From dce3c2b430a4e48adeea9fafcf3f24ab9efc63bd Mon Sep 17 00:00:00 2001
From: Martin Reinecke <martin@mpa-garching.mpg.de>
Date: Fri, 26 Oct 2018 11:58:32 +0200
Subject: [PATCH] fixes

---
 libsharp/sharp_announce.c |  4 ++--
 libsharp/sharp_core.c     | 15 +++++++++++++--
 libsharp/sharp_core.h     |  2 ++
 libsharp/sharp_core_avx.c |  8 ++++----
 4 files changed, 21 insertions(+), 8 deletions(-)

diff --git a/libsharp/sharp_announce.c b/libsharp/sharp_announce.c
index bc6ee50..7027167 100644
--- a/libsharp/sharp_announce.c
+++ b/libsharp/sharp_announce.c
@@ -40,7 +40,7 @@
 #endif
 
 #include "sharp_announce.h"
-#include "sharp_vecutil.h"
+#include "sharp_core.h"
 
 static void OpenMP_status(void)
   {
@@ -70,7 +70,7 @@ static void MPI_status(void)
   }
 
 static void vecmath_status(void)
-  { printf("Supported vector length: %d\n",VLEN); }
+  { printf("Supported vector length: %d\n",sharp_veclen()); }
 
 void sharp_announce (const char *name)
   {
diff --git a/libsharp/sharp_core.c b/libsharp/sharp_core.c
index f052555..8d75893 100644
--- a/libsharp/sharp_core.c
+++ b/libsharp/sharp_core.c
@@ -33,7 +33,7 @@
 #include "sharp_core_inc0.c"
 #undef ARCH
 
-#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6)
+#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=5)
 void inner_loop_avx (sharp_job *job, const int *ispair,const double *cth,
   const double *sth, int llim, int ulim, sharp_Ylmgen_C *gen, int mi,
   const int *mlim);
@@ -43,7 +43,7 @@ void inner_loop (sharp_job *job, const int *ispair,const double *cth,
   const double *sth, int llim, int ulim, sharp_Ylmgen_C *gen, int mi,
   const int *mlim)
   {
-#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6)
+#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=5)
   __builtin_cpu_init();
   if (__builtin_cpu_supports("avx"))
     inner_loop_avx (job, ispair, cth, sth, llim, ulim, gen, mi, mlim);
@@ -51,3 +51,14 @@ void inner_loop (sharp_job *job, const int *ispair,const double *cth,
 #endif
     inner_loop_default (job, ispair, cth, sth, llim, ulim, gen, mi, mlim);
   }
+
+int sharp_veclen(void)
+  {
+#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=5)
+  __builtin_cpu_init();
+  if (__builtin_cpu_supports("avx"))
+    return 4;
+  else
+#endif
+    return VLEN;
+  }
diff --git a/libsharp/sharp_core.h b/libsharp/sharp_core.h
index 1e86488..f641125 100644
--- a/libsharp/sharp_core.h
+++ b/libsharp/sharp_core.h
@@ -43,6 +43,8 @@ void inner_loop (sharp_job *job, const int *ispair,const double *cth,
   const double *sth, int llim, int ulim, sharp_Ylmgen_C *gen, int mi,
   const int *mlim);
 
+int sharp_veclen(void);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/libsharp/sharp_core_avx.c b/libsharp/sharp_core_avx.c
index dc6ee48..a250b49 100644
--- a/libsharp/sharp_core_avx.c
+++ b/libsharp/sharp_core_avx.c
@@ -1,14 +1,14 @@
-#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6)
+#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=5)
 // if we arrive here, we can benefit from an additional AVX version
-#warning entering gcc and x86_64 specific code branch
+// #warning entering gcc and x86_64 specific code branch
 
 #define ARCH _avx
-#define __AVX__
+//#define __AVX__
 #pragma GCC push_options
 #pragma GCC target("avx")
 #include "sharp_core_inc0.c"
 #pragma GCC pop_options
-#undef __AVX__
+//#undef __AVX__
 #undef ARCH
 
 #endif