From 6f59791d099167c947afe777e8aec2419ef3eda1 Mon Sep 17 00:00:00 2001
From: yejianwu <yejianwu@xiaomi.com>
Date: Wed, 18 Jul 2018 16:33:03 +0800
Subject: [PATCH] fix lws overflow

---
 mace/kernels/opencl/depthwise_conv.cc | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mace/kernels/opencl/depthwise_conv.cc b/mace/kernels/opencl/depthwise_conv.cc
index ec358d09..ca44be2f 100644
--- a/mace/kernels/opencl/depthwise_conv.cc
+++ b/mace/kernels/opencl/depthwise_conv.cc
@@ -38,7 +38,7 @@ std::vector<uint32_t> LocalWS(const uint32_t *gws, const uint32_t kwg_size) {
                                   kwg_size / lws[1]);
     }
   }
-  lws[0] = std::max<uint32_t>(lws[0], 1);
+  lws[0] = std::max<uint32_t>(std::min<uint32_t>(lws[0], kwg_size / lws[1]), 1);
   const uint32_t lws_size = lws[0] * lws[1];
   lws[2] = std::min<uint32_t>((cache_size / kernel_cache_size / lws_size) * 4,
                               gws[2]);
-- 
GitLab