Add explicit -mpopcnt to avx2 and avx512 builds.

Zig adds native support for the CPU being built on.  So if we support
popcnt then we don't need an explicit -mpopcnt.  However adding any -m
options appears to disable that native support, so -mavx2 doesn't
enable -mpopcnt, yet our auto-detection previously claimed it was
necessary.  This means building on a CPU with popcnt but no avx2 will
fail to build (with zig) when targetting the avx2 platform.

Therefore the only reliable way to check is the combinations actively
in use.  So if we want to compile code using popcnt + avx2 then we
have to test -mpopcnt -mavx2 together.

It's problematic and tricky to manage in autoconf.

Fixes #109

Author: jkbonfield

configure.ac

@@ -129,68 +129,33 @@ dnl AC_CHECK_LIB([bsc], [bsc_compress], [
 dnl 	LIBS="-lbsc $LIBS"
 dnl 	AC_DEFINE([HAVE_LIBBSC],1,[Define to 1 if you have the libbsc library.])])
 
-dnl Count parts needed to build rANS_static32x16pr_sse4.c
-sse4_prerequisites=""
-
-dnl Check if we can use our SSSE3 implementations of rANS 32x16 codec.
-HTS_CHECK_COMPILE_FLAGS_NEEDED([ssse3], [-mssse3], [AC_LANG_PROGRAM([[
-	  #ifdef __x86_64__
-	  #include "x86intrin.h"
-	  #endif
-	]],[[
-	  #ifdef __x86_64__
-	  __m128i a = _mm_set_epi32(1, 2, 3, 4), b = _mm_set_epi32(4, 3, 2, 1);
-	  __m128i c = _mm_shuffle_epi8(a, b);
-	  return *((char *) &c);
-	  #endif
-	]])], [
-        MSSSE3="$flags_needed"
-	sse4_prerequisites="o$sse4_prerequisites"
-	AC_SUBST([MSSSE3])
-        AC_DEFINE([HAVE_SSSE3],1,[Defined to 1 if rANS source using SSSE3 can be compiled.])
-])
-
-dnl Check if we can use popcnt instructions
-HTS_CHECK_COMPILE_FLAGS_NEEDED([popcnt], [-mpopcnt], [AC_LANG_PROGRAM([[
-	  #ifdef __x86_64__
-	  #include "x86intrin.h"
-	  #endif
-	]],[[
-	  #ifdef __x86_64__
-	  unsigned int i = _mm_popcnt_u32(1);
-	  return i != 1;
-	  #endif
-	]])], [
-        MPOPCNT="$flags_needed"
-	sse4_prerequisites="o$sse4_prerequisites"
-	AC_SUBST([MPOPCNT])
-        AC_DEFINE([HAVE_POPCNT],1,[Defined to 1 if rANS source using popcnt can be compiled.])
-])
-
-dnl Check if we can use our SSE4.1 too.  This *may* always imply SSSE3?
-dnl It may be easier just to target an old era of cpu than -mssse3 -msse4.1
-dnl -mpopcnt.  Eg -march=nehalem.  I don't know how wide spread that is.
-HTS_CHECK_COMPILE_FLAGS_NEEDED([sse4.1], [-msse4.1], [AC_LANG_PROGRAM([[
+dnl Check if we can use our SSE4.1 too.
+dnl Our SSE4 codec uses SSE4.1, SSSE3 (shuffle) and POPCNT, so we check all 3
+dnl together.  This helps Zig builds which don't work well if we test each
+dnl individually.
+HTS_CHECK_COMPILE_FLAGS_NEEDED([sse4.1], [-msse4.1 -mssse3 -mpopcnt], [AC_LANG_PROGRAM([[
 	  #ifdef __x86_64__
 	  #include "x86intrin.h"
 	  #endif
 	]],[[
 	  #ifdef __x86_64__
 	  __m128i a = _mm_set_epi32(1, 2, 3, 4), b = _mm_set_epi32(4, 3, 2, 1);
-	  __m128i c = _mm_max_epu32(a, b);
-	  return *((char *) &c);
+	  __m128i c = _mm_shuffle_epi8(_mm_max_epu32(a, b), b);
+	  return _mm_popcnt_u32(*((char *) &c));
 	  #endif
 	]])], [
         MSSE4_1="$flags_needed"
-	sse4_prerequisites="o$sse4_prerequisites"
+	build_rans_sse4=yes
 	AC_SUBST([MSSE4_1])
         AC_DEFINE([HAVE_SSE4_1],1,[Defined to 1 if rANS source using SSE4.1 can be compiled.])
+        AC_DEFINE([HAVE_SSSE3],1,[Defined to 1 if rANS source using SSSE3 can be compiled.])
+        AC_DEFINE([HAVE_POPCNT],1,[Defined to 1 if rANS source using popcnt can be compiled.])
 ])
-AM_CONDITIONAL([RANS_32x16_SSE4],[test "x$sse4_prerequisites" = "xooo"])
+AM_CONDITIONAL([RANS_32x16_SSE4],[test "$build_rans_sse4" = yes])
 
 dnl Check if we can use our AVX2 implementations.
 build_rans_avx2=no
-HTS_CHECK_COMPILE_FLAGS_NEEDED([avx2], [-mavx2], [AC_LANG_PROGRAM([[
+HTS_CHECK_COMPILE_FLAGS_NEEDED([avx2], [-mavx2 -mpopcnt], [AC_LANG_PROGRAM([[
 	  #ifdef __x86_64__
 	  #include "x86intrin.h"
 	  #endif
@@ -199,34 +164,36 @@ HTS_CHECK_COMPILE_FLAGS_NEEDED([avx2], [-mavx2], [AC_LANG_PROGRAM([[
 	  __m256i a = _mm256_set_epi32(1, 2, 3, 4, 5, 6, 7, 8);
 	  __m256i b = _mm256_add_epi32(a, a);
 	  long long c = _mm256_extract_epi64(b, 0);
-	  return (int) c;
+	  return _mm_popcnt_u32((int)c);
 	  #endif
 	]])], [
         MAVX2="$flags_needed"
 	build_rans_avx2=yes
 	AC_SUBST([MAVX2])
         AC_DEFINE([HAVE_AVX2],1,[Defined to 1 if rANS source using AVX2 can be compiled.])
+        AC_DEFINE([HAVE_POPCNT],1,[Defined to 1 if rANS source using popcnt can be compiled.])
 ])
 AM_CONDITIONAL([RANS_32x16_AVX2],[test "$build_rans_avx2" = yes])
 
-dnl Check also if we have AVX512.  If so this overrides AVX2
+dnl Check also if we have AVX512.
 build_rans_avx512=no
-HTS_CHECK_COMPILE_FLAGS_NEEDED([avx512f], [-mavx512f], [AC_LANG_PROGRAM([[
+HTS_CHECK_COMPILE_FLAGS_NEEDED([avx512f], [-mavx512f -mpopcnt], [AC_LANG_PROGRAM([[
 	  #ifdef __x86_64__
 	  #include "x86intrin.h"
 	  #endif
 	]],[[
 	  #ifdef __x86_64__
 	  __m512i a = _mm512_set1_epi32(1);
 	  __m512i b = _mm512_add_epi32(a, a);
-	  return *((char *) &b);
+	  return _mm_popcnt_u32(*((char *) &b));
 	  #endif
 	]])], [
         MAVX512="$flags_needed"
 	build_rans_avx512=yes
 	AC_SUBST([MAVX512])
         AC_DEFINE([HAVE_AVX512],1,[Defined to 1 if rANS source using AVX512F can be compiled.])
 ])
+        AC_DEFINE([HAVE_POPCNT],1,[Defined to 1 if rANS source using popcnt can be compiled.])
 AM_CONDITIONAL([RANS_32x16_AVX512],[test "$build_rans_avx512" = yes])
 
 AC_SUBST([HTSCODECS_SIMD_SRC])

htscodecs/Makefile.am

@@ -72,7 +72,7 @@ noinst_LTLIBRARIES =
 if RANS_32x16_SSE4
 noinst_LTLIBRARIES += librANS_static32x16pr_sse4.la
 librANS_static32x16pr_sse4_la_SOURCES = rANS_static32x16pr_sse4.c
-librANS_static32x16pr_sse4_la_CFLAGS = @MSSE4_1@ @MSSSE3@ @MPOPCNT@
+librANS_static32x16pr_sse4_la_CFLAGS = @MSSE4_1@
 libhtscodecs_la_LIBADD += librANS_static32x16pr_sse4.la
 endif
 if RANS_32x16_AVX2

m4/hts_check_compile_flags_needed.m4

@@ -50,7 +50,7 @@ AC_CACHE_CHECK([_AC_LANG compiler flags needed for $1], CACHEVAR, [
     [ax_check_save_flags=$[]_AC_LANG_PREFIX[]FLAGS
      _AC_LANG_PREFIX[]FLAGS="$[]_AC_LANG_PREFIX[]FLAGS $6 $2"
      AC_LINK_IFELSE([m4_default([$3],[AC_LANG_PROGRAM()])],
-       [AS_VAR_SET(CACHEVAR,[$2])],
+       [AS_VAR_SET(CACHEVAR,["$2"])],
        [AS_VAR_SET(CACHEVAR,[unsupported])])
      _AC_LANG_PREFIX[]FLAGS=$ax_check_save_flags])])
 AS_VAR_IF(CACHEVAR,unsupported, [