From cd1f3be365119725b9b7845d1c13928e92261c10 Mon Sep 17 00:00:00 2001
From: Paul Baumeister <p.baumeister@fz-juelich.de>
Date: Mon, 29 Jul 2019 15:31:32 +0200
Subject: [PATCH] difficult to link, fails to run due to OpenMPI-error
---
.../source/IterativeSolver/KKRmat_mod.F90 | 9 +++--
source/KKRnano/source/Makefile | 39 +++++++++++++------
.../source/NonCollinearMagnetism_mod.F90 | 2 +-
.../datastructures/ChebMeshData_mod.F90 | 2 +-
4 files changed, 35 insertions(+), 17 deletions(-)
diff --git a/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90 b/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90
index 73a40cee9..fc05a235f 100644
--- a/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90
+++ b/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90
@@ -220,7 +220,7 @@ module KKRmat_mod
use IterativeSolver_mod, only: IterativeSolver, solve
use DirectSolver_mod, only: DirectSolver, solve
#ifdef has_tfQMRgpu
- ! use tfqmrgpu, only: tfqmrgpu_bsrsv_complete ! all-in-one GPU solver interface for rapid integration
+ use tfqmrgpu, only: tfqmrgpu_bsrsv_complete ! all-in-one GPU solver interface for rapid integration
#endif
use SparseMatrixDescription_mod, only: dump
use InitialGuess_mod, only: InitialGuess, load, store
@@ -261,7 +261,6 @@ module KKRmat_mod
double complex :: tracek ! LLY
#ifdef has_tfQMRgpu
- external :: tfqmrgpu_bsrsv_complete ! subroutine
integer :: o = 6
integer :: ierr = 1
integer :: iterations, lda
@@ -393,9 +392,9 @@ module KKRmat_mod
! store the initial guess
call store(iguess_data, op%mat_X, ik=ikpoint, is=ispin, ie=ienergy)
-#ifdef has_tfQMRgpu
case (5) ! GPU solver
+#ifdef has_tfQMRgpu
iterations = 1000
residuum = 1e-7
lda = size(op%mat_A, 1)
@@ -405,8 +404,10 @@ module KKRmat_mod
op%bsr_X%RowStart, op%bsr_X%ColIndex, op%mat_X, 'n', & !! X (out)
op%bsr_B%RowStart, op%bsr_B%ColIndex, op%mat_B, 'n', & !! B (in)
iterations, residuum, o, ierr)
+#else
+ warn(6, "GPU solver needs -D has_tfQMRgpu (Problem is not solved) solver_type ="+solver_type)
#endif
-
+
case default
warn(6, "No solver selected! Problem is not solved, solver_type ="+solver_type)
endselect ! solver_type
diff --git a/source/KKRnano/source/Makefile b/source/KKRnano/source/Makefile
index a87e05bc7..aa5958da0 100644
--- a/source/KKRnano/source/Makefile
+++ b/source/KKRnano/source/Makefile
@@ -7,19 +7,20 @@ TYPE ?= nodebug
### the user may specify SMP=openmp as a command line argument to make
SMP ?= none
-### can we make use of the tfQMRgpu library
-tfQMRgpu ?= no
-
PROGRAM = kkr.exe
# Path to put object files and module files
-BUILDDIR = $(HOME)/build
+BUILDDIR = ./build
EXTRA_FLAGS ?=
+### can we make use of the tfQMRgpu library?
+tfQMRgpu ?= no
ifeq ($(tfQMRgpu),yes)
EXTRA_FLAGS += -D has_tfQMRgpu
+ CUDA_PATH=-L/usr/local/zam/CUDA/cuda-9.1/lib64
endif
+
FC90FLAGS =
FCFLAGS =
PPFLAGS =
@@ -98,7 +99,10 @@ ifeq ($(PLATFORM),gfortran)
FC90FLAGS = -ffree-line-length-0
PPFLAGS += -D NOLOGGING
# FCFLAGS += -D TASKLOCAL_FILES ### breaks when TASKLOCAL_FILES is used
- LDFLAGS += -L /usr/local/atlas/lib -llapack -lf77blas -lcblas -latlas
+# LDFLAGS += -L /usr/local/atlas/lib -llapack -latlas
+ LDFLAGS += -L${MKLROOT}/lib/intel64 -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5 ###-lpthread
+# # LDFLAGS += -lgfortran
+ ### -lf77blas -lcblas -latlas
endif
@@ -377,23 +381,36 @@ SRCSFPP = $(foreach DIR,$(DIRS),$(wildcard $(DIR)/*.F))
SRCS90FPP = $(foreach DIR,$(DIRS),$(wildcard $(DIR)/*.F90))
SRCS95TMPL = $(foreach DIR,$(DIRS),$(wildcard $(DIR)/*.F95))
+OBJS=
+
+LINKER= $(FC90)
+ifeq ($(tfQMRgpu),yes)
+ OBJS += tfqmrgpu_Fortran.o
+ OBJS += tfqmrgpu_Fortran_wrappers.o
+ OBJS += tfqmrgpu.o
+ SRCS90FPP += GPU/tfqmrgpu_Fortran.F90
+# LDFLAGS += $(CUDA_PATH) -gfortran -lcudart -lnvToolsExt -lm -lcurand \
+# -L/usr/lib64/gcc/x86_64-suse-linux/4.8 \
+# -L/usr/lib -lstdc++
+ LDFLAGS += $(CUDA_PATH) -lcudart -lnvToolsExt -lm -lcurand
+ LDFLAGS += -L/usr/lib -lstdc++
+# # LINKER= gfortran
+endif
+
+
# notdir extracts only filename
-OBJS = $(notdir ${SRCS:.f=.o})
+OBJS += $(notdir ${SRCS:.f=.o})
OBJS += $(notdir ${SRCS90:.f90=.o})
OBJS += $(notdir ${SRCSFPP:.F=.o})
OBJS += $(notdir ${SRCS90FPP:.F90=.o})
OBJS += $(notdir ${SRCS95TMPL:.F95=.o})
-ifeq ($(tfQMRgpu),yes)
- OBJS += tfqmrgpu_Fortran.o tfqmrgpu_Fortran_wrappers.o tfqmrgpu.o
- LDFLAGS += -L/usr/local/cuda/lib64 -lcudart -lnvToolsExt -lm -lcurand
-endif
.PHONY: all
all: $(PROGRAM)
$(PROGRAM): $(OBJS)
- $(FC90) $(FCFLAGS) -o $(PROGRAM) $(addprefix $(BUILDDIR)/,$(OBJS)) $(LDFLAGS)
+ $(LINKER) -o $(PROGRAM) $(addprefix $(BUILDDIR)/,$(OBJS)) $(LDFLAGS)
%.o: %.f
$(FC) $(FCFLAGS) $(EXTRA_FLAGS) -c $< -o $(BUILDDIR)/$@
diff --git a/source/KKRnano/source/NonCollinearMagnetism_mod.F90 b/source/KKRnano/source/NonCollinearMagnetism_mod.F90
index fea34a3c3..d1e75662f 100644
--- a/source/KKRnano/source/NonCollinearMagnetism_mod.F90
+++ b/source/KKRnano/source/NonCollinearMagnetism_mod.F90
@@ -1043,7 +1043,7 @@ END IF
! klo and khi now bracket the input value of x.
h=xa(khi)-xa(klo)
! The xa's must be distinct.
-IF (h == 0.d0) PAUSE 'bad xa input in splint'
+IF (h == 0.d0) STOP 'bad xa input in splint' !! used to be PAUSE
! Cubic spline polynomial is now evaluated.
a = (xa(khi)-x)/h
b = (x-xa(klo))/h
diff --git a/source/KKRnano/source/datastructures/ChebMeshData_mod.F90 b/source/KKRnano/source/datastructures/ChebMeshData_mod.F90
index 2ed1cdf35..95358e8fc 100644
--- a/source/KKRnano/source/datastructures/ChebMeshData_mod.F90
+++ b/source/KKRnano/source/datastructures/ChebMeshData_mod.F90
@@ -674,7 +674,7 @@ end if
! klo and khi now bracket the input value of x.
h=xa(khi)-xa(klo)
! the xa's must be distinct.
-if (h == 0.d0) pause 'bad xa input in splint'
+if (h == 0.d0) STOP 'bad xa input in splint' !! used to be PAUSE
! cubic spline polynomial is now evaluated.
a = (xa(khi)-x)/h
b = (x-xa(klo))/h
--
GitLab