s390x: vectorize crc32

Use vector extensions when compiling for s390x and binutils knows
about them. At runtime, check whether kernel supports vector
extensions (it has to be not just the CPU, but also the kernel) and
choose between the regular and the vectorized implementations.

Author: iii-i

CMakeLists.txt

@@ -26,6 +26,7 @@ option(ZLIB_BUILD_MINIZIP "Enable building libminizip contrib library" OFF)
 option(ZLIB_INSTALL "Enable installation of zlib" ON)
 option(ZLIB_PREFIX "prefix for all types and library functions, see zconf.h.in"
        OFF)
+option(CRC32VX "Enable building S390-CRC32VX implementation" ON)
 mark_as_advanced(ZLIB_PREFIX)
 
 if(WIN32)
@@ -120,6 +121,56 @@ set(ZLIB_PC ${zlib_BINARY_DIR}/zlib.pc)
 configure_file(${zlib_SOURCE_DIR}/zlib.pc.cmakein ${ZLIB_PC} @ONLY)
 configure_file(${zlib_BINARY_DIR}/zconf.h.cmakein ${zlib_BINARY_DIR}/zconf.h)
 
+#
+# Check for IBM S390X extensions
+#
+if(CRC32VX)
+
+    # preset the compiler specific flags
+    if (CMAKE_C_COMPILER_ID STREQUAL "Clang")
+        set(VGFMAFLAG "-fzvector")
+    else()
+        set(VGFMAFLAG "-mzarch")
+    endif()
+
+    set(S390X_VX_TEST
+        "#ifndef __s390x__ \n\
+        #error \n\
+        #endif \n\
+        #include <vecintrin.h> \n\
+        int main(void) { \
+            unsigned long long a __attribute__((vector_size(16))) = { 0 }; \
+           unsigned long long b __attribute__((vector_size(16))) = { 0 }; \
+           unsigned char c __attribute__((vector_size(16))) = { 0 }; \
+           c = vec_gfmsum_accum_128(a, b, c); \
+           return c[0]; \
+        }")
+
+    # cflags already contains a valid march
+    set(CMAKE_REQUIRED_FLAGS "${VGFMAFLAG}")
+    check_c_source_compiles("${S390X_VX_TEST}" HAS_S390X_VX_SUPPORT)
+    set(CMAKE_REQUIRED_FLAGS)
+
+    # or set march for our compile units
+    if(NOT HAS_S390X_VX_SUPPORT)
+        set(CMAKE_REQUIRED_FLAGS "${VGFMAFLAG} -march=z13")
+        check_c_source_compiles("${S390X_VX_TEST}" HAS_Z13_S390X_VX_SUPPORT)
+        set(CMAKE_REQUIRED_FLAGS )
+        list(APPEND VGFMAFLAG "-march=z13")
+    endif()
+
+    # prepare compiling for s390x
+    if(HAS_S390X_VX_SUPPORT OR HAS_Z13_S390X_VX_SUPPORT)
+        set(S390X_SRCS
+            contrib/s390x/s390x-functable.c
+            contrib/s390x/crc32-vx.c)
+        set_source_files_properties(
+            contrib/s390x/crc32-vx.c
+            PROPERTIES COMPILE_OPTIONS "${VGFMAFLAG}")
+        add_definitions(-DHAVE_S390X_VX)
+    endif()
+endif()
+
 # ============================================================================
 # zlib
 # ============================================================================
@@ -152,7 +203,8 @@ set(ZLIB_SRCS
     inffast.c
     trees.c
     uncompr.c
-    zutil.c)
+    zutil.c
+    ${S390X_SRCS})
 
 if(WIN32)
     set(zlib_static_suffix "s")

Makefile.in

@@ -27,6 +27,7 @@ LDFLAGS=
 TEST_LIBS=-L. libz.a
 LDSHARED=$(CC)
 CPP=$(CC) -E
+VGFMAFLAG=
 
 STATICLIB=libz.a
 SHAREDLIB=libz.so
@@ -164,6 +165,12 @@ adler32.o: $(SRCDIR)adler32.c
 crc32.o: $(SRCDIR)crc32.c
 	$(CC) $(CFLAGS) $(ZINC) -c -o $@ $(SRCDIR)crc32.c
 
+crc32-vx.o: $(SRCDIR)contrib/s390x/crc32-vx.c
+	$(CC) $(CFLAGS) $(VGFMAFLAG) $(ZINC) -c -o $@ $(SRCDIR)contrib/s390x/crc32-vx.c
+
+s390x-functable.o: $(SRCDIR)contrib/s390x/s390x-functable.c
+	$(CC) $(CFLAGS) $(VGFMAFLAG) $(ZINC) -c -o $@ $(SRCDIR)contrib/s390x/s390x-functable.c
+
 deflate.o: $(SRCDIR)deflate.c
 	$(CC) $(CFLAGS) $(ZINC) -c -o $@ $(SRCDIR)deflate.c
 
@@ -214,6 +221,16 @@ crc32.lo: $(SRCDIR)crc32.c
 	$(CC) $(SFLAGS) $(ZINC) -DPIC -c -o objs/crc32.o $(SRCDIR)crc32.c
 	-@mv objs/crc32.o $@
 
+crc32-vx.lo: $(SRCDIR)contrib/s390x/crc32-vx.c
+	-@mkdir objs 2>/dev/null || test -d objs
+	$(CC) $(SFLAGS) $(VGFMAFLAG) $(ZINC) -DPIC -c -o objs/crc32-vx.o $(SRCDIR)contrib/s390x/crc32-vx.c
+	-@mv objs/crc32-vx.o $@
+
+s390x-functable.lo: $(SRCDIR)contrib/s390x/s390x-functable.c
+	-@mkdir objs 2>/dev/null || test -d objs
+	$(CC) $(SFLAGS) $(VGFMAFLAG) $(ZINC) -DPIC -c -o objs/s390x-functable.o $(SRCDIR)contrib/s390x/s390x-functable.c
+	-@mv objs/s390x-functable.o $@
+
 deflate.lo: $(SRCDIR)deflate.c
 	-@mkdir objs 2>/dev/null || test -d objs
 	$(CC) $(SFLAGS) $(ZINC) -DPIC -c -o objs/deflate.o $(SRCDIR)deflate.c

configure

@@ -870,6 +870,55 @@ EOF
   fi
 fi
 
+# check for ibm s390x build
+HAVE_S390X=0
+
+# preset the compiler specific flags
+if test $clang -eq 1; then
+  VGFMAFLAG=-fzvector
+else
+  VGFMAFLAG=-mzarch
+fi
+
+cat > $test.c <<EOF
+#ifndef __s390x__
+#error
+#endif
+#include <vecintrin.h>
+int main(void) {
+    unsigned long long a __attribute__((vector_size(16))) = { 0 };
+    unsigned long long b __attribute__((vector_size(16))) = { 0 };
+    unsigned char c __attribute__((vector_size(16))) = { 0 };
+    c = vec_gfmsum_accum_128(a, b, c);
+    return c[0];
+}
+EOF
+
+# cflags already contains a valid march
+if try $CC -c $CFLAGS $VGFMAFLAG $test.c; then
+  echo "Checking for s390x build ... Yes." | tee -a configure.log
+  HAVE_S390X=1
+# or set march for our compile units
+elif try $CC -c $CFLAGS $VGFMAFLAG -march=z13 $test.c; then
+  echo "Checking for s390x build (march=z13) ... Yes." | tee -a configure.log
+  HAVE_S390X=1
+  VGFMAFLAG="$VGFMAFLAG -march=z13"
+# else we are not on s390x
+else 
+  echo "Checking for s390x build ... No." | tee -a configure.log
+fi
+
+# prepare compiling for s390x
+if test $HAVE_S390X -eq 1; then
+  CFLAGS="$CFLAGS -DHAVE_S390X_VX"
+  SFLAGS="$SFLAGS -DHAVE_S390X_VX"
+  OBJC="$OBJC crc32-vx.o s390x-functable.o"
+  PIC_OBJC="$PIC_OBJC crc32-vx.lo s390x-functable.lo"
+else
+# this is not a s390x build 
+  VGFMAFLAG=""
+fi
+
 # show the results in the log
 echo >> configure.log
 echo ALL = $ALL >> configure.log
@@ -901,6 +950,8 @@ echo mandir = $mandir >> configure.log
 echo prefix = $prefix >> configure.log
 echo sharedlibdir = $sharedlibdir >> configure.log
 echo uname = $uname >> configure.log
+echo HAVE_S390X = $HAVE_S390X >> configure.log
+echo VGFMAFLAG = $VGFMAFLAG >> configure.log
 
 # update Makefile with the configure results
 sed < ${SRCDIR}Makefile.in "
@@ -912,6 +963,7 @@ sed < ${SRCDIR}Makefile.in "
 /^LDFLAGS *=/s#=.*#=$LDFLAGS#
 /^LDSHARED *=/s#=.*#=$LDSHARED#
 /^CPP *=/s#=.*#=$CPP#
+/^VGFMAFLAG *=/s#=.*#=$VGFMAFLAG#
 /^STATICLIB *=/s#=.*#=$STATICLIB#
 /^SHAREDLIB *=/s#=.*#=$SHAREDLIB#
 /^SHAREDLIBV *=/s#=.*#=$SHAREDLIBV#

contrib/functable/functable.h

@@ -0,0 +1,23 @@
+#ifndef Z_FUNCTABLE_H__
+#define Z_FUNCTABLE_H__
+
+#include "../../zutil.h"
+#include "../../zonce.h"
+
+unsigned long ZLIB_INTERNAL s390_crc32_vx(unsigned long crc, const unsigned char FAR *buf,
+                            z_size_t len);
+
+struct zfunctable_s {
+  unsigned long hwcap;
+  unsigned long (*crc32_z)(unsigned long crc, const unsigned char FAR *buf,
+                           z_size_t len);
+  /* int (*deflate)(z_streamp strm, int flush); */
+  /* int (*inflate)(z_streamp strm, int flush); */
+};
+
+extern struct zfunctable_s ZLIB_INTERNAL arch_functable;
+extern once_t ZLIB_INTERNAL  arch_functable_init_done;
+/* to be implemented by architecture specific code */
+void ZLIB_INTERNAL arch_functable_init(void);
+
+#endif