portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-07-12 19:27:05 +01:00
Code Releases Activity

Finally starting to get decent performance on Volta

Browse Source
This commit is contained in:
Peter Boyle
2018-07-13 12:06:18 -04:00
parent 2cc07450f4
commit b2b5137d28
5 changed files with 203 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@ -215,6 +215,14 @@ public:
mult(&phi(), &U(mu), &chi());
}
#ifdef GPU_VEC
static accelerator_inline void copyLinkGpu(int lane,
SiteDoubledGaugeField & UU,
const SiteDoubledGaugeField &U)
{
auto U_l = extractLane(lane,U);
insertLane(lane,UU,U_l);
}
static accelerator_inline void multLinkGpu(int lane,
typename SiteHalfSpinor::scalar_object &phi,
const SiteDoubledGaugeField &U,
@ -224,6 +232,17 @@ public:
auto U_l = extractLane(lane,U(mu));
phi() = U_l * chi();
}
#else
static accelerator_inline void multLinkGpu(int lane,
SiteHalfSpinor &phi,
const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi,
int mu)
{
auto U_l = U(mu);
phi() = U_l * chi();
}
#endif
static accelerator_inline void multLinkProp(SitePropagator &phi,
const SiteDoubledGaugeField &U,
@ -364,7 +383,13 @@ public:
}
mult(&phi(), &UU(), &chi());
}
#ifdef GPU_VEC
static accelerator_inline void copyLinkGpu(int lane,
SiteDoubledGaugeField & UU,
const SiteDoubledGaugeField &U)
{
UU = U;
}
static accelerator_inline void multLinkGpu(int lane,
typename SiteHalfSpinor::scalar_object &phi,
const SiteDoubledGaugeField &U,
@ -374,6 +399,17 @@ public:
auto U_l = U(mu);
phi() = U_l * chi();
}
#else
static accelerator_inline void multLinkGpu(int lane,
SiteHalfSpinor &phi,
const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi,
int mu)
{
auto U_l = U(mu);
phi() = U_l * chi();
}
#endif
static accelerator_inline void multLinkProp(SitePropagator &phi,
const SiteDoubledGaugeField &U,