]> git.sesse.net Git - nageru/blobdiff - jpeg_frame_view.cpp
Change from operating point 3 to 2 (more laptop-friendly debugging).
[nageru] / jpeg_frame_view.cpp
index 65fccbb0102e6877fa121ed7f9df240838c53a3d..5ef2f0b8ab3a66fb661ebb384eb2717420c5423c 100644 (file)
 
 #include "defs.h"
 #include "post_to_main_thread.h"
+#include "video_stream.h"
 
 using namespace movit;
 using namespace std;
 
-string filename_for_frame(unsigned stream_idx, int64_t pts);
-
-struct JPEGID {
-       unsigned stream_idx;
-       int64_t pts;
-};
 bool operator< (const JPEGID &a, const JPEGID &b) {
        return make_pair(a.stream_idx, a.pts) < make_pair(b.stream_idx, b.pts);
 }
@@ -61,46 +56,63 @@ shared_ptr<Frame> decode_jpeg(const string &filename)
 
        jpeg_read_header(&dinfo, true);
 
-       if (dinfo.num_components != 3 ||
-            dinfo.comp_info[0].h_samp_factor != 2 ||
-            dinfo.comp_info[0].v_samp_factor != 2 ||
-            dinfo.comp_info[1].h_samp_factor != 1 ||
-            dinfo.comp_info[1].v_samp_factor != 2 ||
-            dinfo.comp_info[2].h_samp_factor != 1 ||
-            dinfo.comp_info[2].v_samp_factor != 2) {
-               fprintf(stderr, "Not 4:2:2 JPEG! (%d components, Y=%dx%d, Cb=%dx%d, Cr=%dx%d)\n",
+       if (dinfo.num_components != 3) {
+               fprintf(stderr, "Not a color JPEG. (%d components, Y=%dx%d, Cb=%dx%d, Cr=%dx%d)\n",
                        dinfo.num_components,
                        dinfo.comp_info[0].h_samp_factor, dinfo.comp_info[0].v_samp_factor,
                        dinfo.comp_info[1].h_samp_factor, dinfo.comp_info[1].v_samp_factor,
                        dinfo.comp_info[2].h_samp_factor, dinfo.comp_info[2].v_samp_factor);
                exit(1);
        }
+       if (dinfo.comp_info[0].h_samp_factor != dinfo.max_h_samp_factor ||
+           dinfo.comp_info[0].v_samp_factor != dinfo.max_v_samp_factor ||  // Y' must not be subsampled.
+           dinfo.comp_info[1].h_samp_factor != dinfo.comp_info[2].h_samp_factor ||
+           dinfo.comp_info[1].v_samp_factor != dinfo.comp_info[2].v_samp_factor ||  // Cb and Cr must be identically subsampled.
+           (dinfo.max_h_samp_factor % dinfo.comp_info[1].h_samp_factor) != 0 ||
+           (dinfo.max_v_samp_factor % dinfo.comp_info[1].v_samp_factor) != 0) {  // No 2:3 subsampling or other weirdness.
+               fprintf(stderr, "Unsupported subsampling scheme. (Y=%dx%d, Cb=%dx%d, Cr=%dx%d)\n",
+                       dinfo.comp_info[0].h_samp_factor, dinfo.comp_info[0].v_samp_factor,
+                       dinfo.comp_info[1].h_samp_factor, dinfo.comp_info[1].v_samp_factor,
+                       dinfo.comp_info[2].h_samp_factor, dinfo.comp_info[2].v_samp_factor);
+               exit(1);
+       }
        dinfo.raw_data_out = true;
 
        jpeg_start_decompress(&dinfo);
 
        frame->width = dinfo.output_width;
        frame->height = dinfo.output_height;
+       frame->chroma_subsampling_x = dinfo.max_h_samp_factor / dinfo.comp_info[1].h_samp_factor;
+       frame->chroma_subsampling_y = dinfo.max_v_samp_factor / dinfo.comp_info[1].v_samp_factor;
+
+       unsigned h_mcu_size = DCTSIZE * dinfo.max_h_samp_factor;
+       unsigned v_mcu_size = DCTSIZE * dinfo.max_v_samp_factor;
+       unsigned mcu_width_blocks = (dinfo.output_width + h_mcu_size - 1) / h_mcu_size;
+       unsigned mcu_height_blocks = (dinfo.output_height + v_mcu_size - 1) / v_mcu_size;
 
-       unsigned chroma_width_blocks = (dinfo.output_width + 15) / 16;
-       unsigned width_blocks = chroma_width_blocks * 2;
-       unsigned height_blocks = (dinfo.output_height + 15) / 16;
+       unsigned luma_width_blocks = mcu_width_blocks * dinfo.comp_info[0].h_samp_factor;
+       unsigned chroma_width_blocks = mcu_width_blocks * dinfo.comp_info[1].h_samp_factor;
+       unsigned luma_height_blocks = mcu_height_blocks * dinfo.comp_info[0].v_samp_factor;
+       unsigned chroma_height_blocks = mcu_height_blocks * dinfo.comp_info[1].v_samp_factor;
 
        // TODO: Decode into a PBO.
-       frame->y.reset(new uint8_t[width_blocks * (height_blocks * 2) * DCTSIZE2]);
-       frame->cb.reset(new uint8_t[chroma_width_blocks * (height_blocks * 2) * DCTSIZE2]);
-       frame->cr.reset(new uint8_t[chroma_width_blocks * (height_blocks * 2) * DCTSIZE2]);
+       frame->y.reset(new uint8_t[luma_width_blocks * luma_height_blocks * DCTSIZE2]);
+       frame->cb.reset(new uint8_t[chroma_width_blocks * chroma_height_blocks * DCTSIZE2]);
+       frame->cr.reset(new uint8_t[chroma_width_blocks * chroma_height_blocks * DCTSIZE2]);
+       frame->pitch_y = luma_width_blocks * DCTSIZE;
+       frame->pitch_chroma = chroma_width_blocks * DCTSIZE;
 
-       JSAMPROW yptr[16], cbptr[16], crptr[16];
+       JSAMPROW yptr[v_mcu_size], cbptr[v_mcu_size], crptr[v_mcu_size];
        JSAMPARRAY data[3] = { yptr, cbptr, crptr };
-       for (unsigned y = 0; y < height_blocks; ++y) {
-               for (unsigned yy = 0; yy < DCTSIZE * 2; ++yy) {
-                       yptr[yy] = frame->y.get() + (y * DCTSIZE * 2 + yy) * width_blocks * DCTSIZE;
-                       cbptr[yy] = frame->cb.get() + (y * DCTSIZE * 2 + yy) * chroma_width_blocks * DCTSIZE;
-                       crptr[yy] = frame->cr.get() + (y * DCTSIZE * 2 + yy) * chroma_width_blocks * DCTSIZE;
+       for (unsigned y = 0; y < mcu_height_blocks; ++y) {
+               // NOTE: The last elements of cbptr/crptr will be unused for vertically subsampled chroma.
+               for (unsigned yy = 0; yy < v_mcu_size; ++yy) {
+                       yptr[yy] = frame->y.get() + (y * DCTSIZE * dinfo.max_v_samp_factor + yy) * frame->pitch_y;
+                       cbptr[yy] = frame->cb.get() + (y * DCTSIZE * dinfo.comp_info[1].v_samp_factor + yy) * frame->pitch_chroma;
+                       crptr[yy] = frame->cr.get() + (y * DCTSIZE * dinfo.comp_info[1].v_samp_factor + yy) * frame->pitch_chroma;
                }
 
-               jpeg_read_raw_data(&dinfo, data, /*num_lines=*/16);
+               jpeg_read_raw_data(&dinfo, data, v_mcu_size);
        }
 
        (void) jpeg_finish_decompress(&dinfo);
@@ -130,6 +142,34 @@ void prune_cache()
        }
 }
 
+shared_ptr<Frame> decode_jpeg_with_cache(JPEGID id, CacheMissBehavior cache_miss_behavior, bool *did_decode)
+{
+       *did_decode = false;
+       {
+               unique_lock<mutex> lock(cache_mu);
+               auto it = cache.find(id);
+               if (it != cache.end()) {
+                       it->second.last_used = event_counter++;
+                       return it->second.frame;
+               }
+       }
+
+       if (cache_miss_behavior == RETURN_NULLPTR_IF_NOT_IN_CACHE) {
+               return nullptr;
+       }
+
+       *did_decode = true;
+       shared_ptr<Frame> frame = decode_jpeg(filename_for_frame(id.stream_idx, id.pts));
+
+       unique_lock<mutex> lock(cache_mu);
+       cache[id] = LRUFrame{ frame, event_counter++ };
+
+       if (cache.size() > CACHE_SIZE) {
+               prune_cache();
+       }
+       return frame;
+}
+
 void jpeg_decoder_thread()
 {
        size_t num_decoded = 0, num_dropped = 0;
@@ -138,9 +178,9 @@ void jpeg_decoder_thread()
        for ( ;; ) {
                JPEGID id;
                JPEGFrameView *dest;
-               shared_ptr<Frame> frame;
+               CacheMissBehavior cache_miss_behavior = DECODE_IF_NOT_IN_CACHE;
                {
-                       unique_lock<mutex> lock(cache_mu);
+                       unique_lock<mutex> lock(cache_mu);  // TODO: Perhaps under another lock?
                        any_pending_decodes.wait(lock, [] {
                                return !pending_decodes.empty();
                        });
@@ -148,17 +188,6 @@ void jpeg_decoder_thread()
                        dest = pending_decodes.front().second;
                        pending_decodes.pop_front();
 
-                       auto it = cache.find(id);
-                       if (it != cache.end()) {
-                               frame = it->second.frame;
-                               it->second.last_used = event_counter++;
-                       }
-               }
-
-               if (frame == nullptr) {
-                       // Not found in the cache, so we need to do a decode or drop the request.
-                       // Prune the queue if there are too many pending for this destination.
-                       // TODO: Could we get starvation here?
                        size_t num_pending = 0;
                        for (const pair<JPEGID, JPEGFrameView *> &decode : pending_decodes) {
                                if (decode.second == dest) {
@@ -166,18 +195,20 @@ void jpeg_decoder_thread()
                                }
                        }
                        if (num_pending > 3) {
-                               ++num_dropped;
-                               continue;
+                               cache_miss_behavior = RETURN_NULLPTR_IF_NOT_IN_CACHE;
                        }
+               }
 
-                       frame = decode_jpeg(filename_for_frame(id.stream_idx, id.pts));
+               bool found_in_cache;
+               shared_ptr<Frame> frame = decode_jpeg_with_cache(id, cache_miss_behavior, &found_in_cache);
 
-                       unique_lock<mutex> lock(cache_mu);
-                       cache[id] = LRUFrame{ frame, event_counter++ };
+               if (frame == nullptr) {
+                       assert(cache_miss_behavior == RETURN_NULLPTR_IF_NOT_IN_CACHE);
+                       ++num_dropped;
+                       continue;
+               }
 
-                       if (cache.size() > CACHE_SIZE) {
-                               prune_cache();
-                       }
+               if (!found_in_cache) {
                        ++num_decoded;
                        if (num_decoded % 1000 == 0) {
                                fprintf(stderr, "Decoded %zu images, dropped %zu (%.2f%% dropped)\n",
@@ -206,14 +237,11 @@ void JPEGFrameView::initializeGL()
 {
        glDisable(GL_BLEND);
        glDisable(GL_DEPTH_TEST);
-       glDepthMask(GL_FALSE);
        check_error();
 
        static once_flag once;
        call_once(once, [] {
-               CHECK(init_movit(MOVIT_SHADER_DIR, MOVIT_DEBUG_OFF));
                resource_pool = new ResourcePool;
-
                std::thread(&jpeg_decoder_thread).detach();
        });
 
@@ -221,7 +249,6 @@ void JPEGFrameView::initializeGL()
        ImageFormat image_format;
        image_format.color_space = COLORSPACE_sRGB;
        image_format.gamma_curve = GAMMA_sRGB;
-       YCbCrFormat ycbcr_format;
        ycbcr_format.luma_coefficients = YCBCR_REC_709;
        ycbcr_format.full_range = false;
        ycbcr_format.num_levels = 256;
@@ -255,8 +282,11 @@ void JPEGFrameView::resizeGL(int width, int height)
 
 void JPEGFrameView::paintGL()
 {
-       //glClearColor(0.0f, 1.0f, 0.0f, 1.0f);
-       //glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
+       if (current_frame == nullptr) {
+               glClearColor(0.0f, 0.0f, 0.0f, 1.0f);
+               glClear(GL_COLOR_BUFFER_BIT);
+               return;
+       }
 
        check_error();
        chain->render_to_screen();
@@ -266,15 +296,17 @@ void JPEGFrameView::setDecodedFrame(std::shared_ptr<Frame> frame)
 {
        post_to_main_thread([this, frame] {
                current_frame = frame;
-               int width_blocks = (frame->width + 15) / 16;
+               ycbcr_format.chroma_subsampling_x = frame->chroma_subsampling_x;
+               ycbcr_format.chroma_subsampling_y = frame->chroma_subsampling_y;
+               ycbcr_input->change_ycbcr_format(ycbcr_format);
                ycbcr_input->set_width(frame->width);
                ycbcr_input->set_height(frame->height);
                ycbcr_input->set_pixel_data(0, frame->y.get());
                ycbcr_input->set_pixel_data(1, frame->cb.get());
                ycbcr_input->set_pixel_data(2, frame->cr.get());
-               ycbcr_input->set_pitch(0, width_blocks * 16);
-               ycbcr_input->set_pitch(1, width_blocks * 8);
-               ycbcr_input->set_pitch(2, width_blocks * 8);
+               ycbcr_input->set_pitch(0, frame->pitch_y);
+               ycbcr_input->set_pitch(1, frame->pitch_chroma);
+               ycbcr_input->set_pitch(2, frame->pitch_chroma);
                update();
        });
 }