1
0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-15 06:17:05 +01:00

Merge branch 'develop' into hisq_fat_links

This commit is contained in:
david clarke
2024-02-23 14:47:15 -07:00
20 changed files with 780 additions and 264 deletions

View File

@ -152,6 +152,7 @@ public:
#ifdef RNG_FAST_DISCARD
static void Skip(RngEngine &eng,uint64_t site)
{
#if 0
/////////////////////////////////////////////////////////////////////////////////////
// Skip by 2^40 elements between successive lattice sites
// This goes by 10^12.
@ -162,9 +163,9 @@ public:
// tens of seconds per trajectory so this is clean in all reasonable cases,
// and margin of safety is orders of magnitude.
// We could hack Sitmo to skip in the higher order words of state if necessary
//
// Replace with 2^30 ; avoid problem on large volumes
//
//
// Replace with 2^30 ; avoid problem on large volumes
//
/////////////////////////////////////////////////////////////////////////////////////
// uint64_t skip = site+1; // Old init Skipped then drew. Checked compat with faster init
const int shift = 30;
@ -179,6 +180,9 @@ public:
assert((skip >> shift)==site); // check for overflow
eng.discard(skip);
#else
eng.discardhi(site);
#endif
// std::cout << " Engine " <<site << " state " <<eng<<std::endl;
}
#endif

View File

@ -69,7 +69,7 @@ public:
/*! Construct stout smearing object from explicitly specified rho matrix */
Smear_Stout(const std::vector<double>& rho_)
: OwnedBase{new Smear_APE<Gimpl>(rho_)}, SmearBase{OwnedBase.get()} {
std::cout << GridLogDebug << "Stout smearing constructor : Smear_Stout(const std::vector<double>& " << rho_ << " )" << std::endl
std::cout << GridLogDebug << "Stout smearing constructor : Smear_Stout(const std::vector<double>& " << rho_ << " )" << std::endl;
assert(Nc == 3 && "Stout smearing currently implemented only for Nc==3");
}

View File

@ -218,6 +218,10 @@ public:
// -------------------------------------------------
// misc
// -------------------------------------------------
void discardhi(uint64_t z) {
_s[3] += z;
encrypt_counter();
}
// req: 26.5.1.4 Random number engine requirements, p.908 table 117, row 9
// Advances es state ei to ei+z by any means equivalent to z
@ -387,4 +391,4 @@ private:
#undef MIXK
#undef MIX2
#endif
#endif

View File

@ -7,6 +7,8 @@ uint32_t accelerator_threads=2;
uint32_t acceleratorThreads(void) {return accelerator_threads;};
void acceleratorThreads(uint32_t t) {accelerator_threads = t;};
#define ENV_LOCAL_RANK_PALS "PALS_LOCAL_RANKID"
#define ENV_RANK_PALS "PALS_RANKID"
#define ENV_LOCAL_RANK_OMPI "OMPI_COMM_WORLD_LOCAL_RANK"
#define ENV_RANK_OMPI "OMPI_COMM_WORLD_RANK"
#define ENV_LOCAL_RANK_SLURM "SLURM_LOCALID"
@ -228,8 +230,17 @@ void acceleratorInit(void)
{
rank = atoi(localRankStr);
}
if ((localRankStr = getenv(ENV_LOCAL_RANK_PALS)) != NULL)
{
rank = atoi(localRankStr);
}
if ((localRankStr = getenv(ENV_RANK_OMPI )) != NULL) { world_rank = atoi(localRankStr);}
if ((localRankStr = getenv(ENV_RANK_MVAPICH)) != NULL) { world_rank = atoi(localRankStr);}
if ((localRankStr = getenv(ENV_RANK_PALS )) != NULL) { world_rank = atoi(localRankStr);}
char hostname[HOST_NAME_MAX+1];
gethostname(hostname, HOST_NAME_MAX+1);
if ( rank==0 ) printf(" acceleratorInit world_rank %d is host %s \n",world_rank,hostname);
auto devices = cl::sycl::device::get_devices();
for(int d = 0;d<devices.size();d++){
@ -241,9 +252,10 @@ void acceleratorInit(void)
printf("AcceleratorSyclInit: " #prop ": " FMT" \n",devices[d].get_info<cl::sycl::info::device::prop>());
#define GPU_PROP(prop) GPU_PROP_FMT(prop,"%ld");
if ( world_rank == 0) {
GPU_PROP_STR(vendor);
GPU_PROP_STR(version);
GPU_PROP_STR(vendor);
GPU_PROP_STR(version);
// GPU_PROP_STR(device_type);
/*
GPU_PROP(max_compute_units);
@ -259,7 +271,8 @@ void acceleratorInit(void)
GPU_PROP(single_fp_config);
*/
// GPU_PROP(double_fp_config);
GPU_PROP(global_mem_size);
GPU_PROP(global_mem_size);
}
}
if ( world_rank == 0 ) {