remove host_conv_add

host/host_tensor/include/host_conv.hpp

@@ -101,115 +101,6 @@ void host_direct_convolution(const Tensor<TIn>& in,
     }
 }
 
-template <typename TIn,
-          typename TWei,
-          typename TOut,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-void host_direct_convolution_add(const Tensor<TIn>& in,
-                                 const Tensor<TWei>& wei,
-                                 const Tensor<TOut>& add,
-                                 Tensor<TOut>& out,
-                                 const ConvStrides& conv_strides,
-                                 const ConvDilations& conv_dilations,
-                                 const InLeftPads& in_left_pads,
-                                 const InRightPads&,
-                                 const ConvTensorLayout layout = ConvTensorLayout::NCHW,
-                                 const ck::index_t activ_type  = 0)
-{
-    using namespace ck;
-
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-
-    auto f_nchw = [&](auto n, auto k, auto ho, auto wo) {
-        double v = 0;
-        for(int c = 0; c < wei.mDesc.GetLengths()[1]; ++c)
-        {
-            for(int y = 0; y < wei.mDesc.GetLengths()[2]; ++y)
-            {
-                int hi = ho * conv_strides[I0] + y * conv_dilations[I0] - in_left_pads[I0];
-                for(int x = 0; x < wei.mDesc.GetLengths()[3]; ++x)
-                {
-                    int wi = wo * conv_strides[I1] + x * conv_dilations[I1] - in_left_pads[I1];
-                    if(hi >= 0 && hi < in.mDesc.GetLengths()[2] && wi >= 0 &&
-                       wi < in.mDesc.GetLengths()[3])
-                    {
-                        v += static_cast<const double>(in(n, c, hi, wi)) *
-                             static_cast<const double>(wei(k, c, y, x));
-                    }
-                }
-            }
-        }
-
-        index_t hox2 = ho * 2;
-        index_t wox2 = wo * 2;
-
-        v = activ(v, activ_type);
-
-        out(n, k, hox2, wox2)         = v + add(n, k, hox2, wox2);
-        out(n, k, hox2, wox2 + 1)     = v + add(n, k, hox2, wox2 + 1);
-        out(n, k, hox2 + 1, wox2)     = v + add(n, k, hox2 + 1, wox2);
-        out(n, k, hox2 + 1, wox2 + 1) = v + add(n, k, hox2 + 1, wox2 + 1);
-    };
-
-    auto f_nhwc = [&](auto n, auto ho, auto wo, auto k) {
-        double v = 0;
-        for(int c = 0; c < wei.mDesc.GetLengths()[3]; ++c)
-        {
-            for(int y = 0; y < wei.mDesc.GetLengths()[1]; ++y)
-            {
-                int hi = ho * conv_strides[I0] + y * conv_dilations[I0] - in_left_pads[I0];
-                for(int x = 0; x < wei.mDesc.GetLengths()[2]; ++x)
-                {
-                    int wi = wo * conv_strides[I1] + x * conv_dilations[I1] - in_left_pads[I1];
-                    if(hi >= 0 && hi < in.mDesc.GetLengths()[1] && wi >= 0 &&
-                       wi < in.mDesc.GetLengths()[2])
-                    {
-                        v += static_cast<const double>(in(n, hi, wi, c)) *
-                             static_cast<const double>(wei(k, y, x, c));
-                    }
-                }
-            }
-        }
-
-        index_t hox2 = ho * 2;
-        index_t wox2 = wo * 2;
-
-        v = activ(v, activ_type);
-
-        out(n, k, hox2, wox2)         = v + add(n, k, hox2, wox2);
-        out(n, k, hox2, wox2 + 1)     = v + add(n, k, hox2, wox2 + 1);
-        out(n, k, hox2 + 1, wox2)     = v + add(n, k, hox2 + 1, wox2);
-        out(n, k, hox2 + 1, wox2 + 1) = v + add(n, k, hox2 + 1, wox2 + 1);
-    };
-
-    if(layout == ConvTensorLayout::NCHW)
-    {
-        make_ParallelTensorFunctor(f_nchw,
-                                   out.mDesc.GetLengths()[0],
-                                   out.mDesc.GetLengths()[1],
-                                   out.mDesc.GetLengths()[2] / 2,
-                                   out.mDesc.GetLengths()[3] /
-                                       2)(std::thread::hardware_concurrency());
-    }
-    else if(layout == ConvTensorLayout::NHWC)
-    {
-        make_ParallelTensorFunctor(f_nhwc,
-                                   out.mDesc.GetLengths()[0],
-                                   out.mDesc.GetLengths()[1],
-                                   out.mDesc.GetLengths()[2] / 2,
-                                   out.mDesc.GetLengths()[3] /
-                                       2)(std::thread::hardware_concurrency());
-    }
-    else
-    {
-        throw std::runtime_error("wrong! not supported layout");
-    }
-}
-
 template <typename TIn, typename TWei, typename TOut, typename InLeftPads, typename InRightPads>
 void host_winograd_3x3_convolution(const Tensor<TIn>& in_nchw,
                                    const Tensor<TWei>& wei_kcyx,