Skip to content

Commit

Permalink
removes libXSMM test routines
Browse files Browse the repository at this point in the history
  • Loading branch information
@danielpeter
danielpeter committed Jun 20, 2024
1 parent 15bba52 commit 005b6ed
Showing 1 changed file with 0 additions and 168 deletions.
168 changes: 0 additions & 168 deletions src/specfem3D/compute_forces_crust_mantle_Dev.F90
View file
Original file line number Diff line number Diff line change
Expand Up @@ -245,37 +245,12 @@ subroutine compute_forces_crust_mantle_Dev( NSPEC_STR_OR_ATT,NGLOB,NSPEC_ATT, &
! for incompressible fluid flow, Cambridge University Press (2002),
! pages 386 and 389 and Figure 8.3.1

#ifdef DANIEL_TEST_LOOP
! loop over single x/y/z-component, to test if cache utilization is better
! x-comp
!DIR$ FORCEINLINE
call mxm5_3comp_singleA_1(hprime_xx,m1,dummyx_loc,tempx1,m2)
!DIR$ FORCEINLINE
call mxm5_3comp_3dmat_singleB_1(dummyx_loc,m1,hprime_xxT,m1,tempx2,NGLLX)
!DIR$ FORCEINLINE
call mxm5_3comp_singleB_1(dummyx_loc,m2,hprime_xxT,tempx3,m1)
! y-comp
!DIR$ FORCEINLINE
call mxm5_3comp_singleA_1(hprime_xx,m1,dummyy_loc,tempy1,m2)
!DIR$ FORCEINLINE
call mxm5_3comp_3dmat_singleB_1(dummyy_loc,m1,hprime_xxT,m1,tempy2,NGLLX)
!DIR$ FORCEINLINE
call mxm5_3comp_singleB_1(dummyy_loc,m2,hprime_xxT,tempy3,m1)
! z-comp
!DIR$ FORCEINLINE
call mxm5_3comp_singleA_1(hprime_xx,m1,dummyz_loc,tempz1,m2)
!DIR$ FORCEINLINE
call mxm5_3comp_3dmat_singleB_1(dummyz_loc,m1,hprime_xxT,m1,tempz2,NGLLX)
!DIR$ FORCEINLINE
call mxm5_3comp_singleB_1(dummyz_loc,m2,hprime_xxT,tempz3,m1)
#else
! computes 1. matrix multiplication for tempx1,..
call mxm5_3comp_singleA(hprime_xx,m1,dummyx_loc,dummyy_loc,dummyz_loc,tempx1,tempy1,tempz1,m2)
! computes 2. matrix multiplication for tempx2,..
call mxm5_3comp_3dmat_singleB(dummyx_loc,dummyy_loc,dummyz_loc,m1,hprime_xxT,m1,tempx2,tempy2,tempz2,NGLLX)
! computes 3. matrix multiplication for tempx3,..
call mxm5_3comp_singleB(dummyx_loc,dummyy_loc,dummyz_loc,m2,hprime_xxT,tempx3,tempy3,tempz3,m1)
#endif

!
! compute either isotropic, transverse isotropic or anisotropic elements
Expand Down Expand Up @@ -334,37 +309,12 @@ subroutine compute_forces_crust_mantle_Dev( NSPEC_STR_OR_ATT,NGLOB,NSPEC_ATT, &
! for incompressible fluid flow, Cambridge University Press (2002),
! pages 386 and 389 and Figure 8.3.1

#ifdef DANIEL_TEST_LOOP
! loop over single x/y/z-component, to test if cache utilization is better
! x-comp
!DIR$ FORCEINLINE
call mxm5_3comp_singleA_1(hprimewgll_xxT,m1,tempx1,newtempx1,m2)
!DIR$ FORCEINLINE
call mxm5_3comp_3dmat_singleB_1(tempx2,m1,hprimewgll_xx,m1,newtempx2,NGLLX)
!DIR$ FORCEINLINE
call mxm5_3comp_singleB_1(tempx3,m2,hprimewgll_xx,newtempx3,m1)
! y-comp
!DIR$ FORCEINLINE
call mxm5_3comp_singleA_1(hprimewgll_xxT,m1,tempy1,newtempy1,m2)
!DIR$ FORCEINLINE
call mxm5_3comp_3dmat_singleB_1(tempy2,m1,hprimewgll_xx,m1,newtempy2,NGLLX)
!DIR$ FORCEINLINE
call mxm5_3comp_singleB_1(tempy3,m2,hprimewgll_xx,newtempy3,m1)
! z-comp
!DIR$ FORCEINLINE
call mxm5_3comp_singleA_1(hprimewgll_xxT,m1,tempz1,newtempz1,m2)
!DIR$ FORCEINLINE
call mxm5_3comp_3dmat_singleB_1(tempz2,m1,hprimewgll_xx,m1,newtempz2,NGLLX)
!DIR$ FORCEINLINE
call mxm5_3comp_singleB_1(tempz3,m2,hprimewgll_xx,newtempz3,m1)
#else
! computes 1. matrix multiplication for newtempx1,..
call mxm5_3comp_singleA(hprimewgll_xxT,m1,tempx1,tempy1,tempz1,newtempx1,newtempy1,newtempz1,m2)
! computes 2. matrix multiplication for tempx2,..
call mxm5_3comp_3dmat_singleB(tempx2,tempy2,tempz2,m1,hprimewgll_xx,m1,newtempx2,newtempy2,newtempz2,NGLLX)
! computes 3. matrix multiplication for newtempx3,..
call mxm5_3comp_singleB(tempx3,tempy3,tempz3,m2,hprimewgll_xx,newtempx3,newtempy3,newtempz3,m1)
#endif

! sums contributions
DO_LOOP_IJK
Expand Down Expand Up @@ -840,124 +790,6 @@ pure subroutine mxm5_3comp_3dmat_singleB(A1,A2,A3,n1,B,n2,C1,C2,C3,n3)

end subroutine mxm5_3comp_3dmat_singleB


!--------------------------------------------------------------------------------------------

#ifdef DANIEL_TEST_LOOP

! loops over single x/y/z-component
! test if cache utilization is better

subroutine mxm5_3comp_singleA_1(A,n1,B,C,n3)
use constants_solver, only: CUSTOM_REAL
#ifdef USE_XSMM
use my_libxsmm, only: libxsmm_smm_5_25_5
#endif
implicit none
integer,intent(in) :: n1,n3
real(kind=CUSTOM_REAL),dimension(n1,5),intent(in) :: A
real(kind=CUSTOM_REAL),dimension(5,n3),intent(in) :: B
real(kind=CUSTOM_REAL),dimension(n1,n3),intent(out) :: C
! local parameters
integer :: i,j
#ifdef USE_XSMM
! matrix-matrix multiplication C = alpha A * B + beta C
! with A(n1,n2) 5x5-matrix, B(n2,n3) 5x25-matrix and C(n1,n3) 5x25-matrix
! static version using MNK="5 25, 5" ALPHA=1 BETA=0
call libxsmm_smm_5_25_5(a=A, b=B, c=C)
return
#endif
! matrix-matrix multiplication
do j = 1,n3
!dir$ ivdep
do i = 1,n1
C(i,j) = A(i,1) * B(1,j) &
+ A(i,2) * B(2,j) &
+ A(i,3) * B(3,j) &
+ A(i,4) * B(4,j) &
+ A(i,5) * B(5,j)
enddo
enddo

end subroutine mxm5_3comp_singleA_1


subroutine mxm5_3comp_singleB_1(A,n1,B,C,n3)
use constants_solver, only: CUSTOM_REAL
#ifdef USE_XSMM
use my_libxsmm, only: libxsmm_smm_25_5_5
#endif
implicit none
integer,intent(in) :: n1,n3
real(kind=CUSTOM_REAL),dimension(n1,5),intent(in) :: A
real(kind=CUSTOM_REAL),dimension(5,n3),intent(in) :: B
real(kind=CUSTOM_REAL),dimension(n1,n3),intent(out) :: C
! local parameters
integer :: i,j
#ifdef USE_XSMM
! matrix-matrix multiplication C = alpha A * B + beta C
! with A(n1,n2) 25x5-matrix, B(n2,n3) 5x5-matrix and C(n1,n3) 25x5-matrix
! static version
call libxsmm_smm_25_5_5(a=A, b=B, c=C)
return
#endif
! matrix-matrix multiplication
do j = 1,n3
!dir$ ivdep
do i = 1,n1
C(i,j) = A(i,1) * B(1,j) &
+ A(i,2) * B(2,j) &
+ A(i,3) * B(3,j) &
+ A(i,4) * B(4,j) &
+ A(i,5) * B(5,j)
enddo
enddo
end subroutine mxm5_3comp_singleB_1


subroutine mxm5_3comp_3dmat_singleB_1(A,n1,B,n2,C,n3)
use constants_solver, only: CUSTOM_REAL
#if defined(XSMM_FORCE_EVEN_IF_SLOWER) || ( defined(XSMM) && defined(__MIC__) )
use my_libxsmm, only: libxsmm_smm_5_5_5
#endif
implicit none
integer,intent(in) :: n1,n2,n3
real(kind=CUSTOM_REAL),dimension(n1,5,n3),intent(in) :: A
real(kind=CUSTOM_REAL),dimension(5,n2),intent(in) :: B
real(kind=CUSTOM_REAL),dimension(n1,n2,n3),intent(out) :: C
! local parameters
integer :: i,j,k
#if defined(XSMM_FORCE_EVEN_IF_SLOWER) || ( defined(XSMM) && defined(__MIC__) )
! matrix-matrix multiplication C = alpha A * B + beta C
! with A(n1,n2,n4) 5x5x5-matrix, B(n2,n3) 5x5-matrix and C(n1,n3,n4) 5x5x5-matrix
call libxsmm_smm_5_5_5(a=A(1,1,1), b=B, c=C(1,1,1))
call libxsmm_smm_5_5_5(a=A(1,1,2), b=B, c=C(1,1,2))
call libxsmm_smm_5_5_5(a=A(1,1,3), b=B, c=C(1,1,3))
call libxsmm_smm_5_5_5(a=A(1,1,4), b=B, c=C(1,1,4))
call libxsmm_smm_5_5_5(a=A(1,1,5), b=B, c=C(1,1,5))
return
#endif
! matrix-matrix multiplication
do k = 1,n3
do j = 1,n2
!dir$ ivdep
do i = 1,n1
C(i,j,k) = A(i,1,k) * B(1,j) &
+ A(i,2,k) * B(2,j) &
+ A(i,3,k) * B(3,j) &
+ A(i,4,k) * B(4,j) &
+ A(i,5,k) * B(5,j)
enddo
enddo
enddo
end subroutine mxm5_3comp_3dmat_singleB_1
#endif


!--------------------------------------------------------------------------------------------



end subroutine compute_forces_crust_mantle_Dev


Expand Down

0 comments on commit 005b6ed

Please sign in to comment.