Do not attempt to build the program that becomes the kernel twice. This could have been leading to failures on initialising cl.

ocl.c

@@ -185,7 +185,7 @@ void patch_opcodes(char *w, unsigned remaining)
 
 _clState *initCl(unsigned int gpu, char *name, size_t nameSize)
 {
-	bool patchbfi = false;
+	bool patchbfi = false, prog_built = false;
 	cl_int status = 0;
 	unsigned int i;
 
@@ -526,6 +526,8 @@ build:
 		return NULL;
 	}
 
+	prog_built = true;
+
 	status = clGetProgramInfo( clState->program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t)*numDevices, binary_sizes, NULL );
 	if (unlikely(status != CL_SUCCESS)) {
 		applog(LOG_ERR, "Error: Getting program info CL_PROGRAM_BINARY_SIZES. (clGetPlatformInfo)");
@@ -624,6 +626,7 @@ built:
 	applog(LOG_INFO, "Initialising kernel %s with%s BFI_INT, %d vectors and worksize %d",
 	       filename, patchbfi ? "" : "out", clState->preferred_vwidth, clState->work_size);
 
+	if (!prog_built) {
 		/* create a cl program executable for all the devices specified */
 		status = clBuildProgram(clState->program, 1, &devices[gpu], NULL, NULL, NULL);
 		if (status != CL_SUCCESS) {
@@ -636,6 +639,7 @@ built:
 			applog(LOG_INFO, "%s", log);
 			return NULL;
 		}
+	}
 
 	/* get a kernel object handle for a kernel with the given name */
 	clState->kernel = clCreateKernel(clState->program, "search", &status);