From 2c2dbf2c3f7c69d059435dd638f90d3220748cc2 Mon Sep 17 00:00:00 2001
From: Tom Deakin <tom.deakin@bristol.ac.uk>
Date: Thu, 16 Jul 2015 18:06:07 +0100
Subject: [PATCH] You have to specify block size in CUDA, so make sure
 ARRAY_SIZE divides 1024

---
 cuda-stream.cu | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/cuda-stream.cu b/cuda-stream.cu
index cb2edfb..bf3d5af 100644
--- a/cuda-stream.cu
+++ b/cuda-stream.cu
@@ -131,6 +131,16 @@ int main(int argc, char *argv[])
         << "Version: " << VERSION_STRING << std::endl
         << "Implementation: CUDA" << std::endl << std::endl;
 
+    if (ARRAY_SIZE % 1024 != 0)
+    {
+        unsigned int OLD_ARRAY_SIZE = ARRAY_SIZE;
+        ARRAY_SIZE -= ARRAY_SIZE % 1024;
+        std::cout
+            << "Warning: array size must divide 1024" << std::endl
+            << "Resizing array from " << OLD_ARRAY_SIZE
+            << " to " << ARRAY_SIZE << std::endl;
+    }
+
     try
     {
         parseArguments(argc, argv);