Integrate SIMD optimisations for zlib

third_party/zlib/deflate.h

Index: third_party/zlib/deflate.h
diff --git a/third_party/zlib/deflate.h b/third_party/zlib/deflate.h
index 2fe6fd6670f5f6742655c736d08fd2c89b58dc6c..aff61a51f314fca12e38f1d27f7b67003750ccec 100644
--- a/third_party/zlib/deflate.h
+++ b/third_party/zlib/deflate.h
@@ -107,6 +107,10 @@ typedef struct internal_state {
     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) */
@@ -344,4 +348,27 @@ void ZLIB_INTERNAL _tr_stored_block OF((deflate_state *s, charf *buf,
               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 */