Skip to content

GitLab

Commit 5ae53d29 authored by Andreas Marek's avatar Andreas Marek
Browse files

Test kernel

parent 21669f76
Hide whitespace changes
Inline Side-by-side
Showing with 90 additions and 1306 deletions
.gitlab-ci.yml
View file @ 5ae53d29
This source diff could not be displayed because it is too large. You can view the blob instead.
src/elpa2/compute_hh_trafo.F90
View file @ 5ae53d29
......@@ -77,13 +77,13 @@
use real_generic_simple_kernel !, only : double_hh_trafo_generic_simple
#endif
!#if defined(WITH_REAL_GENERIC_SIMPLE_BLOCK4_KERNEL) && !(defined(USE_ASSUMED_SIZE))
! use real_generic_simple_block4_kernel !, only : double_hh_trafo_generic_simple
!#endif
#if defined(WITH_REAL_GENERIC_SIMPLE_BLOCK4_KERNEL) && !(defined(USE_ASSUMED_SIZE))
use real_generic_simple_block4_kernel !, only : double_hh_trafo_generic_simple
#endif
!#if defined(WITH_REAL_GENERIC_SIMPLE_BLOCK6_KERNEL) && !(defined(USE_ASSUMED_SIZE))
! use real_generic_simple_block6_kernel !, only : double_hh_trafo_generic_simple
!#endif
#if defined(WITH_REAL_GENERIC_SIMPLE_BLOCK6_KERNEL) && !(defined(USE_ASSUMED_SIZE))
use real_generic_simple_block6_kernel !, only : double_hh_trafo_generic_simple
#endif
#if defined(WITH_REAL_GENERIC_KERNEL) && !(defined(USE_ASSUMED_SIZE))
use real_generic_kernel !, only : double_hh_trafo_generic
......@@ -1331,37 +1331,37 @@
w(:,4) = bcast_buffer(1:nbw,j+off-3)
#ifdef WITH_OPENMP
!#ifdef USE_ASSUMED_SIZE
#ifdef USE_ASSUMED_SIZE
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1,j+off+a_off-3,istripe,my_thread), w, nbw, nl, stripe_width, nbw)
!#else
! call quad_hh_trafo_&
! &MATH_DATATYPE&
! &_generic_simple_4hv_&
! &PRECISION&
! & (a(1:stripe_width,j+off+a_off-3:j+off+a_off+nbw-1,istripe,my_thread), w(1:nbw,1:6), nbw, nl, &
! stripe_width, nbw)
!#endif
#else
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1:stripe_width,j+off+a_off-3:j+off+a_off+nbw-1,istripe,my_thread), w(1:nbw,1:6), nbw, nl, &
stripe_width, nbw)
#endif
#else
!#ifdef USE_ASSUMED_SIZE
#ifdef USE_ASSUMED_SIZE
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1,j+off+a_off-3,istripe), w, nbw, nl, stripe_width, nbw)
!#else
! call quad_hh_trafo_&
! &MATH_DATATYPE&
! &_generic_simple_4hv_&
! &PRECISION&
! & (a(1:stripe_width,j+off+a_off-3:j+off+a_off+nbw-1,istripe), w(1:nbw,1:6), nbw, nl, &
! stripe_width, nbw)
!#endif
#else
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1:stripe_width,j+off+a_off-3:j+off+a_off+nbw-1,istripe), w(1:nbw,1:6), nbw, nl, &
stripe_width, nbw)
#endif
#endif
enddo
......@@ -1450,36 +1450,36 @@
#ifdef WITH_OPENMP
!#ifdef USE_ASSUMED_SIZE
#ifdef USE_ASSUMED_SIZE
call hexa_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_6hv_&
&PRECISION&
& (a(1,j+off+a_off-5,istripe,my_thread), w, nbw, nl, stripe_width, nbw)
!#else
! call hexa_hh_trafo_&
! &MATH_DATATYPE&
! &_sse_6hv_&
! &PRECISION&
! & (a(1:stripe_width,j+off+a_off-5:j+off+a_off-1,istripe,my_thread), w(1:nbw,1:6), &
! nbw, nl, stripe_width, nbw)
!#endif
#else
call hexa_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_6hv_&
&PRECISION&
& (a(1:stripe_width,j+off+a_off-5:j+off+a_off-1,istripe,my_thread), w(1:nbw,1:6), &
nbw, nl, stripe_width, nbw)
#endif
#else /* WITH_OPENMP */
!#ifdef USE_ASSUMED_SIZE
#ifdef USE_ASSUMED_SIZE
call hexa_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_6hv_&
&PRECISION&
& (a(1,j+off+a_off-5,istripe), w, nbw, nl, stripe_width, nbw)
!#else
! call hexa_hh_trafo_&
! &MATH_DATATYPE&
! &_generic_simple_6hv_&
! &PRECISION&
! & (a(1:stripe_width,j+off+a_off-5:j+off+a_off+nbw-1,istripe), w(1:nbw,1:6), &
! nbw, nl, stripe_width, nbw)
!#endif
#else
call hexa_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_6hv_&
&PRECISION&
& (a(1:stripe_width,j+off+a_off-5:j+off+a_off+nbw-1,istripe), w(1:nbw,1:6), &
nbw, nl, stripe_width, nbw)
#endif
#endif /* WITH_OPENMP */
enddo
do jj = j, 4, -4
......@@ -1489,38 +1489,38 @@
w(:,4) = bcast_buffer(1:nbw,jj+off-3)
#ifdef WITH_OPENMP
!#ifdef USE_ASSUMED_SIZE
#ifdef USE_ASSUMED_SIZE
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1,jj+off+a_off-3,istripe,my_thread), w, nbw, nl, stripe_width, nbw)
!#else
! call quad_hh_trafo_&
! &MATH_DATATYPE&
! &_generic_simple_4hv_&
! &PRECISION&
! & (a(1:stripe_width,jj+off+a_off-3:jj+off+a_off+nbw-1,istripe,my_thread), &
! w(1:nbw,1:6), nbw, nl, stripe_width, nbw)
!#endif
#else
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1:stripe_width,jj+off+a_off-3:jj+off+a_off+nbw-1,istripe,my_thread), &
w(1:nbw,1:6), nbw, nl, stripe_width, nbw)
#endif
#else /* WITH_OPENMP */
!#ifdef USE_ASSUMED_SIZE
#ifdef USE_ASSUMED_SIZE
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1,jj+off+a_off-3,istripe), w, &
nbw, nl, stripe_width, nbw)
!#else
! call quad_hh_trafo_&
! &MATH_DATATYPE&
! &_generic_simple_4hv_&
! &PRECISION&
! & (a(1:stripe_width,jj+off+a_off-3:jj+off+a_off+nbw-1,istripe), &
! w(1:nbw,1:6), nbw, nl, stripe_width, nbw)
!#endif
#else
call quad_hh_trafo_&
&MATH_DATATYPE&
&_generic_simple_4hv_&
&PRECISION&
& (a(1:stripe_width,jj+off+a_off-3:jj+off+a_off+nbw-1,istripe), &
w(1:nbw,1:6), nbw, nl, stripe_width, nbw)
#endif
#endif /* WITH_OPENMP */
enddo
......
src/elpa2/kernels/real_simple_block4.F90
View file @ 5ae53d29
......@@ -59,18 +59,18 @@
#endif
#include "config-f90.h"
!#ifndef USE_ASSUMED_SIZE
!module real_generic_simple_block4_kernel
!
! private
! public quad_hh_trafo_real_generic_simple_4hv_double
!
!#ifdef WANT_SINGLE_PRECISION_REAL
! public quad_hh_trafo_real_generic_simple_4hv_single
!#endif
!
! contains
!#endif
#ifndef USE_ASSUMED_SIZE
module real_generic_simple_block4_kernel
private
public quad_hh_trafo_real_generic_simple_4hv_double
#ifdef WANT_SINGLE_PRECISION_REAL
public quad_hh_trafo_real_generic_simple_4hv_single
#endif
contains
#endif
#define REALCASE 1
#define DOUBLE_PRECISION 1
......@@ -88,7 +88,7 @@
#undef SINGLE_PRECISION
#endif
!#ifndef USE_ASSUMED_SIZE
!end module real_generic_simple_block4_kernel
!#endif
#ifndef USE_ASSUMED_SIZE
end module real_generic_simple_block4_kernel
#endif
! --------------------------------------------------------------------------------------------------
src/elpa2/kernels/real_simple_block6.F90
View file @ 5ae53d29
......@@ -60,18 +60,18 @@
#endif
#include "config-f90.h"
!#ifndef USE_ASSUMED_SIZE
!module real_generic_simple_block6_kernel
!
! private
! public hexa_hh_trafo_real_generic_simple_6hv_double
!
!#ifdef WANT_SINGLE_PRECISION_REAL
! public hexa_hh_trafo_real_generic_simple_6hv_single
!#endif
!
! contains
!#endif
#ifndef USE_ASSUMED_SIZE
module real_generic_simple_block6_kernel
private
public hexa_hh_trafo_real_generic_simple_6hv_double
#ifdef WANT_SINGLE_PRECISION_REAL
public hexa_hh_trafo_real_generic_simple_6hv_single
#endif
contains
#endif
#define REALCASE 1
#define DOUBLE_PRECISION 1
......@@ -89,7 +89,7 @@
#undef SINGLE_PRECISION
#endif
!#ifndef USE_ASSUMED_SIZE
!end module real_generic_simple_block6_kernel
!#endif
#ifndef USE_ASSUMED_SIZE
end module real_generic_simple_block6_kernel
#endif
! --------------------------------------------------------------------------------------------------
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment