Split insert_string_sse into separate file in arch folder.

diff --git a/arch/x86/Makefile.in b/arch/x86/Makefile.in
index d91411ed776bad7ba50ea0b6d6861d99fc6516af..3604ba85226d728b34b463ffa6e232965f3df3b9 100644 (file)
--- a/arch/x86/Makefile.in
+++ b/arch/x86/Makefile.in
@@ -10,7 +10,7 @@ INCLUDES=
 SRCDIR=
 SRCTOP=
 
-all: x86.o x86.lo fill_window_sse.o fill_window_sse.lo deflate_quick.o deflate_quick.lo crc_folding.o crc_folding.lo
+all: x86.o x86.lo fill_window_sse.o fill_window_sse.lo deflate_quick.o deflate_quick.lo insert_string_sse.o insert_string_sse.lo crc_folding.o crc_folding.lo
 
 x86.o:
        $(CC) $(CFLAGS) $(INCLUDES) -c -o $@ $(SRCDIR)/x86.c
@@ -30,6 +30,12 @@ deflate_quick.o:
 deflate_quick.lo:
        $(CC) $(SFLAGS) -msse4 -DPIC $(INCLUDES) -c -o $@ $(SRCDIR)/deflate_quick.c
 
+insert_string_sse.o:
+       $(CC) $(CFLAGS) -msse4 $(INCLUDES) -c -o $@ $(SRCDIR)/insert_string_sse.c
+
+insert_string_sse.lo:
+       $(CC) $(SFLAGS) -msse4 -DPIC $(INCLUDES) -c -o $@ $(SRCDIR)/insert_string_sse.c
+
 crc_folding.o:
        $(CC) $(CFLAGS) -mpclmul -msse4 $(INCLUDES) -c -o $@ $(SRCDIR)/crc_folding.c
 

diff --git a/arch/x86/insert_string_sse.c b/arch/x86/insert_string_sse.c
new file mode 100644 (file)
index 0000000..4bbda50
--- /dev/null
+++ b/arch/x86/insert_string_sse.c
@@ -0,0 +1,47 @@
+/* insert_string_sse -- insert_string variant using SSE4.2's CRC instructions
+ *
+ * Copyright (C) 1995-2013 Jean-loup Gailly and Mark Adler
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ *
+ */
+
+#include "deflate.h"
+
+/* ===========================================================================
+ * Insert string str in the dictionary and set match_head to the previous head
+ * of the hash chain (the most recent string with same hash key). Return
+ * the previous length of the hash chain.
+ * IN  assertion: all calls to to INSERT_STRING are made with consecutive
+ *    input characters and the first MIN_MATCH bytes of str are valid
+ *    (except for the last MIN_MATCH-1 bytes of the input file).
+ */
+#ifdef X86_SSE4_2_CRC_HASH
+extern inline Pos insert_string_sse(deflate_state *const s, const Pos str, uInt count) {
+    Pos ret = 0;
+    uInt idx;
+    unsigned *ip, val, h = 0;
+
+    for (idx = 0; idx < count; idx++) {
+        ip = (unsigned *)&s->window[str+idx];
+        val = *ip;
+        h = 0;
+
+        if (s->level >= 6)
+            val &= 0xFFFFFF;
+
+#ifdef _MSC_VER
+        h = _mm_crc32_u32(h, val);
+#else
+        __asm__ __volatile__ (
+            "crc32 %1,%0\n\t"
+            : "+r" (h)
+            : "r" (val)
+        );
+#endif
+
+        ret = s->prev[(str+idx) & s->w_mask] = s->head[h & s->hash_mask];
+        s->head[h & s->hash_mask] = str+idx;
+    }
+    return ret;
+}
+#endif

diff --git a/configure b/configure
index 5b48d9144617cfbc64f09dd8b2ca41e67f45a15e..00526ff3a25d6ffe10cfe0f0e4024c5994216956 100755 (executable)
--- a/configure
+++ b/configure
@@ -703,6 +703,8 @@ case "${ARCH}" in
 
             CFLAGS="${CFLAGS} -DX86_SSE4_2_CRC_HASH"
             SFLAGS="${SFLAGS} -DX86_SSE4_2_CRC_HASH"
+            ARCH_STATIC_OBJS="${ARCH_STATIC_OBJS} insert_string_sse.o"
+            ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} insert_string_sse.lo"
 
             if test ${HAVE_PCLMULQDQ_INTRIN} -eq 1; then
                 CFLAGS="${CFLAGS} -DX86_PCLMULQDQ_CRC"

diff --git a/deflate_p.h b/deflate_p.h
index 25923a7228b24f61f80832c3f6a1dc544a0c4dc2..7fee6f02b27aba0e5245b8aa0f3cf1e495edef81 100644 (file)
--- a/deflate_p.h
+++ b/deflate_p.h
@@ -31,35 +31,9 @@ void flush_pending(z_stream *strm);
  *    input characters and the first MIN_MATCH bytes of str are valid
  *    (except for the last MIN_MATCH-1 bytes of the input file).
  */
-#ifdef X86_SSE4_2_CRC_HASH
-local inline Pos insert_string_sse(deflate_state *const s, const Pos str, uInt count) {
-    Pos ret = 0;
-    uInt idx;
-    unsigned *ip, val, h = 0;
-
-    for (idx = 0; idx < count; idx++) {
-        ip = (unsigned *)&s->window[str+idx];
-        val = *ip;
-        h = 0;
-
-        if (s->level >= 6)
-            val &= 0xFFFFFF;
-
-#ifdef _MSC_VER
-        h = _mm_crc32_u32(h, val);
-#else
-        __asm__ __volatile__ (
-            "crc32 %1,%0\n\t"
-            : "+r" (h)
-            : "r" (val)
-        );
-#endif
 
-        ret = s->prev[(str+idx) & s->w_mask] = s->head[h & s->hash_mask];
-        s->head[h & s->hash_mask] = str+idx;
-    }
-    return ret;
-}
+#ifdef X86_SSE4_2_CRC_HASH
+extern Pos insert_string_sse(deflate_state *const s, const Pos str, uInt count);
 #endif
 
 local inline Pos insert_string_c(deflate_state *const s, const Pos str, uInt count) {

diff --git a/win32/Makefile.msc b/win32/Makefile.msc
index e79ce5267655826315f027a206715f5012baae76..b6375ccb4f8aa37d034c4ebf3774cca6f3d5c3ec 100644 (file)
--- a/win32/Makefile.msc
+++ b/win32/Makefile.msc
@@ -31,7 +31,7 @@ DEFFILE = zlib.def
 WITH_GZFILEOP =
 
 OBJS = adler32.obj compress.obj crc32.obj deflate.obj deflate_fast.obj deflate_quick.obj deflate_slow.obj \
-       infback.obj inflate.obj inftrees.obj inffast.obj match.obj trees.obj uncompr.obj zutil.obj x86.obj fill_window_sse.obj crc_folding.obj
+       infback.obj inflate.obj inftrees.obj inffast.obj match.obj trees.obj uncompr.obj zutil.obj x86.obj fill_window_sse.obj insert_string_sse.obj crc_folding.obj
 !if "$(WITH_GZFILEOP)" != ""
 WFLAGS = $(WFLAGS) -DWITH_GZFILEOP
 OBJS = $(OBJS) gzclose.obj gzlib.obj gzread.obj gzwrite.obj