Changed build architecture from 2.0 to 3.0.

Also switched from 'sm_11_atomic_functions.h' to 'sm_20_atomic_functions.h' as the former is no longer included in version 9.1 of the CUDA Toolkit.

Makefile

@@ -27,7 +27,7 @@ WORDS = 24
 
 CXXFLAGS=-g -Wall -O2
 CPPFLAGS=-DWORDS=$(WORDS) -I$(LIBEVENT)/include -I$(GMP)/include -UVERBOSE
-NVCCOPTS=-g -arch sm_20 --ptxas-options=-v -O2
+NVCCOPTS=-g -arch sm_30 --ptxas-options=-v -O2
 NVCC=nvcc $(NVCCOPTS)
 
 CXXFILES = dlrho.cc gen_N.cc

dlrho.cc

@@ -221,7 +221,7 @@ static PDLHandle* p_dl_fork_start(const ZZ &p, const ZZ_p &target,
 
 	cudaError_t cudares = cudaSetDevice(deviceid);
 	if (cudares != cudaSuccess) {
-	    cerr << "Error setting CUDA device\n";
+	    cerr << "Error setting CUDA device: " << cudaGetErrorString(cudares) << "\n";
 	    exit(1);
 	}
 	int res = p_dl(target_p, base_p, exp, fvec, label);

dpstream.cu

@@ -20,7 +20,7 @@
 #ifndef __DPSTREAM_H__
 #define __DPSTREAM_H__
 
-#include <sm_11_atomic_functions.h>
+#include <sm_20_atomic_functions.h>
 #include <unistd.h>
 
 /* Routines to allow CUDA threads to write distinguished points into