diff options
| author | svn_bwardell <svn_bwardell@105869f7-3296-0410-a4ea-f4349344b45a> | 2011-08-07 12:41:50 +0000 |
|---|---|---|
| committer | svn_bwardell <svn_bwardell@105869f7-3296-0410-a4ea-f4349344b45a> | 2011-08-07 12:41:50 +0000 |
| commit | 96caf45931a90c080adc8073762d72aa2a80ac76 (patch) | |
| tree | 48ccb4894a0bdd3dbf62826e8c0e05641d1f4236 | |
| parent | 1d6878bcc13a1695043cb680f7ab83be28048419 (diff) | |
Fix definition of kifthen for architectures where blend instructions are not available.
git-svn-id: https://svn.cct.lsu.edu/repos/numrel/LSUThorns/Vectors/trunk@20 105869f7-3296-0410-a4ea-f4349344b45a
| -rw-r--r-- | src/vectors-4-SSE.h | 15 | ||||
| -rw-r--r-- | src/vectors-4-default.h | 9 | ||||
| -rw-r--r-- | src/vectors-8-SSE2.h | 11 | ||||
| -rw-r--r-- | src/vectors-8-default.h | 9 |
4 files changed, 34 insertions, 10 deletions
diff --git a/src/vectors-4-SSE.h b/src/vectors-4-SSE.h index 8319c49..7948032 100644 --- a/src/vectors-4-SSE.h +++ b/src/vectors-4-SSE.h @@ -303,6 +303,12 @@ static const union { #ifdef __SSE4_1__ # define k4ifthen(x,y,z) (_mm_blendv_ps(y,z,x)) #else +#include <math.h> +#ifdef __cplusplus +#define SGN(x) std::signbit(x) +#else +#define SGN(x) signbit(x) +#endif # define k4ifthen(x,y,z) \ ({ \ CCTK_REAL4_VEC const xx=(x_); \ @@ -311,10 +317,9 @@ static const union { CCTK_REAL4_VEC const y=yy; \ CCTK_REAL4_VEC const zz=(z_); \ CCTK_REAL4_VEC const z=zz; \ - CCTK_REAL4_VEC const c = _mm_and_ps(x,k4sign_mask); \ - vec4_set(vec4_elt0(not vec4_elt0(c) ? y : z), \ - vec4_elt1(not vec4_elt1(c) ? y : z), \ - vec4_elt2(not vec4_elt2(c) ? y : z), \ - vec4_elt3(not vec4_elt3(c) ? y : z)); \ + vec4_set(SGN(vec4_elt0(x)) ? vec4_elt0(z) : vec4_elt0(y), \ + SGN(vec4_elt1(x)) ? vec4_elt1(z) : vec4_elt1(y), \ + SGN(vec4_elt2(x)) ? vec4_elt2(z) : vec4_elt2(y), \ + SGN(vec4_elt3(x)) ? vec4_elt3(z) : vec4_elt3(y)); \ }) #endif diff --git a/src/vectors-4-default.h b/src/vectors-4-default.h index b59db7c..6f66ff9 100644 --- a/src/vectors-4-default.h +++ b/src/vectors-4-default.h @@ -81,4 +81,11 @@ #define k4pow(x,a) (powf(x,a)) #define k4sqrt(x) (sqrtf(x)) -#define k4ifthen(x,y,z) ((x)>=0?(y):(z)) +#include <math.h> +#ifdef __cplusplus +#define SGN(x) std::signbit(x) +#else +#define SGN(x) signbit(x) +#endif + +#define k4ifthen(x,y,z) (SGN(x)?(z):(y)) diff --git a/src/vectors-8-SSE2.h b/src/vectors-8-SSE2.h index 15a45ed..3bd5753 100644 --- a/src/vectors-8-SSE2.h +++ b/src/vectors-8-SSE2.h @@ -225,6 +225,12 @@ static const union { r; \ }) #else +#include <math.h> +#ifdef __cplusplus +#define SGN(x) std::signbit(x) +#else +#define SGN(x) signbit(x) +#endif # define k8ifthen(x_,y_,z_) \ ({ \ CCTK_REAL8_VEC const xx=(x_); \ @@ -233,8 +239,7 @@ static const union { CCTK_REAL8_VEC const y=yy; \ CCTK_REAL8_VEC const zz=(z_); \ CCTK_REAL8_VEC const z=zz; \ - CCTK_REAL8_VEC const c = _mm_and_pd(x,k8sign_mask); \ - vec8_set(vec8_elt0(not vec8_elt0(c) ? y : z), \ - vec8_elt1(not vec8_elt1(c) ? y : z)); \ + vec8_set(SGN(vec8_elt0(x)) ? vec8_elt0(z) : vec8_elt0(y), \ + SGN(vec8_elt1(x)) ? vec8_elt1(z) : vec8_elt1(y)); \ }) #endif diff --git a/src/vectors-8-default.h b/src/vectors-8-default.h index 6f81b91..cdce1ef 100644 --- a/src/vectors-8-default.h +++ b/src/vectors-8-default.h @@ -81,4 +81,11 @@ #define k8pow(x,a) (pow(x,a)) #define k8sqrt(x) (sqrt(x)) -#define k8ifthen(x,y,z) ((x)>=0?(y):(z)) +#include <math.h> +#ifdef __cplusplus +#define SGN(x) std::signbit(x) +#else +#define SGN(x) signbit(x) +#endif + +#define k8ifthen(x,y,z) (SGN(x)?(z):(y)) |