From 3021f0fb0cb44a9f4499644701590cbb73185b37 Mon Sep 17 00:00:00 2001
From: hightower8083 <igor.andriyash@gmail.com>
Date: Wed, 18 Jan 2017 17:00:05 +0100
Subject: [PATCH] The fix in SCAN_INTERVALS_SOURCE source is introduced, which
fixes a compiler crash on AMD Radeon crad on Apple platform. For this
structed data pointers, ldata[:][:].value, are replaced by temporal variables
tmp and tmp_aux before passing to SCAN_EXPR. To avoid leftover-worning
tmp_aux was added to_PREFIX_WORDS.
---
pyopencl/scan.py | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/pyopencl/scan.py b/pyopencl/scan.py
index 13aa7820..1f8d5d2c 100644
--- a/pyopencl/scan.py
+++ b/pyopencl/scan.py
@@ -295,8 +295,11 @@ void ${kernel_name}(
if (LID_0 == 0 && unit_base != interval_begin)
{
+ scan_type tmp = ldata[K][WG_SIZE - 1].value;
+ scan_type tmp_aux = ldata[0][0].value;
+
ldata[0][0].value = SCAN_EXPR(
- ldata[K][WG_SIZE - 1].value, ldata[0][0].value,
+ tmp, tmp_aux,
%if is_segmented:
(l_segment_start_flags[0][0])
%else:
@@ -760,7 +763,7 @@ _PREFIX_WORDS = set("""
group_base seg_end my_val DEBUG ARGS
ints_to_store ints_per_wg scan_types_per_int linear_index
linear_scan_data_idx dest src store_base wrapped_scan_type
- dummy scan_tmp
+ dummy scan_tmp tmp_aux
LID_2 LID_1 LID_0
LDIM_0 LDIM_1 LDIM_2
--
GitLab