Rewrite vectorisation infrastructure:

Use "k" prefix more consistently in arithmetic macros.

Improve vector code generation patterns.

Move all vectorisation run-time out of Kranc and into a new thorn LSUThorns/Vectors, so that non-Kranc thorns can also use it.

13 files changed, 27 insertions, 8 deletions

diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/interface.ccl b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/interface.ccl
index efbdaa6..ccc3785 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/interface.ccl
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/interface.ccl
@@ -7,7 +7,6 @@
 implements: GenericFD
 
 INCLUDE HEADER: GenericFD.h in GenericFD.h
-INCLUDE HEADER: Vectors.hh in Vectors.hh
 INCLUDE HEADER: sbp_calc_coeffs.h in sbp_calc_coeffs.h
 
 USES INCLUDE: Boundary.h
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/GenericFD.h b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/GenericFD.h
index 568c70f..7d8a13d 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/GenericFD.h
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/GenericFD.h
@@ -28,10 +28,18 @@
     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */
 
+#include "cctk.h"
+
 #ifdef __cplusplus
 extern "C" {
 #endif
 
+#ifdef __cplusplus
+#  ifdef CCTK_CXX_RESTRICT
+#    define restrict CCTK_CXX_RESTRICT
+#  endif
+#endif
+
 #ifndef NOPRECOMPUTE
 #define PRECOMPUTE
 #endif
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/MathematicaCompat.h b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/MathematicaCompat.h
index ee6a3b7..2f00da9 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/MathematicaCompat.h
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/MathematicaCompat.h
@@ -31,13 +31,15 @@
 #define Tanh(x)       (tanh(x))
 
 #ifdef KRANC_C
-#  define Sign(x)     ((x)<0?-1:+1)
+#  define Sign(x)     (copysign(1.0,(x)))
 #  define ToReal(x)   ((CCTK_REAL)(x))
 #else
 #  define Sign(x)     (sgn(x))
 #  define ToReal(x)   (real((x),kind(khalf)))
 #endif
 
+#if 0
+
 /* TODO: use fma(x,y,z) to implement fmadd and friends?  Note that fma
    may be unsupported, or may be slow.  */
 
@@ -46,6 +48,7 @@
 /* #define fnmadd(x,y,z) (-(z)-(x)*(y)) */
 /* #define fnmsub(x,y,z) (+(z)-(x)*(y)) */
 
+#define fpos(x)   (+(x))
 #define fneg(x)   (-(x))
 #define fmul(x,y) ((x)*(y))
 #define fdiv(x,y) ((x)/(y))
@@ -65,6 +68,8 @@
 #define kpow(x,y)  (pow(x,y))
 #define ksqrt(x)   (sqrt(x))
 
+#endif
+
 #ifdef KRANC_C
 #  define E           M_E
 #  define Pi          M_PI
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-SSE2-direct.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-SSE2-direct.hh
index 12cd6e8..12cd6e8 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-SSE2-direct.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-SSE2-direct.hh
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-SSE2.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-SSE2.hh
index 4a4eea6..b74fac0 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-SSE2.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-SSE2.hh
@@ -6,16 +6,23 @@
 
 // Vector type corresponding to CCTK_REAL
 struct CCTK_REAL_VEC {
-  // The underlying scalar and vector types
+  // Underlying scalar and vector types
   typedef double S;
   typedef __m128d V;
+  
+  // Payload
   V v;
   
+  // Empty constructur
+  inline CCTK_REAL_VEC() { }
+  
   // Convert from and to the underlying vector type
   inline CCTK_REAL_VEC(V const v_): v(v_) { }
   inline operator V const() const { return v; }
   
-  inline CCTK_REAL_VEC() { }
+  // Convert from the underlying scalar type
+  inline CCTK_REAL_VEC(S const& a): v(_mm_set1_pd(a)) { }
+  inline CCTK_REAL_VEC(int const& a): v(_mm_set1_pd(S(a))) { }
   
   // Copy constructor
   inline CCTK_REAL_VEC(CCTK_REAL_VEC const& x): v(x) { }
@@ -175,8 +182,8 @@ DEFINE_FUNCTION_VR_V(pow,vec_set(pow(vec_elt0(x),a),pow(vec_elt1(x),a)))
 #undef Sign
 #define Sign(x) (42)
 
-#undef ToReal
-#define ToReal(x) vec_set1(x)
+// #undef ToReal
+// #define ToReal(x) vec_set1(x)
 
 #if defined(__PGI) && defined (__amd64__)
 // Special case for PGI 9.0.4 to avoid an internal compiler error
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-VSX-direct.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-VSX-direct.hh
index 7e06017..7e06017 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-VSX-direct.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-VSX-direct.hh
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-VSX.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-VSX.hh
index f591647..3fc97f6 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-VSX.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-VSX.hh
@@ -208,5 +208,5 @@ DEFINE_FUNCTION_V_V(sqrt,vec_set(sqrt(vec_elt0(x)),sqrt(vec_elt1(x))))
 #undef Sign
 #define Sign(x) (42)
 
-#undef ToReal
-#define ToReal(x) (vec_set1(x))
+// #undef ToReal
+// #define ToReal(x) (vec_set1(x))
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-default.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-default.hh
index f928ed8..f928ed8 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-default.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-default.hh
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-define.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-define.hh
index f5c0b22..f5c0b22 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-define.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-define.hh
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-outdated.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-outdated.hh
index df83b3a..df83b3a 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-outdated.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-outdated.hh
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-pseudo.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-pseudo.hh
index f439c9b..f439c9b 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-pseudo.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-pseudo.hh
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-undefine.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-undefine.hh
index 0d950c7..0d950c7 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors-undefine.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors-undefine.hh
diff --git a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors.hh b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors.hh
index d32afb2..d32afb2 100644
--- a/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/Vectors.hh
+++ b/Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/old/Vectors.hh