ + encode_thread = thread([this]{ //SDL_GL_MakeCurrent(window, context); - QOpenGLContext *context = create_context(); + QOpenGLContext *context = create_context(this->surface); eglBindAPI(EGL_OPENGL_API); if (!make_current(context, this->surface)) { printf("display=%p surface=%p context=%p curr=%p err=%d\n", eglGetCurrentDisplay(), this->surface, context, eglGetCurrentContext(), eglGetError()); exit(1); } - copy_thread_func(); + encode_thread_func(); }); } -H264Encoder::~H264Encoder() +H264EncoderImpl::~H264EncoderImpl() { - { - unique_lock lock(storage_task_queue_mutex); - storage_thread_should_quit = true; - storage_task_queue_changed.notify_all(); - } - { - unique_lock lock(frame_queue_mutex); - copy_thread_should_quit = true; - frame_queue_nonempty.notify_one(); - } - storage_thread.join(); - copy_thread.join(); - - release_encode(); - deinit_va(); - - av_write_trailer(avctx); - avformat_free_context(avctx); + shutdown(); } -bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) +bool H264EncoderImpl::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) { + assert(!is_shutdown); { // Wait until this frame slot is done encoding. - std::unique_lock lock(storage_task_queue_mutex); + unique_lock lock(storage_task_queue_mutex); storage_task_queue_changed.wait(lock, [this]{ return storage_thread_should_quit || (srcsurface_status[current_storage_frame % SURFACE_NUM] == SRC_SURFACE_FREE); }); if (storage_thread_should_quit) return false; } @@ -1825,125 +1734,314 @@ bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) *y_tex = surf->y_tex; *cbcr_tex = surf->cbcr_tex; - VASurfaceID surface = surf->src_surface; - VAStatus va_status = vaDeriveImage(va_dpy, surface, &surf->surface_image); - CHECK_VASTATUS(va_status, "vaDeriveImage"); - - VABufferInfo buf_info; - buf_info.mem_type = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME; // or VA_SURFACE_ATTRIB_MEM_TYPE_KERNEL_DRM? - va_status = vaAcquireBufferHandle(va_dpy, surf->surface_image.buf, &buf_info); - CHECK_VASTATUS(va_status, "vaAcquireBufferHandle"); - - // Create Y image. - surf->y_egl_image = EGL_NO_IMAGE_KHR; - EGLint y_attribs[] = { - EGL_WIDTH, frame_width, - EGL_HEIGHT, frame_height, - EGL_LINUX_DRM_FOURCC_EXT, fourcc_code('R', '8', ' ', ' '), - EGL_DMA_BUF_PLANE0_FD_EXT, EGLint(buf_info.handle), - EGL_DMA_BUF_PLANE0_OFFSET_EXT, EGLint(surf->surface_image.offsets[0]), - EGL_DMA_BUF_PLANE0_PITCH_EXT, EGLint(surf->surface_image.pitches[0]), - EGL_NONE - }; + VAStatus va_status = vaDeriveImage(va_dpy, surf->src_surface, &surf->surface_image); + CHECK_VASTATUS(va_status, "vaDeriveImage"); + + if (use_zerocopy) { + VABufferInfo buf_info; + buf_info.mem_type = VA_SURFACE_ATTRIB_MEM_TYPE_DRM_PRIME; // or VA_SURFACE_ATTRIB_MEM_TYPE_KERNEL_DRM? + va_status = vaAcquireBufferHandle(va_dpy, surf->surface_image.buf, &buf_info); + CHECK_VASTATUS(va_status, "vaAcquireBufferHandle"); + + // Create Y image. + surf->y_egl_image = EGL_NO_IMAGE_KHR; + EGLint y_attribs[] = { + EGL_WIDTH, frame_width, + EGL_HEIGHT, frame_height, + EGL_LINUX_DRM_FOURCC_EXT, fourcc_code('R', '8', ' ', ' '), + EGL_DMA_BUF_PLANE0_FD_EXT, EGLint(buf_info.handle), + EGL_DMA_BUF_PLANE0_OFFSET_EXT, EGLint(surf->surface_image.offsets[0]), + EGL_DMA_BUF_PLANE0_PITCH_EXT, EGLint(surf->surface_image.pitches[0]), + EGL_NONE + }; + + surf->y_egl_image = eglCreateImageKHR(eglGetCurrentDisplay(), EGL_NO_CONTEXT, EGL_LINUX_DMA_BUF_EXT, NULL, y_attribs); + assert(surf->y_egl_image != EGL_NO_IMAGE_KHR); + + // Associate Y image to a texture. + glBindTexture(GL_TEXTURE_2D, *y_tex); + glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, surf->y_egl_image); + + // Create CbCr image. + surf->cbcr_egl_image = EGL_NO_IMAGE_KHR; + EGLint cbcr_attribs[] = { + EGL_WIDTH, frame_width, + EGL_HEIGHT, frame_height, + EGL_LINUX_DRM_FOURCC_EXT, fourcc_code('G', 'R', '8', '8'), + EGL_DMA_BUF_PLANE0_FD_EXT, EGLint(buf_info.handle), + EGL_DMA_BUF_PLANE0_OFFSET_EXT, EGLint(surf->surface_image.offsets[1]), + EGL_DMA_BUF_PLANE0_PITCH_EXT, EGLint(surf->surface_image.pitches[1]), + EGL_NONE + }; + + surf->cbcr_egl_image = eglCreateImageKHR(eglGetCurrentDisplay(), EGL_NO_CONTEXT, EGL_LINUX_DMA_BUF_EXT, NULL, cbcr_attribs); + assert(surf->cbcr_egl_image != EGL_NO_IMAGE_KHR); + + // Associate CbCr image to a texture. + glBindTexture(GL_TEXTURE_2D, *cbcr_tex); + glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, surf->cbcr_egl_image); + } - surf->y_egl_image = eglCreateImageKHR(eglGetCurrentDisplay(), EGL_NO_CONTEXT, EGL_LINUX_DMA_BUF_EXT, NULL, y_attribs); - assert(surf->y_egl_image != EGL_NO_IMAGE_KHR); - - // Associate Y image to a texture. - glBindTexture(GL_TEXTURE_2D, *y_tex); - glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, surf->y_egl_image); - - // Create CbCr image. - surf->cbcr_egl_image = EGL_NO_IMAGE_KHR; - EGLint cbcr_attribs[] = { - EGL_WIDTH, frame_width, - EGL_HEIGHT, frame_height, - EGL_LINUX_DRM_FOURCC_EXT, fourcc_code('G', 'R', '8', '8'), - EGL_DMA_BUF_PLANE0_FD_EXT, EGLint(buf_info.handle), - EGL_DMA_BUF_PLANE0_OFFSET_EXT, EGLint(surf->surface_image.offsets[1]), - EGL_DMA_BUF_PLANE0_PITCH_EXT, EGLint(surf->surface_image.pitches[1]), - EGL_NONE - }; + return true; +} + +void H264EncoderImpl::add_audio(int64_t pts, vector audio) +{ + assert(!is_shutdown); + { + unique_lock lock(frame_queue_mutex); + pending_audio_frames[pts] = move(audio); + } + frame_queue_nonempty.notify_all(); +} - surf->cbcr_egl_image = eglCreateImageKHR(eglGetCurrentDisplay(), EGL_NO_CONTEXT, EGL_LINUX_DMA_BUF_EXT, NULL, cbcr_attribs); - assert(surf->cbcr_egl_image != EGL_NO_IMAGE_KHR); +void H264EncoderImpl::end_frame(RefCountedGLsync fence, int64_t pts, const vector &input_frames) +{ + assert(!is_shutdown); - // Associate CbCr image to a texture. - glBindTexture(GL_TEXTURE_2D, *cbcr_tex); - glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, surf->cbcr_egl_image); + if (!use_zerocopy) { + GLSurface *surf = &gl_surfaces[current_storage_frame % SURFACE_NUM]; - return true; + glPixelStorei(GL_PACK_ROW_LENGTH, 0); + check_error(); + + glBindBuffer(GL_PIXEL_PACK_BUFFER, surf->pbo); + check_error(); + + glBindTexture(GL_TEXTURE_2D, surf->y_tex); + check_error(); + glGetTexImage(GL_TEXTURE_2D, 0, GL_RED, GL_UNSIGNED_BYTE, BUFFER_OFFSET(surf->y_offset)); + check_error(); + + glBindTexture(GL_TEXTURE_2D, surf->cbcr_tex); + check_error(); + glGetTexImage(GL_TEXTURE_2D, 0, GL_RG, GL_UNSIGNED_BYTE, BUFFER_OFFSET(surf->cbcr_offset)); + check_error(); + + glBindTexture(GL_TEXTURE_2D, 0); + check_error(); + glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); + check_error(); + + glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT | GL_CLIENT_MAPPED_BUFFER_BARRIER_BIT); + check_error(); + fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0); + check_error(); + } + + { + unique_lock lock(frame_queue_mutex); + pending_video_frames[current_storage_frame] = PendingFrame{ fence, input_frames, pts }; + ++current_storage_frame; + } + frame_queue_nonempty.notify_all(); } -void H264Encoder::end_frame(GLsync fence) +void H264EncoderImpl::shutdown() { + if (is_shutdown) { + return; + } + { unique_lock lock(frame_queue_mutex); - pending_frames[current_storage_frame++] = fence; + encode_thread_should_quit = true; + frame_queue_nonempty.notify_all(); + } + encode_thread.join(); + { + unique_lock lock(storage_task_queue_mutex); + storage_thread_should_quit = true; + frame_queue_nonempty.notify_all(); + storage_task_queue_changed.notify_all(); } - frame_queue_nonempty.notify_one(); + storage_thread.join(); + + release_encode(); + deinit_va(); + is_shutdown = true; } -void H264Encoder::copy_thread_func() +void H264EncoderImpl::encode_thread_func() { - for ( ;; ) { - GLsync fence; - encoding2display_order(current_frame_encoding, intra_period, intra_idr_period, ip_period, - ¤t_frame_display, ¤t_frame_type); - if (current_frame_type == FRAME_IDR) { + int64_t last_dts = -1; + int gop_start_display_frame_num = 0; + for (int encoding_frame_num = 0; ; ++encoding_frame_num) { + PendingFrame frame; + int pts_lag; + int frame_type, display_frame_num; + encoding2display_order(encoding_frame_num, intra_period, intra_idr_period, ip_period, + &display_frame_num, &frame_type, &pts_lag); + if (frame_type == FRAME_IDR) { numShortTerm = 0; current_frame_num = 0; - current_IDR_display = current_frame_display; + gop_start_display_frame_num = display_frame_num; } { unique_lock lock(frame_queue_mutex); - frame_queue_nonempty.wait(lock, [this]{ return copy_thread_should_quit || pending_frames.count(current_frame_display) != 0; }); - if (copy_thread_should_quit) return; - fence = pending_frames[current_frame_display]; - pending_frames.erase(current_frame_display); + frame_queue_nonempty.wait(lock, [this, display_frame_num]{ + return encode_thread_should_quit || pending_video_frames.count(display_frame_num) != 0; + }); + if (encode_thread_should_quit && pending_video_frames.count(display_frame_num) == 0) { + // We have queued frames that were supposed to be B-frames, + // but will be no P-frame to encode them against. Encode them all + // as P-frames instead. Note that this happens under the mutex, + // but nobody else uses it at this point, since we're shutting down, + // so there's no contention. + encode_remaining_frames_as_p(encoding_frame_num, gop_start_display_frame_num, last_dts); + return; + } else { + frame = move(pending_video_frames[display_frame_num]); + pending_video_frames.erase(display_frame_num); + } + } + + // Determine the dts of this frame. + int64_t dts; + if (pts_lag == -1) { + assert(last_dts != -1); + dts = last_dts + (TIMEBASE / MAX_FPS); + } else { + dts = frame.pts - pts_lag; + } + last_dts = dts; + + encode_frame(frame, encoding_frame_num, display_frame_num, gop_start_display_frame_num, frame_type, frame.pts, dts); + } +} + +void H264EncoderImpl::encode_remaining_frames_as_p(int encoding_frame_num, int gop_start_display_frame_num, int64_t last_dts) +{ + if (pending_video_frames.empty()) { + return; + } + + for (auto &pending_frame : pending_video_frames) { + int display_frame_num = pending_frame.first; + assert(display_frame_num > 0); + PendingFrame frame = move(pending_frame.second); + int64_t dts = last_dts + (TIMEBASE / MAX_FPS); + printf("Finalizing encode: Encoding leftover frame %d as P-frame instead of B-frame.\n", display_frame_num); + encode_frame(frame, encoding_frame_num++, display_frame_num, gop_start_display_frame_num, FRAME_P, frame.pts, dts); + last_dts = dts; + } +} + +namespace { + +void memcpy_with_pitch(uint8_t *dst, const uint8_t *src, size_t src_width, size_t dst_pitch, size_t height) +{ + if (src_width == dst_pitch) { + memcpy(dst, src, src_width * height); + } else { + for (size_t y = 0; y < height; ++y) { + const uint8_t *sptr = src + y * src_width; + uint8_t *dptr = dst + y * dst_pitch; + memcpy(dptr, sptr, src_width); } + } +} + +} // namespace + +void H264EncoderImpl::encode_frame(H264EncoderImpl::PendingFrame frame, int encoding_frame_num, int display_frame_num, int gop_start_display_frame_num, + int frame_type, int64_t pts, int64_t dts) +{ + // Wait for the GPU to be done with the frame. + GLenum sync_status; + do { + sync_status = glClientWaitSync(frame.fence.get(), 0, 1000000000); + check_error(); + } while (sync_status == GL_TIMEOUT_EXPIRED); + assert(sync_status != GL_WAIT_FAILED); - // Wait for the GPU to be done with the frame. - glClientWaitSync(fence, 0, 0); - glDeleteSync(fence); + // Release back any input frames we needed to render this frame. + frame.input_frames.clear(); - // Unmap the image. - GLSurface *surf = &gl_surfaces[current_frame_display % SURFACE_NUM]; + GLSurface *surf = &gl_surfaces[display_frame_num % SURFACE_NUM]; + VAStatus va_status; + + if (use_zerocopy) { eglDestroyImageKHR(eglGetCurrentDisplay(), surf->y_egl_image); eglDestroyImageKHR(eglGetCurrentDisplay(), surf->cbcr_egl_image); - VAStatus va_status = vaReleaseBufferHandle(va_dpy, surf->surface_image.buf); + va_status = vaReleaseBufferHandle(va_dpy, surf->surface_image.buf); CHECK_VASTATUS(va_status, "vaReleaseBufferHandle"); - va_status = vaDestroyImage(va_dpy, surf->surface_image.image_id); - CHECK_VASTATUS(va_status, "vaDestroyImage"); + } else { + unsigned char *surface_p = nullptr; + vaMapBuffer(va_dpy, surf->surface_image.buf, (void **)&surface_p); - VASurfaceID surface = surf->src_surface; + unsigned char *va_y_ptr = (unsigned char *)surface_p + surf->surface_image.offsets[0]; + memcpy_with_pitch(va_y_ptr, surf->y_ptr, frame_width, surf->surface_image.pitches[0], frame_height); - // Schedule the frame for encoding. - va_status = vaBeginPicture(va_dpy, context_id, surface); - CHECK_VASTATUS(va_status, "vaBeginPicture"); + unsigned char *va_cbcr_ptr = (unsigned char *)surface_p + surf->surface_image.offsets[1]; + memcpy_with_pitch(va_cbcr_ptr, surf->cbcr_ptr, (frame_width / 2) * sizeof(uint16_t), surf->surface_image.pitches[1], frame_height / 2); - if (current_frame_type == FRAME_IDR) { - render_sequence(); - render_picture(); - if (h264_packedheader) { - render_packedsequence(); - render_packedpicture(); - } - } else { - //render_sequence(); - render_picture(); + va_status = vaUnmapBuffer(va_dpy, surf->surface_image.buf); + CHECK_VASTATUS(va_status, "vaUnmapBuffer"); + } + + va_status = vaDestroyImage(va_dpy, surf->surface_image.image_id); + CHECK_VASTATUS(va_status, "vaDestroyImage"); + + // Schedule the frame for encoding. + VASurfaceID va_surface = surf->src_surface; + va_status = vaBeginPicture(va_dpy, context_id, va_surface); + CHECK_VASTATUS(va_status, "vaBeginPicture"); + + if (frame_type == FRAME_IDR) { + render_sequence(); + render_picture(frame_type, display_frame_num, gop_start_display_frame_num); + if (h264_packedheader) { + render_packedsequence(); + render_packedpicture(); } - render_slice(); - - va_status = vaEndPicture(va_dpy, context_id); - CHECK_VASTATUS(va_status, "vaEndPicture"); - - // so now the data is done encoding (well, async job kicked off)... - // we send that to the storage thread - storage_task_enqueue(current_frame_display, current_frame_encoding, current_frame_type); - - update_ReferenceFrames(); - ++current_frame_encoding; + } else { + //render_sequence(); + render_picture(frame_type, display_frame_num, gop_start_display_frame_num); } + render_slice(encoding_frame_num, display_frame_num, gop_start_display_frame_num, frame_type); + + va_status = vaEndPicture(va_dpy, context_id); + CHECK_VASTATUS(va_status, "vaEndPicture"); + + // so now the data is done encoding (well, async job kicked off)... + // we send that to the storage thread + storage_task tmp; + tmp.display_order = display_frame_num; + tmp.frame_type = frame_type; + tmp.pts = pts; + tmp.dts = dts; + storage_task_enqueue(move(tmp)); + + update_ReferenceFrames(frame_type); +} + +// Proxy object. +H264Encoder::H264Encoder(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd) + : impl(new H264EncoderImpl(surface, va_display, width, height, httpd)) {} + +// Must be defined here because unique_ptr<> destructor needs to know the impl. +H264Encoder::~H264Encoder() {} + +void H264Encoder::add_audio(int64_t pts, vector audio) +{ + impl->add_audio(pts, audio); } + +bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) +{ + return impl->begin_frame(y_tex, cbcr_tex); +} + +void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const vector &input_frames) +{ + impl->end_frame(fence, pts, input_frames); +} + +void H264Encoder::shutdown() +{ + impl->shutdown(); +} + +// Real class.