flate

commit 3c040f8374e0155983ffe86633b6050a16b13499
parent 54ebdaf31afea94fab7aad2224526e8a2d6cca05
Author: nsz <nszabolcs@gmail.com>
Date:   Tue, 12 May 2009 00:29:33 +0200

decode_block optimization
Diffstat:
Makefile
 | 
2
+-
inflate.c
 | 
40
++++++++++++++++++++++++++++------------

2 files changed, 29 insertions(+), 13 deletions(-)
diff --git a/Makefile b/Makefile
@@ -18,7 +18,7 @@ clean:
 
 gcov: inflate.c
 	gcc -fprofile-arcs -ftest-coverage -pg -g -Wall $<
-	cat b.dat | ./a.out > /dev/null
+	cat d.dat | ./a.out > /dev/null
 	gcov -b $< > /dev/null
 	gprof a.out > $<.gprof
 	gcc -g -Wall $<
diff --git a/inflate.c b/inflate.c
@@ -1,10 +1,3 @@
-/*
-TODO:
-	check int types
-	better io model
-	better error handling
-*/
-
 #include <stdlib.h>
 #include <string.h>
 
@@ -414,13 +407,36 @@ static void decode_block(Stream *s, Huff *lhuff, Huff *dhuff) {
 			dist = distbase[sym] + getbits(s, distbits[sym]);
 			if (s->error != FlateOk)
 				return;
-			/* copy match */
-			/* TODO: unroll 3-4 loops */
-			for (i = 0; i < len; i++) {
-				s->win[s->pos] = s->win[(s->pos - dist) & 0x7ff];
-				s->pos++;
+			/* copy match, loop unroll for common case */
+			if (s->pos + len <= WinSize) {
+				uint pos = s->pos;
+
+				while (len > 4) {
+					s->win[pos] = s->win[(pos - dist) & 0x7ff];
+					pos++;
+					s->win[pos] = s->win[(pos - dist) & 0x7ff];
+					pos++;
+					s->win[pos] = s->win[(pos - dist) & 0x7ff];
+					pos++;
+					s->win[pos] = s->win[(pos - dist) & 0x7ff];
+					pos++;
+					len -= 4;
+				}
+				while (len > 0) {
+					s->win[pos] = s->win[(pos - dist) & 0x7ff];
+					pos++;
+					len--;
+				}
+				s->pos = pos;
 				if (!checkpos(s))
 					return;
+			} else {
+				for (i = 0; i < len; i++) {
+					s->win[s->pos] = s->win[(s->pos - dist) & 0x7ff];
+					s->pos++;
+					if (!checkpos(s))
+						return;
+				}
 			}
 		}
 	}