Commit 5d72d309 authored by Joubert, Wayne's avatar Joubert, Wayne
Browse files

nuccor_kernels f/c pgi crest now works (using workarounds).

parent 55b59e83
......@@ -82,7 +82,7 @@ def check_results(path_to_results):
for line in lines:
match = re.findall(r'^FINAL.*? num_failures ([0-9]+)$', line)
match = re.findall(r'^ *FINAL.*? num_failures +([0-9]+) *$', line)
if len(match) == 1:
num_failed = match[0]
......
......@@ -181,7 +181,8 @@ def make_batch_script(batch_recursive_mode, path_to_workspace,
joblaunchcommand = '$EXECUTABLE 1> std.out.txt 2> std.err.txt'
input_file = 'sizes_1128.txt'
input_file = 'sizes_1.txt'
#input_file = 'sizes_10.txt'
#input_file = 'sizes_1.txt'
rg_array = [('__jobname__', jobname),
('__walltime__', walltime),
......
......@@ -120,6 +120,9 @@ function make_nuccor_kernels
#--------------------
if [ $COMPILER_TYPE = PGI ] ; then
#--------------------
# TODO: fix this
echo "Compiling with PGI compiler, C version ..."
EXECUTABLE=nuccor_dgemm_titan_c_pgi
module load cudatoolkit
......@@ -165,6 +168,8 @@ function make_nuccor_kernels
module swap PrgEnv-cray PrgEnv-pgi
module unload craype-accel-nvidia35
rm -f *.o
#--------------------
fi
#--------------------
......@@ -202,10 +207,22 @@ function make_nuccor_kernels
#--------------------
echo "Compiling with $COMPILER_TYPE compiler, $COMPILER_LANG version ..."
module load pgi
module load cuda
local CFLAGS="-fast -acc -ta=nvidia:rdc -Minfo=accel"
local CUDAFLAGS="-L$CUDA_DIR/lib64/ -lcublas_static"
local CUDAFLAGS="$CUDAFLAGS -lcudart -lculibos -ldl -lpthread"
pgcc -c get_wall_time.c
pgf90 -c nuccor_dgemm.F90 $CFLAGS $CUDAFLAGS
pgf90 -o $EXECUTABLE nuccor_dgemm.o get_wall_time.o \
$CFLAGS $CUDAFLAGS $LAPACK_FLAGS -lcudaforblas
module unload cuda
module unload pgi
rm -f *.o
#--------------------
else
......
......@@ -250,8 +250,10 @@
!$acc end host_data
!$acc exit data delete(a3)
!$acc exit data delete(a2)
!$acc exit data
! !$acc exit data copyout(b)
! vvv TODO: FIX
! !$acc exit data
!$acc exit data copyout(b)
! ^^^ TODO: FIX
!$acc wait(0)
endif
......@@ -280,8 +282,10 @@
!$acc end host_data
!$acc exit data delete(a3)
!$acc exit data delete(a2)
!$acc exit data
! !$acc exit data copyout(b)
! vvv TODO: FIX
! !$acc exit data
!$acc exit data copyout(b)
! ^^^ TODO: FIX
!$acc wait(0)
endif
......
......@@ -236,9 +236,10 @@ void compute_product_gpu(Matrix* c, Matrix* a1, Matrix* a2, Matrix* a3,
#pragma acc exit data delete(a3_data[:m*n])
#pragma acc exit data delete(a2_data[:m*m])
/*
#pragma acc exit data copyout(bp_data[:m*n])
*/
// vvv TODO: FIX
//#pragma acc exit data
#pragma acc exit data copyout(bp_data[:m*n])
// ^^^ TODO: FIX
#pragma acc wait(0)
}
......
../Source/Inputs
\ No newline at end of file
../../Source/Common_Scripts/build_executable.x
\ No newline at end of file
../../Source/Common_Scripts/check_executable.x
\ No newline at end of file
../../Source/Common_Scripts/lsf.template.x
\ No newline at end of file
../../Source/Common_Scripts/pbs.template.x
\ No newline at end of file
../../Source/Common_Scripts/submit_executable.x
\ No newline at end of file
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