portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-18 15:57:05 +01:00
Code Releases Activity

HIP runs sensible

Browse Source
This commit is contained in:
Peter Boyle
2020-09-16 03:35:03 +01:00
parent ecd3f890f5
commit 81441e98f4
4 changed files with 44 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Grid/lattice/Lattice_arith.h
View File

@ -60,9 +60,9 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const Lattice<obj3> &rhs){
autoView( lhs_v , lhs, AcceleratorRead);
autoView( rhs_v , rhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto lhs_t=lhs_v(ss);
auto rhs_t=rhs_v(ss);
auto tmp =ret_v(ss);
mac(&tmp,&lhs_t,&rhs_t);
coalescedWrite(ret_v[ss],tmp);
});
@ -124,7 +124,7 @@ void mac(Lattice<obj1> &ret,const Lattice<obj2> &lhs,const obj3 &rhs){
autoView( ret_v , ret, AcceleratorWrite);
autoView( lhs_v , lhs, AcceleratorRead);
accelerator_for(ss,lhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto tmp =ret_v(ss);
auto lhs_t=lhs_v(ss);
mac(&tmp,&lhs_t,&rhs);
coalescedWrite(ret_v[ss],tmp);
@ -182,7 +182,7 @@ void mac(Lattice<obj1> &ret,const obj2 &lhs,const Lattice<obj3> &rhs){
autoView( ret_v , ret, AcceleratorWrite);
autoView( rhs_v , lhs, AcceleratorRead);
accelerator_for(ss,rhs_v.size(),obj1::Nsimd(),{
decltype(coalescedRead(obj1())) tmp;
auto tmp =ret_v(ss);
auto rhs_t=rhs_v(ss);
mac(&tmp,&lhs,&rhs_t);
coalescedWrite(ret_v[ss],tmp);

2
Grid/threads/Accelerator.cc
View File

@ -127,7 +127,7 @@ void acceleratorInit(void)
printf("AcceleratorHipInit: ========================\n");
printf("AcceleratorHipInit: Device identifier: %s\n", prop.name);
GPU_PROP_FMT(totalGlobalMem,"%lld");
GPU_PROP_FMT(totalGlobalMem,"%lu");
// GPU_PROP(managedMemory);
GPU_PROP(isMultiGpuBoard);
GPU_PROP(warpSize);