From 7ef0d334cad13a505baf5d25ec2a8cc4fabe793b Mon Sep 17 00:00:00 2001
From: Steven Robertson <steven@strobe.cc>
Date: Sun, 12 Sep 2010 14:06:07 -0400
Subject: [PATCH] ...except I missed the file that actually contained the new
 method

---
 cuburn/cuda.py | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/cuburn/cuda.py b/cuburn/cuda.py
index ecfc088..f7c9fca 100644
--- a/cuburn/cuda.py
+++ b/cuburn/cuda.py
@@ -110,3 +110,17 @@ class LaunchContext(object):
                 all_okay = False
         return all_okay
 
+    def get_per_thread(self, name, dtype, shaped=False):
+        """
+        Convenience function to get the contents of the global memory variable
+        ``name`` from the device as a numpy array of type ``dtype``, as might
+        be stored by _PTXStdLib.store_per_thread. If ``shaped`` is True, the
+        array will be 3D, as (cta_no, warp_no, lane_no).
+        """
+        if shaped:
+            shape = (self.nctas, self.warps_per_cta, 32)
+        else:
+            shape = self.nthreads
+        dp, l = self.mod.get_global(name)
+        return cuda.from_device(dp, shape, dtype)
+