//#include "sysdeps.h"
#include "h264encode.h"
+#include <movit/util.h>
#include <EGL/eglplatform.h>
#include <X11/X.h>
#include <X11/Xlib.h>
#include <assert.h>
#include <epoxy/egl.h>
+extern "C" {
#include <libavcodec/avcodec.h>
-#include <libavformat/avio.h>
-#include <libavutil/mathematics.h>
+#include <libavformat/avformat.h>
+#include <libavutil/channel_layout.h>
+#include <libavutil/frame.h>
#include <libavutil/rational.h>
+#include <libavutil/samplefmt.h>
+}
#include <libdrm/drm_fourcc.h>
-#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <fcntl.h>
#include <va/va.h>
+#include <va/va_drm.h>
#include <va/va_drmcommon.h>
#include <va/va_enc_h264.h>
#include <va/va_x11.h>
+#include <algorithm>
#include <condition_variable>
+#include <cstdint>
+#include <map>
+#include <memory>
#include <mutex>
#include <queue>
#include <string>
#include <thread>
+#include <utility>
#include "context.h"
+#include "defs.h"
+#include "flags.h"
+#include "httpd.h"
#include "timebase.h"
+using namespace std;
+
class QOpenGLContext;
class QSurface;
exit(1); \
}
+#define BUFFER_OFFSET(i) ((char *)NULL + (i))
+
//#include "loadsurface.h"
#define NAL_REF_IDC_NONE 0
#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 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 = 3;
-static int rc_mode = -1;
-static int rc_default_modes[] = {
+#define MAX_NUM_REF1 16 // Seemingly a hardware-fixed value, not related to SURFACE_NUM
+#define MAX_NUM_REF2 32 // Seemingly a hardware-fixed value, not related to SURFACE_NUM
+
+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,
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
using namespace std;
+// H.264 video comes out in encoding order (e.g. with two B-frames:
+// 0, 3, 1, 2, 6, 4, 5, etc.), but uncompressed video needs to
+// come in the right order. Since we do everything, including waiting
+// for the frames to come out of OpenGL, in encoding order, we need
+// a reordering buffer for uncompressed frames so that they come out
+// correctly. We go the super-lazy way of not making it understand
+// anything about the true order (which introduces some extra latency,
+// though); we know that for N B-frames we need at most (N-1) frames
+// in the reorder buffer, and can just sort on that.
+//
+// The class also deals with keeping a freelist as needed.
+class FrameReorderer {
+public:
+ FrameReorderer(unsigned queue_length, int width, int height);
+
+ // Returns the next frame to insert with its pts, if any. Otherwise -1 and nullptr.
+ // Does _not_ take ownership of data; a copy is taken if needed.
+ // The returned pointer is valid until the next call to reorder_frame, or destruction.
+ // As a special case, if queue_length == 0, will just return pts and data (no reordering needed).
+ pair<int64_t, const uint8_t *> reorder_frame(int64_t pts, const uint8_t *data);
+
+ // The same as reorder_frame, but without inserting anything. Used to empty the queue.
+ pair<int64_t, const uint8_t *> get_first_frame();
+
+ bool empty() const { return frames.empty(); }
+
+private:
+ unsigned queue_length;
+ int width, height;
+
+ priority_queue<pair<int64_t, uint8_t *>> frames;
+ stack<uint8_t *> freelist; // Includes the last value returned from reorder_frame.
+
+ // Owns all the pointers. Normally, freelist and frames could do this themselves,
+ // except priority_queue doesn't work well with movable-only types.
+ vector<unique_ptr<uint8_t[]>> owner;
+};
+
+FrameReorderer::FrameReorderer(unsigned queue_length, int width, int height)
+ : queue_length(queue_length), width(width), height(height)
+{
+ for (unsigned i = 0; i < queue_length; ++i) {
+ owner.emplace_back(new uint8_t[width * height * 2]);
+ freelist.push(owner.back().get());
+ }
+}
+
+pair<int64_t, const uint8_t *> FrameReorderer::reorder_frame(int64_t pts, const uint8_t *data)
+{
+ if (queue_length == 0) {
+ return make_pair(pts, data);
+ }
+
+ assert(!freelist.empty());
+ uint8_t *storage = freelist.top();
+ freelist.pop();
+ memcpy(storage, data, width * height * 2);
+ frames.emplace(-pts, storage); // Invert pts to get smallest first.
+
+ if (frames.size() >= queue_length) {
+ return get_first_frame();
+ } else {
+ return make_pair(-1, nullptr);
+ }
+}
+
+pair<int64_t, const uint8_t *> FrameReorderer::get_first_frame()
+{
+ assert(!frames.empty());
+ pair<int64_t, uint8_t *> storage = frames.top();
+ frames.pop();
+ int64_t pts = storage.first;
+ freelist.push(storage.second);
+ return make_pair(-pts, storage.second); // Re-invert pts (see reorder_frame()).
+}
+
+class H264EncoderImpl {
+public:
+ H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd);
+ ~H264EncoderImpl();
+ void add_audio(int64_t pts, vector<float> audio);
+ bool begin_frame(GLuint *y_tex, GLuint *cbcr_tex);
+ RefCountedGLsync end_frame(int64_t pts, const vector<RefCountedFrame> &input_frames);
+ void shutdown();
+ void open_output_file(const std::string &filename);
+ void close_output_file();
+
+private:
+ struct storage_task {
+ unsigned long long display_order;
+ int frame_type;
+ vector<float> audio;
+ int64_t pts, dts;
+ };
+ struct PendingFrame {
+ RefCountedGLsync fence;
+ vector<RefCountedFrame> input_frames;
+ int64_t pts;
+ };
+
+ // So we never get negative dts.
+ int64_t global_delay() const {
+ return int64_t(ip_period - 1) * (TIMEBASE / MAX_FPS);
+ }
+
+ 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 add_packet_for_uncompressed_frame(int64_t pts, const uint8_t *data);
+ 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 encode_audio(const vector<float> &audio,
+ vector<float> *audio_queue,
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations);
+ void encode_audio_one_frame(const float *audio,
+ size_t num_samples, // In each channel.
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations);
+ 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();
+ void enable_zerocopy_if_possible();
+ 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> 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<int, PendingFrame> pending_video_frames; // under frame_queue_mutex
+ map<int64_t, vector<float>> pending_audio_frames; // under frame_queue_mutex
+ QSurface *surface;
+
+ AVCodecContext *context_audio_file;
+ AVCodecContext *context_audio_stream = nullptr; // nullptr = don't code separate audio for stream.
+
+ vector<float> audio_queue_file;
+ vector<float> audio_queue_stream;
+
+ AVFrame *audio_frame = nullptr;
+ HTTPD *httpd;
+ unique_ptr<FrameReorderer> reorderer;
+
+ 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[MAX_NUM_REF1], RefPicList0_P[MAX_NUM_REF2], RefPicList0_B[MAX_NUM_REF2], RefPicList1_B[MAX_NUM_REF2];
+
+ // 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;
+
+ unique_ptr<Mux> file_mux; // To local disk.
+};
+
+// 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)
{
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) {
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;
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 */
{
}
-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 */
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;
}
}
-static int
-build_packed_pic_buffer(unsigned char **header_buffer)
+int H264EncoderImpl::build_packed_pic_buffer(unsigned char **header_buffer)
{
bitstream bs;
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;
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;
//
// 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_
-// <frame_rate>, 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
#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;
*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;
}
-static const char *rc_to_string(int rcmode)
+static const char *rc_to_string(int rc_mode)
{
switch (rc_mode) {
case VA_RC_NONE:
}
}
-#if 0
-static int process_cmdline(int argc, char *argv[])
+void H264EncoderImpl::enable_zerocopy_if_possible()
{
- 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");
- }
-
-
- 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;
+ if (global_flags.uncompressed_video_to_http) {
+ fprintf(stderr, "Disabling zerocopy H.264 encoding due to --uncompressed_video_to_http.\n");
+ use_zerocopy = false;
+ } else {
+ use_zerocopy = true;
+ }
}
-#endif
-
-static Display *x11_display;
-static Window x11_window;
-VADisplay
-va_open_display(void)
+VADisplay H264EncoderImpl::va_open_display(const string &va_display)
{
- 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;
+ }
+ enable_zerocopy_if_possible();
+ 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;
+ }
+ enable_zerocopy_if_possible();
+ 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;
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");
}
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 */
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));
break;
}
}
-
- printf("RateControl mode: %s\n", rc_to_string(rc_mode));
}
config_attrib[config_attrib_num].type = VAConfigAttribRateControl;
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;
}
}
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++;
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;
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++) {
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<class T, class C>
+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++;
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*/;
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;
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;
}
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 < MAX_NUM_REF1; 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];
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);
}
-static int render_packedpicture(void)
+int H264EncoderImpl::render_packedpicture()
{
VAEncPackedHeaderParameterBuffer packedheader_param_buffer;
VABufferID packedpic_para_bufid, packedpic_data_bufid, render_id[2];
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];
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));
- for (i = refpiclist0_max; i < 32; i++) {
+ for (i = refpiclist0_max; i < MAX_NUM_REF2; i++) {
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;
memcpy(slice_param.RefPicList0, RefPicList0_B, refpiclist0_max*sizeof(VAPictureH264));
- for (i = refpiclist0_max; i < 32; i++) {
+ for (i = refpiclist0_max; i < MAX_NUM_REF2; i++) {
slice_param.RefPicList0[i].picture_id = VA_INVALID_SURFACE;
slice_param.RefPicList0[i].flags = VA_PICTURE_H264_INVALID;
}
memcpy(slice_param.RefPicList1, RefPicList1_B, refpiclist1_max*sizeof(VAPictureH264));
- for (i = refpiclist1_max; i < 32; i++) {
+ for (i = refpiclist1_max; i < MAX_NUM_REF2; i++) {
slice_param.RefPicList1[i].picture_id = VA_INVALID_SURFACE;
slice_param.RefPicList1[i].flags = VA_PICTURE_H264_INVALID;
}
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 &&
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 H264EncoderImpl::save_codeddata(storage_task task)
{
- VACodedBufferSegment *buf_list = NULL;
- VAStatus va_status;
- unsigned int coded_size = 0;
+ VACodedBufferSegment *buf_list = NULL;
+ VAStatus va_status;
- string data;
+ string data;
- const int64_t global_delay = (ip_period - 1) * (TIMEBASE / frame_rate); // 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<const char *>(buf_list->buf), buf_list->size);
+ buf_list = (VACodedBufferSegment *) buf_list->next;
+ }
+ vaUnmapBuffer(va_dpy, gl_surfaces[task.display_order % SURFACE_NUM].coded_buf);
- 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<const char *>(buf_list->buf), buf_list->size);
- buf_list = (VACodedBufferSegment *) buf_list->next;
+ {
+ // Add video.
+ AVPacket pkt;
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.buf = nullptr;
+ pkt.data = reinterpret_cast<uint8_t *>(&data[0]);
+ pkt.size = data.size();
+ pkt.stream_index = 0;
+ if (task.frame_type == FRAME_IDR) {
+ pkt.flags = AV_PKT_FLAG_KEY;
+ } else {
+ pkt.flags = 0;
+ }
+ //pkt.duration = 1;
+ if (file_mux) {
+ file_mux->add_packet(pkt, task.pts + global_delay(), task.dts + global_delay());
+ }
+ if (!global_flags.uncompressed_video_to_http) {
+ 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<float> audio;
+ {
+ unique_lock<mutex> 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);
+ }
- frame_size += coded_size;
- }
- 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.pts = av_rescale_q(task.pts + global_delay, AVRational{1, TIMEBASE}, avstream_video->time_base);
- pkt.dts = av_rescale_q(task.dts + global_delay, AVRational{1, TIMEBASE}, avstream_video->time_base);
- pkt.data = reinterpret_cast<uint8_t *>(&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;
- av_interleaved_write_frame(avctx, &pkt);
- }
- // 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<float> audio;
- {
- unique_lock<mutex> lock(frame_queue_mutex);
- if (pending_audio_frames.empty()) break;
- 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_FLT;
- frame->channel_layout = AV_CH_LAYOUT_STEREO;
-
- unique_ptr<float[]> 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];
- }
+ if (context_audio_stream) {
+ encode_audio(audio, &audio_queue_file, audio_pts, context_audio_file, { file_mux.get() });
+ encode_audio(audio, &audio_queue_stream, audio_pts, context_audio_stream, { httpd });
+ } else {
+ encode_audio(audio, &audio_queue_file, audio_pts, context_audio_file, { httpd, file_mux.get() });
+ }
- AVPacket pkt;
- av_init_packet(&pkt);
- pkt.data = nullptr;
- pkt.size = 0;
- int got_output;
- avcodec_encode_audio2(avstream_audio->codec, &pkt, frame, &got_output);
- if (got_output) {
- pkt.pts = av_rescale_q(audio_pts + global_delay, AVRational{1, TIMEBASE}, avstream_audio->time_base);
- pkt.dts = pkt.pts;
- pkt.stream_index = 1;
- av_interleaved_write_frame(avctx, &pkt);
- }
- // TODO: Delayed frames.
- avcodec_free_frame(&frame);
- }
+ if (audio_pts == task.pts) break;
+ }
+}
-#if 0
- printf("\r "); /* return back to startpoint */
- switch (encode_order % 4) {
- case 0:
- printf("|");
- break;
- case 1:
- printf("/");
- break;
- case 2:
- printf("-");
- break;
- case 3:
- printf("\\");
- break;
- }
- printf("%08lld", encode_order);
- printf("(%06d bytes coded)", coded_size);
-#endif
+void H264EncoderImpl::encode_audio(
+ const vector<float> &audio,
+ vector<float> *audio_queue,
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations)
+{
+ if (ctx->frame_size == 0) {
+ // No queueing needed.
+ assert(audio_queue->empty());
+ assert(audio.size() % 2 == 0);
+ encode_audio_one_frame(&audio[0], audio.size() / 2, audio_pts, ctx, destinations);
+ return;
+ }
- return 0;
+ audio_queue->insert(audio_queue->end(), audio.begin(), audio.end());
+ size_t sample_num;
+ for (sample_num = 0;
+ sample_num + ctx->frame_size * 2 <= audio_queue->size();
+ sample_num += ctx->frame_size * 2) {
+ encode_audio_one_frame(&(*audio_queue)[sample_num],
+ ctx->frame_size,
+ audio_pts,
+ ctx,
+ destinations);
+ }
+ audio_queue->erase(audio_queue->begin(), audio_queue->begin() + sample_num);
}
+void H264EncoderImpl::encode_audio_one_frame(
+ const float *audio,
+ size_t num_samples,
+ int64_t audio_pts,
+ AVCodecContext *ctx,
+ const vector<PacketDestination *> &destinations)
+{
+ audio_frame->nb_samples = num_samples;
+ audio_frame->channel_layout = AV_CH_LAYOUT_STEREO;
+
+ unique_ptr<float[]> planar_samples;
+ unique_ptr<int32_t[]> int_samples;
+
+ if (ctx->sample_fmt == AV_SAMPLE_FMT_FLTP) {
+ audio_frame->format = AV_SAMPLE_FMT_FLTP;
+ planar_samples.reset(new float[num_samples * 2]);
+ avcodec_fill_audio_frame(audio_frame, 2, AV_SAMPLE_FMT_FLTP, (const uint8_t*)planar_samples.get(), num_samples * 2 * sizeof(float), 0);
+ for (size_t i = 0; i < num_samples; ++i) {
+ planar_samples[i] = audio[i * 2 + 0];
+ planar_samples[i + num_samples] = audio[i * 2 + 1];
+ }
+ } else {
+ assert(ctx->sample_fmt == AV_SAMPLE_FMT_S32);
+ int_samples.reset(new int32_t[num_samples * 2]);
+ int ret = avcodec_fill_audio_frame(audio_frame, 2, AV_SAMPLE_FMT_S32, (const uint8_t*)int_samples.get(), num_samples * 2 * sizeof(int32_t), 1);
+ if (ret < 0) {
+ fprintf(stderr, "avcodec_fill_audio_frame() failed with %d\n", ret);
+ exit(1);
+ }
+ for (size_t i = 0; i < num_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 = 0;
+ avcodec_encode_audio2(ctx, &pkt, audio_frame, &got_output);
+ if (got_output) {
+ pkt.stream_index = 1;
+ pkt.flags = AV_PKT_FLAG_KEY;
+ for (PacketDestination *dest : destinations) {
+ dest->add_packet(pkt, audio_pts + global_delay(), audio_pts + global_delay());
+ }
+ }
+ // TODO: Delayed frames.
+ av_frame_unref(audio_frame);
+ av_free_packet(&pkt);
+}
// this is weird. but it seems to put a new frame onto the queue
-void H264Encoder::storage_task_enqueue(storage_task task)
+void H264EncoderImpl::storage_task_enqueue(storage_task task)
{
- std::unique_lock<std::mutex> lock(storage_task_queue_mutex);
+ unique_lock<mutex> 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<std::mutex> lock(storage_task_queue_mutex);
+ unique_lock<mutex> 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();
}
save_codeddata(move(current));
{
- std::unique_lock<std::mutex> lock(storage_task_queue_mutex);
+ unique_lock<mutex> 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);
return 0;
}
+namespace {
-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)
+void init_audio_encoder(const string &codec_name, int bit_rate, AVCodecContext **ctx)
{
- 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);
+ AVCodec *codec_audio = avcodec_find_encoder_by_name(codec_name.c_str());
+ if (codec_audio == nullptr) {
+ fprintf(stderr, "ERROR: Could not find codec '%s'\n", codec_name.c_str());
exit(1);
}
- AVCodec *codec_video = avcodec_find_encoder(AV_CODEC_ID_H264);
- avstream_video = avformat_new_stream(avctx, codec_video);
- if (avstream_video == nullptr) {
- fprintf(stderr, "%s: avformat_new_stream() failed\n", output_filename);
- exit(1);
+
+ AVCodecContext *context_audio = avcodec_alloc_context3(codec_audio);
+ context_audio->bit_rate = bit_rate;
+ context_audio->sample_rate = OUTPUT_FREQUENCY;
+
+ // Choose sample format; we currently only support these two
+ // (see encode_audio), so we're a bit picky.
+ const AVSampleFormat *ptr = codec_audio->sample_fmts;
+ for ( ; *ptr != -1; ++ptr) {
+ if (*ptr == AV_SAMPLE_FMT_FLTP || *ptr == AV_SAMPLE_FMT_S32) {
+ context_audio->sample_fmt = *ptr;
+ break;
+ }
}
- avstream_video->time_base = AVRational{1, TIMEBASE};
- avstream_video->codec->width = width;
- avstream_video->codec->height = height;
- avstream_video->codec->time_base = AVRational{1, TIMEBASE};
- avstream_video->codec->ticks_per_frame = 1; // or 2?
-
- AVCodec *codec_audio = avcodec_find_encoder(AV_CODEC_ID_MP3);
- avstream_audio = avformat_new_stream(avctx, codec_audio);
- if (avstream_audio == nullptr) {
- fprintf(stderr, "%s: avformat_new_stream() failed\n", output_filename);
+ if (*ptr == -1) {
+ fprintf(stderr, "ERROR: Audio codec does not support fltp or s32 sample formats\n");
exit(1);
}
- avstream_audio->time_base = AVRational{1, TIMEBASE};
- avstream_audio->codec->bit_rate = 256000;
- avstream_audio->codec->sample_rate = 48000;
- avstream_audio->codec->sample_fmt = AV_SAMPLE_FMT_FLTP;
- avstream_audio->codec->channels = 2;
- avstream_audio->codec->channel_layout = AV_CH_LAYOUT_STEREO;
- avstream_audio->codec->time_base = AVRational{1, TIMEBASE};
-
- /* open it */
- if (avcodec_open2(avstream_audio->codec, codec_audio, NULL) < 0) {
- fprintf(stderr, "Could not open codec\n");
+
+ 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 '%s'\n", codec_name.c_str());
exit(1);
}
- if (avformat_write_header(avctx, NULL) < 0) {
- fprintf(stderr, "%s: avformat_write_header() failed\n", output_filename);
- exit(1);
+ *ctx = context_audio;
+}
+
+} // namespace
+
+H264EncoderImpl::H264EncoderImpl(QSurface *surface, const string &va_display, int width, int height, HTTPD *httpd)
+ : current_storage_frame(0), surface(surface), httpd(httpd)
+{
+ init_audio_encoder(AUDIO_OUTPUT_CODEC_NAME, DEFAULT_AUDIO_OUTPUT_BIT_RATE, &context_audio_file);
+
+ if (!global_flags.stream_audio_codec_name.empty()) {
+ init_audio_encoder(global_flags.stream_audio_codec_name,
+ global_flags.stream_audio_codec_bitrate, &context_audio_stream);
}
+ audio_frame = av_frame_alloc();
+
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();
+
+ if (global_flags.uncompressed_video_to_http) {
+ reorderer.reset(new FrameReorderer(ip_period - 1, frame_width, frame_height));
+ }
- init_va();
+ init_va(va_display);
setup_encode();
// No frames are ready yet.
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<mutex> lock(storage_task_queue_mutex);
- storage_thread_should_quit = true;
- storage_task_queue_changed.notify_all();
- }
- {
- unique_lock<mutex> lock(frame_queue_mutex);
- copy_thread_should_quit = true;
- frame_queue_nonempty.notify_one();
- }
- storage_thread.join();
- copy_thread.join();
-
- release_encode();
- deinit_va();
+ shutdown();
+ av_frame_free(&audio_frame);
- av_write_trailer(avctx);
- avformat_free_context(avctx);
+ // TODO: Destroy context.
}
-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<std::mutex> lock(storage_task_queue_mutex);
+ unique_lock<mutex> lock(storage_task_queue_mutex);
+ if (srcsurface_status[current_storage_frame % SURFACE_NUM] != SRC_SURFACE_FREE) {
+ fprintf(stderr, "Warning: Slot %d (for frame %d) is still encoding, rendering has to wait for H.264 encoder\n",
+ current_storage_frame % SURFACE_NUM, current_storage_frame);
+ }
storage_task_queue_changed.wait(lock, [this]{ return storage_thread_should_quit || (srcsurface_status[current_storage_frame % SURFACE_NUM] == SRC_SURFACE_FREE); });
+ srcsurface_status[current_storage_frame % SURFACE_NUM] = SRC_SURFACE_IN_ENCODING;
if (storage_thread_should_quit) return false;
}
*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
- };
-
- 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);
+ 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);
+ }
return true;
}
-void H264Encoder::add_audio(int64_t pts, std::vector<float> audio)
+void H264EncoderImpl::add_audio(int64_t pts, vector<float> audio)
{
+ assert(!is_shutdown);
{
unique_lock<mutex> 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<RefCountedFrame> &input_frames)
+RefCountedGLsync H264EncoderImpl::end_frame(int64_t pts, const vector<RefCountedFrame> &input_frames)
{
+ assert(!is_shutdown);
+
+ if (!use_zerocopy) {
+ GLSurface *surf = &gl_surfaces[current_storage_frame % SURFACE_NUM];
+
+ 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();
+ }
+
+ RefCountedGLsync fence = RefCountedGLsync(GL_SYNC_GPU_COMMANDS_COMPLETE, /*flags=*/0);
+ check_error();
+ glFlush(); // Make the H.264 thread see the fence as soon as possible.
+ check_error();
+
{
unique_lock<mutex> 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();
+ return fence;
}
-void H264Encoder::copy_thread_func()
+void H264EncoderImpl::shutdown()
+{
+ if (is_shutdown) {
+ return;
+ }
+
+ {
+ unique_lock<mutex> lock(frame_queue_mutex);
+ encode_thread_should_quit = true;
+ frame_queue_nonempty.notify_all();
+ }
+ encode_thread.join();
+ {
+ unique_lock<mutex> lock(storage_task_queue_mutex);
+ storage_thread_should_quit = true;
+ frame_queue_nonempty.notify_all();
+ storage_task_queue_changed.notify_all();
+ }
+ storage_thread.join();
+
+ release_encode();
+ deinit_va();
+ is_shutdown = true;
+}
+
+void H264EncoderImpl::open_output_file(const std::string &filename)
+{
+ AVFormatContext *avctx = avformat_alloc_context();
+ avctx->oformat = av_guess_format(NULL, filename.c_str(), NULL);
+ assert(filename.size() < sizeof(avctx->filename) - 1);
+ strcpy(avctx->filename, filename.c_str());
+
+ string url = "file:" + filename;
+ int ret = avio_open2(&avctx->pb, url.c_str(), AVIO_FLAG_WRITE, &avctx->interrupt_callback, NULL);
+ if (ret < 0) {
+ char tmp[AV_ERROR_MAX_STRING_SIZE];
+ fprintf(stderr, "%s: avio_open2() failed: %s\n", filename.c_str(), av_make_error_string(tmp, sizeof(tmp), ret));
+ exit(1);
+ }
+
+ file_mux.reset(new Mux(avctx, frame_width, frame_height, Mux::CODEC_H264, TIMEBASE, DEFAULT_AUDIO_OUTPUT_BIT_RATE));
+}
+
+void H264EncoderImpl::close_output_file()
+{
+ file_mux.reset();
+}
+
+void H264EncoderImpl::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<mutex> 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);
+ 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;
+ }
+
+ if (global_flags.uncompressed_video_to_http) {
+ // Add frames left in reorderer.
+ while (!reorderer->empty()) {
+ pair<int64_t, const uint8_t *> output_frame = reorderer->get_first_frame();
+ add_packet_for_uncompressed_frame(output_frame.first, output_frame.second);
}
+ }
+}
- // Wait for the GPU to be done with the frame.
- glClientWaitSync(frame.fence.get(), 0, 0);
+void H264EncoderImpl::add_packet_for_uncompressed_frame(int64_t pts, const uint8_t *data)
+{
+ AVPacket pkt;
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.buf = nullptr;
+ pkt.data = const_cast<uint8_t *>(data);
+ pkt.size = frame_width * frame_height * 2;
+ pkt.stream_index = 0;
+ pkt.flags = AV_PKT_FLAG_KEY;
+ httpd->add_packet(pkt, pts, pts);
+}
- // Release back any input frames we needed to render this frame.
- frame.input_frames.clear();
+namespace {
- // Unmap the image.
- GLSurface *surf = &gl_surfaces[current_frame_display % SURFACE_NUM];
+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);
+
+ // Release back any input frames we needed to render this frame.
+ frame.input_frames.clear();
+
+ 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);
+
+ 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);
- VASurfaceID surface = surf->src_surface;
+ 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);
- // Schedule the frame for encoding.
- va_status = vaBeginPicture(va_dpy, context_id, surface);
- CHECK_VASTATUS(va_status, "vaBeginPicture");
+ va_status = vaUnmapBuffer(va_dpy, surf->surface_image.buf);
+ CHECK_VASTATUS(va_status, "vaUnmapBuffer");
- if (current_frame_type == FRAME_IDR) {
- render_sequence();
- render_picture();
- if (h264_packedheader) {
- render_packedsequence();
- render_packedpicture();
+ if (global_flags.uncompressed_video_to_http) {
+ // Add uncompressed video. (Note that pts == dts here.)
+ // Delay needs to match audio.
+ pair<int64_t, const uint8_t *> output_frame = reorderer->reorder_frame(pts + global_delay(), reinterpret_cast<uint8_t *>(surf->y_ptr));
+ if (output_frame.second != nullptr) {
+ add_packet_for_uncompressed_frame(output_frame.first, output_frame.second);
}
- } 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;
- int64_t dts;
- if (pts_lag == -1) {
- assert(last_dts != -1);
- dts = last_dts + (TIMEBASE / frame_rate);
- } else {
- dts = pts - pts_lag;
- }
- last_dts = dts;
+ va_status = vaDestroyImage(va_dpy, surf->surface_image.image_id);
+ CHECK_VASTATUS(va_status, "vaDestroyImage");
- // 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;
+ // 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();
+ }
+ } 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<float> audio)
+{
+ impl->add_audio(pts, audio);
+}
+
+bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex)
+{
+ return impl->begin_frame(y_tex, cbcr_tex);
+}
+
+RefCountedGLsync H264Encoder::end_frame(int64_t pts, const vector<RefCountedFrame> &input_frames)
+{
+ return impl->end_frame(pts, input_frames);
+}
+
+void H264Encoder::shutdown()
+{
+ impl->shutdown();
+}
+
+void H264Encoder::open_output_file(const std::string &filename)
+{
+ impl->open_output_file(filename);
+}
+
+void H264Encoder::close_output_file()
+{
+ impl->close_output_file();
}
projects / nageru / blobdiff