update performance data in README

README.md

@@ -28,19 +28,19 @@ Paddle-Moible是PaddlePaddle组织下的项目，是一个致力于嵌入式平
 
 |mobilenet arm v7|1线程|2线程|4线程|
 |------------|----|-----|-----|
-|麒麟960(ms)|110.586|70.897|47.474|
+|麒麟960(ms)|110.586|63.285|38.215|
 |||||
 |mobilenetssd arm v7|1线程|2线程|4线程|
-|麒麟960(ms)|222.124|138.952|90.856|
+|麒麟960(ms)|220.248|128.473|79.334|
 |||||
 |googlenet(v1) arm v7|1线程|2线程|4线程|
-|麒麟960(ms)|348.018|240.304|169.998|
+|麒麟960(ms)|341.965|228.724|161.531|
 |||||
 |squeezenet arm v7|1线程|2线程|4线程|
-|麒麟960(ms)|84.685|56.544|38.833|
+|麒麟960(ms)|84.080|55.641|37.182|
 |||||
 |yolo arm v7|1线程|2线程|4线程|
-|麒麟960(ms)|131.831|88.990|60.905|
+|麒麟960(ms)|129.445|80.627|50.936|
 
     arm cpu是paddle-mobile的主要支持方向，cpu的通用性一直是其优势。嵌入式深度学习，需要大量的cpu汇编实现。我们正在紧锣密鼓的编码，为的是能充分硬件的每一点加速能力。
     arm cpu的优化工作还在进行中，现在使用了常规的cpu优化。在arm a73上paddle-mobile arm-v7现在单核运行一次mobilenet1.0是110+ms，显然这不是我们的最终目标，我们正在用大量的汇编改写，后续性能仍会有巨大提升空间, 目前只支持armv7, 未来我们也会支持armv8。

src/io/executor.cpp

@@ -192,8 +192,14 @@ void Executor<Dtype, P>::LoadMemory(const framework::VarDesc var_desc,
     }
     *data += (memory_size * sizeof(uint8_t));
   } else {
-    for (int n = 0; n < memory_size * type_size; ++n) {
-      static_cast<char *>(memory)[n] = (*data)[n];
+    for (int n = 0; n < memory_size; n++) {
+      float value;
+      memcpy(&value, *data + n * type_size, type_size);
+      if (value < 1e-30 && value > -1e-30) {
+        static_cast<float *>(memory)[n] = 0.0;
+      } else {
+        static_cast<float *>(memory)[n] = value;
+      }
     }
     (*data) += (sizeof(char) * memory_size * type_size);
   }

test/net/test_squeezenet.cpp

@@ -18,7 +18,7 @@ limitations under the License. */
 
 int main() {
   paddle_mobile::PaddleMobile<paddle_mobile::CPU> paddle_mobile;
-  paddle_mobile.SetThreadNum(2);
+  paddle_mobile.SetThreadNum(4);
   //  ../../../test/models/googlenet
   //  ../../../test/models/mobilenet
   auto time1 = time();

test/net/test_yolo.cpp

@@ -18,7 +18,7 @@ limitations under the License. */
 
 int main() {
   paddle_mobile::PaddleMobile<paddle_mobile::CPU> paddle_mobile;
-  paddle_mobile.SetThreadNum(2);
+  paddle_mobile.SetThreadNum(4);
   //  ../../../test/models/googlenet
   //  ../../../test/models/mobilenet
   auto time1 = time();

test/operators/test_fusion_conv_add_bn_relu_op.cpp

@@ -12,6 +12,7 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
+#include <iostream>
 #include "../test_include.h"
 #include "operators/fusion_conv_add_bn_relu_op.h"