Automatically choose all available Cuda compute targets > 50

Makefile

@@ -24,13 +24,18 @@
 # Choose OpenCL device
 # Valid values: CPU, GPU, CUDA, OCLGPU
 
+OVERLAP = ON
+
 ifeq ($(DEVICE), $(filter $(DEVICE),GPU CUDA))
-TEST_CUDA := $(shell ./test_cuda.sh nvcc "$(GPU_INCLUDE_PATH)" "$(GPU_LIBRARY_PATH)")
+TARGETS_SUPPORTED := $(shell ./test_cuda.sh nvcc "$(GPU_INCLUDE_PATH)" "$(GPU_LIBRARY_PATH)" "$(TARGETS)")
 # if user specifies DEVICE=CUDA it will be used (wether the test succeeds or not)
 # if user specifies DEVICE=GPU the test result determines wether CUDA will be used or not
-ifeq ($(DEVICE)$(TEST_CUDA),GPUyes)
-override DEVICE:=CUDA
+ifeq ($(TARGETS_SUPPORTED),)
+$(error Cuda verification failed)
 endif
+override TARGETS:=$(TARGETS_SUPPORTED)
+export
+override DEVICE:=CUDA
 endif
 ifeq ($(DEVICE),CUDA)
 override DEVICE:=GPU

Makefile.Cuda

@@ -79,8 +79,8 @@ else
 		NWI=-DN16WI
 		TARGET:=$(TARGET)_16wi
 	else ifeq ($(DEVICE), GPU)
-		NWI=-DN64WI
-		TARGET:=$(TARGET)_64wi
+		NWI=-DN128WI
+		TARGET:=$(TARGET)_128wi
 	endif
 endif
 

Makefile.OpenCL

@@ -133,8 +133,8 @@ else
 		NWI=-DN16WI
 		TARGET:=$(TARGET)_16wi
 	else ifeq ($(DEVICE), GPU)
-		NWI=-DN64WI
-		TARGET:=$(TARGET)_64wi
+		NWI=-DN128WI
+		TARGET:=$(TARGET)_128wi
 	endif
 endif
 

host/src/getparameters.cpp

@@ -198,10 +198,8 @@ int parse_dpf(
 							if(mypars->ligandfile) free(mypars->ligandfile);
 							if(strincmp(argstr,"empty",5) != 0){
 								if(check_path && !has_absolute_path(argstr)){
-									len = strlen(argstr);
-									mypars->ligandfile = (char*)malloc((dpf_path.size()+len+1)*sizeof(char));
-									mypars->ligandfile[dpf_path.size()] = '\0'; // make sure first part to copy is terminated
-									strncat(strncpy(mypars->ligandfile, dpf_path.c_str(), dpf_path.size()), argstr, len);
+									mypars->ligandfile = (char*)malloc((dpf_path.size()+strlen(argstr)+1)*sizeof(char));
+									strcat(strcpy(mypars->ligandfile, dpf_path.c_str()), argstr);
 								} else mypars->ligandfile = strdup(argstr);
 							}
 						}
@@ -211,10 +209,8 @@ int parse_dpf(
 							sscanf(line.c_str(),"%*s %255s",argstr);
 							if(mypars->flexresfile) free(mypars->flexresfile);
 							if(check_path && !has_absolute_path(argstr)){
-								len = strlen(argstr);
-								mypars->flexresfile = (char*)malloc((dpf_path.size()+len+1)*sizeof(char));
-								mypars->flexresfile[dpf_path.size()] = '\0'; // make sure first part to copy is terminated
-								strncat(strncpy(mypars->flexresfile, dpf_path.c_str(), dpf_path.size()), argstr, len);
+								mypars->flexresfile = (char*)malloc((dpf_path.size()+strlen(argstr)+1)*sizeof(char));
+								strcat(strcpy(mypars->flexresfile, dpf_path.c_str()), argstr);
 							} else mypars->flexresfile = strdup(argstr);
 						}
 						break;
@@ -224,10 +220,8 @@ int parse_dpf(
 							// Add the .fld file
 							if(mypars->fldfile) free(mypars->fldfile);
 							if(check_path && !has_absolute_path(argstr)){
-								len = strlen(argstr);
-								mypars->fldfile = (char*)malloc((dpf_path.size()+len+1)*sizeof(char));
-								mypars->fldfile[dpf_path.size()] = '\0'; // make sure first part to copy is terminated
-								strncat(strncpy(mypars->fldfile, dpf_path.c_str(), dpf_path.size()), argstr, len);
+								mypars->fldfile = (char*)malloc((dpf_path.size()+strlen(argstr)+1)*sizeof(char));
+								strcat(strcpy(mypars->fldfile, dpf_path.c_str()), argstr);
 							} else mypars->fldfile = strdup(argstr); // this allows using the dpf to set up all parameters but the ligand
 							// Filling mygrid according to the specified fld file
 							if (get_gridinfo(mypars->fldfile, mygrid) != 0)

test_cuda.sh

@@ -3,12 +3,31 @@
 
 current_dir=`pwd`
 script_dir=`dirname $0`
+CUDA_VERSION=`nvcc --version | grep release | awk '{ print $(NF-1) }' | sed "s/,//g"`
+if [[ $CUDA_VERSION != "" ]]; then
+	printf "Using Cuda %s\n" $CUDA_VERSION >&2
+else
+	printf "Error: nvcc command is not working.\n" >&2
+	exit 1
+fi
+if [[ "$4" != "" ]]; then
+	for T in $4; do
+		TARGET_SUPPORTED=`nvcc --list-gpu-arch | grep $T`
+		if [[ $TARGET_SUPPORTED == "" ]]; then
+			printf "Error: Specified compute target <$T> not supported by installed Cuda version.\n" >&2
+			exit 1
+		fi
+	done
+	TARGETS="$4"
+else
+	TARGETS=`nvcc --list-gpu-arch | awk -F'_' '{if(\$2>50) print \$2}' | tr "\n" " "`
+fi
+printf "Compiling for targets: %s\n" "$TARGETS" >&2
 cd "$script_dir"
 if [[ ! -f "test_cuda" ]]; then
 	$1 -I$2 -L$3 -lcuda -lcudart -o test_cuda test_cuda.cpp &> /dev/null
-	test -e test_cuda && echo yes || echo no
+	test -e test_cuda && echo $TARGETS
 else
-	test -e test_cuda && echo yes || echo no
+	test -e test_cuda && echo $TARGETS
 fi
 cd "$current_dir"
-