]> git.sesse.net Git - nageru/blobdiff - h264encode.cpp
Make the license blurb read-only.
[nageru] / h264encode.cpp
index a417c30fb8fbb5dbda794816d44d9d383790bc9a..02ada023b182879a0f0bf3fea15d3302dcad541b 100644 (file)
@@ -22,6 +22,7 @@ extern "C" {
 #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>
@@ -109,20 +110,21 @@ class H264EncoderImpl {
 public:
        H264EncoderImpl(QSurface *surface, int width, int height, HTTPD *httpd);
        ~H264EncoderImpl();
-       void add_audio(int64_t pts, std::vector<float> audio);  // Needs to come before end_frame() of same pts.
+       void add_audio(int64_t pts, vector<float> audio);
        bool begin_frame(GLuint *y_tex, GLuint *cbcr_tex);
-       void end_frame(RefCountedGLsync fence, int64_t pts, const std::vector<RefCountedFrame> &input_frames);
+       void end_frame(RefCountedGLsync fence, int64_t pts, const vector<RefCountedFrame> &input_frames);
+       void shutdown();
 
 private:
        struct storage_task {
                unsigned long long display_order;
                int frame_type;
-               std::vector<float> audio;
+               vector<float> audio;
                int64_t pts, dts;
        };
        struct PendingFrame {
                RefCountedGLsync fence;
-               std::vector<RefCountedFrame> input_frames;
+               vector<RefCountedFrame> input_frames;
                int64_t pts;
        };
 
@@ -154,22 +156,24 @@ private:
        void update_ReferenceFrames(int frame_type);
        int update_RefPicList(int frame_type);
 
-       std::thread encode_thread, storage_thread;
+       bool is_shutdown = false;
 
-       std::mutex storage_task_queue_mutex;
-       std::condition_variable storage_task_queue_changed;
+       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
-       std::queue<storage_task> storage_task_queue;  // 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
 
-       std::mutex frame_queue_mutex;
-       std::condition_variable frame_queue_nonempty;
+       mutex frame_queue_mutex;
+       condition_variable frame_queue_nonempty;
        bool encode_thread_should_quit = false;  // under frame_queue_mutex
 
        int current_storage_frame;
 
-       std::map<int, PendingFrame> pending_video_frames;  // under frame_queue_mutex
-       std::map<int64_t, std::vector<float>> pending_audio_frames;  // under frame_queue_mutex
+       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;
@@ -228,7 +232,6 @@ private:
        int frame_height_mbaligned;
 };
 
-
 // Supposedly vaRenderPicture() is supposed to destroy the buffer implicitly,
 // but if we don't delete it here, we get leaks. The GStreamer implementation
 // does the same.
@@ -824,149 +827,6 @@ static const char *rc_to_string(int rc_mode)
     }
 }
 
-#if 0
-static int process_cmdline(int argc, char *argv[])
-{
-    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 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 * MAX_FPS / 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;
-}
-#endif
-
 VADisplay H264EncoderImpl::va_open_display(void)
 {
     x11_display = XOpenDisplay(NULL);
@@ -1210,68 +1070,15 @@ int H264EncoderImpl::setup_encode()
     return 0;
 }
 
-
-
-#define partition(ref, field, key, ascending)   \
-    while (i <= j) {                            \
-        if (ascending) {                        \
-            while (ref[i].field < key)          \
-                i++;                            \
-            while (ref[j].field > key)          \
-                j--;                            \
-        } else {                                \
-            while (ref[i].field > key)          \
-                i++;                            \
-            while (ref[j].field < key)          \
-                j--;                            \
-        }                                       \
-        if (i <= j) {                           \
-            tmp = ref[i];                       \
-            ref[i] = ref[j];                    \
-            ref[j] = tmp;                       \
-            i++;                                \
-            j--;                                \
-        }                                       \
-    }                                           \
-
-static void sort_one(VAPictureH264 ref[], int left, int right,
-                     int ascending, int frame_idx)
-{
-    int i = left, j = right;
-    unsigned int key;
-    VAPictureH264 tmp;
-
-    if (frame_idx) {
-        key = ref[(left + right) / 2].frame_idx;
-        partition(ref, frame_idx, key, ascending);
-    } else {
-        key = ref[(left + right) / 2].TopFieldOrderCnt;
-        partition(ref, TopFieldOrderCnt, (signed int)key, ascending);
-    }
-    
-    /* recursion */
-    if (left < j)
-        sort_one(ref, left, j, ascending, frame_idx);
-    
-    if (i < right)
-        sort_one(ref, i, right, ascending, frame_idx);
-}
-
-static void sort_two(VAPictureH264 ref[], int left, int right, unsigned int key, unsigned int frame_idx,
-                     int partition_ascending, int list0_ascending, int list1_ascending)
+// Given a list like 1 9 3 0 2 8 4 and a pivot element 3, will produce
+//
+//   2 1 0 [3] 4 8 9
+template<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);
 }
 
 void H264EncoderImpl::update_ReferenceFrames(int frame_type)
@@ -1297,21 +1104,25 @@ void H264EncoderImpl::update_ReferenceFrames(int frame_type)
 
 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 (frame_type == FRAME_P) {
         memcpy(RefPicList0_P, ReferenceFrames, numShortTerm * sizeof(VAPictureH264));
-        sort_one(RefPicList0_P, 0, numShortTerm-1, 0, 1);
-    }
-    
-    if (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;
@@ -1841,26 +1652,12 @@ H264EncoderImpl::H264EncoderImpl(QSurface *surface, int width, int height, HTTPD
 
 H264EncoderImpl::~H264EncoderImpl()
 {
-       {
-               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();
+       shutdown();
 }
 
 bool H264EncoderImpl::begin_frame(GLuint *y_tex, GLuint *cbcr_tex)
 {
+       assert(!is_shutdown);
        {
                // Wait until this frame slot is done encoding.
                unique_lock<mutex> lock(storage_task_queue_mutex);
@@ -1925,6 +1722,7 @@ bool H264EncoderImpl::begin_frame(GLuint *y_tex, GLuint *cbcr_tex)
 
 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);
@@ -1934,6 +1732,7 @@ void H264EncoderImpl::add_audio(int64_t pts, vector<float> audio)
 
 void H264EncoderImpl::end_frame(RefCountedGLsync fence, int64_t pts, const vector<RefCountedFrame> &input_frames)
 {
+       assert(!is_shutdown);
        {
                unique_lock<mutex> lock(frame_queue_mutex);
                pending_video_frames[current_storage_frame] = PendingFrame{ fence, input_frames, pts };
@@ -1942,6 +1741,31 @@ void H264EncoderImpl::end_frame(RefCountedGLsync fence, int64_t pts, const vecto
        frame_queue_nonempty.notify_all();
 }
 
+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::encode_thread_func()
 {
        int64_t last_dts = -1;
@@ -2067,7 +1891,7 @@ H264Encoder::H264Encoder(QSurface *surface, int width, int height, HTTPD *httpd)
 // Must be defined here because unique_ptr<> destructor needs to know the impl.
 H264Encoder::~H264Encoder() {}
 
-void H264Encoder::add_audio(int64_t pts, std::vector<float> audio)
+void H264Encoder::add_audio(int64_t pts, vector<float> audio)
 {
        impl->add_audio(pts, audio);
 }
@@ -2077,9 +1901,14 @@ bool H264Encoder::begin_frame(GLuint *y_tex, GLuint *cbcr_tex)
        return impl->begin_frame(y_tex, cbcr_tex);
 }
 
-void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const std::vector<RefCountedFrame> &input_frames)
+void H264Encoder::end_frame(RefCountedGLsync fence, int64_t pts, const vector<RefCountedFrame> &input_frames)
 {
        impl->end_frame(fence, pts, input_frames);
 }
 
+void H264Encoder::shutdown()
+{
+       impl->shutdown();
+}
+
 // Real class.