diff options
| author | eschnett <eschnett@105869f7-3296-0410-a4ea-f4349344b45a> | 2012-05-03 19:57:56 +0000 |
|---|---|---|
| committer | eschnett <eschnett@105869f7-3296-0410-a4ea-f4349344b45a> | 2012-05-03 19:57:56 +0000 |
| commit | 8ce5889dc25dc3ddb5c036e4f98d775b671a1420 (patch) | |
| tree | 7f62e0eca05046abd40519a312559af7d36bec03 | |
| parent | 41227f75a7f75d7d0f984d143f01814989953893 (diff) | |
Support partial stores for the Double Hummer architecture (Blue Gene/P)
git-svn-id: https://svn.cct.lsu.edu/repos/numrel/LSUThorns/Vectors/trunk@58 105869f7-3296-0410-a4ea-f4349344b45a
| -rw-r--r-- | src/vectors-8-DoubleHummer.h | 19 |
1 files changed, 19 insertions, 0 deletions
diff --git a/src/vectors-8-DoubleHummer.h b/src/vectors-8-DoubleHummer.h index 3366a0a..b0a1adf 100644 --- a/src/vectors-8-DoubleHummer.h +++ b/src/vectors-8-DoubleHummer.h @@ -121,6 +121,25 @@ #define vec8_storeu(p,x) (__stfpd(&(p),x)) // this may not work #define vec8_store_nta(p,x) (__stfpd(&(p),x)) // this doesn't avoid the cache +// Store a partial vector (aligned and non-temporal) +#define vec8_store_partial_prepare(i,imin,imax) \ + bool const v8stp_lo = (i)>=(imin); \ + bool const v8stp_hi = (i)+CCTK_REAL_VEC_SIZE-1<(imax) +#define vec8_store_nta_partial(p_,x_) \ + ({ \ + CCTK_REAL8& p__=(p_); \ + CCTK_REAL8& p=p__; \ + CCTK_REAL8_VEC const x__=(x_); \ + CCTK_REAL8_VEC const x=x__; \ + if (CCTK_BUILTIN_EXPECT(v8stp_lo and v8stp_hi, true)) { \ + vec8_store(p,x); \ + } else if (v8stp_lo) { \ + (&p)[0]=vec8_elt0(x); \ + } else if (v8stp_hi) { \ + (&p)[1]=vec8_elt1(x); \ + } \ + }) + // Store a lower or higher partial vector (aligned and non-temporal); // the non-temporal hint is probably ignored #define vec8_store_nta_partial_lo(p,x,n) ((&(p))[0]=vec8_elt0(x)) |