[Framework]Update Malloc implementation to reduce memory usage of Tensor #3959

lite/backends/arm/math/prior_box.cc

@@ -21,7 +21,7 @@ namespace lite {
 namespace arm {
 namespace math {
 
-const int MALLOC_ALIGN = 64;
+const int MALLOC_ALIGN = 16;
 
 void* fast_malloc(size_t size) {
   size_t offset = sizeof(void*) + MALLOC_ALIGN - 1;

lite/backends/host/target_wrapper.cc

@@ -19,7 +19,7 @@
 namespace paddle {
 namespace lite {
 
-const int MALLOC_ALIGN = 64;
+const int MALLOC_ALIGN = 16;
 
 void* TargetWrapper<TARGET(kHost)>::Malloc(size_t size) {
   size_t offset = sizeof(void*) + MALLOC_ALIGN - 1;

lite/tests/kernels/prior_box_compute_test.cc

@@ -21,7 +21,7 @@
 namespace paddle {
 namespace lite {
 
-const int MALLOC_ALIGN = 64;
+const int MALLOC_ALIGN = 16;
 
 void* fast_malloc(size_t size) {
   size_t offset = sizeof(void*) + MALLOC_ALIGN - 1;