summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--CMakeLists.txt18
-rw-r--r--Makefile.in16
-rw-r--r--arch/aarch64/Makefile.in49
-rw-r--r--arch/aarch64/crc32_acle.c81
-rw-r--r--arch/aarch64/insert_string_acle.c50
-rwxr-xr-xconfigure27
-rw-r--r--crc32.c2
7 files changed, 230 insertions, 13 deletions
diff --git a/CMakeLists.txt b/CMakeLists.txt
index a84da4a..4dac05f 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -58,7 +58,7 @@ option(WITH_OPTIM "Build with optimisation" ON)
option(WITH_NEW_STRATEGIES "Use new strategies" ON)
option(WITH_NATIVE_INSTRUCTIONS
"Instruct the compiler to use the full instruction set on this host (gcc/clang -march=native)" OFF)
-if("${ARCH}" MATCHES "arm")
+if("${ARCH}" MATCHES "arm" OR "${ARCH}" MATCHES "aarch64")
option(WITH_ACLE "Build with ACLE CRC" OFF)
endif()
@@ -125,6 +125,8 @@ else()
endif()
if("${ARCH}" MATCHES "arm")
set(ACLEFLAG "-march=armv8-a+crc")
+ elseif("${ARCH}" MATCHES "aarch64")
+ set(ACLEFLAG "-march=armv8-a+crc")
endif()
else(NOT NATIVEFLAG)
set(SSE2FLAG ${NATIVEFLAG})
@@ -132,6 +134,8 @@ else()
set(PCLMULFLAG ${NATIVEFLAG})
if("${ARCH}" MATCHES "arm")
set(ACLEFLAG "${NATIVEFLAG}")
+ elseif("${ARCH}" MATCHES "aarch64")
+ set(ACLEFLAG "${NATIVEFLAG}")
endif()
endif(NOT NATIVEFLAG)
endif()
@@ -140,7 +144,7 @@ add_feature_info(ZLIB_COMPAT ZLIB_COMPAT "Provide a zlib-compatible API")
add_feature_info(WITH_GZFILEOP WITH_GZFILEOP "Compile with support for gzFile-related functions")
add_feature_info(WITH_OPTIM WITH_OPTIM "Build with optimisation")
add_feature_info(WITH_NEW_STRATEGIES WITH_NEW_STRATEGIES "Use new strategies")
-if("${ARCH}" MATCHES "arm")
+if("${ARCH}" MATCHES "arm" OR "${ARCH}" MATCHES "aarch64")
add_feature_info(WITH_ACLE WITH_ACLE "Build with ACLE CRC")
endif()
@@ -333,6 +337,9 @@ if("${ARCH}" MATCHES "x86_64" OR "${ARCH}" MATCHES "AMD64")
elseif("${ARCH}" MATCHES "arm")
set(ARCHDIR "arch/arm")
add_definitions(-DUNALIGNED_OK -DUNROLL_LESS)
+elseif("${ARCH}" MATCHES "aarch64")
+ set(ARCHDIR "arch/aarch64")
+ add_definitions(-DUNALIGNED_OK -DUNROLL_LESS)
else()
set(ARCHDIR "arch/x86")
add_definitions(-DX86 -DUNALIGNED_OK -DUNROLL_LESS)
@@ -346,6 +353,13 @@ if(WITH_OPTIM)
add_definitions("-DARM_ACLE_CRC_HASH")
add_feature_info(ACLE_CRC 1 "Support CRC hash generation using the ACLE instruction set, using \"${ACLEFLAG}\"")
endif()
+ elseif("${ARCH}" MATCHES "aarch64")
+ if(WITH_ACLE)
+ set(ZLIB_ARCH_SRCS ${ZLIB_ARCH_SRCS} ${ARCHDIR}/crc32_acle.c ${ARCHDIR}/insert_string_acle.c)
+ add_definitions("-DARM_ACLE_CRC_HASH")
+ add_feature_info(ACLE_CRC 1 "Support CRC hash generation using the ACLE instruction set, using \"${ACLEFLAG}\"")
+ set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${ACLEFLAG}")
+ endif()
elseif("${ARCHDIR}" MATCHES "arch/x86")
add_definitions("-DX86_CPUID")
set(ZLIB_ARCH_SRCS ${ZLIB_ARCH_SRCS} ${ARCHDIR}/x86.c)
diff --git a/Makefile.in b/Makefile.in
index 9ad7512..12d824e 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -160,37 +160,37 @@ endif
endif
example$(EXE): example.o $(STATICLIB)
- $(CC) $(LDFLAGS) -o $@ example.o $(TEST_LIBS)
+ $(CC) $(LDFLAGS) -o $@ example.o $(TEST_LIBS) $(LDSHAREDLIBC)
ifneq ($(STRIP),)
$(STRIP) $@
endif
minigzip$(EXE): minigzip.o $(STATICLIB)
- $(CC) $(LDFLAGS) -o $@ minigzip.o $(TEST_LIBS)
+ $(CC) $(LDFLAGS) -o $@ minigzip.o $(TEST_LIBS) $(LDSHAREDLIBC)
ifneq ($(STRIP),)
$(STRIP) $@
endif
examplesh$(EXE): example.o $(SHAREDTARGET)
- $(CC) $(LDFLAGS) -o $@ example.o $(SHAREDTARGET)
+ $(CC) $(LDFLAGS) -o $@ example.o $(SHAREDTARGET) $(LDSHAREDLIBC)
ifneq ($(STRIP),)
$(STRIP) $@
endif
minigzipsh$(EXE): minigzip.o $(SHAREDTARGET)
- $(CC) $(LDFLAGS) -o $@ minigzip.o $(SHAREDTARGET)
+ $(CC) $(LDFLAGS) -o $@ minigzip.o $(SHAREDTARGET) $(LDSHAREDLIBC)
ifneq ($(STRIP),)
$(STRIP) $@
endif
example64$(EXE): example64.o $(STATICLIB)
- $(CC) $(LDFLAGS) -o $@ example64.o $(TEST_LIBS)
+ $(CC) $(LDFLAGS) -o $@ example64.o $(TEST_LIBS) $(LDSHAREDLIBC)
ifneq ($(STRIP),)
$(STRIP) $@
endif
minigzip64$(EXE): minigzip64.o $(STATICLIB)
- $(CC) $(LDFLAGS) -o $@ minigzip64.o $(TEST_LIBS)
+ $(CC) $(LDFLAGS) -o $@ minigzip64.o $(TEST_LIBS) $(LDSHAREDLIBC)
ifneq ($(STRIP),)
$(STRIP) $@
endif
@@ -323,6 +323,8 @@ inffast.o: $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h $(SRCDIR)/inftrees.h $(SRC
inftrees.o: $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h $(SRCDIR)/inftrees.h
trees.o: $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h $(SRCDIR)/trees.h
zutil.o: $(SRCDIR)/zutil.h $(SRCDIR)/gzguts.h $(SRCDIR)/zlib.h zconf.h
+arch/aarch64/crc32_acle.o: zconf.h
+arch/aarch64/insert_string_acle.o: $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h
arch/arm/crc32_acle.o: zconf.h
arch/arm/insert_string_acle.o: $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h
arch/x86/crc_folding.o: $(SRCDIR)/arch/x86/crc_folding.h $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h
@@ -344,6 +346,8 @@ inffast.lo: $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h $(SRCDIR)/inftrees.h $(SR
inftrees.lo: $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h $(SRCDIR)/inftrees.h
trees.lo: $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h $(SRCDIR)/trees.h
zutil.lo: $(SRCDIR)/zutil.h $(SRCDIR)/gzguts.h $(SRCDIR)/zlib.h zconf.h
+arch/aarch64/crc32_acle.lo: zconf.h
+arch/aarch64/insert_string_acle.lo: $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h
arch/arm/crc32_acle.lo: zconf.h
arch/arm/insert_string_acle.lo: $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h
arch/x86/crc_folding.lo: $(SRCDIR)/arch/x86/crc_folding.h $(SRCDIR)/deflate.h $(SRCDIR)/zutil.h $(SRCDIR)/zlib.h zconf.h
diff --git a/arch/aarch64/Makefile.in b/arch/aarch64/Makefile.in
new file mode 100644
index 0000000..a2c96bd
--- /dev/null
+++ b/arch/aarch64/Makefile.in
@@ -0,0 +1,49 @@
+# Makefile for zlib
+# Copyright (C) 1995-2013 Jean-loup Gailly, Mark Adler
+# For conditions of distribution and use, see copyright notice in zlib.h
+
+CC=
+CFLAGS=
+SFLAGS=
+INCLUDES=
+
+SRCDIR=.
+SRCTOP=../..
+TOPDIR=$(SRCTOP)
+
+all: crc32_acle.o crc32_acle.lo insert_string_acle.o insert_string_acle.lo
+
+crc32_acle.o: $(SRCDIR)/crc32_acle.c
+ $(CC) $(CFLAGS) $(INCLUDES) -c -o $@ $(SRCDIR)/crc32_acle.c
+
+crc32_acle.lo: $(SRCDIR)/crc32_acle.c
+ $(CC) $(SFLAGS) $(INCLUDES) -c -o $@ $(SRCDIR)/crc32_acle.c
+
+insert_string_acle.o: $(SRCDIR)/insert_string_acle.c
+ $(CC) $(CFLAGS) $(INCLUDES) -c -o $@ $(SRCDIR)/insert_string_acle.c
+
+insert_string_acle.lo: $(SRCDIR)/insert_string_acle.c
+ $(CC) $(SFLAGS) $(INCLUDES) -c -o $@ $(SRCDIR)/insert_string_acle.c
+
+mostlyclean: clean
+clean:
+ rm -f *.o *.lo *~
+ rm -rf objs
+ rm -f *.gcda *.gcno *.gcov
+
+distclean:
+ rm -f Makefile
+
+depend:
+ makedepend -Y -- $(CFLAGS) -- $(SRCDIR)/*.c
+ makedepend -Y -a -o.lo -- $(SFLAGS) -- $(SRCDIR)/*.c
+ @sed "s=^$(SRCDIR)/\([a-zA-Z0-9_]*\.\(lo\|o\):\)=\1=g" < Makefile > Makefile.tmp
+ @mv -f Makefile.tmp Makefile
+
+# DO NOT DELETE THIS LINE -- make depend depends on it.
+
+crc32_acle.o: $(TOPDIR)/zconf.h
+insert_string_acle.o: $(SRCTOP)/deflate.h $(SRCTOP)/zutil.h $(SRCTOP)/zlib.h $(TOPDIR)/zconf.h
+
+crc32_acle.lo: $(TOPDIR)/zconf.h
+insert_string_acle.lo: $(SRCTOP)/deflate.h $(SRCTOP)/zutil.h $(SRCTOP)/zlib.h $(TOPDIR)/zconf.h
diff --git a/arch/aarch64/crc32_acle.c b/arch/aarch64/crc32_acle.c
new file mode 100644
index 0000000..e43cbb5
--- /dev/null
+++ b/arch/aarch64/crc32_acle.c
@@ -0,0 +1,81 @@
+/* crc32_acle.c -- compute the CRC-32 of a data stream
+ * Copyright (C) 1995-2006, 2010, 2011, 2012 Mark Adler
+ * Copyright (C) 2016 Yang Zhang
+ * For conditions of distribution and use, see copyright notice in zlib.h
+ *
+*/
+
+#ifdef __ARM_FEATURE_CRC32
+#include <arm_acle.h>
+#include "zconf.h"
+#ifdef __linux__
+# include <stddef.h>
+#endif
+
+uint32_t crc32_acle(uint32_t crc, const unsigned char *buf, size_t len)
+{
+ register uint32_t c;
+ register const uint16_t *buf2;
+ register const uint32_t *buf4;
+ register const uint64_t *buf8;
+
+ c = ~crc;
+ if (len && ((ptrdiff_t)buf & 1)) {
+ c = __crc32b(c, *buf++);
+ len--;
+ }
+
+ if ((len > 2) && ((ptrdiff_t)buf & 2)) {
+ buf2 = (const uint16_t *) buf;
+ c = __crc32h(c, *buf2++);
+ len -= 2;
+ buf4 = (const uint32_t *) buf2;
+ } else {
+ buf4 = (const uint32_t *) buf;
+ }
+
+ if ((len > 4) && ((ptrdiff_t)buf & 4)) {
+ c = __crc32b(c, *buf4++);
+ len -= 4;
+ }
+
+ buf8 = (const uint64_t *) buf4;
+
+#ifndef UNROLL_LESS
+ while (len >= 32) {
+ c = __crc32b(c, *buf8++);
+ c = __crc32b(c, *buf8++);
+ c = __crc32b(c, *buf8++);
+ c = __crc32b(c, *buf8++);
+ len -= 32;
+ }
+#endif
+
+ while (len >= 8) {
+ c = __crc32b(c, *buf8++);
+ len -= 8;
+ }
+
+ if (len >= 4) {
+ buf4 = (const uint32_t *) buf8;
+ c = __crc32w(c, *buf4++);
+ len -= 4;
+ buf2 = (const uint16_t *) buf4;
+ } else {
+ buf2 = (const uint16_t *) buf8;
+ }
+
+ if (len >= 2) {
+ c = __crc32h(c, *buf2++);
+ len -= 2;
+ }
+
+ if (len) {
+ buf = (const unsigned char *) buf2;
+ c = __crc32b(c, *buf);
+ }
+
+ c = ~c;
+ return c;
+}
+#endif
diff --git a/arch/aarch64/insert_string_acle.c b/arch/aarch64/insert_string_acle.c
new file mode 100644
index 0000000..985b726
--- /dev/null
+++ b/arch/aarch64/insert_string_acle.c
@@ -0,0 +1,50 @@
+/* insert_string_acle.c -- insert_string variant using ACLE'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
+ *
+ */
+
+#ifdef __ARM_FEATURE_CRC32
+#include <arm_acle.h>
+#endif
+#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 ARM_ACLE_CRC_HASH
+Pos insert_string_acle(deflate_state *const s, const Pos str, unsigned int count) {
+ Pos p, lp;
+
+ if (unlikely(count == 0)) {
+ return s->prev[str & s->w_mask];
+ }
+
+ lp = str + count - 1; /* last position */
+
+ for (p = str; p <= lp; p++) {
+ unsigned *ip, val, h, hm;
+
+ ip = (unsigned *)&s->window[p];
+ val = *ip;
+
+ if (s->level >= TRIGGER_LEVEL)
+ val &= 0xFFFFFF;
+
+ h = __crc32w(0, val);
+ hm = h & s->hash_mask;
+
+ if (s->head[hm] != p) {
+ s->prev[p & s->w_mask] = s->head[hm];
+ s->head[hm] = p;
+ }
+ }
+ return s->prev[lp & s->w_mask];
+}
+#endif
diff --git a/configure b/configure
index 903698a..256c421 100755
--- a/configure
+++ b/configure
@@ -254,6 +254,10 @@ if test "$gcc" -eq 1 && ($cc $CFLAGS -c $test.c) >> configure.log 2>&1; then
ARCH=native
fi
fi ;;
+ aarch64)
+ if test "${uname}" = "elf"; then
+ uname=aarch64
+ fi ;;
esac
CFLAGS="${CFLAGS--O3}"
if test -n "${ARCHS}"; then
@@ -376,6 +380,9 @@ if test "$gcc" -eq 1 && ($cc $CFLAGS -c $test.c) >> configure.log 2>&1; then
AR="/usr/bin/libtool"
fi
ARFLAGS="-o" ;;
+ aarch64) LDSHARED=${LDSHARED-"$cc"}
+ LDSHAREDFLAGS="-shared -Wl,-soname,libz.so.1 -Wl,--version-script,${SRCDIR}/zlib.map"
+ LDSHAREDLIBC="-Wl,--start-group -lc -lrdimon -Wl,--end-group" ;;
*) LDSHARED=${LDSHARED-"$cc"}
LDSHAREDFLAGS="-shared" ;;
esac
@@ -480,7 +487,7 @@ if test $shared -eq 1; then
echo Checking for shared library support... | tee -a configure.log
# we must test in two steps (cc then ld), required at least on SunOS 4.x
if try $CC -w -c $SFLAGS $test.c &&
- try $LDSHARED $LDSHAREDFLAGS $LDFLAGS -o $test$shared_ext $test.o; then
+ try $LDSHARED $LDSHAREDFLAGS $LDFLAGS -o $test$shared_ext $test.o $LDSHAREDLIBC; then
echo Building shared library $SHAREDTARGET with $CC. | tee -a configure.log
elif test -z "$old_cc" -a -z "$old_cflags"; then
echo No shared library support. | tee -a configure.log
@@ -545,7 +552,7 @@ int main(void) {
return 0;
}
EOF
- if try $CC $CFLAGS -o $test $test.c; then
+ if try $CC $CFLAGS -o $test $test.c $LDSHAREDLIBC; then
echo "Checking for fseeko... Yes." | tee -a configure.log
else
CFLAGS="${CFLAGS} -DNO_FSEEKO"
@@ -562,7 +569,7 @@ cat > $test.c <<EOF
#include <errno.h>
int main() { return strlen(strerror(errno)); }
EOF
-if try $CC $CFLAGS -o $test $test.c; then
+if try $CC $CFLAGS -o $test $test.c $LDSHAREDLIBC; then
echo "Checking for strerror... Yes." | tee -a configure.log
else
CFLAGS="${CFLAGS} -DNO_STRERROR"
@@ -704,7 +711,7 @@ int main(void)
return 0;
}
EOF
-if try ${CC} ${CFLAGS} $test.c; then
+if try ${CC} ${CFLAGS} $test.c $LDSHAREDLIBC; then
echo "Checking for __builtin_ctzl ... Yes." | tee -a configure.log
CFLAGS="$CFLAGS -DHAVE_BUILTIN_CTZL"
SFLAGS="$SFLAGS -DHAVE_BUILTIN_CTZL"
@@ -843,6 +850,18 @@ case "${ARCH}" in
esac
;;
+ # 64-bit ARM specific optimizations
+ aarch64)
+ ARCHDIR=arch/aarch64
+ if test $buildacle -eq 1; then
+ CFLAGS="-march=armv8-a+crc ${CFLAGS} -DARM_ACLE_CRC_HASH"
+ SFLAGS="-march=armv8-a+crc ${SFLAGS} -DARM_ACLE_CRC_HASH"
+ ARCH_STATIC_OBJS="${ARCH_STATIC_OBJS} crc32_acle.o insert_string_acle.o"
+ ARCH_SHARED_OBJS="${ARCH_SHARED_OBJS} crc32_acle.lo insert_string_acle.lo"
+ fi
+ CFLAGS="${CFLAGS} -DUNALIGNED_OK -DUNROLL_LESS"
+ SFLAGS="${SFLAGS} -DUNALIGNED_OK -DUNROLL_LESS"
+ ;;
esac
echo "ARCH: ${ARCH}"
diff --git a/crc32.c b/crc32.c
index 000a868..8ad1414 100644
--- a/crc32.c
+++ b/crc32.c
@@ -23,7 +23,7 @@
# endif
#elif defined(__linux__)
# include <endian.h>
-#elif defined(__APPLE__) || defined(__arm__)
+#elif defined(__APPLE__) || defined(__arm__) || defined(__aarch64__)
# include <machine/endian.h>
#elif defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__bsdi__) || defined(__DragonFly__)
# include <sys/endian.h>
generated by cgit v1.2.3 (git 2.25.1) at 2025-10-03 09:02:18 +0000