From 39edcf61e162d7f3a7449e05bfedccd1301fe34e Mon Sep 17 00:00:00 2001
From: "Safonov, Igor" <igor.safonov@intel.com>
Date: Thu, 7 Apr 2022 10:21:18 -0700
Subject: [PATCH] src: cpu: reorder: simple: fix reverse order between memory
 with different block size

---
 src/cpu/reorder/simple_reorder.hpp | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/src/cpu/reorder/simple_reorder.hpp b/src/cpu/reorder/simple_reorder.hpp
index e6034a0728c..8c5454ed758 100644
--- a/src/cpu/reorder/simple_reorder.hpp
+++ b/src/cpu/reorder/simple_reorder.hpp
@@ -1355,10 +1355,11 @@ struct simple_reorder_impl<SIMPLE_REORDER_TEMPL_CALL,
                     for (int c = 0; c < block_i; ++c) {
                         o[o_off + c] = _qz_a1b0<type_i, type_o>()(i[i_off + c]);
                     }
-                    if (order_keep && b + 1 == nb) {
+                    if (b + 1 == nb) {
                         // zero padding
-                        const auto pad_size
-                                = blksize_16 - ((nb - 1) * blksize_i);
+                        const auto pad_size = order_keep
+                                ? blksize_16 - ((nb - 1) * blksize_i)
+                                : blksize_i;
                         const auto pad_start = block_i + o_off;
                         const auto pad_end = pad_size + o_off;
                         PRAGMA_OMP_SIMD()
@@ -1379,10 +1380,11 @@ struct simple_reorder_impl<SIMPLE_REORDER_TEMPL_CALL,
                         o[o_off + c] = _qz<type_i, type_o>()(
                                 i[i_off + c], o[o_off + c], alpha, beta);
                     }
-                    if (order_keep && b + 1 == nb) {
+                    if (b + 1 == nb) {
                         // zero padding
-                        const auto pad_size
-                                = blksize_16 - ((nb - 1) * blksize_i);
+                        const auto pad_size = order_keep
+                                ? blksize_16 - ((nb - 1) * blksize_i)
+                                : blksize_i;
                         const auto pad_start = block_i + o_off;
                         const auto pad_end = pad_size + o_off;
                         PRAGMA_OMP_SIMD()