Updates libjpeg-turbo to 1.1.90 (r677)

third_party/libjpeg_turbo/jdhuff.c

 /*
  * jdhuff.c
  *
  * Copyright (C) 1991-1997, Thomas G. Lane.
+ * Copyright (C) 2009-2011, D. R. Commander.
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
  * This file contains Huffman entropy decoding routines.
  *
  * Much of the complexity here has to do with supporting input suspension.
  * If the data source module demands suspension, we want to be able to back
  * up to the start of the current MCU.  To do this, we copy state variables
  * into local working storage, and update them back to the permanent
  * storage only upon successful completion of an MCU.
  */
 
-/* Modifications:
- * Copyright (C)2007 Sun Microsystems, Inc.
- * Copyright (C)2009-2010 D. R. Commander
- *
- * This library is free software and may be redistributed and/or modified under
- * the terms of the wxWindows Library License, Version 3.1 or (at your option)
- * any later version.  The full license is in the LICENSE.txt file included
- * with this distribution.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * wxWindows Library License for more details.
- */
-
 #define JPEG_INTERNALS
 #include "jinclude.h"
 #include "jpeglib.h"
-#include "jdhuff.h"             /* Declarations shared with jdphuff.c */
+#include "jdhuff.h"             /* Declarations shared with jdphuff.c */
+#include "jpegcomp.h"
 
 
 /*
  * Expanded entropy decoder object for Huffman decoding.
  *
  * The savable_state subrecord contains fields that change within an MCU,
  * but must not be updated permanently until we complete the MCU.
  */
 
 typedef struct {
   int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
 } savable_state;
 
 /* This macro is to work around compilers with missing or broken
  * structure assignment.  You'll need to fix this code if you have
  * such a compiler and you change MAX_COMPS_IN_SCAN.
  */
 
 #ifndef NO_STRUCT_ASSIGN
 #define ASSIGN_STATE(dest,src)  ((dest) = (src))
 #else
 #if MAX_COMPS_IN_SCAN == 4
 #define ASSIGN_STATE(dest,src)  \
-        ((dest).last_dc_val[0] = (src).last_dc_val[0], \
-         (dest).last_dc_val[1] = (src).last_dc_val[1], \
-         (dest).last_dc_val[2] = (src).last_dc_val[2], \
-         (dest).last_dc_val[3] = (src).last_dc_val[3])
+        ((dest).last_dc_val[0] = (src).last_dc_val[0], \
+         (dest).last_dc_val[1] = (src).last_dc_val[1], \
+         (dest).last_dc_val[2] = (src).last_dc_val[2], \
+         (dest).last_dc_val[3] = (src).last_dc_val[3])
 #endif
 #endif
 
 
 typedef struct {
   struct jpeg_entropy_decoder pub; /* public fields */
 
   /* These fields are loaded into local variables at start of each MCU.
    * In case of suspension, we exit WITHOUT updating them.
    */
-  bitread_perm_state bitstate;  /* Bit buffer at start of MCU */
-  savable_state saved;          /* Other state at start of MCU */
+  bitread_perm_state bitstate;  /* Bit buffer at start of MCU */
+  savable_state saved;          /* Other state at start of MCU */
 
   /* These fields are NOT loaded into local working state. */
-  unsigned int restarts_to_go;  /* MCUs left in this restart interval */
+  unsigned int restarts_to_go;  /* MCUs left in this restart interval */
 
   /* Pointers to derived tables (these workspaces have image lifespan) */
   d_derived_tbl * dc_derived_tbls[NUM_HUFF_TBLS];
   d_derived_tbl * ac_derived_tbls[NUM_HUFF_TBLS];
 
   /* Precalculated info set up by start_pass for use in decode_mcu: */
 
   /* Pointers to derived tables to be used for each block within an MCU */
   d_derived_tbl * dc_cur_tbls[D_MAX_BLOCKS_IN_MCU];
   d_derived_tbl * ac_cur_tbls[D_MAX_BLOCKS_IN_MCU];
@@ skipping 24 matching lines @@
       cinfo->Ah != 0 || cinfo->Al != 0)
     WARNMS(cinfo, JWRN_NOT_SEQUENTIAL);
 
   for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
     compptr = cinfo->cur_comp_info[ci];
     dctbl = compptr->dc_tbl_no;
     actbl = compptr->ac_tbl_no;
     /* Compute derived values for Huffman tables */
     /* We may do this more than once for a table, but it's not expensive */
     jpeg_make_d_derived_tbl(cinfo, TRUE, dctbl,
-                            & entropy->dc_derived_tbls[dctbl]);
+                            & entropy->dc_derived_tbls[dctbl]);
     jpeg_make_d_derived_tbl(cinfo, FALSE, actbl,
-                            & entropy->ac_derived_tbls[actbl]);
+                            & entropy->ac_derived_tbls[actbl]);
     /* Initialize DC predictions to 0 */
     entropy->saved.last_dc_val[ci] = 0;
   }
 
   /* Precalculate decoding info for each block in an MCU of this scan */
   for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
     ci = cinfo->MCU_membership[blkn];
     compptr = cinfo->cur_comp_info[ci];
     /* Precalculate which table to use for each block */
     entropy->dc_cur_tbls[blkn] = entropy->dc_derived_tbls[compptr->dc_tbl_no];
     entropy->ac_cur_tbls[blkn] = entropy->ac_derived_tbls[compptr->ac_tbl_no];
     /* Decide whether we really care about the coefficient values */
     if (compptr->component_needed) {
       entropy->dc_needed[blkn] = TRUE;
       /* we don't need the ACs if producing a 1/8th-size image */
-      entropy->ac_needed[blkn] = (compptr->DCT_scaled_size > 1);
+      entropy->ac_needed[blkn] = (compptr->_DCT_scaled_size > 1);
     } else {
       entropy->dc_needed[blkn] = entropy->ac_needed[blkn] = FALSE;
     }
   }
 
   /* Initialize bitread state variables */
   entropy->bitstate.bits_left = 0;
   entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */
   entropy->pub.insufficient_data = FALSE;
 
   /* Initialize restart counter */
   entropy->restarts_to_go = cinfo->restart_interval;
 }
 
 
 /*
  * Compute the derived values for a Huffman table.
  * This routine also performs some validation checks on the table.
  *
  * Note this is also used by jdphuff.c.
  */
 
 GLOBAL(void)
 jpeg_make_d_derived_tbl (j_decompress_ptr cinfo, boolean isDC, int tblno,
-                         d_derived_tbl ** pdtbl)
+                         d_derived_tbl ** pdtbl)
 {
   JHUFF_TBL *htbl;
   d_derived_tbl *dtbl;
   int p, i, l, si, numsymbols;
   int lookbits, ctr;
   char huffsize[257];
   unsigned int huffcode[257];
   unsigned int code;
 
   /* Note that huffsize[] and huffcode[] are filled in code-length order,
    * paralleling the order of the symbols themselves in htbl->huffval[].
    */
 
   /* Find the input Huffman table */
   if (tblno < 0 || tblno >= NUM_HUFF_TBLS)
     ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
   htbl =
     isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];
   if (htbl == NULL)
     ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);
 
   /* Allocate a workspace if we haven't already done so. */
   if (*pdtbl == NULL)
     *pdtbl = (d_derived_tbl *)
       (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-                                  SIZEOF(d_derived_tbl));
+                                  SIZEOF(d_derived_tbl));
   dtbl = *pdtbl;
-  dtbl->pub = htbl;             /* fill in back link */
-
+  dtbl->pub = htbl;             /* fill in back link */
+  
   /* Figure C.1: make table of Huffman code length for each symbol */
 
   p = 0;
   for (l = 1; l <= 16; l++) {
     i = (int) htbl->bits[l];
-    if (i < 0 || p + i > 256)   /* protect against table overrun */
+    if (i < 0 || p + i > 256)   /* protect against table overrun */
       ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
     while (i--)
       huffsize[p++] = (char) l;
   }
   huffsize[p] = 0;
   numsymbols = p;
-
+  
   /* Figure C.2: generate the codes themselves */
   /* We also validate that the counts represent a legal Huffman code tree. */
-
+  
   code = 0;
   si = huffsize[0];
   p = 0;
   while (huffsize[p]) {
     while (((int) huffsize[p]) == si) {
       huffcode[p++] = code;
       code++;
     }
     /* code is now 1 more than the last code used for codelength si; but
      * it must still fit in si bits, since no code is allowed to be all ones.
      */
     if (((INT32) code) >= (((INT32) 1) << si))
       ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
     code <<= 1;
     si++;
   }
 
   /* Figure F.15: generate decoding tables for bit-sequential decoding */
 
   p = 0;
   for (l = 1; l <= 16; l++) {
     if (htbl->bits[l]) {
       /* valoffset[l] = huffval[] index of 1st symbol of code length l,
        * minus the minimum code of length l
        */
       dtbl->valoffset[l] = (INT32) p - (INT32) huffcode[p];
       p += htbl->bits[l];
       dtbl->maxcode[l] = huffcode[p-1]; /* maximum code of length l */
     } else {
-      dtbl->maxcode[l] = -1;    /* -1 if no codes of this length */
+      dtbl->maxcode[l] = -1;    /* -1 if no codes of this length */
     }
   }
   dtbl->valoffset[17] = 0;
   dtbl->maxcode[17] = 0xFFFFFL; /* ensures jpeg_huff_decode terminates */
 
   /* Compute lookahead tables to speed up decoding.
    * First we set all the table entries to 0, indicating "too long";
    * then we iterate through the Huffman codes that are short enough and
    * fill in all the entries that correspond to bit sequences starting
    * with that code.
    */
 
    for (i = 0; i < (1 << HUFF_LOOKAHEAD); i++)
      dtbl->lookup[i] = (HUFF_LOOKAHEAD + 1) << HUFF_LOOKAHEAD;
 
   p = 0;
   for (l = 1; l <= HUFF_LOOKAHEAD; l++) {
     for (i = 1; i <= (int) htbl->bits[l]; i++, p++) {
       /* l = current code's length, p = its index in huffcode[] & huffval[]. */
       /* Generate left-justified code followed by all possible bit sequences */
       lookbits = huffcode[p] << (HUFF_LOOKAHEAD-l);
       for (ctr = 1 << (HUFF_LOOKAHEAD-l); ctr > 0; ctr--) {
-        dtbl->lookup[lookbits] = (l << HUFF_LOOKAHEAD) | htbl->huffval[p];
-        lookbits++;
+        dtbl->lookup[lookbits] = (l << HUFF_LOOKAHEAD) | htbl->huffval[p];
+        lookbits++;
       }
     }
   }
 
   /* Validate symbols as being reasonable.
    * For AC tables, we make no check, but accept all byte values 0..255.
    * For DC tables, we require the symbols to be in range 0..15.
    * (Tighter bounds could be applied depending on the data depth and mode,
    * but this is sufficient to ensure safe decoding.)
    */
   if (isDC) {
     for (i = 0; i < numsymbols; i++) {
       int sym = htbl->huffval[i];
       if (sym < 0 || sym > 15)
-        ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+        ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
     }
   }
 }
 
 
 /*
  * Out-of-line code for bit fetching (shared with jdphuff.c).
  * See jdhuff.h for info about usage.
  * Note: current values of get_buffer and bits_left are passed as parameters,
  * but are returned in the corresponding fields of the state struct.
  *
  * On most machines MIN_GET_BITS should be 25 to allow the full 32-bit width
  * of get_buffer to be used.  (On machines with wider words, an even larger
  * buffer could be used.)  However, on some machines 32-bit shifts are
  * quite slow and take time proportional to the number of places shifted.
  * (This is true with most PC compilers, for instance.)  In this case it may
  * be a win to set MIN_GET_BITS to the minimum value of 15.  This reduces the
  * average shift distance at the cost of more calls to jpeg_fill_bit_buffer.
  */
 
 #ifdef SLOW_SHIFT_32
-#define MIN_GET_BITS  15        /* minimum allowable value */
+#define MIN_GET_BITS  15        /* minimum allowable value */
 #else
 #define MIN_GET_BITS  (BIT_BUF_SIZE-7)
 #endif
 
 
 GLOBAL(boolean)
 jpeg_fill_bit_buffer (bitread_working_state * state,
-                      register bit_buf_type get_buffer, register int bits_left,
-                      int nbits)
+                      register bit_buf_type get_buffer, register int bits_left,
+                      int nbits)
 /* Load up the bit buffer to a depth of at least nbits */
 {
   /* Copy heavily used state fields into locals (hopefully registers) */
   register const JOCTET * next_input_byte = state->next_input_byte;
   register size_t bytes_in_buffer = state->bytes_in_buffer;
   j_decompress_ptr cinfo = state->cinfo;
 
   /* Attempt to load at least MIN_GET_BITS bits into get_buffer. */
   /* (It is assumed that no request will be for more than that many bits.) */
   /* We fail to do so only if we hit a marker or are forced to suspend. */
 
-  if (cinfo->unread_marker == 0) {      /* cannot advance past a marker */
+  if (cinfo->unread_marker == 0) {      /* cannot advance past a marker */
     while (bits_left < MIN_GET_BITS) {
       register int c;
 
       /* Attempt to read a byte */
       if (bytes_in_buffer == 0) {
-        if (! (*cinfo->src->fill_input_buffer) (cinfo))
-          return FALSE;
-        next_input_byte = cinfo->src->next_input_byte;
-        bytes_in_buffer = cinfo->src->bytes_in_buffer;
+        if (! (*cinfo->src->fill_input_buffer) (cinfo))
+          return FALSE;
+        next_input_byte = cinfo->src->next_input_byte;
+        bytes_in_buffer = cinfo->src->bytes_in_buffer;
       }
       bytes_in_buffer--;
       c = GETJOCTET(*next_input_byte++);
 
       /* If it's 0xFF, check and discard stuffed zero byte */
       if (c == 0xFF) {
-        /* Loop here to discard any padding FF's on terminating marker,
-         * so that we can save a valid unread_marker value.  NOTE: we will
-         * accept multiple FF's followed by a 0 as meaning a single FF data
-         * byte.  This data pattern is not valid according to the standard.
-         */
-        do {
-          if (bytes_in_buffer == 0) {
-            if (! (*cinfo->src->fill_input_buffer) (cinfo))
-              return FALSE;
-            next_input_byte = cinfo->src->next_input_byte;
-            bytes_in_buffer = cinfo->src->bytes_in_buffer;
-          }
-          bytes_in_buffer--;
-          c = GETJOCTET(*next_input_byte++);
-        } while (c == 0xFF);
+        /* Loop here to discard any padding FF's on terminating marker,
+         * so that we can save a valid unread_marker value.  NOTE: we will
+         * accept multiple FF's followed by a 0 as meaning a single FF data
+         * byte.  This data pattern is not valid according to the standard.
+         */
+        do {
+          if (bytes_in_buffer == 0) {
+            if (! (*cinfo->src->fill_input_buffer) (cinfo))
+              return FALSE;
+            next_input_byte = cinfo->src->next_input_byte;
+            bytes_in_buffer = cinfo->src->bytes_in_buffer;
+          }
+          bytes_in_buffer--;
+          c = GETJOCTET(*next_input_byte++);
+        } while (c == 0xFF);
 
-        if (c == 0) {
-          /* Found FF/00, which represents an FF data byte */
-          c = 0xFF;
-        } else {
-          /* Oops, it's actually a marker indicating end of compressed data.
-           * Save the marker code for later use.
-           * Fine point: it might appear that we should save the marker into
-           * bitread working state, not straight into permanent state.  But
-           * once we have hit a marker, we cannot need to suspend within the
-           * current MCU, because we will read no more bytes from the data
-           * source.  So it is OK to update permanent state right away.
-           */
-          cinfo->unread_marker = c;
-          /* See if we need to insert some fake zero bits. */
-          goto no_more_bytes;
-        }
+        if (c == 0) {
+          /* Found FF/00, which represents an FF data byte */
+          c = 0xFF;
+        } else {
+          /* Oops, it's actually a marker indicating end of compressed data.
+           * Save the marker code for later use.
+           * Fine point: it might appear that we should save the marker into
+           * bitread working state, not straight into permanent state.  But
+           * once we have hit a marker, we cannot need to suspend within the
+           * current MCU, because we will read no more bytes from the data
+           * source.  So it is OK to update permanent state right away.
+           */
+          cinfo->unread_marker = c;
+          /* See if we need to insert some fake zero bits. */
+          goto no_more_bytes;
+        }
       }
 
       /* OK, load c into get_buffer */
       get_buffer = (get_buffer << 8) | c;
       bits_left += 8;
     } /* end while */
   } else {
   no_more_bytes:
     /* We get here if we've read the marker that terminates the compressed
      * data segment.  There should be enough bits in the buffer register
      * to satisfy the request; if so, no problem.
      */
     if (nbits > bits_left) {
       /* Uh-oh.  Report corrupted data to user and stuff zeroes into
        * the data stream, so that we can produce some kind of image.
        * We use a nonvolatile flag to ensure that only one warning message
        * appears per data segment.
        */
       if (! cinfo->entropy->insufficient_data) {
-        WARNMS(cinfo, JWRN_HIT_MARKER);
-        cinfo->entropy->insufficient_data = TRUE;
+        WARNMS(cinfo, JWRN_HIT_MARKER);
+        cinfo->entropy->insufficient_data = TRUE;
       }
       /* Fill the buffer with zero bits */
       get_buffer <<= MIN_GET_BITS - bits_left;
       bits_left = MIN_GET_BITS;
     }
   }
 
   /* Unload the local registers */
   state->next_input_byte = next_input_byte;
   state->bytes_in_buffer = bytes_in_buffer;
   state->get_buffer = get_buffer;
   state->bits_left = bits_left;
 
   return TRUE;
 }
 
 
+/* Macro version of the above, which performs much better but does not
+   handle markers.  We have to hand off any blocks with markers to the
+   slower routines. */
+
+#define GET_BYTE \
+{ \
+  register int c0, c1; \
+  c0 = GETJOCTET(*buffer++); \
+  c1 = GETJOCTET(*buffer); \
+  /* Pre-execute most common case */ \
+  get_buffer = (get_buffer << 8) | c0; \
+  bits_left += 8; \
+  if (c0 == 0xFF) { \
+    /* Pre-execute case of FF/00, which represents an FF data byte */ \
+    buffer++; \
+    if (c1 != 0) { \
+      /* Oops, it's actually a marker indicating end of compressed data. */ \
+      cinfo->unread_marker = c1; \
+      /* Back out pre-execution and fill the buffer with zero bits */ \
+      buffer -= 2; \
+      get_buffer &= ~0xFF; \
+    } \
+  } \
+}
+
+#if __WORDSIZE == 64 || defined(_WIN64)
+
+/* Pre-fetch 48 bytes, because the holding register is 64-bit */
+#define FILL_BIT_BUFFER_FAST \
+  if (bits_left < 16) { \
+    GET_BYTE GET_BYTE GET_BYTE GET_BYTE GET_BYTE GET_BYTE \
+  }
+
+#else
+
+/* Pre-fetch 16 bytes, because the holding register is 32-bit */
+#define FILL_BIT_BUFFER_FAST \
+  if (bits_left < 16) { \
+    GET_BYTE GET_BYTE \
+  }
+
+#endif
+
+
 /*
  * Out-of-line code for Huffman code decoding.
  * See jdhuff.h for info about usage.
  */
 
 GLOBAL(int)
 jpeg_huff_decode (bitread_working_state * state,
-                  register bit_buf_type get_buffer, register int bits_left,
-                  d_derived_tbl * htbl, int min_bits)
+                  register bit_buf_type get_buffer, register int bits_left,
+                  d_derived_tbl * htbl, int min_bits)
 {
   register int l = min_bits;
   register INT32 code;
 
   /* HUFF_DECODE has determined that the code is at least min_bits */
   /* bits long, so fetch that many bits in one swoop. */
 
   CHECK_BIT_BUFFER(*state, l, return -1);
   code = GET_BITS(l);
 
   /* Collect the rest of the Huffman code one bit at a time. */
   /* This is per Figure F.16 in the JPEG spec. */
 
   while (code > htbl->maxcode[l]) {
     code <<= 1;
     CHECK_BIT_BUFFER(*state, 1, return -1);
     code |= GET_BITS(1);
     l++;
   }
 
   /* Unload the local registers */
   state->get_buffer = get_buffer;
   state->bits_left = bits_left;
 
   /* With garbage input we may reach the sentinel value l = 17. */
 
   if (l > 16) {
     WARNMS(state->cinfo, JWRN_HUFF_BAD_CODE);
-    return 0;                   /* fake a zero as the safest result */
+    return 0;                   /* fake a zero as the safest result */
   }
 
   return htbl->pub->huffval[ (int) (code + htbl->valoffset[l]) ];
 }
 
 
 /*
  * Figure F.12: extend sign bit.
  * On some machines, a shift and add will be faster than a table lookup.
  */
@@ skipping 99 matching lines @@
 
     if (entropy->ac_needed[blkn]) {
 
       /* Section F.2.2.2: decode the AC coefficients */
       /* Since zeroes are skipped, output area must be cleared beforehand */
       for (k = 1; k < DCTSIZE2; k++) {
         HUFF_DECODE(s, br_state, actbl, return FALSE, label2);
 
         r = s >> 4;
         s &= 15;
-
+      
         if (s) {
           k += r;
           CHECK_BIT_BUFFER(br_state, s, return FALSE);
           r = GET_BITS(s);
           s = HUFF_EXTEND(r, s);
           /* Output coefficient in natural (dezigzagged) order.
            * Note: the extra entries in jpeg_natural_order[] will save us
            * if k >= DCTSIZE2, which could happen if the data is corrupted.
            */
           (*block)[jpeg_natural_order[k]] = (JCOEF) s;
@@ skipping 27 matching lines @@
     }
   }
 
   /* Completed MCU, so update state */
   BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
   ASSIGN_STATE(entropy->saved, state);
   return TRUE;
 }
 
 
-/***************************************************************/
-
-#define ADD_BYTE  {                                     \
-  int val0 = *(buffer++);                               \
-  int val1 = *(buffer);                                 \
-                                                        \
-  bits_left += 8;                                       \
-  get_buffer = (get_buffer << 8) | (val0);              \
-  if (val0 == 0xFF) {                                   \
-    buffer++;                                           \
-    if (val1 != 0) {                                    \
-      buffer   -= 2;                                    \
-      get_buffer      &= ~0xFF;                         \
-    }                                                   \
-  }                                                     \
-}
-
-/***************************************************************/
-
-#if __WORDSIZE == 64 || defined(_WIN64)
-
-#define ENSURE_SHORT \
-  if (bits_left < 16) { \
-    ADD_BYTE ADD_BYTE ADD_BYTE ADD_BYTE ADD_BYTE ADD_BYTE \
-  }
-
-#else
-
-#define ENSURE_SHORT  if (bits_left < 16) { ADD_BYTE ADD_BYTE }
-
-#endif
-
-/***************************************************************/
-
-#define HUFF_DECODE_FAST(symbol, size, htbl) { \
-  ENSURE_SHORT \
-  symbol = PEEK_BITS(HUFF_LOOKAHEAD); \
-  symbol = htbl->lookup[symbol]; \
-  size = symbol >> 8; \
-  bits_left -= size; \
-  symbol = symbol & ((1 << HUFF_LOOKAHEAD) - 1); \
-  if (size == HUFF_LOOKAHEAD + 1) { \
-    symbol = (get_buffer >> bits_left) & ((1 << (size)) - 1); \
-    while (symbol > htbl->maxcode[size]) { \
-      symbol <<= 1; \
-      symbol |= GET_BITS(1); \
-      size++; \
-    } \
-    symbol = htbl->pub->huffval[ (int) (symbol + htbl->valoffset[size]) & 0xFF ]; \
-  } \
-}
-
-/***************************************************************/
-
 LOCAL(boolean)
 decode_mcu_fast (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 {
   huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
   BITREAD_STATE_VARS;
   JOCTET *buffer;
   int blkn;
   savable_state state;
   /* Outer loop handles each block in the MCU */
 
   /* Load up working state */
   BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
   buffer = (JOCTET *) br_state.next_input_byte;
   ASSIGN_STATE(state, entropy->saved);
 
   for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
     JBLOCKROW block = MCU_data[blkn];
     d_derived_tbl * dctbl = entropy->dc_cur_tbls[blkn];
     d_derived_tbl * actbl = entropy->ac_cur_tbls[blkn];
     register int s, k, r, l;
 
     HUFF_DECODE_FAST(s, l, dctbl);
     if (s) {
-      ENSURE_SHORT
+      FILL_BIT_BUFFER_FAST
       r = GET_BITS(s);
       s = HUFF_EXTEND(r, s);
     }
 
     if (entropy->dc_needed[blkn]) {
       int ci = cinfo->MCU_membership[blkn];
       s += state.last_dc_val[ci];
       state.last_dc_val[ci] = s;
       (*block)[0] = (JCOEF) s;
     }
 
     if (entropy->ac_needed[blkn]) {
 
       for (k = 1; k < DCTSIZE2; k++) {
         HUFF_DECODE_FAST(s, l, actbl);
         r = s >> 4;
         s &= 15;
-
+      
         if (s) {
           k += r;
-          ENSURE_SHORT
+          FILL_BIT_BUFFER_FAST
           r = GET_BITS(s);
           s = HUFF_EXTEND(r, s);
           (*block)[jpeg_natural_order[k]] = (JCOEF) s;
         } else {
           if (r != 15) break;
           k += 15;
         }
       }
 
     } else {
 
       for (k = 1; k < DCTSIZE2; k++) {
         HUFF_DECODE_FAST(s, l, actbl);
         r = s >> 4;
         s &= 15;
 
         if (s) {
           k += r;
-          ENSURE_SHORT
+          FILL_BIT_BUFFER_FAST
           DROP_BITS(s);
         } else {
           if (r != 15) break;
           k += 15;
         }
       }
     }
   }
 
+  if (cinfo->unread_marker != 0) {
+    cinfo->unread_marker = 0;
+    return FALSE;
+  }
+
   br_state.bytes_in_buffer -= (buffer - br_state.next_input_byte);
   br_state.next_input_byte = buffer;
   BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
   ASSIGN_STATE(entropy->saved, state);
   return TRUE;
 }
 
 
 /*
  * Decode and return one MCU's worth of Huffman-compressed coefficients.
  * The coefficients are reordered from zigzag order into natural array order,
  * but are not dequantized.
  *
  * The i'th block of the MCU is stored into the block pointed to by
  * MCU_data[i].  WE ASSUME THIS AREA HAS BEEN ZEROED BY THE CALLER.
  * (Wholesale zeroing is usually a little faster than retail...)
  *
  * Returns FALSE if data source requested suspension.  In that case no
  * changes have been made to permanent state.  (Exception: some output
  * coefficients may already have been assigned.  This is harmless for
  * this module, since we'll just re-assign them on the next call.)
  */
 
-#define BUFSIZE (DCTSIZE2 * 2u)
+#define BUFSIZE (DCTSIZE2 * 2)
 
 METHODDEF(boolean)
 decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
 {
   huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
   int usefast = 1;
 
   /* Process restart marker if needed; may have to suspend */
   if (cinfo->restart_interval) {
     if (entropy->restarts_to_go == 0)
       if (! process_restart(cinfo))
-        return FALSE;
+        return FALSE;
     usefast = 0;
   }
 
-  if (cinfo->src->bytes_in_buffer < BUFSIZE * cinfo->blocks_in_MCU)
+  if (cinfo->src->bytes_in_buffer < BUFSIZE * cinfo->blocks_in_MCU
+    || cinfo->unread_marker != 0)
     usefast = 0;
 
   /* If we've run out of data, just leave the MCU set to zeroes.
    * This way, we return uniform gray for the remainder of the segment.
    */
   if (! entropy->pub.insufficient_data) {
 
     if (usefast) {
-      if (!decode_mcu_fast(cinfo, MCU_data)) return FALSE;
+      if (!decode_mcu_fast(cinfo, MCU_data)) goto use_slow;
     }
     else {
+      use_slow:
       if (!decode_mcu_slow(cinfo, MCU_data)) return FALSE;
     }
 
   }
 
   /* Account for restart interval (no-op if not using restarts) */
   entropy->restarts_to_go--;
 
   return TRUE;
 }
 
 
 /*
  * Module initialization routine for Huffman entropy decoding.
  */
 
 GLOBAL(void)
 jinit_huff_decoder (j_decompress_ptr cinfo)
 {
   huff_entropy_ptr entropy;
   int i;
 
   entropy = (huff_entropy_ptr)
     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
-                                SIZEOF(huff_entropy_decoder));
+                                SIZEOF(huff_entropy_decoder));
   cinfo->entropy = (struct jpeg_entropy_decoder *) entropy;
   entropy->pub.start_pass = start_pass_huff_decoder;
   entropy->pub.decode_mcu = decode_mcu;
 
   /* Mark tables unallocated */
   for (i = 0; i < NUM_HUFF_TBLS; i++) {
     entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
   }
 }