Skip to content

GitLab

Commit 8264ec0d authored by Andreas Marek's avatar Andreas Marek
Browse files

Remove all references to AVX512 single precision until they are implemented

parent 8a167d7f
Hide whitespace changes
Inline Side-by-side
Showing with 63 additions and 64 deletions
Makefile.am
View file @ 8264ec0d
......@@ -145,10 +145,9 @@ endif
if WITH_REAL_AVX512_BLOCK2_KERNEL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_2hv_double_precision.c
if WANT_SINGLE_PRECISION_REAL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_2hv_single_precision.c
endif
#if WANT_SINGLE_PRECISION_REAL
# libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_2hv_single_precision.c
#endif
endif
......@@ -175,9 +174,9 @@ endif
if WITH_REAL_AVX512_BLOCK4_KERNEL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_4hv_double_precision.c
if WANT_SINGLE_PRECISION_REAL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_4hv_single_precision.c
endif
#if WANT_SINGLE_PRECISION_REAL
# libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_4hv_single_precision.c
#endif
endif
......@@ -205,9 +204,9 @@ endif
if WITH_REAL_AVX512_BLOCK6_KERNEL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_6hv_double_precision.c
if WANT_SINGLE_PRECISION_REAL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_6hv_single_precision.c
endif
#if WANT_SINGLE_PRECISION_REAL
# libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_real_avx512_6hv_single_precision.c
#endif
endif
......@@ -235,9 +234,9 @@ endif
if WITH_COMPLEX_AVX512_BLOCK1_KERNEL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx512_1hv_double_precision.c
if WANT_SINGLE_PRECISION_COMPLEX
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx512_1hv_single_precision.c
endif
#if WANT_SINGLE_PRECISION_COMPLEX
# libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx512_1hv_single_precision.c
#endif
endif
if WITH_COMPLEX_SSE_BLOCK2_KERNEL
......@@ -263,9 +262,9 @@ endif
if WITH_COMPLEX_AVX512_BLOCK2_KERNEL
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx512_2hv_double_precision.c
if WANT_SINGLE_PRECISION_COMPLEX
libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx512_2hv_single_precision.c
endif
#if WANT_SINGLE_PRECISION_COMPLEX
# libelpa@SUFFIX@_private_la_SOURCES += src/elpa2_kernels/elpa2_kernels_complex_avx512_2hv_single_precision.c
#endif
endif
.cu.lo:
......
src/mod_compute_hh_trafo_complex.F90
View file @ 8264ec0d
......@@ -539,33 +539,33 @@ module compute_hh_trafo_complex
#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
#endif /* WITH_COMPLEX_SSE_BLOCK2_KERNEL */
#if defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL)
#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
if ( (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK2) ) then
#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
ttt = mpi_wtime()
do j = ncols, 2, -2
w(:,1) = bcast_buffer(1:nbw,j+off)
w(:,2) = bcast_buffer(1:nbw,j+off-1)
#ifdef WITH_OPENMP
call double_hh_trafo_complex_avx512_2hv_single(a(1,j+off+a_off-1,istripe,my_thread), &
w, nbw, nl, stripe_width, nbw)
#else
call double_hh_trafo_complex_avx512_2hv_single(a(1,j+off+a_off-1,istripe), &
w, nbw, nl, stripe_width, nbw)
#endif
enddo
#ifdef WITH_OPENMP
if (j==1) call single_hh_trafo_complex_avx512_1hv_single(a(1,1+off+a_off,istripe,my_thread), &
bcast_buffer(1,off+1), nbw, nl, stripe_width)
#else
if (j==1) call single_hh_trafo_complex_avx512_1hv_single(a(1,1+off+a_off,istripe), &
bcast_buffer(1,off+1), nbw, nl, stripe_width)
#endif
#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
endif ! (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK2)
#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
#endif /* WITH_COMPLEX_AVX512_BLOCK2_KERNEL */
!#if defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL)
!#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
! if ( (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK2) ) then
!#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
! ttt = mpi_wtime()
! do j = ncols, 2, -2
! w(:,1) = bcast_buffer(1:nbw,j+off)
! w(:,2) = bcast_buffer(1:nbw,j+off-1)
!#ifdef WITH_OPENMP
! call double_hh_trafo_complex_avx512_2hv_single(a(1,j+off+a_off-1,istripe,my_thread), &
! w, nbw, nl, stripe_width, nbw)
!#else
! call double_hh_trafo_complex_avx512_2hv_single(a(1,j+off+a_off-1,istripe), &
! w, nbw, nl, stripe_width, nbw)
!#endif
! enddo
!#ifdef WITH_OPENMP
! if (j==1) call single_hh_trafo_complex_avx512_1hv_single(a(1,1+off+a_off,istripe,my_thread), &
! bcast_buffer(1,off+1), nbw, nl, stripe_width)
!#else
! if (j==1) call single_hh_trafo_complex_avx512_1hv_single(a(1,1+off+a_off,istripe), &
! bcast_buffer(1,off+1), nbw, nl, stripe_width)
!#endif
!#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
! endif ! (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK2)
!#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
!#endif /* WITH_COMPLEX_AVX512_BLOCK2_KERNEL */
#if defined(WITH_COMPLEX_GENERIC_SIMPLE_KERNEL)
......@@ -710,27 +710,27 @@ module compute_hh_trafo_complex
#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
#endif /* WITH_COMPLEX_AVX_BLOCK1_KERNEL || WITH_COMPLEX_AVX_BLOCK1_KERNEL */
#if defined(WITH_COMPLEX_AVX512_BLOCK1_KERNEL)
#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
if ( (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK1)) then
#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
ttt = mpi_wtime()
#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL) || (defined(WITH_ONE_SPECIFIC_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL) )
do j = ncols, 1, -1
#ifdef WITH_OPENMP
call single_hh_trafo_complex_avx512_1hv_single(a(1,j+off+a_off,istripe,my_thread), &
bcast_buffer(1,j+off),nbw,nl,stripe_width)
#else
call single_hh_trafo_complex_avx512_1hv_single(a(1,j+off+a_off,istripe), &
bcast_buffer(1,j+off),nbw,nl,stripe_width)
#endif
enddo
#endif /* defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL) || (defined(WITH_ONE_SPECIFIC_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL) ) */
#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
endif ! ( (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK1))
#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
#endif /* WITH_COMPLEX_AVX512_BLOCK1_KERNEL */
!#if defined(WITH_COMPLEX_AVX512_BLOCK1_KERNEL)
!#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
! if ( (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK1)) then
!#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
! ttt = mpi_wtime()
!#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL) || (defined(WITH_ONE_SPECIFIC_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL) )
! do j = ncols, 1, -1
!#ifdef WITH_OPENMP
! call single_hh_trafo_complex_avx512_1hv_single(a(1,j+off+a_off,istripe,my_thread), &
! bcast_buffer(1,j+off),nbw,nl,stripe_width)
!#else
! call single_hh_trafo_complex_avx512_1hv_single(a(1,j+off+a_off,istripe), &
! bcast_buffer(1,j+off),nbw,nl,stripe_width)
!#endif
! enddo
!#endif /* defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL) || (defined(WITH_ONE_SPECIFIC_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX512_BLOCK2_KERNEL) ) */
!
!#if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL)
! endif ! ( (THIS_COMPLEX_ELPA_KERNEL .eq. COMPLEX_ELPA_KERNEL_AVX512_BLOCK1))
!#endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
!#endif /* WITH_COMPLEX_AVX512_BLOCK1_KERNEL */
#ifdef WITH_OPENMP
if (my_thread==1) then
......
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