Add ability to scale to arbitrary factors

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 acb0e2c1a19888d758cd2d46893fca9e2b93a477..fe6c1683e241c4cc51c5db9300d6f711593c7fab 100644
--- a/webrtc/media/base/videoadapter.cc
+++ b/webrtc/media/base/videoadapter.cc
@@ -11,15 +11,19 @@
 #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 {
+const int kResolutionRequiredDivisor = 4;
 
 struct Fraction {
   int numerator;
@@ -46,8 +50,32 @@ int roundUp(int valueToRound, int multiple, int maxValue) {
 }
 
 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];
+  while (true) {
+    const float scale =
+        best_scale.numerator / static_cast<float>(best_scale.denominator);
+    float test_num_pixels = input_num_pixels * scale * scale;
+    float diff = target_num_pixels - test_num_pixels;
+    if (diff < 0) {

[magjed_webrtc, 2016/12/06 16:35:08]

nit: I think this logic is cleaner:
if (test_num_pixels <= target_num_pixels)
  return best_scale;
...modify best_scale...

[kthelgason, 2016/12/07 12:35:25]

Acknowledged.

+      if (best_scale.numerator == 1) {
+        best_scale.numerator = 3;
+        best_scale.denominator *= 4;
+      } else {
+        best_scale.numerator = 1;
+        best_scale.denominator /= 2;
+      }
+    } else {
+      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);
@@ -58,7 +86,7 @@ Fraction FindScaleLessThanOrEqual(int input_num_pixels, int target_num_pixels) {
     }
     if (diff < best_distance) {
       best_distance = diff;
-      best_scale = fraction;
+      best_scale = rtc::Optional<Fraction>(fraction);
       if (best_distance == 0) {  // Found exact match.
         break;
       }
@@ -67,9 +95,9 @@ Fraction FindScaleLessThanOrEqual(int input_num_pixels, int target_num_pixels) {
   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.
@@ -93,21 +121,30 @@ Fraction FindScaleLargerThan(int input_num_pixels,
   return best_scale;
 }
 
-Fraction FindScale(int input_num_pixels,
-                   int max_pixel_count_step_up,
-                   int max_pixel_count) {
+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 = FindScaleLargerThan(
+    const Fraction scale = FindOptimizedScaleLargerThan(
         input_num_pixels, max_pixel_count_step_up, &resulting_pixel_count);
     if (resulting_pixel_count <= max_pixel_count)
-      return scale;
+      return rtc::Optional<Fraction>(scale);
   }
   // Return largest scale below |max_pixel_count|.
-  return FindScaleLessThanOrEqual(input_num_pixels, 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) {
+  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 {
@@ -211,23 +248,25 @@ bool VideoAdapter::AdaptFrameResolution(int in_width,
     *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);
-  RTC_DCHECK_EQ(0, *cropped_width % scale.denominator);
-  RTC_DCHECK_EQ(0, *cropped_height % scale.denominator);
+  // scale factor. Make sure the resulting dimensions are a multiple of 4
+  // to be nice to hardware encoders.
+  *cropped_width = roundUp(
+      *cropped_width, scale.denominator * kResolutionRequiredDivisor, in_width);
+  *cropped_height =
+      roundUp(*cropped_height, scale.denominator * kResolutionRequiredDivisor,
+              in_height);
+  RTC_DCHECK_EQ(
+      0, *cropped_width % scale.denominator * kResolutionRequiredDivisor);
+  RTC_DCHECK_EQ(
+      0, *cropped_height % scale.denominator * kResolutionRequiredDivisor);
 
   // Calculate final output size.
   *out_width = *cropped_width / scale.denominator * scale.numerator;
   *out_height = *cropped_height / scale.denominator * scale.numerator;
-
   ++frames_out_;
   if (scale.numerator != scale.denominator)
     ++frames_scaled_;