Cleaup Near-lossless code.

Cleaup Near-lossless code
- Simplified and refactored the code.
- Removed the requirement (TODO) to allocate the buffer of size WxH and work
  with buffer of size 3xW.
- Disabled the Near-lossless prr-processing for small icon images (W < 64 and H < 64).

Change-Id: Id7ee90c90622368d5528de4dd14fd5ead593bb1b

src/enc/near_lossless.c

@@ -21,6 +21,10 @@
 #include "./vp8enci.h"
 
 #ifdef WEBP_EXPERIMENTAL_FEATURES
+
+#define MIN_DIM_FOR_NEAR_LOSSLESS 64
+#define MAX_LIMIT_BITS             5
+
 // Computes quantized pixel value and distance from original value.
 static void GetValAndDistance(int a, int initial, int bits,
                               int* const val, int* const distance) {
@@ -29,20 +33,21 @@ static void GetValAndDistance(int a, int initial, int bits,
   *distance = 2 * abs(a - *val);
 }
 
-// Quantizes values {a, a+(1<<bits), a-(1<<bits)}, checks if in [min, max] range
+// Clamps the value to range [0, 255].
-// and returns the nearest one.
+static int Clamp8b(int val) {
-static int FindClosestDiscretized(int a, int bits, int min, int max) {
+  const int min_val = 0;
+  const int max_val = 0xff;
+  return (val < min_val) ? min_val : (val > max_val) ? max_val : val;
+}
+
+// Quantizes values {a, a+(1<<bits), a-(1<<bits)} and returns the nearest one.
+static int FindClosestDiscretized(int a, int bits) {
   int best_val = a, i;
   int min_distance = 256;
 
   for (i = -1; i <= 1; ++i) {
-    int val = a + i * (1 << bits);
     int candidate, distance;
-    if (val < 0) {
+    const int val = Clamp8b(a + i * (1 << bits));
-      val = 0;
-    } else if (val > 255) {
-      val = 255;
-    }
     GetValAndDistance(a, val, bits, &candidate, &distance);
     if (i != 0) {
       ++distance;
@@ -50,7 +55,7 @@ static int FindClosestDiscretized(int a, int bits, int min, int max) {
     // Smallest distance but favor i == 0 over i == -1 and i == 1
     // since that keeps the overall intensity more constant in the
     // images.
-    if (distance < min_distance && candidate >= min && candidate <= max) {
+    if (distance < min_distance) {
       min_distance = distance;
       best_val = candidate;
     }
@@ -59,30 +64,37 @@ static int FindClosestDiscretized(int a, int bits, int min, int max) {
 }
 
 // Applies FindClosestDiscretized to all channels of pixel.
-static uint32_t ClosestDiscretizedArgb(uint32_t a, int bits,
+static uint32_t ClosestDiscretizedArgb(uint32_t a, int bits) {
-                                       uint32_t min, uint32_t max) {
+  return
-  return (FindClosestDiscretized(a >> 24, bits, min >> 24, max >> 24) << 24) |
+      (FindClosestDiscretized(a >> 24, bits) << 24) |
-         (FindClosestDiscretized((a >> 16) & 0xff, bits,
+      (FindClosestDiscretized((a >> 16) & 0xff, bits) << 16) |
-                                 (min >> 16) & 0xff,
+      (FindClosestDiscretized((a >> 8) & 0xff, bits) << 8) |
-                                 (max >> 16) & 0xff) << 16) |
+      (FindClosestDiscretized(a & 0xff, bits));
-         (FindClosestDiscretized((a >> 8) & 0xff, bits,
-                                 (min >> 8) & 0xff,
-                                 (max >> 8) & 0xff) << 8) |
-         (FindClosestDiscretized(a & 0xff, bits, min & 0xff, max & 0xff));
 }
 
 // Checks if distance between corresponding channel values of pixels a and b
-// exceeds given limit.
+// is within the given limit.
-static int IsFar(uint32_t a, uint32_t b, int limit) {
+static int IsNear(uint32_t a, uint32_t b, int limit) {
   int k;
   for (k = 0; k < 4; ++k) {
-    const int delta = (int)((a >> (k * 8)) & 0xff) -
+    const int delta =
-                      (int)((b >> (k * 8)) & 0xff);
+        (int)((a >> (k * 8)) & 0xff) - (int)((b >> (k * 8)) & 0xff);
     if (delta >= limit || delta <= -limit) {
-      return 1;
+      return 0;
     }
   }
-  return 0;
+  return 1;
+}
+
+static int IsSmooth(const uint32_t* const prev_row,
+                    const uint32_t* const curr_row,
+                    const uint32_t* const next_row,
+                    int ix, int limit) {
+  // Check that all pixels in 4-connected neighborhood are smooth.
+  return (IsNear(curr_row[ix], curr_row[ix - 1], limit) &&
+          IsNear(curr_row[ix], curr_row[ix + 1], limit) &&
+          IsNear(curr_row[ix], prev_row[ix], limit) &&
+          IsNear(curr_row[ix], next_row[ix], limit));
 }
 
 // Adjusts pixel values of image with given maximum error.
@@ -90,39 +102,37 @@ static void NearLossless(int xsize, int ysize, uint32_t* argb,
                         int limit_bits, uint32_t* copy_buffer) {
   int x, y;
   const int limit = 1 << limit_bits;
-  memcpy(copy_buffer, argb, xsize * ysize * sizeof(argb[0]));
+  uint32_t* prev_row = copy_buffer;
+  uint32_t* curr_row = prev_row + xsize;
+  uint32_t* next_row = curr_row + xsize;
+  memcpy(copy_buffer, argb, xsize * 2 * sizeof(argb[0]));
 
-  for (y = 0; y < ysize; ++y) {
+  for (y = 1; y < ysize - 1; ++y) {
-    const int offset = y * xsize;
+    uint32_t* const curr_argb_row = argb + y * xsize;
-    for (x = 0; x < xsize; ++x) {
+    uint32_t* const next_argb_row = curr_argb_row + xsize;
-      const int ix = offset + x;
+    memcpy(next_row, next_argb_row, xsize * sizeof(argb[0]));
-      // Check that all pixels in 4-connected neighborhood are smooth.
+    for (x = 1; x < xsize - 1; ++x) {
-      int smooth_area = 1;
+      if (!IsSmooth(prev_row, curr_row, next_row, x, limit)) {
-      if (x != 0 && IsFar(copy_buffer[ix], copy_buffer[ix - 1], limit)) {
+        curr_argb_row[x] = ClosestDiscretizedArgb(curr_row[x], limit_bits);
-        smooth_area = 0;
-      } else if (y != 0 &&
-                 IsFar(copy_buffer[ix], copy_buffer[ix - xsize], limit)) {
-        smooth_area = 0;
-      } else if (x != xsize - 1 &&
-                 IsFar(copy_buffer[ix], copy_buffer[ix + 1], limit)) {
-        smooth_area = 0;
-      } else if (y != ysize - 1 &&
-                 IsFar(copy_buffer[ix], copy_buffer[ix + xsize], limit)) {
-        smooth_area = 0;
-      }
-      if (!smooth_area) {
-        argb[ix] = ClosestDiscretizedArgb(argb[ix], limit_bits, 0, 0xffffffff);
       }
     }
+    {
+      // Three-way swap.
+      uint32_t* const temp = prev_row;
+      prev_row = curr_row;
+      curr_row = next_row;
+      next_row = temp;
+    }
   }
 }
 
 static int QualityToLimitBits(int quality) {
-  return 5 - (quality + 12) / 25;
+  // quality mapping 0..12 -> 5
+  //                 13..100 -> 4..1
+  return MAX_LIMIT_BITS - (quality + 12) / 25;
 }
 #endif  // WEBP_EXPERIMENTAL_FEATURES
 
-// TODO(vikasa): optimize memory to O(xsize)
 int VP8ApplyNearLossless(int xsize, int ysize, uint32_t* argb, int quality) {
 #ifndef WEBP_EXPERIMENTAL_FEATURES
   (void)xsize;
@@ -132,16 +142,20 @@ int VP8ApplyNearLossless(int xsize, int ysize, uint32_t* argb, int quality) {
 #else
   int i;
   uint32_t* const copy_buffer =
-      (uint32_t *)WebPSafeMalloc(xsize * ysize, sizeof(*copy_buffer));
+      (uint32_t*)WebPSafeMalloc(xsize * 3, sizeof(*copy_buffer));
-  // quality mapping 0..12 -> 5
-  //                 13..100 -> 4..1
   const int limit_bits = QualityToLimitBits(quality);
   assert(argb != NULL);
   assert(limit_bits >= 0);
-  assert(limit_bits < 31);
+  assert(limit_bits <= MAX_LIMIT_BITS);
   if (copy_buffer == NULL) {
     return 0;
   }
+  // For small icon images, don't attempt to apply near-lossless compression.
+  if (xsize < MIN_DIM_FOR_NEAR_LOSSLESS && ysize < MIN_DIM_FOR_NEAR_LOSSLESS) {
+    WebPSafeFree(copy_buffer);
+    return 1;
+  }
+
   for (i = limit_bits; i != 0; --i) {
     NearLossless(xsize, ysize, argb, i, copy_buffer);
   }