Fix a bug in frame parallel decode and add a unit test for that.

A flush bug is discovered during putting frame parallel decoder
into Android. This test will expose that bug.

Change-Id: Ia047f27972f4da0471649f79f1f91e7695297473

test/test-data.sha1

@@ -665,3 +665,5 @@ f97088c7359fc8d3d5aa5eafe57bc7308b3ee124  vp90-2-20-big_superframe-01.webm
 47d7d409785afa33b123376de0c907336e6c7bd7  vp90-2-20-big_superframe-01.webm.md5
 65ade6d2786209582c50d34cfe22b3cdb033abaf  vp90-2-20-big_superframe-02.webm
 7c0ed8d04c4d06c5411dd2e5de2411d37f092db5  vp90-2-20-big_superframe-02.webm.md5
+667ec8718c982aef6be07eb94f083c2efb9d2d16  vp90-2-07-frame_parallel-1.webm
+bfc82bf848e9c05020d61e3ffc1e62f25df81d19  vp90-2-07-frame_parallel-1.webm.md5

test/test.mk

@@ -31,6 +31,7 @@ LIBVPX_TEST_SRCS-$(CONFIG_VP8_ENCODER) += keyframe_test.cc
 
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += external_frame_buffer_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += user_priv_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9_DECODER) += vp9_frame_parallel_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += active_map_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += borders_test.cc
 LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += cpu_speed_test.cc
@@ -683,6 +684,8 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-06-bilinear.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-06-bilinear.webm.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-07-frame_parallel.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-07-frame_parallel.webm.md5
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-07-frame_parallel-1.webm
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-07-frame_parallel-1.webm.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-08-tile-4x1.webm
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-08-tile-4x1.webm.md5
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += vp90-2-08-tile-4x4.webm

test/test_vectors.cc

@@ -154,7 +154,8 @@ const char *const kVP9TestVectors[] = {
   "vp90-2-03-size-226x210.webm", "vp90-2-03-size-226x224.webm",
   "vp90-2-03-size-226x226.webm", "vp90-2-03-deltaq.webm",
   "vp90-2-05-resize.ivf", "vp90-2-06-bilinear.webm",
-  "vp90-2-07-frame_parallel.webm", "vp90-2-08-tile_1x2_frame_parallel.webm",
+  "vp90-2-07-frame_parallel.webm", "vp90-2-07-frame_parallel-1.webm",
+  "vp90-2-08-tile_1x2_frame_parallel.webm",
   "vp90-2-08-tile_1x2.webm", "vp90-2-08-tile_1x4_frame_parallel.webm",
   "vp90-2-08-tile_1x4.webm", "vp90-2-08-tile_1x8_frame_parallel.webm",
   "vp90-2-08-tile_1x8.webm", "vp90-2-08-tile-4x4.webm",

test/vp9_frame_parallel_test.cc

@@ -0,0 +1,122 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+#include "third_party/googletest/src/include/gtest/gtest.h"
+#include "./vpx_config.h"
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/ivf_video_source.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+#if CONFIG_WEBM_IO
+#include "test/webm_video_source.h"
+#endif
+#include "vpx_mem/vpx_mem.h"
+
+namespace {
+
+using std::string;
+
+#if CONFIG_WEBM_IO
+
+struct FileList {
+  const char *name;
+  // md5 sum for decoded frames which does not include skipped frames.
+  const char *expected_md5;
+  const int pause_frame_num;
+};
+
+// Decodes |filename| with |num_threads|. Pause at the specified frame_num,
+// seek to next key frame and then continue decoding until the end. Return
+// the md5 of the decoded frames which does not include skipped frames.
+string DecodeFile(const string &filename, int num_threads, int pause_num) {
+  libvpx_test::WebMVideoSource video(filename);
+  video.Init();
+  int in_frames = 0;
+  int out_frames = 0;
+
+  vpx_codec_dec_cfg_t cfg = {0};
+  cfg.threads = num_threads;
+  vpx_codec_flags_t flags = 0;
+  flags |= VPX_CODEC_USE_FRAME_THREADING;
+  libvpx_test::VP9Decoder decoder(cfg, flags, 0);
+
+  libvpx_test::MD5 md5;
+  video.Begin();
+
+  do {
+    ++in_frames;
+    const vpx_codec_err_t res =
+        decoder.DecodeFrame(video.cxdata(), video.frame_size());
+    if (res != VPX_CODEC_OK) {
+      EXPECT_EQ(VPX_CODEC_OK, res) << decoder.DecodeError();
+      break;
+    }
+
+    // Pause at specified frame number.
+    if (in_frames == pause_num) {
+      // Flush the decoder and then seek to next key frame.
+      decoder.DecodeFrame(NULL, 0);
+      video.SeekToNextKeyFrame();
+    } else {
+      video.Next();
+    }
+
+    // Flush the decoder at the end of the video.
+    if (!video.cxdata())
+      decoder.DecodeFrame(NULL, 0);
+
+    libvpx_test::DxDataIterator dec_iter = decoder.GetDxData();
+    const vpx_image_t *img;
+
+    // Get decompressed data
+    while ((img = dec_iter.Next())) {
+      ++out_frames;
+      md5.Add(img);
+    }
+  } while (video.cxdata() != NULL);
+
+  EXPECT_EQ(in_frames, out_frames) <<
+      "Input frame count does not match output frame count";
+
+  return string(md5.Get());
+}
+
+void DecodeFiles(const FileList files[]) {
+  for (const FileList *iter = files; iter->name != NULL; ++iter) {
+    SCOPED_TRACE(iter->name);
+    for (int t = 2; t <= 8; ++t) {
+      EXPECT_EQ(iter->expected_md5,
+                DecodeFile(iter->name, t, iter->pause_frame_num))
+          << "threads = " << t;
+    }
+  }
+}
+
+TEST(VP9MultiThreadedFrameParallel, PauseSeekResume) {
+  // vp90-2-07-frame_parallel-1.webm is a 40 frame video file with
+  // one key frame for every ten frames.
+  static const FileList files[] = {
+    { "vp90-2-07-frame_parallel-1.webm",
+      "6ea7c3875d67252e7caf2bc6e75b36b1", 6},
+    { "vp90-2-07-frame_parallel-1.webm",
+      "4bb634160c7356a8d7d4299b6dc83a45", 12},
+    { "vp90-2-07-frame_parallel-1.webm",
+      "89772591e6ef461f9fa754f916c78ed8", 26},
+    { NULL, NULL, 0},
+  };
+  DecodeFiles(files);
+}
+
+#endif  // CONFIG_WEBM_IO
+}  // namespace

vp9/decoder/vp9_decoder.c

@@ -272,6 +272,8 @@ int vp9_receive_compressed_data(VP9Decoder *pbi,
       cm->frame_refs[0].buf->corrupted = 1;
   }
 
+  pbi->ready_for_new_data = 0;
+
   // Check if the previous frame was a frame without any references to it.
   // Release frame buffer if not decoding in frame parallel mode.
   if (!pbi->frame_parallel_decode && cm->new_fb_idx >= 0
@@ -296,6 +298,7 @@ int vp9_receive_compressed_data(VP9Decoder *pbi,
 
   if (setjmp(cm->error.jmp)) {
     cm->error.setjmp = 0;
+    pbi->ready_for_new_data = 1;
 
     // We do not know if the missing frame(s) was supposed to update
     // any of the reference buffers, but we act conservative and
@@ -354,8 +357,6 @@ int vp9_receive_compressed_data(VP9Decoder *pbi,
     vp9_swap_current_and_last_seg_map(cm);
   }
 
-  pbi->ready_for_new_data = 0;
-
   cm->error.setjmp = 0;
   return retcode;
 }
@@ -370,12 +371,12 @@ int vp9_get_raw_frame(VP9Decoder *pbi, YV12_BUFFER_CONFIG *sd,
   if (pbi->ready_for_new_data == 1)
     return ret;
 
+  pbi->ready_for_new_data = 1;
+
   /* no raw frame to show!!! */
   if (pbi->common.show_frame == 0)
     return ret;
 
-  pbi->ready_for_new_data = 1;
-
 #if CONFIG_VP9_POSTPROC
   ret = vp9_post_proc_frame(&pbi->common, sd, flags);
 #else

vp9/vp9_dx_iface.c

@@ -443,6 +443,8 @@ static vpx_codec_err_t decode_one(vpx_codec_alg_priv_t *ctx,
           &ctx->frame_workers[ctx->next_submit_worker_id],
           &ctx->frame_workers[ctx->last_submit_worker_id]);
 
+    frame_worker_data->pbi->ready_for_new_data = 0;
+
     // Copy the compressed data into worker's internal buffer.
     // TODO(hkuang): Will all the workers allocate the same size
     // as the size of the first intra frame be better? This will
@@ -757,10 +759,10 @@ static vpx_image_t *decoder_get_frame(vpx_codec_alg_priv_t *ctx,
           (ctx->next_output_worker_id + 1) % ctx->num_frame_workers;
       // Wait for the frame from worker thread.
       winterface->sync(worker);
-      ++ctx->available_threads;
       if (vp9_get_raw_frame(frame_worker_data->pbi, &sd, &flags) == 0) {
         VP9_COMMON *const cm = &frame_worker_data->pbi->common;
         RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
+        ++ctx->available_threads;
         release_last_output_frame(ctx);
         ctx->last_show_frame = frame_worker_data->pbi->common.new_fb_idx;
         yuvconfig2image(&ctx->img, &sd, frame_worker_data->user_priv);