Add AVX detection and usage of -mavx[2] flags.

Force GCC to use Clang assembler always on MAC OS X for AVX support.

Makefile.in

@@ -239,8 +239,8 @@ $(TRANSP_OBJS) $(CRYPTO_OBJS) $(ZLIB_OBJS) $(BZLIB_OBJS) $(XXHASH_OBJS) $(BLAKE2
 @CRYPTO_COMPAT_OBJS@ $(CRYPTO_ASM_OBJS) $(ARCHIVEOBJS) $(PJPGOBJS) $(DISPACKOBJS)
 
 DEBUG_LINK = $(GPP) -pthread @LIBBSCGEN_OPT@ @EXTRA_OPT_FLAGS@ -fopenmp -fPIC
-DEBUG_COMPILE = $(GCC) -g -c @EXTRA_OPT_FLAGS@ -fPIC
+DEBUG_COMPILE = $(GCC) -g -c @EXTRA_OPT_FLAGS@ -fPIC @USE_CLANG_AS@
-DEBUG_COMPILE_cpp = $(GPP) -g -c @EXTRA_OPT_FLAGS@ -fPIC
+DEBUG_COMPILE_cpp = $(GPP) -g -c @EXTRA_OPT_FLAGS@ -fPIC @USE_CLANG_AS@
 DEBUG_VEC_FLAGS = 
 DEBUG_LOOP_OPTFLAGS = 
 DEBUG_GEN_OPT = -O -fno-omit-frame-pointer @LIBBSCGEN_OPT@ -fopenmp
@@ -248,8 +248,8 @@ DEBUG_CPPFLAGS = $(COMMON_CPPFLAGS)
 DEBUG_FPTR_FLAG = 
 
 RELEASE_LINK = $(GPP) -pthread @LIBBSCGEN_OPT@ @EXTRA_OPT_FLAGS@ -fopenmp -fPIC
-RELEASE_COMPILE = $(GCC) -c @EXTRA_OPT_FLAGS@ -fPIC -fvisibility=hidden
+RELEASE_COMPILE = $(GCC) -c @EXTRA_OPT_FLAGS@ -fPIC -fvisibility=hidden @USE_CLANG_AS@
-RELEASE_COMPILE_cpp = $(GPP) -c @EXTRA_OPT_FLAGS@ -fPIC -fvisibility=hidden
+RELEASE_COMPILE_cpp = $(GPP) -c @EXTRA_OPT_FLAGS@ -fPIC -fvisibility=hidden @USE_CLANG_AS@
 RELEASE_VEC_FLAGS = $(COMMON_VEC_FLAGS)
 RELEASE_LOOP_OPTFLAGS = $(COMMON_LOOP_OPTFLAGS)
 RELEASE_CPPFLAGS = $(COMMON_CPPFLAGS) -DNDEBUG

config

@@ -40,7 +40,9 @@ ${prog} [<options>]
 			Enable building against an alternate Bzip2 and library installation.
 --with-libarchive=<path to libarchive installation tree> (Default: System)
 			Enable building against an alternate libarchive installation.
---no-sse-detect		Do NOT attempt to probe the system's SSE/AVX capability for build flags. 
+--no-sse-detect		Do NOT attempt to probe the system's SSE capability for build flags. 
+			Implies '--no-avx-detect' below.
+--no-avx-detect		Do NOT attempt to probe the system's AVX apability for build flags.
 --no-1.3-archive-compat Disable compatibility with compressed archives created with Pcompress
 			version 1.3 (default: retain compatibility). Hash formats changed from
 			version 1.3 to 1.4 so this option is required if files created using
@@ -86,6 +88,7 @@ zlib_prefix=
 bzlib_prefix=
 libarchive_prefix=
 sse_detect=1
+avx_detect=1
 sse_opt_flags="-msse2"
 crypto_compat_objs='\$\(CRYPTO_COMPAT_OBJS\)'
 crypto_compat_flags="-D__HASH_COMPATIBILITY_"
@@ -195,6 +198,7 @@ do
 	;;
 	--no-sse-detect)
 		sse_detect=0
+		avx_detect=0
 	;;
 	--no-1.3-archive-compat)
 		crypto_compat_objs=""
@@ -254,8 +258,6 @@ then
 	exit 1
 fi
 
-# SSE Detection
-${ECHO} -n "Checking for CPU SSE version ... "
 if [ $sse_detect -eq 1 ]
 then
 	${GCC} -o sse_level ./utils/sse_level.c ./utils/cpuid.c -I./utils
@@ -266,6 +268,12 @@ then
 		echo ""
 		exit 1
 	fi
+fi
+
+if [ $sse_detect -eq 1 ]
+then
+	# SSE Detection
+	${ECHO} -n "Checking for CPU SSE version ... "
 	sse_ver=`./sse_level`
 	if [ $? -ne 0 ]
 	then
@@ -276,10 +284,30 @@ then
 		exit 1
 	fi
 	echo $sse_ver
-	rm -f sse_level
 	sse_opt_flags="-m${sse_ver}"
 fi
 
+if [ $avx_detect -eq 1 -a $sse_detect -eq 1 ]
+then
+	${ECHO} -n "Checking for CPU AVX version ... "
+	avx_ver=`./sse_level --avx`
+	if [ $? -ne 0 ]
+	then
+		rm -f sse_level
+		echo "ERROR:"
+		echo "SSE/AVX version detection utility. Try configuring with --no-sse-check option."
+		echo ""
+		exit 1
+	fi
+	if [ "x$avx_ver" = "x" ]
+	then
+		echo None
+	else
+		echo $avx_ver
+		sse_opt_flags="${sse_opt_flags} -m${avx_ver}"
+	fi
+fi
+rm -f sse_level
 
 echo $plat | egrep 'x86_64|amd64' > /dev/null
 if [ $? -eq 0 ]

utils/sse_level.c

@@ -23,24 +23,48 @@
  */
 
 #include <stdio.h>
+#include <stdlib.h>
 #include <utils.h>
 #include <cpuid.h>
 
+void
+usage(void)
+{
+	printf("Usage: sse_level [--avx]\n");
+	exit(1);
+}
+
 int
-main(void)
+main(int argc, char *argv[])
 {
 	processor_cap_t pc;
+	int avx_detect = 0;
 	cpuid_basic_identify(&pc);
-	if (pc.sse_level == 3 && pc.sse_sub_level == 1) {
+
-		printf("ssse%d", pc.sse_level);
+	if (argc > 1) {
-		pc.sse_sub_level = 0;
+		if (strcmp(argv[1], "--avx") == 0)
-	} else {
+			avx_detect = 1;
-		printf("sse%d", pc.sse_level);
+		else
+			usage();
 	}
-	if (pc.sse_sub_level > 0)
+	if (!avx_detect) {
-		printf(".%d\n", pc.sse_sub_level);
+		if (pc.sse_level == 3 && pc.sse_sub_level == 1) {
-	else
+			printf("ssse%d", pc.sse_level);
-		printf("\n");
+			pc.sse_sub_level = 0;
+		} else {
+			printf("sse%d", pc.sse_level);
+		}
+		if (pc.sse_sub_level > 0)
+			printf(".%d\n", pc.sse_sub_level);
+		else
+			printf("\n");
+	} else {
+		if (pc.avx_level == 1)
+			printf("avx\n");
+		else if (pc.avx_level == 2)
+			printf("avx2\n");
+	}
+
 	return (0);
 }