Added support for sm_52 target and flush denormals to zero for small …

…performance boost.

src/Makefile

@@ -26,7 +26,7 @@ RPATH =
 LIBXGPU = libxgpu.dll
 else
 
-NVCCFLAGS = -Xcompiler -D_REENTRANT -Xcompiler -Wall -Xcompiler -Wno-unused-function
+NVCCFLAGS = -Xcompiler -D_REENTRANT -Xcompiler -Wall -Xcompiler -Wno-unused-function -ftz=true
 
 ifneq ($(strip $(OSTYPE)),osx)
 NVCCFLAGS += -Xcompiler -fopenmp 
@@ -55,6 +55,10 @@ ifeq ($(strip $(CUDA_ARCH)),sm_50)
 NVCCFLAGS += -maxrregcount=72
 endif
 
+ifeq ($(strip $(CUDA_ARCH)),sm_52)
+NVCCFLAGS += -maxrregcount=72
+endif
+
 #NVCCFLAGS += -Xptxas -dlcm=cg # disable L1 cache
 
 # Possibly (re-)build xgpu_version.h