Correct vectorised fabs() function for Intel

git-svn-id: https://svn.cct.lsu.edu/repos/numrel/LSUThorns/Vectors/trunk@5 105869f7-3296-0410-a4ea-f4349344b45a

2 files changed, 12 insertions, 2 deletions

diff --git a/src/vectors-intel-4.h b/src/vectors-intel-4.h
index 4549a70..73c90be 100644
--- a/src/vectors-intel-4.h
+++ b/src/vectors-intel-4.h
@@ -110,6 +110,11 @@ static const union {
   __m128   v;
 } k4sign_mask_union = {{ 0x80000000U, 0x80000000U, 0x80000000U, 0x80000000U }};
 #define k4sign_mask (k4sign_mask_union.v)
+static const union {
+  unsigned i[4];
+  __m128   v;
+} k4abs_mask_union = {{ 0x7fffffffU, 0x7fffffffU, 0x7fffffffU, 0x7fffffffU }};
+#define k4abs_mask (k4abs_mask_union.v)
 
 // Operators
 #define k4pos(x) (x)
@@ -127,7 +132,7 @@ static const union {
 #define k4nmsub(x,y,z) (k4sub(z,k4mul(x,y)))
 
 // Cheap functions
-#define k4fabs(x)   (_mm_andnot_ps(x,k4sign_mask))
+#define k4fabs(x)   (_mm_and_ps(x,k4abs_mask))
 #define k4fmax(x,y) (_mm_max_ps(x,y))
 #define k4fmin(x,y) (_mm_min_ps(x,y))
 #define k4fnabs(x)  (_mm_or_ps(x,k4sign_mask))
diff --git a/src/vectors-intel-8.h b/src/vectors-intel-8.h
index a9e4764..35dffa6 100644
--- a/src/vectors-intel-8.h
+++ b/src/vectors-intel-8.h
@@ -74,6 +74,11 @@ static const union {
   __m128d            v;
 } k8sign_mask_union = {{ 0x8000000000000000ULL, 0x8000000000000000ULL }};
 #define k8sign_mask (k8sign_mask_union.v)
+static const union {
+  unsigned long long i[2];
+  __m128d            v;
+} k8abs_mask_union = {{ 0x7fffffffffffffffULL, 0x7fffffffffffffffULL }};
+#define k8abs_mask (k8sign_mask_union.v)
 
 // Operators
 #define k8pos(x) (x)
@@ -91,7 +96,7 @@ static const union {
 #define k8nmsub(x,y,z) (k8sub(z,k8mul(x,y)))
 
 // Cheap functions
-#define k8fabs(x)   (_mm_andnot_pd(x,k8sign_mask))
+#define k8fabs(x)   (_mm_and_pd(x,k8abs_mask))
 #define k8fmax(x,y) (_mm_max_pd(x,y))
 #define k8fmin(x,y) (_mm_min_pd(x,y))
 #define k8fnabs(x)  (_mm_or_pd(x,k8sign_mask))