Add Mace version to identify OpenCL cache for not use old cache.

docs/development/memory_layout.rst

@@ -50,11 +50,11 @@ The Input/Output Tensor is stored in NHWC format:
     * - Height-Major Input/Output
       - NHWC
       - [W * C, N * (H+3)/4]
-      - Winograd Convolution format
+      - WinogradTransform and MatMul output format
     * - Width-Major Input/Output
       - NHWC
       - [(W+3)/4 * C, N * H]
-      - Winograd Convolution format
+      - Unused now
 
 Each Pixel of **Image** contains 4 elements. The below table list the
 coordination relation between **Image** and **Buffer**.

mace/core/runtime/opencl/opencl_runtime.cc

@@ -321,7 +321,8 @@ OpenCLRuntime::OpenCLRuntime():
   std::stringstream ss;
   ss << default_platform.getInfo<CL_PLATFORM_NAME>()
      << ", " << default_platform.getInfo<CL_PLATFORM_PROFILE>() << ", "
-     << default_platform.getInfo<CL_PLATFORM_VERSION>();
+     << default_platform.getInfo<CL_PLATFORM_VERSION>() << ", "
+     << MaceVersion();
   platform_info_ = ss.str();
   VLOG(1) << "Using platform: " << platform_info_;
 

mace/python/tools/model_header.jinja2

@@ -27,8 +27,6 @@ namespace {{tag}} {
 
 const unsigned char *LoadModelData(const std::string &model_data_file);
 
-void UnloadModelData(const unsigned char *model_data);
-
 const std::shared_ptr<NetDef> CreateNet();
 
 const std::string ModelName();