Add -Werror at op BUILD and fix warnings.

mace/core/future.h

@@ -30,6 +30,17 @@ struct StatsFuture {
   };
 };
 
+inline void SetFutureDefaultWaitFn(StatsFuture *future) {
+  if (future != nullptr) {
+    future->wait_fn = [](CallStats * stats) {
+      if (stats != nullptr) {
+        stats->start_micros = NowMicros();
+        stats->end_micros = stats->start_micros;
+      }
+    };
+  }
+}
+
 }  // namespace mace
 
 #endif  // MACE_CORE_FUTURE_H_

mace/kernels/eltwise.h

@@ -652,11 +652,14 @@ struct EltwiseFunctor<DeviceType::CPU, float> : EltwiseFunctorBase {
         static_cast<uint32_t>(input0->dim_size() - input1->dim_size());
     if (data_format_ == NCHW) {
       MACE_CHECK(
-          input0->dim_size() == 4 &&
-              (input1->dim_size() == 0 ||
-               input1->dim_size() == 4 && input1->dim(1) == input0->dim(1) &&
-                   (input1->dim(0) == input0->dim(0) || input1->dim(0) == 1) ||
-               input1->dim_size() == 1 && input1->dim(0) == input0->dim(1)),
+          (input0->dim_size() == 4)
+              && ((input1->dim_size() == 0)
+                  || (input1->dim_size() == 4
+                      && input1->dim(1) == input0->dim(1)
+                      && (input1->dim(0) == input0->dim(0)
+                          || input1->dim(0) == 1))
+                  || (input1->dim_size() == 1
+                      && input1->dim(0) == input0->dim(1))),
           "only support broadcast channel dimension");
     } else {
       if (rank_diff > 0 && rank_diff < input0->dim_size()) {

mace/kernels/stack.h

@@ -34,8 +34,6 @@ struct StackFunctor {
   MaceStatus operator()(const std::vector<const Tensor *> &inputs,
                         Tensor *output,
                         StatsFuture *future) {
-    MACE_UNUSED(future);
-
     MACE_CHECK(!inputs.empty(), "stack inputs are empty.");
     std::vector<index_t> input_shape = inputs[0]->shape();
     MACE_CHECK(axis_ >= -(inputs[0]->dim_size() + 1) &&
@@ -51,7 +49,7 @@ struct StackFunctor {
     // On host, no need to map data
     T *output_data = output->mutable_data<T>();
     std::vector<const T *> input_data(inputs.size());
-    for (int i = 0; i < inputs.size(); ++i) {
+    for (size_t i = 0; i < inputs.size(); ++i) {
       input_data[i] = inputs[i]->data<T>();
     }
 
@@ -62,13 +60,14 @@ struct StackFunctor {
         std::accumulate(input_shape.begin() + axis_, input_shape.end(), 1,
                         std::multiplies<index_t>());
     for (index_t h = 0; h < high_dim_elem_size; ++h) {
-      for (index_t i = 0; i < inputs.size(); ++i) {
+      for (size_t i = 0; i < inputs.size(); ++i) {
         memcpy(output_data, input_data[i] + h * low_dim_elem_size,
                sizeof(T) * low_dim_elem_size);
         output_data += low_dim_elem_size;
       }
     }
 
+    SetFutureDefaultWaitFn(future);
     return MACE_SUCCESS;
   }
 

mace/kernels/strided_slice.h

@@ -45,7 +45,6 @@ struct StridedSliceFunctor {
                         const Tensor *strides,
                         Tensor *output,
                         StatsFuture *future) {
-    MACE_UNUSED(future);
     MACE_CHECK(ellipsis_mask_ == 0 && new_axis_mask_ == 0,
                "ellipsis_mask and new_axis_mask are not supported yet.");
 
@@ -144,6 +143,7 @@ struct StridedSliceFunctor {
       }
     }
 
+    SetFutureDefaultWaitFn(future);
     return MACE_SUCCESS;
   }
 

mace/kernels/winograd_transform.h

@@ -95,9 +95,9 @@ struct WinogradInverseTransformFunctorBase {
       : batch_(batch),
         height_(height),
         width_(width),
+        wino_blk_size_(block_size),
         activation_(activation),
-        relux_max_limit_(relux_max_limit),
-        wino_blk_size_(block_size) {}
+        relux_max_limit_(relux_max_limit) {}
 
   const int batch_;
   const int height_;

mace/ops/BUILD

@@ -41,7 +41,10 @@ cc_library(
         ["*.h"],
         exclude = ["ops_test_util.h"],
     ),
-    copts = if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
+    copts = [
+        "-Werror",
+        "-Wextra",
+    ] + if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
         "-DMACE_ENABLE_NEON",
     ]) + if_android_armv7([
         "-mfpu=neon",
@@ -64,7 +67,10 @@ cc_test(
     srcs = glob(
         ["*_test.cc"],
     ),
-    copts = if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
+    copts = [
+        "-Werror",
+        "-Wextra",
+    ] + if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
         "-DMACE_ENABLE_NEON",
     ]) + if_android_armv7([
         "-mfpu=neon",
@@ -88,7 +94,10 @@ cc_test(
     name = "ops_benchmark",
     testonly = 1,
     srcs = glob(["*_benchmark.cc"]),
-    copts = if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
+    copts = [
+        "-Werror",
+        "-Wextra",
+    ] + if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
         "-DMACE_ENABLE_NEON",
     ]) + if_android_armv7([
         "-mfpu=neon",

mace/ops/identity.h

@@ -32,6 +32,7 @@ class IdentityOp : public Operator<D, T> {
     const Tensor *input = this->Input(INPUT);
     Tensor *output = this->Output(OUTPUT);
     output->ReuseTensorBuffer(*input);
+    SetFutureDefaultWaitFn(future);
     return MACE_SUCCESS;
   }
 

mace/ops/shape.h

@@ -42,6 +42,7 @@ class ShapeOp : public Operator<D, T> {
     for (index_t i = 0; i < input->dim_size(); ++i) {
       output_data[i] = input->dim(i);
     }
+    SetFutureDefaultWaitFn(future);
 
     return MACE_SUCCESS;
   }

mace/ops/squeeze.h

@@ -31,8 +31,6 @@ class SqueezeOp : public Operator<D, T> {
         axis_(OperatorBase::GetRepeatedArgs<int>("axis", {})) {}
 
   MaceStatus Run(StatsFuture *future) override {
-    MACE_UNUSED(future);
-
     const Tensor *input = this->Input(INPUT);
     Tensor *output = this->Output(OUTPUT);
 
@@ -47,6 +45,7 @@ class SqueezeOp : public Operator<D, T> {
     output->ReuseTensorBuffer(*input);
     output->Reshape(output_shape);
 
+    SetFutureDefaultWaitFn(future);
     return MACE_SUCCESS;
   }
 

mace/ops/stack.h

@@ -33,7 +33,6 @@ class StackOp : public Operator<D, T> {
   MaceStatus Run(StatsFuture *future) override {
     const std::vector<const Tensor *> &inputs = this->Inputs();
     Tensor *output = this->Output(OUTPUT);
-
     return functor_(inputs, output, future);
   }
 

mace/ops/stack_test.cc

@@ -29,7 +29,7 @@ void TestStack(const std::vector<index_t> &input_shape,
                const std::vector<index_t> &output_shape,
                const std::vector<float> &output) {
   OpsTestNet net;
-  for (int i = 0; i < inputs.size(); ++i) {
+  for (size_t i = 0; i < inputs.size(); ++i) {
     net.AddInputFromArray<CPU, float>(MakeString("Input", i), input_shape,
                                       inputs[i]);
   }
@@ -38,7 +38,7 @@ void TestStack(const std::vector<index_t> &input_shape,
                         .Output("Output")
                         .AddIntArg("axis", axis);
 
-  for (int i = 0; i < inputs.size(); ++i) {
+  for (size_t i = 0; i < inputs.size(); ++i) {
     op_builder.Input(MakeString("Input", i));
   }
   op_builder.Finalize(net.NewOperatorDef());

mace/tools/validation/BUILD

@@ -4,7 +4,10 @@ load("//mace:mace.bzl", "if_openmp_enabled", "if_android")
 cc_binary(
     name = "mace_run_static",
     srcs = ["mace_run.cc"],
-    copts = if_android([
+    copts = [
+        "-Werror",
+        "-Wextra",
+    ] + if_android([
         "-DMACE_ENABLE_OPENCL",
     ]),
     linkopts = if_openmp_enabled(["-fopenmp"]),
@@ -19,7 +22,10 @@ cc_binary(
 cc_binary(
     name = "mace_run_shared",
     srcs = ["mace_run.cc"],
-    copts = if_android([
+    copts = [
+        "-Werror",
+        "-Wextra",
+    ] + if_android([
         "-DMACE_ENABLE_OPENCL",
     ]),
     linkopts = ["-lm", "-pie", "-fPIE"] + if_openmp_enabled(["-fopenmp"]),

tools/converter.py

@@ -1045,7 +1045,7 @@ def run_specific_target(flags, configs, target_abi,
 
 
 def run_mace(flags):
-    configs = format_model_config(flags.config)
+    configs = format_model_config(flags)
 
     target_socs = configs[YAMLKeyword.target_socs]
     if not target_socs or ALL_SOC_TAG in target_socs:
@@ -1159,7 +1159,7 @@ def bm_specific_target(flags, configs, target_abi, target_soc, serial_num):
 
 
 def benchmark_model(flags):
-    configs = format_model_config(flags.config)
+    configs = format_model_config(flags)
 
     target_socs = configs[YAMLKeyword.target_socs]
     if not target_socs or ALL_SOC_TAG in target_socs:
@@ -1211,6 +1211,16 @@ def parse_args():
         default="",
         required=True,
         help="model yaml configuration file path")
+    all_type_parent_parser.add_argument(
+        "--build_type",
+        type=str,
+        default="",
+        help="Model build type, can be ['proto', 'code'].")
+    all_type_parent_parser.add_argument(
+        "--target_abis",
+        type=str,
+        default="",
+        help="Target ABIs, comma seperated list.")
     build_run_parent_parser = argparse.ArgumentParser(add_help=False)
     build_run_parent_parser.add_argument(
         '--address_sanitizer',
@@ -1253,16 +1263,6 @@ def parse_args():
         "--enable_openmp",
         action="store_false",
         help="Enable openmp for multiple thread.")
-    build.add_argument(
-        "--build_type",
-        type=str,
-        default="",
-        help="Model build type, can be ['proto', 'code'].")
-    build.add_argument(
-        "--target_abis",
-        type=str,
-        default="",
-        help="Target ABIs, comma seperated list.")
     run = subparsers.add_parser(
         'run',
         parents=[all_type_parent_parser, run_bm_parent_parser,