From 73d8ca4eb63f0cac370c67c68de352e60de9c990 Mon Sep 17 00:00:00 2001
From: Tim Holy <tim.holy@gmail.com>
Date: Thu, 2 Aug 2012 06:12:22 -0500
Subject: [PATCH] Add performance test for nested looping

This is extremely curious: in a nest of 3 loops, I get a huge
performance gain (10x) by moving the inner loop into a separate
function. If I had to speculate wildly, I suspect this has something
to do with the compiler worrying that the outer loop variables might
be changed by the inner loop? And that putting it inside a function
removes that concern?
---
 test/perf2/looping.jl | 40 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)
 create mode 100644 test/perf2/looping.jl

diff --git a/test/perf2/looping.jl b/test/perf2/looping.jl
new file mode 100644
index 0000000000000..cab42a70c9e86
--- /dev/null
+++ b/test/perf2/looping.jl
@@ -0,0 +1,40 @@
+function testlooping1(p, sz)
+    index = 1        
+    for k = 1:sz[3]
+        for j = 1:sz[2]
+            jkprod = j*k
+            for i = 1:sz[1]
+                p[index] = i*jkprod
+                index += 1
+            end
+        end
+    end
+end
+
+function oneloop(p, index::Int, jkprod::Int, imax::Int)
+    for i = 1:imax
+        p[index] = i*jkprod
+        index += 1
+    end
+    return index
+end
+
+function testlooping2(p, sz)
+    index = 1        
+    for k = 1:sz[3]
+        for j = 1:sz[2]
+            index = oneloop(p, index, j*k, sz[1])
+        end
+    end
+end
+
+sz = (1000,1000,10)
+p = Array(Int, sz)
+szsmall = (10,10,10)
+psmall = Array(Int, szsmall)
+testlooping1(psmall, szsmall)
+@time testlooping1(p, sz)
+p1 = copy(p)
+testlooping2(psmall, szsmall)
+@time testlooping2(p, sz)
+@assert p == p1