comp-chem/NWChem
1
0
Fork 0
mirror of https://github.com/nwchemgit/nwchem.git synced 2026-07-28 22:25:48 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Add unused ifdefs for intel xe openmp offload ccsd(t)

Browse source
This commit is contained in:
Omar Khalil Ahmed 2024-12-05 08:38:05 -08:00
parent a5ebdaeed6
commit 34181bcd60
1 changed files with 0 additions and 78 deletions
Download patch file Download diff file Expand all files Collapse all files

78
src/ccsd/ccsd_trpdrv_openmp_imax.F
View file

@ -185,29 +185,6 @@
double precision :: t_dgemm0, t_dgemm1, t_dgemm_total
double precision :: t_red0, t_red1, t_red_total
#if 0
!$omp interop init(targetsync:obj0)
!$omp interop init(targetsync:obj1)
!$omp interop init(targetsync:obj2)
!$omp interop init(targetsync:obj3)
!$omp interop init(targetsync:obj4)
!$omp interop init(targetsync:obj5)
!$omp interop init(targetsync:obj6)
!$omp interop init(targetsync:obj7)
#endif
#if 0
!$omp interop init(prefer_type("level_zero"),targetsync: obj0)
!$omp interop init(prefer_type("level_zero"),targetsync: obj1)
!$omp interop init(prefer_type("level_zero"),targetsync: obj2)
!$omp interop init(prefer_type("level_zero"),targetsync: obj3)
!$omp interop init(prefer_type("level_zero"),targetsync: obj4)
!$omp interop init(prefer_type("level_zero"),targetsync: obj5)
!$omp interop init(prefer_type("level_zero"),targetsync: obj6)
!$omp interop init(prefer_type("level_zero"),targetsync: obj7)
#endif
#if 1
!$omp interop init(prefer_type("sycl"),targetsync: obj0)
!$omp interop init(prefer_type("sycl"),targetsync: obj1)
!$omp interop init(prefer_type("sycl"),targetsync: obj2)
@ -218,7 +195,6 @@
!$omp interop init(prefer_type("sycl"),targetsync: obj7)
!$omp interop init(prefer_type("level_zero"),targetsync: obj_lev0)
#endif
t_dgemm_total = 0.0
t_red_total = 0.0
@ -255,7 +231,6 @@
! & stat=alloc_error)
! if (alloc_error.ne.0) call errquit('f[1234][tn]',8,MA_ERR)
#if 1
!$omp allocate allocator(omp_target_device_mem_alloc)
allocate( f1n(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_device_mem_alloc)
@ -272,25 +247,6 @@
allocate( f4n(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_device_mem_alloc)
allocate( f4t(1:nvir,1:nvir) )
#else
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f1n(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f1t(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f2n(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f2t(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f3n(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f3t(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f4n(1:nvir,1:nvir) )
!$omp allocate allocator(omp_target_host_mem_alloc)
allocate( f4t(1:nvir,1:nvir) )
#endif
!
! device-only copy of input eorb
!
@ -928,11 +884,7 @@
!$omp end parallel sections
#endif
#if 0
!$omp interop use(obj0)
#endif
#if 1
! "omp interop use(obj_lev0)" generates a device barrier
! on level0 queue. This behavior is specific to Intel and
! works because of the mapping of level0 queues to
@ -944,7 +896,6 @@
! interop use(obj) == device barrier + EventHostSynchronize on host
!$omp interop use(obj_lev0) nowait
#endif
t_dgemm1 = util_wallsec()
t_dgemm_total = t_dgemm_total + (t_dgemm1 - t_dgemm0)
@ -1044,35 +995,6 @@
emp5 = emp5 + emp5k
#else
#if 0
!use two separate calls to C implementation
!$omp target update to(emp4i,emp5i)
call ccsd_trpdrv_omp_reduce_01(f1n, f1t, f2n, f2t,
& f3n, f3t, f4n, f4t,
& eorb,
& ncor, nocc, nvir,
! & emp4i, emp5i,
& eaijk,
& dintc1, dintx1, t1v1)
!$omp target update from(emp4i,emp5i)
if (i.ne.k) then
!$omp target update to(emp4k,emp5k)
call ccsd_trpdrv_omp_reduce_02(f1n, f1t, f2n, f2t,
& f3n, f3t, f4n, f4t,
& eorb,
& ncor, nocc, nvir,
! & emp4k, emp5k,
& eaijk,
& dintc2, dintx2, t1v2)
!$omp target update from(emp4k,emp5k)
end if ! (i.ne.k)
emp4 = emp4 + emp4i
emp5 = emp5 + emp5i
emp4 = emp4 + emp4k
emp5 = emp5 + emp5k
#endif
call ccsd_trpdrv_omp_fbody_reduce_new (f1n, f1t, f2n, f2t,
& f3n, f3t, f4n, f4t,