another 12% better perf for compress

README.md

@@ -31,9 +31,9 @@ Each file is compressed 1000 times.
 **https://github.com/guzba/zippy** compress results:
 File | Time | % Size Reduction
 --- | --- | ---:
-alice29.txt | 6.3391s | 62.33%
-urls.10K | 19.9189s | 67.01%
-rfctest3.gold | 1.1809s | 70.91%
+alice29.txt | 4.4424s | 62.33%
+urls.10K | 16.9259s | 67.01%
+rfctest3.gold | 1.0140s | 70.91%
 randtest3.gold | 0.1285s | 0%
 
 https://github.com/nim-lang/zip compress results: (Requires zlib1.dll)

src/zippy/common.nim

@@ -104,6 +104,9 @@ when defined(release):
 template read32*(p: pointer): uint32 =
   cast[ptr uint32](p)[]
 
+template read64*(p: pointer): uint64 =
+  cast[ptr uint64](p)[]
+
 func adler32*(data: seq[uint8]): uint32 =
   ## See https://github.com/madler/zlib/blob/master/adler32.c
 

src/zippy/deflate.nim

@@ -283,16 +283,22 @@ func lz77Encode(src: seq[uint8]): (seq[uint16], seq[int], seq[int], int) =
 
       prevOffset = offset
 
-      var matchLen: int
-      if (
-        (read32(src[pos - offset].unsafeAddr) xor read32(src[pos].unsafeAddr)
-      ) shl 8) == 0:
-        # The first 3 bytes match (the hash of them got us here so usually do)
-        inc(matchLen, 3)
-        for i in 3 ..< stop - pos:
-          if src[pos - offset + i] != src[pos + i]:
+      var i, matchLen: int
+      while i < stop - pos:
+        let
+          bytesToCheck = min(8, stop - pos - i)
+          v = read64(src[pos - offset + i].unsafeAddr) xor
+            read64(src[pos + i].unsafeAddr)
+        if v == 0:
+          inc(matchLen, bytesToCheck)
+        else:
+          let
+            zeroBits = countTrailingZeroBits(v)
+            matchingBytes = min(zeroBits div 8, bytesToCheck)
+          inc(matchLen, matchingBytes)
+          if matchingBytes < bytesToCheck:
             break
-          inc matchLen
+        inc(i, bytesToCheck)
 
       if matchLen > longestMatchLen:
         longestMatchLen = matchLen

tests/benchmark.nim

@@ -1,5 +1,5 @@
 import std/monotimes, strformat, zip/zlib, zippy
-# import miniz, nimPNG/nimz,
+# import miniz, nimPNG/nimz
 
 const
   zs = [