Add ability to scale to arbitrary factors

webrtc/media/base/videoadapter.cc

 /*
  *  Copyright (c) 2010 The WebRTC 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 "webrtc/media/base/videoadapter.h"
 
 #include <algorithm>
+#include <cmath>
 #include <cstdlib>
 #include <limits>
 
+#include "webrtc/base/arraysize.h"
 #include "webrtc/base/checks.h"
 #include "webrtc/base/logging.h"
+#include "webrtc/base/optional.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 Fraction kScaleFractions[] = {
   {1, 1},
   {3, 4},
   {1, 2},
   {3, 8},
   {1, 4},
   {3, 16},
 };
 
 // Round |valueToRound| to a multiple of |multiple|. Prefer rounding upwards,

[magjed_webrtc, 2016/12/08 09:08:55]

nit: update the variable names in these comments as well

 // 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);
+int roundUp(int value_to_round, int multiple, int max_value) {
+  const int rounded_value =
+      (value_to_round + multiple - 1) / multiple * multiple;
+  return rounded_value <= max_value ? rounded_value
+                                    : (max_value / multiple * multiple);
 }
 
+// Generates a scale factor that makes |input_num_pixels| smaller than
+// |target_num_pixels|. This should only be used after making sure none
+// of the optimized factors are small enough.
 Fraction FindScaleLessThanOrEqual(int input_num_pixels, int target_num_pixels) {
+  // Start searching from the last of the optimal fractions;
+  Fraction best_scale = kScaleFractions[arraysize(kScaleFractions) - 1];
+  const float target_scale =
+      sqrt(target_num_pixels / static_cast<float>(input_num_pixels));
+  do {
+    if (best_scale.numerator % 3 == 0 && best_scale.denominator % 2 == 0) {
+      // Multiply by 2/3
+      best_scale.numerator /= 3;
+      best_scale.denominator /= 2;
+    } else {
+      // Multiply by 3/4
+      best_scale.numerator *= 3;
+      best_scale.denominator *= 4;
+    }
+  } while (best_scale.numerator > (target_scale * best_scale.denominator));
+  return best_scale;
+}
+
+rtc::Optional<Fraction> FindOptimizedScaleLessThanOrEqual(
+    int input_num_pixels,
+    int target_num_pixels) {
   float best_distance = std::numeric_limits<float>::max();
-  Fraction best_scale = {0, 1};  // Default to 0 if nothing matches.
+  rtc::Optional<Fraction> best_scale;
   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 = fraction;
+      best_scale = rtc::Optional<Fraction>(fraction);
       if (best_distance == 0) {  // Found exact match.
         break;
       }
     }
   }
   return best_scale;
 }
 
-Fraction FindScaleLargerThan(int input_num_pixels,
-                             int target_num_pixels,
-                             int* resulting_number_of_pixels) {
+Fraction FindOptimizedScaleLargerThan(int input_num_pixels,
+                                      int target_num_pixels,
+                                      int* resulting_number_of_pixels) {
   float best_distance = std::numeric_limits<float>::max();
   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 = fraction;
       best_number_of_pixels = test_num_pixels;
     }
   }
 
   *resulting_number_of_pixels = static_cast<int>(best_number_of_pixels + .5f);
   return best_scale;
 }
 
+rtc::Optional<Fraction> FindOptimizedScale(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 = FindOptimizedScaleLargerThan(
+        input_num_pixels, max_pixel_count_step_up, &resulting_pixel_count);
+    if (resulting_pixel_count <= max_pixel_count)
+      return rtc::Optional<Fraction>(scale);
+  }
+  // Return largest scale below |max_pixel_count|.
+  return FindOptimizedScaleLessThanOrEqual(input_num_pixels, max_pixel_count);
+}
+
 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|.
+  const rtc::Optional<Fraction> optimized_scale = FindOptimizedScale(
+      input_num_pixels, max_pixel_count_step_up, max_pixel_count);
+  if (optimized_scale)
+    return *optimized_scale;
   return FindScaleLessThanOrEqual(input_num_pixels, max_pixel_count);
 }
-
 }  // namespace
 
 namespace cricket {
 
-VideoAdapter::VideoAdapter()
+VideoAdapter::VideoAdapter(int required_resolution_alignment)
     : frames_in_(0),
       frames_out_(0),
       frames_scaled_(0),
       adaption_changes_(0),
       previous_width_(0),
       previous_height_(0),
+      required_resolution_alignment_(required_resolution_alignment),
       resolution_request_max_pixel_count_(std::numeric_limits<int>::max()),
       resolution_request_max_pixel_count_step_up_(0) {}
 
+VideoAdapter::VideoAdapter() : VideoAdapter(1) {}
+
 VideoAdapter::~VideoAdapter() {}
 
 bool VideoAdapter::KeepFrame(int64_t in_timestamp_ns) {
   rtc::CritScope cs(&critical_section_);
   if (!requested_format_ || requested_format_->interval == 0)
     return true;
 
   if (next_frame_timestamp_ns_) {
     // Time until next frame should be outputted.
     const int64_t time_until_next_frame_ns =
@@ skipping 69 matching lines @@
       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));
   }
-
-  // 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);
+  // scale factor. Make sure the resulting dimensions are aligned correctly
+  // to be nice to hardware encoders.
+  *cropped_width =
+      roundUp(*cropped_width,
+              scale.denominator * required_resolution_alignment_, in_width);
+  *cropped_height =
+      roundUp(*cropped_height,
+              scale.denominator * required_resolution_alignment_, 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;
+  RTC_DCHECK_EQ(0, *out_height % required_resolution_alignment_);
+  RTC_DCHECK_EQ(0, *out_height % required_resolution_alignment_);
 
   ++frames_out_;
   if (scale.numerator != scale.denominator)
     ++frames_scaled_;
 
   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_
@@ skipping 20 matching lines @@
     rtc::Optional<int> max_pixel_count,
     rtc::Optional<int> max_pixel_count_step_up) {
   rtc::CritScope cs(&critical_section_);
   resolution_request_max_pixel_count_ =
       max_pixel_count.value_or(std::numeric_limits<int>::max());
   resolution_request_max_pixel_count_step_up_ =
       max_pixel_count_step_up.value_or(0);
 }
 
 }  // namespace cricket