Bitonic: shrink amount of used local memory in accordance with @inferrna

's comments in #80
inducer · Jul 16, 2015 · 9d6dd85 · 9d6dd85
1 parent 2b98b07
commit 9d6dd85
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/pyopencl/bitonic_sort.py b/pyopencl/bitonic_sort.py
@@ -121,8 +121,8 @@ def __call__(self, arr, idx=None, queue=None, wait_for=None, axis=0):
             if aux:
                 last_evt = knl(
                         queue, (nt,), wg, arr.data, idx.data,
-                        cl.LocalMemory(wg[0]*4*arr.dtype.itemsize),
-                        cl.LocalMemory(wg[0]*4*idx.dtype.itemsize),
+                        cl.LocalMemory(wg[0]*arr.dtype.itemsize),
+                        cl.LocalMemory(wg[0]*idx.dtype.itemsize),
                         wait_for=[last_evt])
             for knl, nt, wg, _ in run_queue[1:]:
                 last_evt = knl(
@@ -184,9 +184,9 @@ def sort_b_prepare_wl(self, argsort, key_dtype, idx_dtype, shape, axis):
 
         available_lmem = dev.local_mem_size
         while True:
-            lmem_size = wg*4*key_dtype.itemsize
+            lmem_size = wg*key_dtype.itemsize
             if argsort:
-                lmem_size += wg*4*idx_dtype.itemsize
+                lmem_size += wg*idx_dtype.itemsize
 
             if lmem_size + 512 > available_lmem:
                 wg //= 2