faster compress, ~5%

src/zippy.nim

@@ -114,11 +114,11 @@ func uncompress(
 
     inflate(src[pos ..< ^8], dst)
 
-    let checksum = cast[ptr uint32](src[src.len - 8].unsafeAddr)[]
+    let checksum = read32(src[src.len - 8].unsafeAddr)
     if checksum != crc32(dst):
       raise newException(ZippyError, "Checksum verification failed")
 
-    let isize = cast[ptr uint32](src[src.len - 4].unsafeAddr)[]
+    let isize = read32(src[src.len - 4].unsafeAddr)
     if isize != dst.len.uint32:
       raise newException(ZippyError, "Size verification failed")
   of dfZlib:

src/zippy/common.nim

@@ -101,6 +101,9 @@ const
 when defined(release):
   {.push checks: off.}
 
+template read32*(p: pointer): uint32 =
+  cast[ptr uint32](p)[]
+
 func adler32*(data: seq[uint8]): uint32 =
   ## See https://github.com/madler/zlib/blob/master/adler32.c
 

src/zippy/deflate.nim

@@ -284,10 +284,15 @@ func lz77Encode(src: seq[uint8]): (seq[uint16], seq[int], seq[int], int) =
       prevOffset = offset
 
       var matchLen: int
-      for i in 0 ..< stop - pos:
-        if src[pos - offset + i] != src[pos + i]:
-          break
-        inc matchLen
+      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]:
+            break
+          inc matchLen
 
       if matchLen > longestMatchLen:
         longestMatchLen = matchLen

tests/benchmark.nim

@@ -1,4 +1,5 @@
-import miniz, nimPNG/nimz, std/monotimes, strformat, zip/zlib, zippy
+import std/monotimes, strformat, zip/zlib, zippy
+# import miniz, nimPNG/nimz,
 
 const
   zs = [