VideoAdapter: Add cropping based on OnOutputFormatRequest()

webrtc/media/base/videoadapter.cc

Index: webrtc/media/base/videoadapter.cc
diff --git a/webrtc/media/base/videoadapter.cc b/webrtc/media/base/videoadapter.cc
index 797a876f2b4049c38049bf1477b5ecb80819e022..9c3837c5729bc4b48531689c36ed2e0a9206afc7 100644
--- a/webrtc/media/base/videoadapter.cc
+++ b/webrtc/media/base/videoadapter.cc
@@ -13,69 +13,77 @@
 #include <algorithm>
 #include <limits>
 
+#include "webrtc/base/checks.h"
 #include "webrtc/base/logging.h"
 #include "webrtc/media/base/mediaconstants.h"
 #include "webrtc/media/base/videocommon.h"
 
 namespace {
 
+struct Fraction {
+  int numerator;
+  int denominator;
+};
+
 // Scale factors optimized for in libYUV that we accept.
 // Must be sorted in decreasing scale factors for FindScaleLargerThan to work.
-const float kScaleFactors[] = {
-    1.f / 1.f,   // Full size.
-    3.f / 4.f,   // 3/4 scale.
-    1.f / 2.f,   // 1/2 scale.
-    3.f / 8.f,   // 3/8 scale.
-    1.f / 4.f,   // 1/4 scale.
-    3.f / 16.f,  // 3/16 scale.
+const Fraction kScaleFractions[] = {
+  {1, 1},
+  {3, 4},
+  {1, 2},
+  {3, 8},
+  {1, 4},
+  {3, 16},
 };
 
-float FindScaleLessThanOrEqual(int width,
-                               int height,
-                               int target_num_pixels,
-                               int* resulting_number_of_pixels) {
+// Round |valueToRound| to a multiple of |multiple|. Prefer rounding upwards,
+// but never more than |maxValue|.
+int roundUp(int valueToRound, int multiple, int maxValue) {
+  const int roundedValue = (valueToRound + multiple - 1) / multiple * multiple;
+  return roundedValue <= maxValue ? roundedValue
+                                  : (maxValue / multiple * multiple);
+}
+
+Fraction FindScaleLessThanOrEqual(int input_num_pixels, int target_num_pixels) {
   float best_distance = std::numeric_limits<float>::max();
-  float best_scale = 0.0f;  // Default to 0 if nothing matches.
-  float pixels = width * height;
-  float best_number_of_pixels = 0.0f;
-  for (const auto& scale : kScaleFactors) {
-    float test_num_pixels = pixels * scale * scale;
+  Fraction best_scale = {0, 1};  // Default to 0 if nothing matches.
+  for (const auto& fraction : kScaleFractions) {
+    const float scale =
+        fraction.numerator / static_cast<float>(fraction.denominator);
+    float test_num_pixels = input_num_pixels * scale * scale;
     float diff = target_num_pixels - test_num_pixels;
     if (diff < 0) {
       continue;
     }
     if (diff < best_distance) {
       best_distance = diff;
-      best_scale = scale;
-      best_number_of_pixels = test_num_pixels;
+      best_scale = fraction;
       if (best_distance == 0) {  // Found exact match.
         break;
       }
     }
   }
-  if (resulting_number_of_pixels) {
-    *resulting_number_of_pixels = static_cast<int>(best_number_of_pixels + .5f);
-  }
   return best_scale;
 }
 
-float FindScaleLargerThan(int width,
-                          int height,
-                          int target_num_pixels,
-                          int* resulting_number_of_pixels) {
+Fraction FindScaleLargerThan(int input_num_pixels,
+                             int target_num_pixels,
+                             int* resulting_number_of_pixels) {
   float best_distance = std::numeric_limits<float>::max();
-  float best_scale = 1.f;  // Default to unscaled if nothing matches.
-  float pixels = width * height;
-  float best_number_of_pixels = pixels;  // Default to input number of pixels.
-  for (const auto& scale : kScaleFactors) {
-    float test_num_pixels = pixels * scale * scale;
+  Fraction best_scale = {1, 1};  // Default to unscaled if nothing matches.
+  // Default to input number of pixels.
+  float best_number_of_pixels = input_num_pixels;
+  for (const auto& fraction : kScaleFractions) {
+    const float scale =
+        fraction.numerator / static_cast<float>(fraction.denominator);
+    float test_num_pixels = input_num_pixels * scale * scale;
     float diff = test_num_pixels - target_num_pixels;
     if (diff <= 0) {
       break;
     }
     if (diff < best_distance) {
       best_distance = diff;
-      best_scale = scale;
+      best_scale = fraction;
       best_number_of_pixels = test_num_pixels;
     }
   }
@@ -84,21 +92,36 @@ float FindScaleLargerThan(int width,
   return best_scale;
 }
 
+Fraction FindScale(int input_num_pixels,
+                   int max_pixel_count_step_up,
+                   int max_pixel_count) {
+  // Try scale just above |max_pixel_count_step_up_|.
+  if (max_pixel_count_step_up > 0) {
+    int resulting_pixel_count;
+    const Fraction scale = FindScaleLargerThan(
+        input_num_pixels, max_pixel_count_step_up, &resulting_pixel_count);
+    if (resulting_pixel_count <= max_pixel_count)
+      return scale;
+  }
+  // Return largest scale below |max_pixel_count|.
+  return FindScaleLessThanOrEqual(input_num_pixels, max_pixel_count);
+}
+
 }  // namespace
 
 namespace cricket {
 
 VideoAdapter::VideoAdapter()
-    : output_num_pixels_(std::numeric_limits<int>::max()),
-      frames_in_(0),
+    : frames_in_(0),
       frames_out_(0),
       frames_scaled_(0),
       adaption_changes_(0),
       previous_width_(0),
       previous_height_(0),
+      input_interval_(0),
       interval_next_frame_(0),
-      format_request_max_pixel_count_(std::numeric_limits<int>::max()),
-      resolution_request_max_pixel_count_(std::numeric_limits<int>::max()) {}
+      resolution_request_max_pixel_count_(std::numeric_limits<int>::max()),
+      resolution_request_max_pixel_count_step_up_(0) {}
 
 VideoAdapter::~VideoAdapter() {}
 
@@ -106,57 +129,43 @@ void VideoAdapter::SetExpectedInputFrameInterval(int64_t interval) {
   // TODO(perkj): Consider measuring input frame rate instead.
   // Frame rate typically varies depending on lighting.
   rtc::CritScope cs(&critical_section_);
-  input_format_.interval = interval;
-}
-
-void VideoAdapter::SetInputFormat(const VideoFormat& format) {
-  bool is_resolution_change = (input_format().width != format.width ||
-                               input_format().height != format.height);
-  int64_t old_input_interval = input_format_.interval;
-  input_format_ = format;
-  output_format_.interval =
-      std::max(output_format_.interval, input_format_.interval);
-  if (old_input_interval != input_format_.interval) {
-    LOG(LS_INFO) << "VAdapt input interval changed from "
-      << old_input_interval << " to " << input_format_.interval;
-  }
-  if (is_resolution_change) {
-    // Trigger the adaptation logic again, to potentially reset the adaptation
-    // state for things like view requests that may not longer be capping
-    // output (or may now cap output).
-    Adapt(std::min(format_request_max_pixel_count_,
-                   resolution_request_max_pixel_count_),
-          0);
-  }
-}
-
-const VideoFormat& VideoAdapter::input_format() const {
-  rtc::CritScope cs(&critical_section_);
-  return input_format_;
+  input_interval_ = interval;
 }
 
-VideoFormat VideoAdapter::AdaptFrameResolution(int in_width, int in_height) {
+void VideoAdapter::AdaptFrameResolution(int in_width,
+                                        int in_height,
+                                        int* cropped_width,
+                                        int* cropped_height,
+                                        int* out_width,
+                                        int* out_height) {
   rtc::CritScope cs(&critical_section_);
   ++frames_in_;
 
-  SetInputFormat(VideoFormat(
-      in_width, in_height, input_format_.interval, input_format_.fourcc));
+  // The max output pixel count is the minimum of the requests from
+  // OnOutputFormatRequest and OnResolutionRequest.
+  int max_pixel_count = resolution_request_max_pixel_count_;
+  if (requested_format_) {
+    max_pixel_count = std::min(
+        max_pixel_count, requested_format_->width * requested_format_->height);
+  }
 
   // Drop the input frame if necessary.
   bool should_drop = false;
-  if (!output_num_pixels_) {
+  if (max_pixel_count == 0) {
     // Drop all frames as the output format is 0x0.
     should_drop = true;
-  } else {
+  } else if (requested_format_ && requested_format_->interval > 0) {
     // Drop some frames based on input fps and output fps.
     // Normally output fps is less than input fps.
-    interval_next_frame_ += input_format_.interval;
-    if (output_format_.interval > 0) {
-      if (interval_next_frame_ >= output_format_.interval) {
-        interval_next_frame_ %= output_format_.interval;
-      } else {
-        should_drop = true;
-      }
+    interval_next_frame_ += input_interval_;
+    if (interval_next_frame_ >= requested_format_->interval) {
+      interval_next_frame_ -= requested_format_->interval;
+      // Reset |interval_next_frame_| if it accumulates too much to avoid
+      // "catching up" behaviour.
+      if (interval_next_frame_ >= requested_format_->interval)
+        interval_next_frame_ = 0;
+    } else {
+      should_drop = true;
     }
   }
   if (should_drop) {
@@ -170,48 +179,79 @@ VideoFormat VideoAdapter::AdaptFrameResolution(int in_width, int in_height) {
                    << " Changes: " << adaption_changes_
                    << " Input: " << in_width
                    << "x" << in_height
-                   << " i" << input_format_.interval
-                   << " Output: i" << output_format_.interval;
+                   << " i" << input_interval_
+                   << " Output: i"
+                   << (requested_format_ ? requested_format_->interval : 0);
     }
 
-    return VideoFormat();  // Drop frame.
+    // Drop frame.
+    *cropped_width = 0;
+    *cropped_height = 0;
+    *out_width = 0;
+    *out_height = 0;
+    return;
+  }
+
+  // Calculate how the input should be cropped.
+  if (!requested_format_ ||
+      requested_format_->width == 0 || requested_format_->height == 0) {
+    *cropped_width = in_width;
+    *cropped_height = in_height;
+  } else {
+    // Adjust |requested_format_| orientation to match input.
+    if ((in_width > in_height) !=
+        (requested_format_->width > requested_format_->height)) {
+      std::swap(requested_format_->width, requested_format_->height);
+    }
+    const float requested_aspect =
+        requested_format_->width /
+        static_cast<float>(requested_format_->height);
+    *cropped_width =
+        std::min(in_width, static_cast<int>(in_height * requested_aspect));
+    *cropped_height =
+        std::min(in_height, static_cast<int>(in_width / requested_aspect));
   }
 
-  const float scale = FindScaleLessThanOrEqual(in_width, in_height,
-                                               output_num_pixels_, nullptr);
-  const int output_width = static_cast<int>(in_width * scale + .5f);
-  const int output_height = static_cast<int>(in_height * scale + .5f);
+  // Find best scale factor.
+  const Fraction scale =
+      FindScale(*cropped_width * *cropped_height,
+                resolution_request_max_pixel_count_step_up_, max_pixel_count);
+
+  // Adjust cropping slightly to get even integer output size and a perfect
+  // scale factor.
+  *cropped_width = roundUp(*cropped_width, scale.denominator, in_width);
+  *cropped_height = roundUp(*cropped_height, scale.denominator, in_height);
+  RTC_DCHECK_EQ(0, *cropped_width % scale.denominator);
+  RTC_DCHECK_EQ(0, *cropped_height % scale.denominator);
+
+  // Calculate final output size.
+  *out_width = *cropped_width / scale.denominator * scale.numerator;
+  *out_height = *cropped_height / scale.denominator * scale.numerator;
 
   ++frames_out_;
-  if (scale != 1)
+  if (scale.numerator != scale.denominator)
     ++frames_scaled_;
 
-  if (previous_width_ && (previous_width_ != output_width ||
-                          previous_height_ != output_height)) {
+  if (previous_width_ && (previous_width_ != *out_width ||
+                          previous_height_ != *out_height)) {
     ++adaption_changes_;
     LOG(LS_INFO) << "Frame size changed: scaled " << frames_scaled_ << " / out "
                  << frames_out_ << " / in " << frames_in_
                  << " Changes: " << adaption_changes_ << " Input: " << in_width
-                 << "x" << in_height << " i" << input_format_.interval
-                 << " Scale: " << scale << " Output: " << output_width << "x"
-                 << output_height << " i" << output_format_.interval;
+                 << "x" << in_height << " i" << input_interval_
+                 << " Scale: " << scale.numerator << "/" << scale.denominator
+                 << " Output: " << *out_width << "x" << *out_height << " i"
+                 << (requested_format_ ? requested_format_->interval : 0);
   }
 
-  output_format_.width = output_width;
-  output_format_.height = output_height;
-  previous_width_ = output_width;
-  previous_height_ = output_height;
-
-  return output_format_;
+  previous_width_ = *out_width;
+  previous_height_ = *out_height;
 }
 
 void VideoAdapter::OnOutputFormatRequest(const VideoFormat& format) {
   rtc::CritScope cs(&critical_section_);
-  format_request_max_pixel_count_ = format.width * format.height;
-  output_format_.interval = format.interval;
-  Adapt(std::min(format_request_max_pixel_count_,
-                 resolution_request_max_pixel_count_),
-        0);
+  requested_format_ = rtc::Optional<VideoFormat>(format);
+  interval_next_frame_ = 0;
 }
 
 void VideoAdapter::OnResolutionRequest(
@@ -220,44 +260,8 @@ void VideoAdapter::OnResolutionRequest(
   rtc::CritScope cs(&critical_section_);
   resolution_request_max_pixel_count_ =
       max_pixel_count.value_or(std::numeric_limits<int>::max());
-  Adapt(std::min(format_request_max_pixel_count_,
-                 resolution_request_max_pixel_count_),
-        max_pixel_count_step_up.value_or(0));
-}
-
-bool VideoAdapter::Adapt(int max_num_pixels, int max_pixel_count_step_up) {
-  float scale_lower =
-      FindScaleLessThanOrEqual(input_format_.width, input_format_.height,
-                               max_num_pixels, &max_num_pixels);
-  float scale_upper =
-      max_pixel_count_step_up > 0
-          ? FindScaleLargerThan(input_format_.width, input_format_.height,
-                                max_pixel_count_step_up,
-                                &max_pixel_count_step_up)
-          : 1.f;
-
-  bool use_max_pixel_count_step_up =
-      max_pixel_count_step_up > 0 && max_num_pixels > max_pixel_count_step_up;
-
-  int old_num_pixels = output_num_pixels_;
-  output_num_pixels_ =
-      use_max_pixel_count_step_up ? max_pixel_count_step_up : max_num_pixels;
-  // Log the new size.
-  float scale = use_max_pixel_count_step_up ? scale_upper : scale_lower;
-  int new_width = static_cast<int>(input_format_.width * scale + .5f);
-  int new_height = static_cast<int>(input_format_.height * scale + .5f);
-
-  bool changed = output_num_pixels_ != old_num_pixels;
-  LOG(LS_INFO) << "OnResolutionRequest: "
-               << " Max pixels: " << max_num_pixels
-               << " Max pixels step up: " << max_pixel_count_step_up
-               << " Output Pixels: " << output_num_pixels_
-               << " Input: " << input_format_.width << "x"
-               << input_format_.height << " Scale: " << scale
-               << " Resolution: " << new_width << "x" << new_height
-               << " Changed: " << (changed ? "true" : "false");
-
-  return changed;
+  resolution_request_max_pixel_count_step_up_ =
+      max_pixel_count_step_up.value_or(0);
 }
 
 }  // namespace cricket