diff options
| author | eschnett <eschnett@105869f7-3296-0410-a4ea-f4349344b45a> | 2012-11-14 19:00:17 +0000 |
|---|---|---|
| committer | eschnett <eschnett@105869f7-3296-0410-a4ea-f4349344b45a> | 2012-11-14 19:00:17 +0000 |
| commit | daf2693619c35d5775091aa8fd2911c4cb7f57d5 (patch) | |
| tree | 06994bbf9188775a0f5168b23c75eec330e78a47 | |
| parent | 2a3929883d01ba26feb60c22b18ec910bc12d1e0 (diff) | |
Allow const QPX vectors
git-svn-id: https://svn.cct.lsu.edu/repos/numrel/LSUThorns/Vectors/trunk@73 105869f7-3296-0410-a4ea-f4349344b45a
| -rw-r--r-- | src/vectors-8-QPX.h | 62 |
1 files changed, 31 insertions, 31 deletions
diff --git a/src/vectors-8-QPX.h b/src/vectors-8-QPX.h index 93a7707..4d4991f 100644 --- a/src/vectors-8-QPX.h +++ b/src/vectors-8-QPX.h @@ -22,7 +22,14 @@ // Vector type corresponding to CCTK_REAL // TODO: Use a typedef to avoid the "const" issue? Or use a struct? -#define CCTK_REAL8_VEC vector4double +// #define CCTK_REAL8_VEC vector4double +struct CCTK_REAL8_VEC { + vector4double v; + CCTK_REAL8_VEC() {} + CCTK_REAL8_VEC(CCTK_REAL8_VEC const& x): v(x.v) {} + CCTK_REAL8_VEC(vector4double v_): v(v_) {} + operator vector4double() const { return v; } +}; // Number of vector elements in a CCTK_REAL_VEC #define CCTK_REAL8_VEC_SIZE 4 @@ -35,13 +42,6 @@ -union k8const_t { - CCTK_REAL8 f[CCTK_REAL8_VEC_SIZE]; - CCTK_REAL8_VEC vf; -}; - - - // Create vectors, extract vector elements #define vec8_set1(a) (vec_splats(a)) @@ -64,7 +64,7 @@ union k8const_t { ({ \ CCTK_REAL8 const& p__=(p_); \ CCTK_REAL8& p = *(CCTK_REAL8*)&p__; \ - CCTK_REAL8_VEC v1, v2, vp; \ + vector4double v1, v2, vp; \ v1 = vec_ld(0,&p); /* load the left part of the vector */ \ v2 = vec_ld(32,&p); /* load the right part of the vector */ \ vp = vec_lvsl(0,&p); /* generate control value */ \ @@ -105,28 +105,28 @@ union k8const_t { // Store a vector to memory (aligned and non-temporal); this stores to // a reference to a scalar #define vec8_store(p,x) (vec_sta(x,0,&(p))) -#define vec8_storeu(p_,x_) \ - ({ \ - CCTK_REAL8 & p__=(p_); \ - CCTK_REAL8_VEC const x__=(x_); \ - CCTK_REAL8 & p=p__; \ - CCTK_REAL8_VEC const x=x__; \ - CCTK_REAL8_VEC v1, v2, v3, vp, m1, m2, m3; \ - /* generate insert masks */ \ - vp = vec_lvsr(0,&p); \ - m1 = k8lfalse; \ - m2 = k8ltrue; \ - m3 = vec_perm(m1,m2,vp); \ - /* get existing data */ \ - v1 = vec_ld(0,&p); \ - v2 = vec_ld(32,&p); \ - /* permute and insert */ \ - v3 = vec_perm(x,x,vp); \ - v1 = vec_sel(v1,v3,m3); \ - v2 = vec_sel(v3,v2,m3); \ - /* store data back */ \ - vec_st(0,&p,v1); \ - vec_st(32,&p,v2); \ +#define vec8_storeu(p_,x_) \ + ({ \ + CCTK_REAL8& p__=(p_); \ + CCTK_REAL8_VEC x__=(x_); \ + CCTK_REAL8& p=p__; \ + CCTK_REAL8_VEC x=x__; \ + CCTK_REAL8_VEC v1, v2, v3, vp, m1, m2, m3; \ + /* generate insert masks */ \ + vp = vec_lvsr(0,&p); \ + m1 = k8lfalse; \ + m2 = k8ltrue; \ + m3 = vec_perm(m1,m2,vp); \ + /* get existing data */ \ + v1 = vec_ld(0,&p); \ + v2 = vec_ld(32,&p); \ + /* permute and insert */ \ + v3 = vec_perm(x,x,vp); \ + v1 = vec_sel(v1,v3,m3); \ + v2 = vec_sel(v3,v2,m3); \ + /* store data back */ \ + vec_st(0,&p,v1); \ + vec_st(32,&p,v2); \ }) #define vec8_store_nta(p,x) (vec_sta(x,0,&(p))) // this doesn't avoid the cache |