From 36b5a00bcbaea66b9d369b70ebd43b98b5bed226 Mon Sep 17 00:00:00 2001
From: M Clark <mclark@nvidia.com>
Date: Tue, 14 Oct 2014 10:45:45 -0700
Subject: [PATCH] Added support for sm_52 target and flush denormals to zero
 for small performance boost.

---
 src/Makefile | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/src/Makefile b/src/Makefile
index f0d6d68..49d47d2 100644
--- a/src/Makefile
+++ b/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