From 16be192f4710fafe2f97687d929b00542d36bae2 Mon Sep 17 00:00:00 2001
From: skal <pascal.massimino@gmail.com>
Date: Sat, 11 Feb 2017 01:51:37 -0800
Subject: [PATCH] VP8LSetBitPos: remove the eos_ setting

This code is ultra-critical for lossless decoding, especially on ARM.
The extra call VP8LIsEndOfStream() was causing unnecessary slow-down.

Now, we check for bitstream-end separately in the main loop.

Change-Id: I739b5d74cc29578e2b712ba99b544fd995ef0e0d
---
 src/dec/vp8l_dec.c           | 11 ++++++-----
 src/utils/bit_reader_utils.h |  3 ++-
 2 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/src/dec/vp8l_dec.c b/src/dec/vp8l_dec.c
index ef359a91..c6c2a010 100644
--- a/src/dec/vp8l_dec.c
+++ b/src/dec/vp8l_dec.c
@@ -1012,12 +1012,13 @@ static int DecodeAlphaData(VP8LDecoder* const dec, uint8_t* const data,
       ok = 0;
       goto End;
     }
-    assert(br->eos_ == VP8LIsEndOfStream(br));
+    br->eos_ = VP8LIsEndOfStream(br);
   }
   // Process the remaining rows corresponding to last row-block.
   ExtractPalettedAlphaRows(dec, row > last_row ? last_row : row);
 
  End:
+  br->eos_ = VP8LIsEndOfStream(br);
   if (!ok || (br->eos_ && pos < end)) {
     ok = 0;
     dec->status_ = br->eos_ ? VP8_STATUS_SUSPENDED
@@ -1094,7 +1095,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
     } else {
       code = ReadSymbol(htree_group->htrees[GREEN], br);
     }
-    if (br->eos_) break;  // early out
+    if (VP8LIsEndOfStream(br)) break;
     if (code < NUM_LITERAL_CODES) {  // Literal
       if (htree_group->is_trivial_literal) {
         *src = htree_group->literal_arb | (code << 8);
@@ -1104,7 +1105,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
         VP8LFillBitWindow(br);
         blue = ReadSymbol(htree_group->htrees[BLUE], br);
         alpha = ReadSymbol(htree_group->htrees[ALPHA], br);
-        if (br->eos_) break;
+        if (VP8LIsEndOfStream(br)) break;
         *src = ((uint32_t)alpha << 24) | (red << 16) | (code << 8) | blue;
       }
     AdvanceByOne:
@@ -1132,7 +1133,7 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
       VP8LFillBitWindow(br);
       dist_code = GetCopyDistance(dist_symbol, br);
       dist = PlaneCodeToDistance(width, dist_code);
-      if (br->eos_) break;
+      if (VP8LIsEndOfStream(br)) break;
       if (src - data < (ptrdiff_t)dist || src_end - src < (ptrdiff_t)length) {
         goto Error;
       } else {
@@ -1169,9 +1170,9 @@ static int DecodeImageData(VP8LDecoder* const dec, uint32_t* const data,
     } else {  // Not reached
       goto Error;
     }
-    assert(br->eos_ == VP8LIsEndOfStream(br));
   }
 
+  br->eos_ = VP8LIsEndOfStream(br);
   if (dec->incremental_ && br->eos_ && src < src_end) {
     RestoreState(dec);
   } else if (!br->eos_) {
diff --git a/src/utils/bit_reader_utils.h b/src/utils/bit_reader_utils.h
index ec3426cd..f67fe98a 100644
--- a/src/utils/bit_reader_utils.h
+++ b/src/utils/bit_reader_utils.h
@@ -155,9 +155,10 @@ static WEBP_INLINE int VP8LIsEndOfStream(const VP8LBitReader* const br) {
 
 // For jumping over a number of bits in the bit stream when accessed with
 // VP8LPrefetchBits and VP8LFillBitWindow.
+// This function does *not* set br->eos_, since it's speed-critical.
+// Use with extreme care!
 static WEBP_INLINE void VP8LSetBitPos(VP8LBitReader* const br, int val) {
   br->bit_pos_ = val;
-  br->eos_ = VP8LIsEndOfStream(br);
 }
 
 // Advances the read buffer by 4 bytes to make room for reading next 32 bits.