portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

fixed lost bracket

Browse Source
This commit is contained in:
Nils Meyer 2020-12-19 03:09:20 +01:00
parent 3f9ae6e7e7
commit 4b882e8056
1 changed files with 18 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
Grid/simd/Fujitsu_A64FX_intrin_double.h
View File

@ -144,38 +144,38 @@ Author: Nils Meyer <nils.meyer@ur.de>
// PREFETCH_CHIMU_L2 (prefetch to L2)
#define PREFETCH_CHIMU_L2_INTERNAL_A64FXd(base) \
{ \
svprfd_vnum(pg1, (void*)(base), (int64_t)0), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)4), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)8), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)(0), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)(4), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)(8), SV_PLDL2STRM); \
}
// PREFETCH_CHIMU_L1 (prefetch to L1)
#define PREFETCH_CHIMU_L1_INTERNAL_A64FXd(base) \
{ \
svprfd_vnum(pg1, (void*)(base), (int64_t)0), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)4), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)8), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)(0), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)(4), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(base), (int64_t)(8), SV_PLDL1STRM); \
}
// PREFETCH_GAUGE_L2 (prefetch to L2)
#define PREFETCH_GAUGE_L2_INTERNAL_A64FXd(A) \
{ \
const auto & ref(U[sUn](A)); baseU = (uint64_t)&ref + 3 * 3 * 64; \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)-4), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)0), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)4), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)8), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)12), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)16), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)20), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)24), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)28), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(-4), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(0), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(4), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(8), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(12), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(16), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(20), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(24), SV_PLDL2STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(28), SV_PLDL2STRM); \
}
// PREFETCH_GAUGE_L1 (prefetch to L1)
#define PREFETCH_GAUGE_L1_INTERNAL_A64FXd(A) \
{ \
const auto & ref(U[sU](A)); baseU = (uint64_t)&ref; \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)0), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)4), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)8), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(0), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(4), SV_PLDL1STRM); \
svprfd_vnum(pg1, (void*)(baseU), (int64_t)(8), SV_PLDL1STRM); \
}
// LOAD_CHI
#define LOAD_CHI_A64FXd(base) \