X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=h264encode.cpp;h=d0da39d08c0925e79ab9c9cef418257c3b95bb2e;hb=fec3f686bf239f9d98782603616613c16f8025b1;hp=d63fe896e5efebfaf717e368226fca20c7a5c6c5;hpb=345b65e8f95fba76164f9ac8e1304138fd984b3d;p=nageru diff --git a/h264encode.cpp b/h264encode.cpp index d63fe89..d0da39d 100644 --- a/h264encode.cpp +++ b/h264encode.cpp @@ -7,11 +7,11 @@ #include #include #include -#include -#include +#include +#include #include +#include #include -#include #include #include #include @@ -20,15 +20,21 @@ #include #include #include +#include +#include #include #include #include #include +#include #include "context.h" +#include "defs.h" #include "httpd.h" #include "timebase.h" +using namespace std; + class QOpenGLContext; class QSurface; @@ -104,21 +110,17 @@ 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 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 intra_idr_period = MAX_FPS; // About a second; more at lower frame rates. Not ideal. static int ip_period = 3; static int rc_mode = -1; static int rc_default_modes[] = { @@ -129,11 +131,7 @@ 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; @@ -151,6 +149,20 @@ typedef struct __bitstream bitstream; using namespace std; +// 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(va_dpy, buffers[i]); + CHECK_VASTATUS(va_status, "vaDestroyBuffer"); + } +} + static unsigned int va_swap32(unsigned int val) { @@ -347,8 +359,8 @@ static void sps_rbsp(bitstream *bs) 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, 1, 8); /* transfer_characteristics (1 = BT.709) */ - bitstream_put_ui(bs, 1, 8); /* matrix_coefficients (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 */ @@ -601,7 +613,7 @@ static int build_packed_slice_buffer(unsigned char **header_buffer) // // 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_ -// , which gives us a frame period N. The decoder can always decode +// 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 @@ -647,9 +659,9 @@ static int build_packed_slice_buffer(unsigned char **header_buffer) #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 *displaying_order, int *frame_type, int *pts_lag) { int encoding_order_gop = 0; @@ -694,7 +706,7 @@ void encoding2display_order( *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 / frame_rate; + *pts_lag = TIMEBASE / MAX_FPS; } else if (((encoding_order_gop - 1) % ip_period) != 0) { /* B frames */ *frame_type = FRAME_B; *displaying_order = encoding_order - 1; @@ -855,7 +867,7 @@ static int process_cmdline(int argc, char *argv[]) } if (frame_bitrate == 0) - frame_bitrate = frame_width * frame_height * 12 * frame_rate / 50; + frame_bitrate = frame_width * frame_height * 12 * MAX_FPS / 50; if (coded_fn == NULL) { struct stat buf; @@ -955,28 +967,23 @@ static int init_va(void) } 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 */ @@ -1006,23 +1013,6 @@ 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)); @@ -1034,8 +1024,6 @@ static int init_va(void) break; } } - - printf("RateControl mode: %s\n", rc_to_string(rc_mode)); } config_attrib[config_attrib_num].type = VAConfigAttribRateControl; @@ -1047,29 +1035,23 @@ 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; } @@ -1078,19 +1060,6 @@ 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++; @@ -1098,28 +1067,6 @@ 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); @@ -1259,12 +1206,12 @@ static void sort_two(VAPictureH264 ref[], int left, int right, unsigned int key, sort_one(ref, j+1, right, list1_ascending, frame_idx); } -static int update_ReferenceFrames(void) +static void 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++; @@ -1274,25 +1221,22 @@ 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) +static int update_RefPicList(int frame_type) { unsigned int current_poc = CurrentCurrPic.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) { + 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); @@ -1367,8 +1311,7 @@ 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");; + render_picture_and_delete(va_dpy, context_id, &render_id[0], 2); if (misc_priv_type != 0) { va_status = vaCreateBuffer(va_dpy, context_id, @@ -1381,19 +1324,19 @@ static int render_sequence(void) 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, &misc_param_tmpbuf, 1); } 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; @@ -1411,7 +1354,7 @@ 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; } @@ -1419,51 +1362,39 @@ static int calc_poc(int pic_order_cnt_lsb) return TopFieldOrderCnt; } -static int render_picture(void) +static int 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; } @@ -1498,8 +1429,7 @@ 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); @@ -1536,8 +1466,7 @@ 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); @@ -1573,28 +1502,27 @@ 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) +static int 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)); @@ -1602,7 +1530,7 @@ 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; @@ -1622,7 +1550,7 @@ 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 && @@ -1631,17 +1559,16 @@ 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(storage_task task) +void H264Encoder::save_codeddata(storage_task task) { VACodedBufferSegment *buf_list = NULL; VAStatus va_status; @@ -1649,7 +1576,7 @@ int H264Encoder::save_codeddata(storage_task task) string data; - const int64_t global_delay = (ip_period - 1) * (TIMEBASE / frame_rate); // So we never get negative dts. + const int64_t global_delay = (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"); @@ -1678,13 +1605,13 @@ int H264Encoder::save_codeddata(storage_task task) 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. - // (They can never be queued to us after the video frame they belong to, only before.) for ( ;; ) { int64_t audio_pts; - std::vector audio; + vector audio; { unique_lock lock(frame_queue_mutex); - if (pending_audio_frames.empty()) break; + 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; @@ -1694,14 +1621,23 @@ int H264Encoder::save_codeddata(storage_task task) AVFrame *frame = avcodec_alloc_frame(); frame->nb_samples = audio.size() / 2; - frame->format = AV_SAMPLE_FMT_FLT; + frame->format = AV_SAMPLE_FMT_S32; frame->channel_layout = AV_CH_LAYOUT_STEREO; - unique_ptr planar_samples(new float[audio.size()]); - avcodec_fill_audio_frame(frame, 2, AV_SAMPLE_FMT_FLTP, (const uint8_t*)planar_samples.get(), audio.size() * sizeof(float), 0); - for (int i = 0; i < frame->nb_samples; ++i) { - planar_samples[i] = audio[i * 2 + 0]; - planar_samples[i + frame->nb_samples] = audio[i * 2 + 1]; + 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; @@ -1716,6 +1652,8 @@ int H264Encoder::save_codeddata(storage_task task) } // TODO: Delayed frames. avcodec_free_frame(&frame); + av_free_packet(&pkt); + if (audio_pts == task.pts) break; } #if 0 @@ -1737,15 +1675,13 @@ int H264Encoder::save_codeddata(storage_task task) 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(storage_task task) { - std::unique_lock lock(storage_task_queue_mutex); + 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(); @@ -1757,9 +1693,9 @@ void H264Encoder::storage_task_thread() 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; + if (storage_thread_should_quit && storage_task_queue.empty()) return; current = move(storage_task_queue.front()); storage_task_queue.pop(); } @@ -1772,7 +1708,7 @@ void H264Encoder::storage_task_thread() 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(); } @@ -1805,35 +1741,14 @@ 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(QSurface *surface, int width, int height, HTTPD *httpd) : current_storage_frame(0), surface(surface), httpd(httpd) { - AVCodec *codec_audio = avcodec_find_encoder(AV_CODEC_ID_MP3); + AVCodec *codec_audio = avcodec_find_encoder(AUDIO_OUTPUT_CODEC); context_audio = avcodec_alloc_context3(codec_audio); - context_audio->bit_rate = 256000; - context_audio->sample_rate = 48000; - context_audio->sample_fmt = AV_SAMPLE_FMT_FLTP; + 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}; @@ -1847,9 +1762,8 @@ H264Encoder::H264Encoder(QSurface *surface, int width, int height, HTTPD *httpd) 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(); setup_encode(); @@ -1861,35 +1775,36 @@ H264Encoder::H264Encoder(QSurface *surface, int width, int height, HTTPD *httpd) 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(&H264Encoder::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() { + { + unique_lock lock(frame_queue_mutex); + 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(); } - { - 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(); @@ -1899,7 +1814,7 @@ bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) { { // 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; } @@ -1959,106 +1874,139 @@ bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex) return true; } -void H264Encoder::add_audio(int64_t pts, std::vector audio) +void H264Encoder::add_audio(int64_t pts, vector audio) { { unique_lock lock(frame_queue_mutex); pending_audio_frames[pts] = move(audio); } - frame_queue_nonempty.notify_one(); + frame_queue_nonempty.notify_all(); } - -void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const std::vector &input_frames) +void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const vector &input_frames) { { unique_lock lock(frame_queue_mutex); pending_video_frames[current_storage_frame] = PendingFrame{ fence, input_frames, pts }; ++current_storage_frame; } - frame_queue_nonempty.notify_one(); + frame_queue_nonempty.notify_all(); } -void H264Encoder::copy_thread_func() +void H264Encoder::encode_thread_func() { int64_t last_dts = -1; - for ( ;; ) { + int gop_start_display_frame_num = 0; + for (int encoding_frame_num = 0; ; ++encoding_frame_num) { PendingFrame frame; int pts_lag; - encoding2display_order(current_frame_encoding, intra_period, intra_idr_period, ip_period, - ¤t_frame_display, ¤t_frame_type, &pts_lag); - if (current_frame_type == FRAME_IDR) { + 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_video_frames.count(current_frame_display) != 0; }); - if (copy_thread_should_quit) return; - frame = move(pending_video_frames[current_frame_display]); - pending_video_frames.erase(current_frame_display); - } - - // Wait for the GPU to be done with the frame. - glClientWaitSync(frame.fence.get(), 0, 0); - - // 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]; - eglDestroyImageKHR(eglGetCurrentDisplay(), surf->y_egl_image); - eglDestroyImageKHR(eglGetCurrentDisplay(), surf->cbcr_egl_image); - VAStatus 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"); - - VASurfaceID surface = surf->src_surface; - - // Schedule the frame for encoding. - va_status = vaBeginPicture(va_dpy, context_id, surface); - CHECK_VASTATUS(va_status, "vaBeginPicture"); - - if (current_frame_type == FRAME_IDR) { - render_sequence(); - render_picture(); - if (h264_packedheader) { - render_packedsequence(); - render_packedpicture(); + 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); } - } else { - //render_sequence(); - render_picture(); } - render_slice(); - - va_status = vaEndPicture(va_dpy, context_id); - CHECK_VASTATUS(va_status, "vaEndPicture"); - // Determine the pts and dts of this frame. - int64_t pts = frame.pts; + // Determine the dts of this frame. int64_t dts; if (pts_lag == -1) { assert(last_dts != -1); - dts = last_dts + (TIMEBASE / frame_rate); + dts = last_dts + (TIMEBASE / MAX_FPS); } else { - dts = pts - pts_lag; + dts = frame.pts - pts_lag; } last_dts = dts; - // 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 = current_frame_display; - tmp.frame_type = current_frame_type; - tmp.pts = pts; - tmp.dts = dts; - storage_task_enqueue(move(tmp)); - - update_ReferenceFrames(); - ++current_frame_encoding; + encode_frame(frame, encoding_frame_num, display_frame_num, gop_start_display_frame_num, frame_type, frame.pts, dts); } } + +void H264Encoder::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; + } +} + +void H264Encoder::encode_frame(H264Encoder::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. + glClientWaitSync(frame.fence.get(), 0, 0); + + // Release back any input frames we needed to render this frame. + frame.input_frames.clear(); + + // Unmap the image. + GLSurface *surf = &gl_surfaces[display_frame_num % SURFACE_NUM]; + eglDestroyImageKHR(eglGetCurrentDisplay(), surf->y_egl_image); + eglDestroyImageKHR(eglGetCurrentDisplay(), surf->cbcr_egl_image); + VAStatus 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"); + + VASurfaceID surface = surf->src_surface; + + // Schedule the frame for encoding. + va_status = vaBeginPicture(va_dpy, context_id, 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(); + } + } 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); +}