Problem in compilation of MINPACK library with CUDA code - c++

Question

I need to solve a nonlinear system of equations and I found the MINPACK library to do the job.
But I need the code to work with CUDA and I'm not able to compile the code.
I'm facing this problem:
nvcc -arch sm_61 -Iinc -I/usr/local/include obj/main.o -o oi -L/usr/local/lib64 -lcuminpack
nvlink error : Undefined reference to 'dpmpar' in 'obj/main.o'
make: *** [makefile:55: oi] Erro 255
My main file is:
main.cu
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#define __cminpack_double__
#include <cminpack-1/cminpack.h>
#define real __cminpack_real__
typedef struct{
real fnorm1, fnorm2;
int info;
real solution[2];
} ResultType;
__cminpack_attr__ int fcn(void *p, int n, const real *x, real *fvec, int iflag){
--fvec;
--x;
fvec[0] = x[0] + x[0]*x[1] - 4.0;
fvec[1] = x[0] + x[1] -3.0;
return 0;
}
__global__ void SolvingSystem(ResultType *pResults){
int info;
real fnorm1, fnorm2, tol;
int n = 2;
real fvec[2];
real x[2];
const int lwa = (n*(3*n + 13))/2;
real wa[50];
tol = sqrt(dpmpar(1));
x[0] = 1.98;
x[1] = 1.02;
printf("Initial Guess: %g, %g\n", x[0], x[1]);
}
int main(int argc, char const *argv[]){
cudaSetDevice(0);
ResultType *reuslt_GPU, *reuslt_CPU;
cudaMalloc((void**)&reuslt_GPU, sizeof(ResultType));
cudaMallocHost((void**)&reuslt_CPU, sizeof(ResultType));
printf("Solving System...\n");
dim3 grid(1, 1, 1);
dim3 block(1, 1, 1);
SolvingSystem<<< grid, block>>>(reuslt_GPU);
cudaDeviceSynchronize();
printf("Done!\n");
return 0;
}
And my makefile is:
makefile
# MakeFile
# Program Name
EXE = oi
# Compilers
NVCC := nvcc
ALL_LIBRARIES := cuminpack
LIBDIR := /usr/local/lib64
# Directories
SRCDIR := src
OBJDIR := obj
INCDIR := inc /usr/local/include
#Flags
NVCCARCHFLAG := -arch sm_61
NVCCFLAGS := -std=c++11
LDFLAGS := $(addprefix -L, $(LIBDIR))
INCLUDES := $(addprefix -I, $(INCDIR))
LIBRARIES := $(addprefix -l, $(ALL_LIBRARIES))
LIBRARIES +=
ALL_CPFLAGS := -MMD
ALL_CPFLAGS += $(NVCCARCHFLAG)
ALL_CPFLAGS += $(NVCCFLAGS)
ALL_LDFLAGS := $(LDFLAGS)
# Files
C_FILES := $(wildcard $(SRCDIR)/*.c)
CU_FILES := $(wildcard $(SRCDIR)/*.cu)
C_OBJ := $(patsubst $(SRCDIR)/%.c, $(OBJDIR)/%.o, $(C_FILES))
CU_OBJ := $(patsubst $(SRCDIR)/%.cu, $(OBJDIR)/%.o, $(CU_FILES))
C_DEP := $(patsubst $(SRCDIR)/%.c, $(OBJDIR)/%.d, $(C_FILES))
CU_DEP := $(patsubst $(SRCDIR)/%.cu, $(OBJDIR)/%.d, $(CU_FILES))
SRC := $(C_FILES) $(CU_FILES)
OBJ := $(C_OBJ) $(CU_OBJ)
DEP := $(C_DEP) $(CU_DEP)
COMPILE.c := $(NVCC) -MMD -g $(INCLUDES) -c
COMPILE.cu := $(NVCC) $(ALL_CPFLAGS) $(INCLUDES) -dc
.PHONY: all clean
all:$(EXE)
$(EXE): $(OBJ)
$(NVCC) $(NVCCARCHFLAG) $(INCLUDES) $^ -o $# $(ALL_LDFLAGS) $(LIBRARIES)
$(OBJDIR)/%.o: $(SRCDIR)/%.c | $(OBJDIR)
$(COMPILE.c) $< -o $#
$(OBJDIR)/%.o: $(SRCDIR)/%.cu | $(OBJDIR)
$(COMPILE.cu) $< -o $#
$(OBJDIR):
#mkdir -p $#
# Cleaning up
clean:
#echo Cleaning up...
#rm -f -r $(OBJDIR)
#rm $(EXE)
# Including dependency
-include $(DEP)
The main file needs to be finished and that's the reason to some unused variables.
I just can't figure out what is causing this error, I think the problem is in cuminpack library, but I don't now how to fix it.
I'm using cminpack-1.3.8 from this website: http://devernay.free.fr/hacks/cminpack/
I installed the library using cmake then make and make install. Then I went to source folder and did make cuda. This last command generated libcuminpack.a file in cuda directory, then I copied this .a file to /usr/local/lib64 where cmake installed the other libraries files.
If anyone could help me I would be very thankful
EDIT
Files from the cminpack library
cminpack.h
...
#ifdef __cplusplus
extern "C" {
#endif /* __cplusplus */
...
#if defined(__CUDA_ARCH__) || defined(__CUDACC__)
#define __cminpack_attr__ __device__
#ifndef __cminpack_real__
#define __cminpack_float__
#define __cminpack_real__ float
#endif
#define __cminpack_type_fcn_nn__ __cminpack_attr__ int fcn_nn
#define __cminpack_type_fcnder_nn__ __cminpack_attr__ int fcnder_nn
#define __cminpack_type_fcn_mn__ __cminpack_attr__ int fcn_mn
#define __cminpack_type_fcnder_mn__ __cminpack_attr__ int fcnder_mn
#define __cminpack_type_fcnderstr_mn__ __cminpack_attr__ int fcnderstr_mn
#define __cminpack_decl_fcn_nn__
#define __cminpack_decl_fcnder_nn__
#define __cminpack_decl_fcn_mn__
#define __cminpack_decl_fcnder_mn__
#define __cminpack_decl_fcnderstr_mn__
#define __cminpack_param_fcn_nn__
#define __cminpack_param_fcnder_nn__
#define __cminpack_param_fcn_mn__
#define __cminpack_param_fcnder_mn__
#define __cminpack_param_fcnderstr_mn__
...
__cminpack_attr__
__cminpack_real__ CMINPACK_EXPORT __cminpack_func__(dpmpar)( int i );
dpmpar.c
#include "cminpack.h"
#include <float.h>
#include "cminpackP.h"
#define DPMPAR(type,X) _DPMPAR(type,X)
#define _DPMPAR(type,X) type ## _ ## X
__cminpack_attr__
real __cminpack_func__(dpmpar)(int i)
{
/* ********** */
/* Function dpmpar */
...

Answer 1

Based on talonmies's comment I was able to compile the code defining the __device__ function dpmpar in main.cu file.
Then I found this post https://forums.developer.nvidia.com/t/external-calls-to-device-functions/17737 where the guy answered that CUDA doesn't have a linker on the device side, so it's not possible to call a __device__ function from a different .cu file.
With this in mind, I copied only the files that I needed from the library and made some modifications, like changing __cminpack_attr__ to __device__ and I rewrote the header file too. So basically I made my own device library (thank's again to talonmies's).
The last thing I did was move all this new files to the same folder where my main.cu and my makefile lives and I rewrote the makefile.
Now the code compile fine!!