tidy

host/host_tensor/include/host_conv.hpp

@@ -14,15 +14,13 @@ void host_direct_convolution(const Tensor<TIn>& in,
                              const ConvStrides& conv_strides,
                              const ConvDilations& conv_dilations,
                              const InLeftPads& in_left_pads,
-                             const InRightPads& in_right_pads,
+                             const InRightPads&,
                              const ConvTensorLayout layout = ConvTensorLayout::NCHW)
 {
     using namespace ck;
 
     constexpr auto I0 = Number<0>{};
     constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
 
     auto f_nchw = [&](auto n, auto k, auto ho, auto wo) {
         double v = 0;
@@ -68,23 +66,25 @@ void host_direct_convolution(const Tensor<TIn>& in,
         out(n, ho, wo, k) = v;
     };
 
-    switch(layout)
+    if(layout == ConvTensorLayout::NCHW)
     {
-    case ConvTensorLayout::NCHW:
         make_ParallelTensorFunctor(f_nchw,
                                    out.mDesc.GetLengths()[0],
                                    out.mDesc.GetLengths()[1],
                                    out.mDesc.GetLengths()[2],
                                    out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-        break;
-    case ConvTensorLayout::NHWC:
+    }
+    else if(layout == ConvTensorLayout::NHWC)
+    {
         make_ParallelTensorFunctor(f_nhwc,
                                    out.mDesc.GetLengths()[0],
                                    out.mDesc.GetLengths()[1],
                                    out.mDesc.GetLengths()[2],
                                    out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-        break;
-    default: throw std::runtime_error("wrong! not supported layout");
+    }
+    else
+    {
+        throw std::runtime_error("wrong! not supported layout");
     }
 }
 
@@ -102,15 +102,13 @@ void host_winograd_3x3_convolution(const Tensor<TIn>& in_nchw,
 
     std::size_t N = in_nchw.mDesc.GetLengths()[0];
     std::size_t C = in_nchw.mDesc.GetLengths()[1];
-    std::size_t HI = in_nchw.mDesc.GetLengths()[2];
-    std::size_t WI = in_nchw.mDesc.GetLengths()[3];
 
     std::size_t K = wei_kcyx.mDesc.GetLengths()[0];
     std::size_t Y = wei_kcyx.mDesc.GetLengths()[2];
     std::size_t X = wei_kcyx.mDesc.GetLengths()[3];
 
-    std::size_t HO = out_nkhw.mDesc.GetLengths()[2];
-    std::size_t WO = out_nkhw.mDesc.GetLengths()[3];
+    std::size_t Ho = out_nkhw.mDesc.GetLengths()[2];
+    std::size_t Wo = out_nkhw.mDesc.GetLengths()[3];
 
     index_t h_pad_low = InLeftPads{}.Get(Number<0>{});
     index_t w_pad_low = InLeftPads{}.Get(Number<1>{});
@@ -118,8 +116,8 @@ void host_winograd_3x3_convolution(const Tensor<TIn>& in_nchw,
     std::size_t HiPerTile = HoPerTile + Y - 1;
     std::size_t WiPerTile = WoPerTile + X - 1;
 
-    std::size_t HTile = (HO + HoPerTile - 1) / HoPerTile;
-    std::size_t WTile = (WO + WoPerTile - 1) / WoPerTile;
+    std::size_t HTile = (Ho + HoPerTile - 1) / HoPerTile;
+    std::size_t WTile = (Wo + WoPerTile - 1) / WoPerTile;
 
     Tensor<double> in_hold({N, C, HTile, WTile, HiPerTile, WiPerTile});
     Tensor<double> in_transform({N, C, HTile, WTile, HiPerTile, WiPerTile});

host/host_tensor/include/host_tensor_generator.hpp

@@ -9,7 +9,7 @@ struct GeneratorTensor_1
     int value = 1;
 
     template <typename... Is>
-    float operator()(Is... is)
+    float operator()(Is...)
     {
         return value;
     }

host/solver/include/conv_igemm_fwd_v6r1_dlops_nchw_kcyx_nkhw.hpp

@@ -99,40 +99,48 @@ struct CompileParameterConvIgemmFwdV6r1DlopsNchwKcyxNkhw
         // clang-format on
     }
 
-    ck::DataTypeEnum_t ABDataTypeEnum;
-    ck::DataTypeEnum_t AccDataTypeEnum;
-    ck::DataTypeEnum_t CDataTypeEnum;
-
-    int BlockSize;
-
-    int GN0;
-    int GK1;
-
-    int GM1PerBlockGM11;
-    int GN1PerBlockGN11;
-    int GK0PerBlock;
-
-    int BM1PerThreadBM11;
-    int BN1PerThreadBN11;
-    int BK0PerThread;
-
-    std::array<int, 2> BM10BN10ThreadClusterBM10Xs;
-    std::array<int, 2> BM10BN10ThreadClusterBN10Xs;
-
-    std::array<int, 5> ABlockTransferThreadSliceLengths_GK0_GM0_GM10_GM11_GK1;
-    std::array<int, 5> ABlockTransferThreadClusterLengths_GK0_GM0_GM10_GM11_GK1;
-    std::array<int, 5> ABlockTransferSrcVectorTensorLengths_GK0_GM0_GM10_GM11_GK1;
-    std::array<int, 5> ABlockTransferDstVectorTensorLengths_GK0_GM0_GM10_GM11_GK1;
-
-    std::array<int, 5> BBlockTransferThreadSliceLengths_GK0_GN0_GN10_GN11_GK1;
-    std::array<int, 5> BBlockTransferThreadClusterLengths_GK0_GN0_GN10_GN11_GK1;
-    std::array<int, 5> BBlockTransferSrcVectorTensorLengths_GK0_GN0_GN10_GN11_GK1;
-    std::array<int, 5> BBlockTransferDstVectorTensorLengths_GK0_GN0_GN10_GN11_GK1;
-
-    int CThreadTransferDstScalarPerVector;
-
-    bool HasMainKBlockLoop;
-    bool HasDoubleTailKBlockLoop;
+    ck::DataTypeEnum_t ABDataTypeEnum  = ck::DataTypeEnum_t::Unknown;
+    ck::DataTypeEnum_t AccDataTypeEnum = ck::DataTypeEnum_t::Unknown;
+    ck::DataTypeEnum_t CDataTypeEnum   = ck::DataTypeEnum_t::Unknown;
+
+    int BlockSize = 1;
+
+    int GN0 = -1;
+    int GK1 = -1;
+
+    int GM1PerBlockGM11 = -1;
+    int GN1PerBlockGN11 = -1;
+    int GK0PerBlock     = -1;
+
+    int BM1PerThreadBM11 = -1;
+    int BN1PerThreadBN11 = -1;
+    int BK0PerThread     = -1;
+
+    std::array<int, 2> BM10BN10ThreadClusterBM10Xs = {-1, -1};
+    std::array<int, 2> BM10BN10ThreadClusterBN10Xs = {-1, -1};
+
+    std::array<int, 5> ABlockTransferThreadSliceLengths_GK0_GM0_GM10_GM11_GK1 = {
+        -1, -1, -1, -1, -1};
+    std::array<int, 5> ABlockTransferThreadClusterLengths_GK0_GM0_GM10_GM11_GK1 = {
+        -1, -1, -1, -1, -1};
+    std::array<int, 5> ABlockTransferSrcVectorTensorLengths_GK0_GM0_GM10_GM11_GK1 = {
+        -1, -1, -1, -1, -1};
+    std::array<int, 5> ABlockTransferDstVectorTensorLengths_GK0_GM0_GM10_GM11_GK1 = {
+        -1, -1, -1, -1, -1};
+
+    std::array<int, 5> BBlockTransferThreadSliceLengths_GK0_GN0_GN10_GN11_GK1 = {
+        -1, -1, -1, -1, -1};
+    std::array<int, 5> BBlockTransferThreadClusterLengths_GK0_GN0_GN10_GN11_GK1 = {
+        -1, -1, -1, -1, -1};
+    std::array<int, 5> BBlockTransferSrcVectorTensorLengths_GK0_GN0_GN10_GN11_GK1 = {
+        -1, -1, -1, -1, -1};
+    std::array<int, 5> BBlockTransferDstVectorTensorLengths_GK0_GN0_GN10_GN11_GK1 = {
+        -1, -1, -1, -1, -1};
+
+    int CThreadTransferDstScalarPerVector = -1;
+
+    bool HasMainKBlockLoop       = false;
+    bool HasDoubleTailKBlockLoop = false;
 };
 
 struct TunableConvIgemmFwdV6r1DlopsNchwKcyxNkhw