Integrate SIMD optimisations for zlib

third_party/zlib/deflate.h

 /* deflate.h -- internal compression state
  * Copyright (C) 1995-2010 Jean-loup Gailly
  * For conditions of distribution and use, see copyright notice in zlib.h
  */
 
 /* WARNING: this file should *not* be used by applications. It is
    part of the implementation of the compression library and is
    subject to change. Applications should only use zlib.h.
  */
 
@@ skipping 89 matching lines @@
     Bytef *pending_buf;  /* output still pending */
     ulg   pending_buf_size; /* size of pending_buf */
     Bytef *pending_out;  /* next pending byte to output to the stream */
     uInt   pending;      /* nb of bytes in the pending buffer */
     int   wrap;          /* bit 0 true for zlib, bit 1 true for gzip */
     gz_headerp  gzhead;  /* gzip header information to write */
     uInt   gzindex;      /* where in extra, name, or comment */
     Byte  method;        /* STORED (for zip only) or DEFLATED */
     int   last_flush;    /* value of flush param for previous deflate call */
 
+#ifdef HAVE_PCLMULQDQ
+    unsigned __attribute__((aligned(16))) crc0[4 * 5];
+#endif
+
                 /* used by deflate.c: */
 
     uInt  w_size;        /* LZ77 window size (32K by default) */
     uInt  w_bits;        /* log2(w_size)  (8..16) */
     uInt  w_mask;        /* w_size - 1 */
 
     Bytef *window;
     /* Sliding window. Input bytes are read into the second half of the window,
      * and move to the first half later to keep a dictionary of at least wSize
      * bytes. With this organization, matches are limited to a distance of
@@ skipping 217 matching lines @@
     s->dyn_ltree[_length_code[len]+LITERALS+1].Freq++; \
     s->dyn_dtree[d_code(dist)].Freq++; \
     flush = (s->last_lit == s->lit_bufsize-1); \
   }
 #else
 # define _tr_tally_lit(s, c, flush) flush = _tr_tally(s, 0, c)
 # define _tr_tally_dist(s, distance, length, flush) \
               flush = _tr_tally(s, distance, length)
 #endif
 
+/* ===========================================================================
+ * Update a hash value with the given input byte
+ * IN  assertion: all calls to to UPDATE_HASH are made with consecutive
+ *    input characters, so that a running hash key can be computed from the
+ *    previous key instead of complete recalculation each time.
+ */
+#ifdef USE_SSE4_2_CRC_HASH

[agl, 2014/09/23 21:41:40]

don't define internal macros in public headers.

+#define UPDATE_HASH(s,h,i) (\
+    {\
+        if (s->level < 6) \
+            h = (3483 * (s->window[i]) +\
+                 23081* (s->window[i+1]) +\
+                 6954 * (s->window[i+2]) +\
+                 20947* (s->window[i+3])) & s->hash_mask;\
+        else\
+            h = (25881* (s->window[i]) +\
+                 24674* (s->window[i+1]) +\
+                 25811* (s->window[i+2])) & s->hash_mask;\
+    })
+#else
+#define UPDATE_HASH(s,h,i) (h = (((h)<<s->hash_shift) ^ (s->window[i + (MIN_MATCH-1)])) & s->hash_mask)
+#endif
+
 #endif /* DEFLATE_H */