libwebp: update to 0.4.1

third_party/libwebp/utils/bit_reader.c

Index: third_party/libwebp/utils/bit_reader.c
diff --git a/third_party/libwebp/utils/bit_reader.c b/third_party/libwebp/utils/bit_reader.c
index 79f64d36e486ce2572f52b6649a00367e3bb26f2..87b4f462152f0bb5c12f741f5b36454a534547d2 100644
--- a/third_party/libwebp/utils/bit_reader.c
+++ b/third_party/libwebp/utils/bit_reader.c
@@ -7,18 +7,16 @@
 // be found in the AUTHORS file in the root of the source tree.
 // -----------------------------------------------------------------------------
 //
-// Boolean decoder
+// Boolean decoder non-inlined methods
 //
 // Author: Skal (pascal.massimino@gmail.com)
 
-#include "./bit_reader.h"
-
-#ifndef USE_RIGHT_JUSTIFY
-#define MK(X) (((range_t)(X) << (BITS)) | (MASK))
-#else
-#define MK(X) ((range_t)(X))
+#ifdef HAVE_CONFIG_H
+#include "../webp/config.h"
 #endif
 
+#include "./bit_reader_inl.h"
+
 //------------------------------------------------------------------------------
 // VP8BitReader
 
@@ -27,12 +25,20 @@ void VP8InitBitReader(VP8BitReader* const br,
   assert(br != NULL);
   assert(start != NULL);
   assert(start <= end);
-  br->range_   = MK(255 - 1);
+  br->range_   = 255 - 1;
   br->buf_     = start;
   br->buf_end_ = end;
   br->value_   = 0;
   br->bits_    = -8;   // to load the very first 8bits
   br->eof_     = 0;
+  VP8LoadNewBytes(br);
+}
+
+void VP8RemapBitReader(VP8BitReader* const br, ptrdiff_t offset) {
+  if (br->buf_ != NULL) {
+    br->buf_ += offset;
+    br->buf_end_ += offset;
+  }
 }
 
 const uint8_t kVP8Log2Range[128] = {
@@ -47,45 +53,35 @@ const uint8_t kVP8Log2Range[128] = {
   0
 };
 
-// range = (range << kVP8Log2Range[range]) + trailing 1's
+// range = ((range - 1) << kVP8Log2Range[range]) + 1
 const range_t kVP8NewRange[128] = {
-  MK(127), MK(127), MK(191), MK(127), MK(159), MK(191), MK(223), MK(127),
-  MK(143), MK(159), MK(175), MK(191), MK(207), MK(223), MK(239), MK(127),
-  MK(135), MK(143), MK(151), MK(159), MK(167), MK(175), MK(183), MK(191),
-  MK(199), MK(207), MK(215), MK(223), MK(231), MK(239), MK(247), MK(127),
-  MK(131), MK(135), MK(139), MK(143), MK(147), MK(151), MK(155), MK(159),
-  MK(163), MK(167), MK(171), MK(175), MK(179), MK(183), MK(187), MK(191),
-  MK(195), MK(199), MK(203), MK(207), MK(211), MK(215), MK(219), MK(223),
-  MK(227), MK(231), MK(235), MK(239), MK(243), MK(247), MK(251), MK(127),
-  MK(129), MK(131), MK(133), MK(135), MK(137), MK(139), MK(141), MK(143),
-  MK(145), MK(147), MK(149), MK(151), MK(153), MK(155), MK(157), MK(159),
-  MK(161), MK(163), MK(165), MK(167), MK(169), MK(171), MK(173), MK(175),
-  MK(177), MK(179), MK(181), MK(183), MK(185), MK(187), MK(189), MK(191),
-  MK(193), MK(195), MK(197), MK(199), MK(201), MK(203), MK(205), MK(207),
-  MK(209), MK(211), MK(213), MK(215), MK(217), MK(219), MK(221), MK(223),
-  MK(225), MK(227), MK(229), MK(231), MK(233), MK(235), MK(237), MK(239),
-  MK(241), MK(243), MK(245), MK(247), MK(249), MK(251), MK(253), MK(127)
+  127, 127, 191, 127, 159, 191, 223, 127,
+  143, 159, 175, 191, 207, 223, 239, 127,
+  135, 143, 151, 159, 167, 175, 183, 191,
+  199, 207, 215, 223, 231, 239, 247, 127,
+  131, 135, 139, 143, 147, 151, 155, 159,
+  163, 167, 171, 175, 179, 183, 187, 191,
+  195, 199, 203, 207, 211, 215, 219, 223,
+  227, 231, 235, 239, 243, 247, 251, 127,
+  129, 131, 133, 135, 137, 139, 141, 143,
+  145, 147, 149, 151, 153, 155, 157, 159,
+  161, 163, 165, 167, 169, 171, 173, 175,
+  177, 179, 181, 183, 185, 187, 189, 191,
+  193, 195, 197, 199, 201, 203, 205, 207,
+  209, 211, 213, 215, 217, 219, 221, 223,
+  225, 227, 229, 231, 233, 235, 237, 239,
+  241, 243, 245, 247, 249, 251, 253, 127
 };
 
-#undef MK
-
 void VP8LoadFinalBytes(VP8BitReader* const br) {
   assert(br != NULL && br->buf_ != NULL);
   // Only read 8bits at a time
   if (br->buf_ < br->buf_end_) {
-#ifndef USE_RIGHT_JUSTIFY
-    br->value_ |= (bit_t)(*br->buf_++) << ((BITS) - 8 - br->bits_);
-#else
-    br->value_ = (bit_t)(*br->buf_++) | (br->value_ << 8);
-#endif
     br->bits_ += 8;
+    br->value_ = (bit_t)(*br->buf_++) | (br->value_ << 8);
   } else if (!br->eof_) {
-#ifdef USE_RIGHT_JUSTIFY
-    // These are not strictly needed, but it makes the behaviour
-    // consistent for both USE_RIGHT_JUSTIFY and !USE_RIGHT_JUSTIFY.
     br->value_ <<= 8;
     br->bits_ += 8;
-#endif
     br->eof_ = 1;
   }
 }
@@ -109,36 +105,50 @@ int32_t VP8GetSignedValue(VP8BitReader* const br, int bits) {
 //------------------------------------------------------------------------------
 // VP8LBitReader
 
-#define MAX_NUM_BIT_READ 25
-
 #define LBITS 64      // Number of bits prefetched.
 #define WBITS 32      // Minimum number of bytes needed after VP8LFillBitWindow.
 #define LOG8_WBITS 4  // Number of bytes needed to store WBITS bits.
 
-static const uint32_t kBitMask[MAX_NUM_BIT_READ] = {
-  0, 1, 3, 7, 15, 31, 63, 127, 255, 511, 1023, 2047, 4095, 8191, 16383, 32767,
-  65535, 131071, 262143, 524287, 1048575, 2097151, 4194303, 8388607, 16777215
+#if !defined(WEBP_FORCE_ALIGNED) && \
+    (defined(__arm__) || defined(_M_ARM) || defined(__aarch64__) || \
+     defined(__i386__) || defined(_M_IX86) || \
+     defined(__x86_64__) || defined(_M_X64))
+#define VP8L_USE_UNALIGNED_LOAD
+#endif
+
+static const uint32_t kBitMask[VP8L_MAX_NUM_BIT_READ + 1] = {
+  0,
+  0x000001, 0x000003, 0x000007, 0x00000f,
+  0x00001f, 0x00003f, 0x00007f, 0x0000ff,
+  0x0001ff, 0x0003ff, 0x0007ff, 0x000fff,
+  0x001fff, 0x003fff, 0x007fff, 0x00ffff,
+  0x01ffff, 0x03ffff, 0x07ffff, 0x0fffff,
+  0x1fffff, 0x3fffff, 0x7fffff, 0xffffff
 };
 
-void VP8LInitBitReader(VP8LBitReader* const br,
-                       const uint8_t* const start,
+void VP8LInitBitReader(VP8LBitReader* const br, const uint8_t* const start,
                        size_t length) {
   size_t i;
+  vp8l_val_t value = 0;
   assert(br != NULL);
   assert(start != NULL);
   assert(length < 0xfffffff8u);   // can't happen with a RIFF chunk.
 
-  br->buf_ = start;
   br->len_ = length;
   br->val_ = 0;
-  br->pos_ = 0;
   br->bit_pos_ = 0;
   br->eos_ = 0;
   br->error_ = 0;
-  for (i = 0; i < sizeof(br->val_) && i < br->len_; ++i) {
-    br->val_ |= ((vp8l_val_t)br->buf_[br->pos_]) << (8 * i);
-    ++br->pos_;
+
+  if (length > sizeof(br->val_)) {
+    length = sizeof(br->val_);
+  }
+  for (i = 0; i < length; ++i) {
+    value |= (vp8l_val_t)start[i] << (8 * i);
   }
+  br->val_ = value;
+  br->pos_ = length;
+  br->buf_ = start;
 }
 
 // Special version that assumes br->pos_ <= br_len_.
@@ -173,13 +183,16 @@ static void ShiftBytes(VP8LBitReader* const br) {
 
 void VP8LFillBitWindow(VP8LBitReader* const br) {
   if (br->bit_pos_ >= WBITS) {
-#if (defined(__x86_64__) || defined(_M_X64))
+    // TODO(jzern): given the fixed read size it may be possible to force
+    //              alignment in this block.
+#if defined(VP8L_USE_UNALIGNED_LOAD)
     if (br->pos_ + sizeof(br->val_) < br->len_) {
       br->val_ >>= WBITS;
       br->bit_pos_ -= WBITS;
       // The expression below needs a little-endian arch to work correctly.
       // This gives a large speedup for decoding speed.
-      br->val_ |= *(const vp8l_val_t*)(br->buf_ + br->pos_) << (LBITS - WBITS);
+      br->val_ |= (vp8l_val_t)*(const uint32_t*)(br->buf_ + br->pos_) <<
+                  (LBITS - WBITS);
       br->pos_ += LOG8_WBITS;
       return;
     }
@@ -192,7 +205,7 @@ void VP8LFillBitWindow(VP8LBitReader* const br) {
 uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits) {
   assert(n_bits >= 0);
   // Flag an error if end_of_stream or n_bits is more than allowed limit.
-  if (!br->eos_ && n_bits < MAX_NUM_BIT_READ) {
+  if (!br->eos_ && n_bits <= VP8L_MAX_NUM_BIT_READ) {
     const uint32_t val =
         (uint32_t)(br->val_ >> br->bit_pos_) & kBitMask[n_bits];
     const int new_bits = br->bit_pos_ + n_bits;
@@ -208,4 +221,3 @@ uint32_t VP8LReadBits(VP8LBitReader* const br, int n_bits) {
 }
 
 //------------------------------------------------------------------------------
-