tidy

CMakeLists.txt

@@ -147,7 +147,6 @@ enable_clang_tidy(
         "\.hpp$"
     EXTRA_ARGS
         -DMIOPEN_USE_CLANG_TIDY
-
 )
 
 include(CppCheck)

composable_kernel/include/tensor_operation/blockwise_gemm_dlops_v2r2.hpp

@@ -71,7 +71,7 @@ struct BlockwiseGemmDlops_km_kn_m0m1n0n1_v2r2_pipeline_2x2
     static constexpr index_t N0 = N / N1;
 
     __host__ __device__ static constexpr auto
-    MakeAKM0M1BlockDescriptor(const AKMBlockDesc& a_k_m_block_desc)
+    MakeAKM0M1BlockDescriptor(const AKMBlockDesc& /* a_k_m_block_desc */)
     {
         const auto a_k_m0_m1_block_desc = transform_dynamic_tensor_descriptor(
             AKMBlockDesc{},
@@ -84,7 +84,7 @@ struct BlockwiseGemmDlops_km_kn_m0m1n0n1_v2r2_pipeline_2x2
     }
 
     __host__ __device__ static constexpr auto
-    MakeBKN0N1BlockDescriptor(const BKNBlockDesc& b_k_n_block_desc)
+    MakeBKN0N1BlockDescriptor(const BKNBlockDesc& /* b_k_n_block_desc */)
     {
         const auto b_k_n0_n1_block_desc = transform_dynamic_tensor_descriptor(
             BKNBlockDesc{},
@@ -194,7 +194,7 @@ struct BlockwiseGemmDlops_km_kn_m0m1n0n1_v2r2_pipeline_2x2
               typename ABlockBuffer,
               typename BBlockBuffer,
               typename CThreadBuffer>
-    __device__ void Run(const CM0M1N0N1ThreadDesc& c_m0_m1_n0_n1_thread_desc,
+    __device__ void Run(const CM0M1N0N1ThreadDesc& /* c_m0_m1_n0_n1_thread_desc */,
                         const ABlockBuffer& a_block_buf,
                         const BBlockBuffer& b_block_buf,
                         CThreadBuffer& c_thread_buf) const

composable_kernel/include/tensor_operation/blockwise_gemm_dlops_v3.hpp

@@ -120,9 +120,6 @@ struct BlockwiseGemmDlops_km_kn_m0m1n0n1_v3
                       "wrong! inconsistent type");
 
         constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-        constexpr auto I2 = Number<2>{};
-        constexpr auto I3 = Number<3>{};
 
         constexpr auto a_block_mtx = BlockMatrixA{};
 

composable_kernel/include/tensor_operation/blockwise_gemm_xdlops.hpp

@@ -270,7 +270,6 @@ struct BlockwiseGemmXdlops_km_kn_m0m1m2n_v1_2x2pipeline
         const index_t waveId    = thread_id / WaveSize;
         const index_t laneId    = thread_id % WaveSize;
         const index_t waveId_m  = waveId / NWaves;
-        const index_t waveId_n  = waveId % NWaves;
 
         if constexpr(xdlops_gemm.IsKReduction)
         {

composable_kernel/include/tensor_operation/gridwise_dynamic_gemm_dlops_v1r2.hpp

@@ -619,17 +619,6 @@ struct GridwiseDynamicGemmDlops_km_kn_mn_v1r2
 
         // output: register to global memory
         {
-            constexpr index_t M11 =
-                M1PerThreadM111 * M11N11ThreadClusterM1100 * M11N11ThreadClusterM1101;
-            constexpr index_t N11 =
-                N1PerThreadN111 * M11N11ThreadClusterN1100 * M11N11ThreadClusterN1101;
-
-            constexpr index_t M10 = MPerBlockM1 / M11;
-            constexpr index_t N10 = NPerBlockN1 / N11;
-
-            constexpr index_t M111 = M1PerThreadM111;
-            constexpr index_t N111 = N1PerThreadN111;
-
             constexpr auto c_m0_m10_m11_n0_n10_n11_thread_desc =
                 make_dynamic_naive_tensor_descriptor_packed_v2(
                     make_tuple(I1,

composable_kernel/include/tensor_operation/gridwise_dynamic_gemm_dlops_v1r3.hpp

@@ -191,12 +191,12 @@ struct GridwiseDynamicGemmDlops_km_kn_mn_v1r3
         const auto M  = a_k0_m_k1_grid_desc.GetLength(I1);
         const auto N  = b_k0_n_k1_grid_desc.GetLength(I1);
         const auto K0 = a_k0_m_k1_grid_desc.GetLength(I0);
-        const auto K1 = a_k0_m_k1_grid_desc.GetLength(I2);
 
         // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
 
         return (M == c_m_n_grid_desc.GetLength(I0) && N == c_m_n_grid_desc.GetLength(I1) &&
                 K0 == b_k0_n_k1_grid_desc.GetLength(I0) &&
+                K1 == a_k0_m_k1_grid_desc.GetLength(I2) &&
                 K1 == b_k0_n_k1_grid_desc.GetLength(I2)) &&
                (M % MPerBlockM1 == 0 && N % NPerBlockN1 == 0 && K0 % KPerBlock == 0);
     }
@@ -608,19 +608,6 @@ struct GridwiseDynamicGemmDlops_km_kn_mn_v1r3
 
         // output: register to global memory
         {
-            constexpr auto M11 =
-                Number<container_reduce(M11N11ThreadClusterM110Xs{}, math::multiplies_v2{}, I1) *
-                       M1PerThreadM111>{};
-            constexpr auto N11 =
-                Number<container_reduce(M11N11ThreadClusterN110Xs{}, math::multiplies_v2{}, I1) *
-                       N1PerThreadN111>{};
-
-            constexpr index_t M10 = MPerBlockM1 / M11;
-            constexpr index_t N10 = NPerBlockN1 / N11;
-
-            constexpr index_t M111 = M1PerThreadM111;
-            constexpr index_t N111 = N1PerThreadN111;
-
             constexpr auto c_m0_m10_m11_n0_n10_n11_thread_desc =
                 make_dynamic_naive_tensor_descriptor_packed_v2(
                     make_tuple(I1,

composable_kernel/include/tensor_operation/gridwise_dynamic_gemm_dlops_v2.hpp

@@ -102,7 +102,6 @@ struct GridwiseDynamicGemmDlops_km_kn_mn_v3
 
 // divide block work by [M, N]
 #if 0
-        const auto k_block_work_num   = K / Number<KPerBlock>{};
         const auto ho_block_work_num  = Ho / Number<HoPerBlock>{};
         const auto wo_block_work_num  = Wo / Number<WoPerBlock>{};
         const auto hwo_block_work_num = ho_block_work_num * wo_block_work_num;
@@ -114,7 +113,6 @@ struct GridwiseDynamicGemmDlops_km_kn_mn_v3
         const index_t wo_block_work_id = hwo_block_work_id - ho_block_work_id * wo_block_work_num;
 #else
         // Hack: this force result into SGPR
-        const index_t k_block_work_num   = __builtin_amdgcn_readfirstlane(K / KPerBlock);
         const index_t ho_block_work_num  = __builtin_amdgcn_readfirstlane(Ho / HoPerBlock);
         const index_t wo_block_work_num  = __builtin_amdgcn_readfirstlane(Wo / WoPerBlock);
         const index_t hwo_block_work_num = ho_block_work_num * wo_block_work_num;

composable_kernel/include/tensor_operation/gridwise_dynamic_gemm_xdlops_v2r3.hpp

@@ -269,11 +269,6 @@ struct GridwiseDynamicGemm_k0mk1_k0nk1_mn_xdlops_v2r3
                                const CM0M1M2NGridDesc& c_m0_m1_m2_n_grid_desc,
                                const CBlockClusterAdaptor& c_block_cluster_adaptor)
     {
-        constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-        constexpr auto I2 = Number<2>{};
-        constexpr auto I3 = Number<3>{};
-
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(
             p_a_grid, a_k0_m_k1_grid_desc.GetElementSpaceSize());
         const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum_t::Global>(

composable_kernel/include/tensor_operation/threadwise_gemm_dlops_v3.hpp

@@ -57,8 +57,6 @@ struct ThreadwiseGemmDlops_km_kn_mn_v3
 
         constexpr auto I0 = Number<0>{};
         constexpr auto I1 = Number<1>{};
-        constexpr auto I2 = Number<2>{};
-        constexpr auto I3 = Number<3>{};
 
         constexpr auto E = ADesc{}.GetLength(I0);
         constexpr auto K = ADesc{}.GetLength(I1);

host/driver_offline/include/device_dynamic_convolution_forward_implicit_gemm_v4r4_dlops_nchw_kcyx_nkhw.hpp

@@ -34,12 +34,6 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4_dlops_nchw_kcyx_nkhw(
     constexpr auto I0 = Number<0>{};
     constexpr auto I1 = Number<1>{};
     constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-    constexpr auto I4 = Number<4>{};
-    constexpr auto I5 = Number<5>{};
-    constexpr auto I6 = Number<6>{};
-    constexpr auto I7 = Number<7>{};
-    constexpr auto I8 = Number<8>{};
 
     DeviceMem in_n_c_hi_wi_device_buf(sizeof(TInWei) * in_n_c_hi_wi.mDesc.GetElementSpace());
     DeviceMem wei_k_c_y_x_device_buf(sizeof(TInWei) * wei_k_c_y_x.mDesc.GetElementSpace());
@@ -198,8 +192,8 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4_dlops_nchw_kcyx_nkhw(
             in_gemmk_gemmn0_gemmn1_grid_move_slice_window_iterator_hacks,
             nrepeat);
 
-        float perf = (float)calculate_convolution_flops(
-                         in_n_c_hi_wi_desc, wei_k_c_y_x_desc, out_n_k_ho_wo_desc) /
+        float perf = static_cast<float>(calculate_convolution_flops(
+                         in_n_c_hi_wi_desc, wei_k_c_y_x_desc, out_n_k_ho_wo_desc)) /
                      (std::size_t(1000) * 1000 * 1000) / ave_time;
 
         std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s" << std::endl;

host/driver_offline/include/device_dynamic_convolution_forward_implicit_gemm_v4r4r2_dlops_nhwc_kyxc_nhwk.hpp

@@ -35,11 +35,6 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4r2_dlops_nhwc_kyxc_nhw
     constexpr auto I1 = Number<1>{};
     constexpr auto I2 = Number<2>{};
     constexpr auto I3 = Number<3>{};
-    constexpr auto I4 = Number<4>{};
-    constexpr auto I5 = Number<5>{};
-    constexpr auto I6 = Number<6>{};
-    constexpr auto I7 = Number<7>{};
-    constexpr auto I8 = Number<8>{};
 
     DeviceMem in_n_hi_wi_c_device_buf(sizeof(TInWei) * in_n_hi_wi_c.mDesc.GetElementSpace());
     DeviceMem wei_k_y_x_c_device_buf(sizeof(TInWei) * wei_k_y_x_c.mDesc.GetElementSpace());
@@ -271,7 +266,7 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4r2_dlops_nhwc_kyxc_nhw
             const auto Y = wei_k_y_x_c_lengths[I1];
             const auto X = wei_k_y_x_c_lengths[I2];
 
-            float perf = (float)(std::size_t(2) * N * K * Ho * Wo * C * Y * X) /
+            float perf = static_cast<float>(std::size_t(2) * N * K * Ho * Wo * C * Y * X) /
                          (std::size_t(1000) * 1000 * 1000) / ave_time;
 
             std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s"

host/driver_offline/include/device_dynamic_convolution_forward_implicit_gemm_v4r4r2_xdlops_nchw_kcyx_nkhw.hpp

@@ -34,12 +34,6 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4r2_xdlops_nchw_kcyx_nk
     constexpr auto I0 = Number<0>{};
     constexpr auto I1 = Number<1>{};
     constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-    constexpr auto I4 = Number<4>{};
-    constexpr auto I5 = Number<5>{};
-    constexpr auto I6 = Number<6>{};
-    constexpr auto I7 = Number<7>{};
-    constexpr auto I8 = Number<8>{};
 
     DeviceMem in_n_c_hi_wi_device_buf(sizeof(TInWei) * in_n_c_hi_wi.mDesc.GetElementSpace());
     DeviceMem wei_k_c_y_x_device_buf(sizeof(TInWei) * wei_k_c_y_x.mDesc.GetElementSpace());
@@ -194,8 +188,8 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4r2_xdlops_nchw_kcyx_nk
                    in_gemmk0_gemmn_gemmk1_grid_move_slice_window_iterator_hacks,
                    nrepeat);
 
-        float perf = (float)calculate_convolution_flops(
-                         in_n_c_hi_wi_desc, wei_k_c_y_x_desc, out_n_k_ho_wo_desc) /
+        float perf = static_cast<float>(calculate_convolution_flops(
+                         in_n_c_hi_wi_desc, wei_k_c_y_x_desc, out_n_k_ho_wo_desc)) /
                      (std::size_t(1000) * 1000 * 1000) / ave_time;
 
         std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s" << std::endl;

host/driver_offline/include/device_dynamic_convolution_forward_implicit_gemm_v4r4r4_xdlops_nhwc_kyxc_nhwk.hpp

@@ -35,11 +35,6 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4r4_xdlops_nhwc_kyxc_nh
     constexpr auto I1 = Number<1>{};
     constexpr auto I2 = Number<2>{};
     constexpr auto I3 = Number<3>{};
-    constexpr auto I4 = Number<4>{};
-    constexpr auto I5 = Number<5>{};
-    constexpr auto I6 = Number<6>{};
-    constexpr auto I7 = Number<7>{};
-    constexpr auto I8 = Number<8>{};
 
     DeviceMem in_n_hi_wi_c_device_buf(sizeof(TInWei) * in_n_hi_wi_c.mDesc.GetElementSpace());
     DeviceMem wei_k_y_x_c_device_buf(sizeof(TInWei) * wei_k_y_x_c.mDesc.GetElementSpace());
@@ -352,7 +347,7 @@ void device_dynamic_convolution_forward_implicit_gemm_v4r4r4_xdlops_nhwc_kyxc_nh
             const auto Y = wei_k_y_x_c_lengths[I1];
             const auto X = wei_k_y_x_c_lengths[I2];
 
-            float perf = (float)(std::size_t(2) * N * K * Ho * Wo * C * Y * X) /
+            float perf = static_cast<float>((std::size_t(2) * N * K * Ho * Wo * C * Y * X)) /
                          (std::size_t(1000) * 1000 * 1000) / ave_time;
 
             std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s"

host/driver_offline/include/device_dynamic_convolution_forward_implicit_gemm_v5r1_dlops_nchw_kcyx_nkhw.hpp

@@ -26,7 +26,7 @@ void device_dynamic_convolution_forward_implicit_gemm_v5r1_dlops_nchw_kcyx_nkhw(
     const Tensor<TInWei>& in_n_c_hi_wi,
     const Tensor<TInWei>& wei_k_c_y_x,
     Tensor<TOut>& out_n_k_ho_wo,
-    ck::index_t nrepeat)
+    ck::index_t /* nrepeat */)
 {
     using namespace ck;
 

host/driver_offline/include/device_dynamic_convolution_forward_implicit_gemm_v6r1_dlops_nchw_kcyx_nkhw.hpp

@@ -232,8 +232,8 @@ void device_dynamic_convolution_forward_implicit_gemm_v6r1_dlops_nchw_kcyx_nkhw(
             in_grid_move_slice_window_iterator_hacks,
             nrepeat);
 
-        float perf = (float)calculate_convolution_flops(
-                         in_desc_n_c_hi_wi, wei_desc_k_c_y_x, out_desc_n_k_ho_wo) /
+        float perf = static_cast<float>(calculate_convolution_flops(
+                         in_desc_n_c_hi_wi, wei_desc_k_c_y_x, out_desc_n_k_ho_wo)) /
                      (std::size_t(1000) * 1000 * 1000) / ave_time;
 
         std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s" << std::endl;

host/driver_offline/include/driver_dynamic_convolution_forward_implicit_gemm_v5r1_dlops_nchw_kcyx_nkhw.hpp

@@ -338,9 +338,10 @@ struct DriverDynamicConvolutionForwardImplicitGemmDlops_v5r1_nchw_kcyx_nkhw_pad
 
             float ave_time = timer.GetElapsedTime() / nrepeat;
 
-            float perf = (float)calculate_convolution_flops(in_n_c_hi_wi_global_desc,
+            float perf =
+                static_cast<float>(calculate_convolution_flops(in_n_c_hi_wi_global_desc,
                                                                wei_k_c_y_x_global_desc,
-                                                            out_n_k0_ho_wo_k1_global_desc) /
+                                                               out_n_k0_ho_wo_k1_global_desc)) /
                 (std::size_t(1000) * 1000 * 1000) / ave_time;
 
             std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s"

host/driver_offline/include/driver_dynamic_convolution_forward_implicit_gemm_v5r1_dlops_nchw_kcyx_nkhw_outpad.hpp

@@ -354,9 +354,10 @@ struct DriverDynamicConvolutionForwardImplicitGemmDlops_v5r1_nchw_kcyx_nkhw_outp
 
             float ave_time = timer.GetElapsedTime() / nrepeat;
 
-            float perf = (float)calculate_convolution_flops(in_n_c_hi_wi_global_desc,
+            float perf =
+                static_cast<float>(calculate_convolution_flops(in_n_c_hi_wi_global_desc,
                                                                wei_k_c_y_x_global_desc,
-                                                            out_n_k0_ho_wo_k1_global_desc) /
+                                                               out_n_k0_ho_wo_k1_global_desc)) /
                 (std::size_t(1000) * 1000 * 1000) / ave_time;
 
             std::cout << "Average time : " << ave_time << " ms, " << perf << " TFlop/s"

host/driver_offline/src/conv_bwd_driver_offline.cpp

@@ -128,10 +128,8 @@ int main(int argc, char* argv[])
 
     std::vector<std::size_t> in_lengths_host(4), wei_lengths_host(4), out_lengths_host(4);
 
-    switch(layout)
+    if(layout == ConvTensorLayout::NCHW)
     {
-    case ConvTensorLayout::NCHW:
-        // NCHW
         in_lengths_host[0]  = static_cast<std::size_t>(N);
         in_lengths_host[1]  = static_cast<std::size_t>(C);
         in_lengths_host[2]  = static_cast<std::size_t>(Hi);
@@ -144,9 +142,9 @@ int main(int argc, char* argv[])
         out_lengths_host[1] = static_cast<std::size_t>(K);
         out_lengths_host[2] = static_cast<std::size_t>(Ho);
         out_lengths_host[3] = static_cast<std::size_t>(Wo);
-        break;
-    case ConvTensorLayout::NHWC:
-        // NHWC
+    }
+    else if(layout == ConvTensorLayout::NHWC)
+    {
         in_lengths_host[0]  = static_cast<std::size_t>(N);
         in_lengths_host[1]  = static_cast<std::size_t>(Hi);
         in_lengths_host[2]  = static_cast<std::size_t>(Wi);
@@ -159,8 +157,10 @@ int main(int argc, char* argv[])
         out_lengths_host[1] = static_cast<std::size_t>(Ho);
         out_lengths_host[2] = static_cast<std::size_t>(Wo);
         out_lengths_host[3] = static_cast<std::size_t>(K);
-        break;
-    default: throw std::runtime_error("wrong! not implemented");
+    }
+    else
+    {
+        throw std::runtime_error("wrong! not implemented");
     }
 
     Tensor<in_data_t> in_host(in_lengths_host);

host/driver_offline/src/conv_fwd_driver_offline.cpp

@@ -467,7 +467,6 @@ int main(int argc, char* argv[])
 
         check_error(out_host, out_device);
 
-#if 0
         if(do_log)
         {
             LogRangeAsType<float>(std::cout << "in : ", in.mData, ",") << std::endl;
@@ -475,6 +474,5 @@ int main(int argc, char* argv[])
             LogRangeAsType<float>(std::cout << "out_host  : ", out_host.mData, ",") << std::endl;
             LogRangeAsType<float>(std::cout << "out_device: ", out_device.mData, ",") << std::endl;
         }
-#endif
     }
 }

host/host_tensor/include/conv_common.hpp

@@ -62,7 +62,7 @@ constexpr auto get_convolution_output_default_4d_tensor_descriptor(
 
 template <class InDesc, class WeiDesc, class OutDesc>
 constexpr std::size_t
-calculate_convolution_flops(const InDesc& in_desc, const WeiDesc& wei_desc, const OutDesc& out_desc)
+calculate_convolution_flops(const InDesc&, const WeiDesc& wei_desc, const OutDesc& out_desc)
 {
     using namespace ck;