From 4b3b52039d57ff2c5c5529fc4a82ae5b2797df35 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Mon, 2 Sep 2013 15:47:23 -0500
Subject: [PATCH] Fix reduction when no_sync_size <= group_size (spotted by
Alex Nitz)
---
pyopencl/reduction.py | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/pyopencl/reduction.py b/pyopencl/reduction.py
index f342909f..489be110 100644
--- a/pyopencl/reduction.py
+++ b/pyopencl/reduction.py
@@ -110,7 +110,15 @@ KERNEL = """//CL//
barrier(CLK_LOCAL_MEM_FENCE);
- if (lid < ${no_sync_size})
+ <%
+ # NB: There's an exact duplicate of this calculation in the
+ # %while loop below.
+
+ new_size = cur_size // 2
+ assert new_size * 2 == cur_size
+ %>
+
+ if (lid < ${new_size})
{
__local volatile out_type *lvdata = ldata;
% while cur_size > 1:
--
GitLab