Fix: add virutal method sync in Runtime

include/bang/bang_runtime.h

@@ -36,7 +36,7 @@ class BangRuntimeObj : public RuntimeObj {
              bool profiling = false) const;
     // double runEvaluation(const Graph &graph, int nWarmups,
     //                      int nEvaluations) const;
-    void sync() const;
+    void sync() const override;
     BangPtr alloc(size_t size) override {
         void *ptr;
         checkBangError(cnrtMalloc(&ptr, size));

include/core/runtime.h

@@ -79,6 +79,7 @@ class RuntimeObj : public std::enable_shared_from_this<RuntimeObj> {
     virtual void copyBlobToCPU(void *dst, const void *src,
                                size_t bytes) const = 0;
     virtual string toString() const = 0;
+    virtual void sync() const {}
 
     map<UidBaseType, bool>
     getCompileTimeComputableAttribute(const Graph &graph) const;

include/cuda/cuda_runtime.h

@@ -27,7 +27,7 @@ class CudaRuntimeObj : public RuntimeObj {
              bool profiling = false) const;
     // double runEvaluation(const Graph &graph, int nWarmups,
     //                      int nEvaluations) const;
-    void sync() const;
+    void sync() const override;
     CudaPtr alloc(size_t size) override {
         void *ptr;
         // printf("Try to cudaMalloc: %lu bytes\n", size);

include/intelcpu/mkl_runtime.h

@@ -29,7 +29,7 @@ class MklRuntimeObj : public CpuRuntimeObj {
     string toString() const override { return "INTELCPU Runtime"; };
     dnnl::engine getEngine() const { return dnnl::engine(engine, true); }
     dnnl::stream getStream() const { return dnnl::stream(stream, true); }
-    void sync() const;
+    void sync() const override;
 };
 
 } // namespace infini

src/core/runtime.cc

@@ -241,7 +241,7 @@ double RuntimeObj::timeNonCtcOperators(const Graph &graph, int warmup,
                     kernel->compute(op, this);
             }
         },
-        [&]() { sync(); }, warmup, repeat);
+        [&]() { this->sync(); }, warmup, repeat);
     return ret;
 }