Merge remote-tracking branch 'origin/develop' into rework_ector_type

host/driver_offline/src/gemm_driver_offline.cpp

@@ -239,14 +239,10 @@ int main(int argc, char* argv[])
     using ab_data_t  = float;
     using acc_data_t = float;
     using c_data_t   = float;
-#elif 0
+#elif 1
     using ab_data_t  = half_t;
     using acc_data_t = float;
     using c_data_t   = half_t;
-#elif 1
-    using ab_data_t  = ushort;
-    using acc_data_t = float;
-    using c_data_t   = ushort;
 #elif 1
     using ab_data_t  = int8_t;
     using acc_data_t = int32_t;

host/host_tensor/include/conv_common.hpp

@@ -74,4 +74,17 @@ calculate_convolution_flops(const InDesc&, const WeiDesc& wei_desc, const OutDes
     return std::size_t(2) * N * K * Ho * Wo * C * Y * X;
 }
 
+template <typename T>
+inline auto activ(T v, const ck::ActivTypeEnum_t activ_type)
+{
+    const T alpha = 0.3;
+    switch(activ_type)
+    {
+    case ck::ActivTypeEnum_t::None: return v;
+    case ck::ActivTypeEnum_t::LeakyRelu: return (v >= 0 ? v : alpha * v);
+    case ck::ActivTypeEnum_t::Sigmoid: return (1 / (1 + exp(-v)));
+    default: throw std::runtime_error("unsupported activ type"); break;
+    }
+}
+
 #endif

host/host_tensor/include/host_tensor.hpp

@@ -257,6 +257,18 @@ struct Tensor
                                        mDesc.GetLengths()[3])(num_thread);
             break;
         }
+        case 5: {
+            auto f = [&](auto i0, auto i1, auto i2, auto i3, auto i4) {
+                (*this)(i0, i1, i2, i3, i4) = g(i0, i1, i2, i3, i4);
+            };
+            make_ParallelTensorFunctor(f,
+                                       mDesc.GetLengths()[0],
+                                       mDesc.GetLengths()[1],
+                                       mDesc.GetLengths()[2],
+                                       mDesc.GetLengths()[3],
+                                       mDesc.GetLengths()[4])(num_thread);
+            break;
+        }
         default: throw std::runtime_error("unspported dimension");
         }
     }