Added vector addition.

.gitignore

@@ -0,0 +1 @@
+/.vscode

README.md

@@ -1,2 +1,3 @@
-# cuda-testbed
+# CUDA Testbed
+
 Simple CUDA acceleration tests and samples.

samples/vector_add.py

@@ -0,0 +1,37 @@
+from argparse import ArgumentParser
+import numpy as np
+import numba as nb
+import time
+
+
+@nb.vectorize(['float32(float32, float32)'], target='cuda')
+def Add(A, B):
+    """Add two vectors A and B of size N and return the result."""
+    return A+B
+
+
+def main(args):
+    N = args.N
+    A = np.ones( N,  dtype = np.float32)
+    B = np.ones( N,  dtype = np.float32)
+    C = np.empty(N,  dtype = np.float32)
+
+    # Interestingly, the function is actually compiled the first time it is used.
+    # Calling Add() early removes the compilation overhead from the actual Add() operation.
+
+    if args.e: 
+        Add(np.float32(0), np.float32(0))
+
+    start = time.time()
+    C = Add(A, B)
+    end = time.time()
+    print(C[:5], C[-5:], 'Completed in {:.4f} s'.format(end-start))
+
+
+if __name__ == '__main__':
+    parser = ArgumentParser('Addition of two vectors in CUDA.')
+    parser.add_argument('--N', type=int, default=32000000, help='Size of vector')
+    parser.add_argument('--e', dest='e', default=False, action='store_true', help='Turn on early compilation')
+    args = parser.parse_args()
+
+    main(args)