Rename GPU into cuda, and merge SPGPU code.

rename-cuda
sfilippone 1 year ago
parent 2732336915
commit 6aa7987d52

@ -67,15 +67,15 @@ UTILMODNAME=@UTILMODNAME@
CBINDLIBNAME=libpsb_cbind.a
GPUD=@GPUD@
GPULD=@GPULD@
LGPU=@LGPU@
CUDAD=@CUDAD@
CUDALD=@CUDALD@
LCUDA=@LCUDA@
SPGPUDIR=@SPGPU_DIR@
SPGPU_INCDIR=@SPGPU_INCDIR@
#SPGPUDIR=@SPGPU_DIR@
#SPGPU_INCDIR=@SPGPU_INCDIR@
SPGPU_LIBS=@SPGPU_LIBS@
SPGPU_DEFINES=@SPGPU_DEFINES@
SPGPU_INCLUDES=@SPGPU_INCLUDES@
#SPGPU_INCLUDES=@SPGPU_INCLUDES@
CUDA_DIR=@CUDA_DIR@
CUDA_DEFINES=@CUDA_DEFINES@
@ -93,4 +93,4 @@ CUDEFINES=@CUDEFINES@
@PSBLASRULES@
PSBGPULDLIBS=$(LGPU) $(SPGPU_LIBS) $(CUDA_LIBS) $(LIBS)
PSBGPULDLIBS=$(LCUDA) $(SPGPU_LIBS) $(CUDA_LIBS) $(PSBLDLIBS) $(LIBS)

@ -1,6 +1,6 @@
include Make.inc
all: dirs based precd kryld utild cbindd extd $(GPUD) libd
all: dirs based precd kryld utild cbindd extd $(CUDAD) libd
@echo "====================================="
@echo "PSBLAS libraries Compilation Successful."
@ -13,18 +13,18 @@ precd: based
utild: based
kryld: precd
extd: based
gpud: extd
cudad: extd
cbindd: based precd kryld utild
libd: based precd kryld utild cbindd extd $(GPULD)
libd: based precd kryld utild cbindd extd $(CUDALD)
$(MAKE) -C base lib
$(MAKE) -C prec lib
$(MAKE) -C krylov lib
$(MAKE) -C util lib
$(MAKE) -C cbind lib
$(MAKE) -C ext lib
gpuld: gpud
$(MAKE) -C gpu lib
cudald: cudad
$(MAKE) -C cuda lib
based:
@ -39,8 +39,8 @@ cbindd:
$(MAKE) -C cbind objs
extd: based
$(MAKE) -C ext objs
gpud: based extd
$(MAKE) -C gpu objs
cudad: based extd
$(MAKE) -C cuda objs
install: all
@ -66,7 +66,7 @@ clean:
$(MAKE) -C util clean
$(MAKE) -C cbind clean
$(MAKE) -C ext clean
$(MAKE) -C gpu clean
$(MAKE) -C cuda clean
check: all
make check -C test/serial
@ -83,7 +83,7 @@ veryclean: cleanlib
cd util && $(MAKE) veryclean
cd cbind && $(MAKE) veryclean
cd ext && $(MAKE) veryclean
cd gpu && $(MAKE) veryclean
cd cuda && $(MAKE) veryclean
cd test/fileread && $(MAKE) clean
cd test/pargen && $(MAKE) clean
cd test/util && $(MAKE) clean

210
configure vendored

@ -653,9 +653,9 @@ ac_subst_vars='am__EXEEXT_FALSE
am__EXEEXT_TRUE
LTLIBOBJS
LIBOBJS
LGPU
GPULD
GPUD
LCUDA
CUDALD
CUDAD
CUDEFINES
CUDA_NVCC
CUDA_SHORT_VERSION
@ -665,10 +665,7 @@ CUDA_INCLUDES
CUDA_DEFINES
CUDA_DIR
EXTRALDLIBS
SPGPU_INCDIR
SPGPU_INCLUDES
SPGPU_DEFINES
SPGPU_DIR
SPGPU_LIBS
SPGPU_FLAGS
METISINCFILE
@ -844,7 +841,6 @@ with_amddir
with_amdincdir
with_amdlibdir
with_cuda
with_spgpu
with_cudacc
'
ac_precious_vars='build_alias
@ -1545,8 +1541,6 @@ Optional Packages:
--with-amdincdir=DIR Specify the directory for AMD includes.
--with-amdlibdir=DIR Specify the directory for AMD library.
--with-cuda=DIR Specify the directory for CUDA library and includes.
--with-spgpu=DIR Specify the directory for SPGPU library and
includes.
--with-cudacc A comma-separated list of CCs to compile to, for
example, --with-cudacc=30,35,37,50,60
@ -10851,181 +10845,16 @@ CPPFLAGS="$SAVE_CPPFLAGS"
CUDA_VERSION="$pac_cv_cuda_version";
CUDA_SHORT_VERSION=$(expr $pac_cv_cuda_version / 1000);
SAVE_LIBS="$LIBS"
SAVE_CPPFLAGS="$CPPFLAGS"
if test "x$pac_cv_have_cuda" == "x"; then
# Check whether --with-cuda was given.
if test ${with_cuda+y}
then :
withval=$with_cuda; pac_cv_cuda_dir=$withval
else $as_nop
pac_cv_cuda_dir=''
fi
ac_ext=c
ac_cpp='$CPP $CPPFLAGS'
ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
ac_compiler_gnu=$ac_cv_c_compiler_gnu
SAVE_LIBS="$LIBS"
SAVE_CPPFLAGS="$CPPFLAGS"
if test "x$pac_cv_cuda_dir" != "x"; then
CUDA_DIR="$pac_cv_cuda_dir"
LIBS="-L$pac_cv_cuda_dir/lib $LIBS"
CUDA_INCLUDES="-I$pac_cv_cuda_dir/include"
CUDA_DEFINES="-DHAVE_CUDA"
CPPFLAGS="$CUDA_INCLUDES $CPPFLAGS"
CUDA_LIBDIR="-L$pac_cv_cuda_dir/lib64 -L$pac_cv_cuda_dir/lib"
if test -f "$pac_cv_cuda_dir/bin/nvcc"; then
CUDA_NVCC="$pac_cv_cuda_dir/bin/nvcc"
else
CUDA_NVCC="nvcc"
fi
fi
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking cuda dir $pac_cv_cuda_dir" >&5
printf %s "checking cuda dir $pac_cv_cuda_dir... " >&6; }
ac_fn_c_check_header_compile "$LINENO" "cuda_runtime.h" "ac_cv_header_cuda_runtime_h" "$ac_includes_default"
if test "x$ac_cv_header_cuda_runtime_h" = xyes
then :
pac_cuda_header_ok=yes
else $as_nop
pac_cuda_header_ok=no; CUDA_INCLUDES=""
fi
if test "x$pac_cuda_header_ok" == "xyes" ; then
CUDA_LIBS="-lcusparse -lcublas -lcudart $CUDA_LIBDIR"
LIBS="$CUDA_LIBS -lm $LIBS";
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking for cudaMemcpy in $CUDA_LIBS" >&5
printf %s "checking for cudaMemcpy in $CUDA_LIBS... " >&6; }
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
/* end confdefs.h. */
/* Override any GCC internal prototype to avoid an error.
Use char because int might match the return type of a GCC
builtin and then its argument prototype would still apply. */
char cudaMemcpy ();
int
main (void)
{
return cudaMemcpy ();
;
return 0;
}
_ACEOF
if ac_fn_c_try_link "$LINENO"
then :
pac_cv_have_cuda=yes;pac_cuda_lib_ok=yes;
else $as_nop
pac_cv_have_cuda=no;pac_cuda_lib_ok=no; CUDA_LIBS=""
fi
rm -f core conftest.err conftest.$ac_objext conftest.beam \
conftest$ac_exeext conftest.$ac_ext
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: result: $pac_cuda_lib_ok" >&5
printf "%s\n" "$pac_cuda_lib_ok" >&6; }
fi
LIBS="$SAVE_LIBS"
CPPFLAGS="$SAVE_CPPFLAGS"
fi
if test "x$pac_cv_have_cuda" == "xyes"; then
# Check whether --with-spgpu was given.
if test ${with_spgpu+y}
then :
withval=$with_spgpu; pac_cv_spgpudir=$withval
else $as_nop
pac_cv_spgpudir=''
fi
ac_ext=c
ac_cpp='$CPP $CPPFLAGS'
ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
ac_link='$CC -o conftest$ac_exeext $CFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
ac_compiler_gnu=$ac_cv_c_compiler_gnu
if test "x$pac_cv_spgpudir" != "x"; then
LIBS="-L$pac_cv_spgpudir/lib $LIBS"
GPU_INCLUDES="-I$pac_cv_spgpudir/include"
CPPFLAGS="$GPU_INCLUDES $CUDA_INCLUDES $CPPFLAGS"
GPU_LIBDIR="-L$pac_cv_spgpudir/lib"
fi
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking spgpu dir $pac_cv_spgpudir" >&5
printf %s "checking spgpu dir $pac_cv_spgpudir... " >&6; }
ac_fn_c_check_header_compile "$LINENO" "core.h" "ac_cv_header_core_h" "$ac_includes_default"
if test "x$ac_cv_header_core_h" = xyes
then :
pac_gpu_header_ok=yes
else $as_nop
pac_gpu_header_ok=no; GPU_INCLUDES=""
fi
if test "x$pac_gpu_header_ok" == "xyes" ; then
GPU_LIBS="-lspgpu $GPU_LIBDIR"
LIBS="$GPU_LIBS $CUDA_LIBS -lm $LIBS";
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking for spgpuCreate in $GPU_LIBS" >&5
printf %s "checking for spgpuCreate in $GPU_LIBS... " >&6; }
cat confdefs.h - <<_ACEOF >conftest.$ac_ext
/* end confdefs.h. */
/* Override any GCC internal prototype to avoid an error.
Use char because int might match the return type of a GCC
builtin and then its argument prototype would still apply. */
char spgpuCreate ();
int
main (void)
{
return spgpuCreate ();
;
return 0;
}
_ACEOF
if ac_fn_c_try_link "$LINENO"
then :
pac_cv_have_spgpu=yes;pac_gpu_lib_ok=yes;
else $as_nop
pac_cv_have_spgpu=no;pac_gpu_lib_ok=no; GPU_LIBS=""
fi
rm -f core conftest.err conftest.$ac_objext conftest.beam \
conftest$ac_exeext conftest.$ac_ext
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: result: $pac_gpu_lib_ok" >&5
printf "%s\n" "$pac_gpu_lib_ok" >&6; }
if test "x$pac_cv_have_spgpu" == "xyes" ; then
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: Have found SPGPU" >&5
printf "%s\n" "$as_me: Have found SPGPU" >&6;}
SPGPULIBNAME="libpsbgpu.a";
SPGPU_DIR="$pac_cv_spgpudir";
SPGPU_DEFINES="-DHAVE_SPGPU";
SPGPU_INCDIR="$SPGPU_DIR/include";
SPGPU_INCLUDES="-I$SPGPU_INCDIR";
SPGPU_LIBS="-lspgpu -L$SPGPU_DIR/lib";
LGPU=-lpsb_gpu
CUDA_DIR="$pac_cv_cuda_dir";
CUDA_DEFINES="-DHAVE_CUDA";
CUDA_INCLUDES="-I$pac_cv_cuda_dir/include"
CUDA_LIBDIR="-L$pac_cv_cuda_dir/lib64 -L$pac_cv_cuda_dir/lib"
FDEFINES="$psblas_cv_define_prepend-DHAVE_GPU $psblas_cv_define_prepend-DHAVE_SPGPU $psblas_cv_define_prepend-DHAVE_CUDA $FDEFINES";
CDEFINES="-DHAVE_SPGPU -DHAVE_CUDA $CDEFINES" ;
fi
fi
fi
LIBS="$SAVE_LIBS"
CPPFLAGS="$SAVE_CPPFLAGS"
if test "x$pac_cv_have_spgpu" == "xyes" ; then
GPUD=gpud;
GPULD=gpuld;
LGPU="-lpsb_gpu";
if test "x$pac_cv_have_cuda" == "xyes" ; then
SPGPU_DEFINES="-DHAVE_SPGPU -DHAVE_GPU";
SPGPU_LIBS="-lspgpu";
CUDAD=cudad;
CUDALD=cudald;
LCUDA="-lpsb_cuda";
EXTRALDLIBS="-lstdc++";
fi
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: At this point GPUTARGET is $GPUD $GPULD" >&5
printf "%s\n" "$as_me: At this point GPUTARGET is $GPUD $GPULD" >&6;}
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: At this point GPUTARGET is $CUDAD $CUDALD" >&5
printf "%s\n" "$as_me: At this point GPUTARGET is $CUDAD $CUDALD" >&6;}
@ -11041,7 +10870,7 @@ fi
if test "x$pac_cv_cudacc" == "x"; then
pac_cv_cudacc="30,35,37,50,60";
fi
CUDEFINES="";
CUDEFINES="--dopt=on";
for cc in `echo $pac_cv_cudacc|sed 's/,/ /gi'`
do
CUDEFINES="$CUDEFINES -gencode arch=compute_$cc,code=sm_$cc";
@ -11057,8 +10886,10 @@ fi
if test "x$pac_cv_ipk_size" != "x4"; then
{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: For CUDA I need psb_ipk_ to be 4 bytes but it is $pac_cv_ipk_size, disabling CUDA/SPGPU" >&5
printf "%s\n" "$as_me: For CUDA I need psb_ipk_ to be 4 bytes but it is $pac_cv_ipk_size, disabling CUDA/SPGPU" >&6;}
GPUD="";
GPULD="";
SPGPU_DEFINES="";
SPGPU_LIBS="";
CUDAD="";
CUDALD="";
CUDEFINES="";
CUDA_INCLUDES="";
CUDA_LIBS="";
@ -11124,9 +10955,9 @@ UTILLIBNAME=libpsb_util.a
PSBLASRULES='
PSBLDLIBS=$(LAPACK) $(BLAS) $(METIS_LIB) $(AMD_LIB) $(LIBS)
CXXDEFINES=$(PSBCXXDEFINES)
CDEFINES=$(PSBCDEFINES)
FDEFINES=$(PSBFDEFINES)
CXXDEFINES=$(PSBCXXDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
CDEFINES=$(PSBCDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
FDEFINES=$(PSBFDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
# These should be portable rules, arent they?
@ -11159,9 +10990,6 @@ FDEFINES=$(PSBFDEFINES)

@ -798,20 +798,22 @@ if test "x$pac_cv_have_cuda" == "xyes"; then
PAC_CHECK_CUDA_VERSION()
CUDA_VERSION="$pac_cv_cuda_version";
CUDA_SHORT_VERSION=$(expr $pac_cv_cuda_version / 1000);
PAC_CHECK_SPGPU()
if test "x$pac_cv_have_spgpu" == "xyes" ; then
GPUD=gpud;
GPULD=gpuld;
LGPU="-lpsb_gpu";
dnl PAC_CHECK_SPGPU()
if test "x$pac_cv_have_cuda" == "xyes" ; then
SPGPU_DEFINES="-DHAVE_SPGPU -DHAVE_GPU";
SPGPU_LIBS="-lspgpu";
CUDAD=cudad;
CUDALD=cudald;
LCUDA="-lpsb_cuda";
EXTRALDLIBS="-lstdc++";
fi
AC_MSG_NOTICE([At this point GPUTARGET is $GPUD $GPULD])
AC_MSG_NOTICE([At this point GPUTARGET is $CUDAD $CUDALD])
PAC_ARG_WITH_CUDACC()
if test "x$pac_cv_cudacc" == "x"; then
pac_cv_cudacc="30,35,37,50,60";
fi
CUDEFINES="";
CUDEFINES="--dopt=on";
for cc in `echo $pac_cv_cudacc|sed 's/,/ /gi'`
do
CUDEFINES="$CUDEFINES -gencode arch=compute_$cc,code=sm_$cc";
@ -826,8 +828,10 @@ fi
if test "x$pac_cv_ipk_size" != "x4"; then
AC_MSG_NOTICE([For CUDA I need psb_ipk_ to be 4 bytes but it is $pac_cv_ipk_size, disabling CUDA/SPGPU])
GPUD="";
GPULD="";
SPGPU_DEFINES="";
SPGPU_LIBS="";
CUDAD="";
CUDALD="";
CUDEFINES="";
CUDA_INCLUDES="";
CUDA_LIBS="";
@ -893,9 +897,9 @@ AC_SUBST(FINCLUDES)
PSBLASRULES='
PSBLDLIBS=$(LAPACK) $(BLAS) $(METIS_LIB) $(AMD_LIB) $(LIBS)
CXXDEFINES=$(PSBCXXDEFINES)
CDEFINES=$(PSBCDEFINES)
FDEFINES=$(PSBFDEFINES)
CXXDEFINES=$(PSBCXXDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
CDEFINES=$(PSBCDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
FDEFINES=$(PSBFDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
# These should be portable rules, arent they?
@ -919,10 +923,10 @@ AC_SUBST(UTILLIBNAME)
AC_SUBST(METISINCFILE)
AC_SUBST(SPGPU_FLAGS)
AC_SUBST(SPGPU_LIBS)
AC_SUBST(SPGPU_DIR)
dnl AC_SUBST(SPGPU_DIR)
AC_SUBST(SPGPU_DEFINES)
AC_SUBST(SPGPU_INCLUDES)
AC_SUBST(SPGPU_INCDIR)
dnl AC_SUBST(SPGPU_INCLUDES)
dnl AC_SUBST(SPGPU_INCDIR)
AC_SUBST(EXTRALDLIBS)
AC_SUBST(CUDA_DIR)
AC_SUBST(CUDA_DEFINES)
@ -932,9 +936,9 @@ AC_SUBST(CUDA_VERSION)
AC_SUBST(CUDA_SHORT_VERSION)
AC_SUBST(CUDA_NVCC)
AC_SUBST(CUDEFINES)
AC_SUBST(GPUD)
AC_SUBST(GPULD)
AC_SUBST(LGPU)
AC_SUBST(CUDAD)
AC_SUBST(CUDALD)
AC_SUBST(LCUDA)
###############################################################################
# the following files will be created by Automake

@ -15,11 +15,10 @@ LDLIBS=$(PSBLDLIBS)
#
#CCOPT= -g
FINCLUDES=$(FMFLAG). $(FMFLAG)$(INCDIR) $(FMFLAG)$(PSBINCDIR) $(FIFLAG).
CINCLUDES=$(SPGPU_INCLUDES) $(CUDA_INCLUDES) -I..
CINCLUDES=$(SPGPU_INCLUDES) $(CUDA_INCLUDES) -I.. -I$(INCDIR)
LIBNAME=libpsb_gpu.a
CUDAOBJS=psi_cuda_c_CopyCooToElg.o psi_cuda_c_CopyCooToHlg.o \
psi_cuda_d_CopyCooToElg.o psi_cuda_d_CopyCooToHlg.o \
psi_cuda_s_CopyCooToElg.o psi_cuda_s_CopyCooToHlg.o \

@ -13,8 +13,8 @@ LDLIBS=$(PSBLDLIBS)
#
#CCOPT= -g
FINCLUDES=$(FMFLAG). $(FMFLAG)$(INCDIR) $(FMFLAG)$(MODDIR) $(FIFLAG).
CINCLUDES=$(SPGPU_INCLUDES) $(CUDA_INCLUDES)
LIBNAME=libpsb_gpu.a
CINCLUDES=$(SPGPU_INCLUDES) $(CUDA_INCLUDES) -I$(INCDIR)
LIBNAME=libpsb_cuda.a
FOBJS=cusparse_mod.o base_cusparse_mod.o \
@ -49,11 +49,18 @@ OBJS=$(COBJS) $(FOBJS)
lib: objs
objs: $(OBJS) iobjs cudaobjs
objs: spgpuinc $(OBJS) iobjs cudaobjs spgpuobjs
/bin/cp -p *$(.mod) $(MODDIR)
/bin/cp -p *.h $(INCDIR)
lib: ilib cudalib
spgpuinc:
$(MAKE) -C spgpu includes
spgpuobjs:
$(MAKE) -C spgpu objs
spgpulib:
$(MAKE) -C spgpu lib
lib: ilib cudalib spgpulib
ar cur $(LIBNAME) $(OBJS)
/bin/cp -p $(LIBNAME) $(LIBDIR)
@ -121,7 +128,7 @@ ilib: objs
cudalib: objs ilib
$(MAKE) -C CUDA lib LIBNAME=$(LIBNAME)
clean: cclean iclean cudaclean
clean: cclean iclean cudaclean spgpuclean
/bin/rm -f $(FOBJS) *$(.mod) *.a
cclean:
@ -130,5 +137,7 @@ iclean:
$(MAKE) -C impl clean
cudaclean:
$(MAKE) -C CUDA clean
spgpuclean:
$(MAKE) -C spgpu clean
veryclean: clean

@ -12,6 +12,9 @@ LDLIBS=$(PSBLDLIBS)
FINCLUDES=$(FMFLAG).. $(FMFLAG)$(MODDIR) $(FMFLAG)$(INCDIR) $(FIFLAG)..
CINCLUDES=-I$(GPU_INCDIR) -I$(CUDA_INCDIR)
LIBNAME=libpsb_gpu.a
CXXDEFINES=$(PSBCXXDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
CDEFINES=$(PSBCDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
FDEFINES=$(PSBFDEFINES) $(SPGPU_DEFINES) $(CUDA_DEFINES)
OBJS= \
psb_d_cp_csrg_from_coo.o \

Some files were not shown because too many files have changed in this diff Show More

Loading…
Cancel
Save