-
Notifications
You must be signed in to change notification settings - Fork 161
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Add Intel Xe GPU support through OpenMP Offload
- Loading branch information
1 parent
03d554f
commit 05f5cb5
Showing
6 changed files
with
1,264 additions
and
5 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,121 @@ | ||
subroutine ccsd_trpdrv_omp_fbody_reduce_new (f1n, f1t, f2n, f2t, | ||
& f3n, f3t, f4n, f4t, | ||
& eorb, | ||
& ncor, nocc, nvir, | ||
& emp4, emp5, | ||
& i, k, | ||
& eaijk, | ||
& dintc1, dintx1, t1v1, | ||
& dintc2, dintx2, t1v2) | ||
|
||
use omp_lib | ||
use iso_fortran_env | ||
implicit none | ||
|
||
double precision, intent(inout) :: emp4, emp5 | ||
double precision, intent(inout) :: f1n(nvir,nvir), f1t(nvir,nvir) | ||
double precision, intent(inout) :: f2n(nvir,nvir), f2t(nvir,nvir) | ||
double precision, intent(inout) :: f3n(nvir,nvir), f3t(nvir,nvir) | ||
double precision, intent(inout) :: f4n(nvir,nvir), f4t(nvir,nvir) | ||
double precision, intent(in) :: eorb(*) | ||
double precision, intent(in) :: eaijk | ||
double precision, intent(in) :: dintc1(nvir), dintx1(nvir) | ||
double precision, intent(in) :: dintc2(nvir), dintx2(nvir) | ||
double precision, intent(in) :: t1v1(nvir), t1v2(nvir) | ||
integer, intent(in) :: ncor, nocc, nvir | ||
integer, intent(in) :: i, k | ||
|
||
double precision :: emp4i,emp5i,emp4k,emp5k, denom | ||
double precision :: f1nbc,f1tbc,f1ncb,f1tcb | ||
double precision :: f2nbc,f2tbc,f2ncb,f2tcb | ||
double precision :: f3nbc,f3tbc,f3ncb,f3tcb | ||
double precision :: f4nbc,f4tbc,f4ncb,f4tcb | ||
double precision :: t1v1b,t1v2b,dintx1c,dintx2c,dintc1c,dintc2c | ||
integer :: b,c | ||
|
||
emp4i = 0.0 | ||
emp5i = 0.0 | ||
emp4k = 0.0 | ||
emp5k = 0.0 | ||
|
||
!$omp target teams distribute parallel do collapse(2) | ||
& reduction(+:emp5i,emp4i,emp5k,emp4k) | ||
& private(f1nbc,f1tbc,f1ncb,f1tcb,f2nbc,f2tbc,f2ncb,f2tcb) | ||
& private(f3nbc,f3tbc,f3ncb,f3tcb,f4nbc,f4tbc,f4ncb,f4tcb) | ||
& private(t1v1b,t1v2b,dintx1c,dintx2c,dintc1c,dintc2c) | ||
& private(denom) firstprivate(eaijk,nvir,ncor,nocc) | ||
do b=1,nvir | ||
do c=1,nvir | ||
denom=-1.0/( eorb(ncor+nocc+b)+eorb(ncor+nocc+c)+eaijk ) | ||
|
||
f1nbc = f1n(b,c); | ||
f1tbc = f1t(b,c); | ||
f1ncb = f1n(c,b); | ||
f1tcb = f1t(c,b); | ||
|
||
f2nbc = f2n(b,c); | ||
f2tbc = f2t(b,c); | ||
f2ncb = f2n(c,b); | ||
f2tcb = f2t(c,b); | ||
|
||
f3nbc = f3n(b,c); | ||
f3tbc = f3t(b,c); | ||
f3ncb = f3n(c,b); | ||
f3tcb = f3t(c,b); | ||
|
||
f4nbc = f4n(b,c); | ||
f4tbc = f4t(b,c); | ||
f4ncb = f4n(c,b); | ||
f4tcb = f4t(c,b); | ||
|
||
t1v1b = t1v1(b); | ||
t1v2b = t1v2(b); | ||
|
||
dintx1c = dintx1(c); | ||
dintx2c = dintx2(c); | ||
dintc1c = dintc1(c); | ||
dintc2c = dintc2(c); | ||
|
||
emp4i = emp4i | ||
& + (denom * (f1tbc+f1ncb+f2tcb+f3nbc+f4ncb) | ||
& * (f1tbc-f2tbc*2-f3tbc*2+f4tbc) | ||
& - denom * (f1nbc+f1tcb+f2ncb+f3ncb) | ||
& * (f1tbc*2-f2tbc-f3tbc+f4tbc*2) | ||
& + denom * 3 * (f1nbc*(f1nbc+f3ncb+f4tcb*2) | ||
& + f2nbc*f2tcb+f3nbc*f4tbc)) | ||
|
||
emp4k = emp4k | ||
& + (denom * (f1nbc+f1tcb+f2ncb+f3tbc+f4tcb) | ||
& * (f1nbc-f2nbc*2-f3nbc*2+f4nbc) | ||
& - denom * (f1tbc+f1ncb+f2tcb+f3tcb) | ||
& * (f1nbc*2-f2nbc-f3nbc+f4nbc*2) | ||
& + denom * 3 * (f1tbc*(f1tbc+f3tcb+f4ncb*2) | ||
& + f2tbc*f2ncb+f3tbc*f4nbc)) | ||
|
||
emp5i = emp5i | ||
& + (denom * t1v1b * dintx1c | ||
& * (f1tbc+f2nbc+f4ncb | ||
& - (f3tbc+f4nbc+f2ncb+f1nbc+f2tbc+f3ncb)*2 | ||
& + (f3nbc+f4tbc+f1ncb)*4) | ||
& + denom * t1v1b * dintc1c | ||
& * (f1nbc+f4nbc+f1tcb -(f2nbc+f3nbc+f2tcb)*2)) | ||
|
||
emp5k = emp5k | ||
& + (denom * t1v2b * dintx2c | ||
& * (f1nbc+f2tbc+f4tcb | ||
& - (f3nbc+f4tbc+f2tcb +f1tbc+f2nbc+f3tcb)*2 | ||
& + (f3tbc+f4nbc+f1tcb)*4) | ||
& + denom * t1v2b * dintc2c | ||
& * (f1tbc+f4tbc+f1ncb -(f2tbc+f3tbc+f2ncb)*2)) | ||
enddo | ||
enddo | ||
!$omp end target teams distribute parallel do | ||
|
||
emp4 = emp4 + emp4i | ||
emp5 = emp5 + emp5i | ||
if (i.ne.k) then | ||
emp4 = emp4 + emp4k | ||
emp5 = emp5 + emp5k | ||
end if ! (i.ne.k) | ||
|
||
end |
Oops, something went wrong.