ncnn/src/layer/vulkan/convolution_vulkan.cpp

// Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.

#include "convolution_vulkan.h"
#include <algorithm>
#include "layer_type.h"
#include "layer_shader_type.h"

namespace ncnn {

DEFINE_LAYER_CREATOR(Convolution_vulkan)

Convolution_vulkan::Convolution_vulkan()
{
    support_vulkan = true;
    support_image_storage = true;

    padding = 0;

    pipeline_convolution = 0;
    pipeline_convolution_1x1s1d1 = 0;
    pipeline_convolution_pack4 = 0;
    pipeline_convolution_pack4_1x1s1d1 = 0;
    pipeline_convolution_pack1to4 = 0;
    pipeline_convolution_pack4to1 = 0;

    pipeline_convolution_pack8 = 0;
    pipeline_convolution_pack8_1x1s1d1 = 0;
    pipeline_convolution_pack1to8 = 0;
    pipeline_convolution_pack4to8 = 0;
    pipeline_convolution_pack8to1 = 0;
    pipeline_convolution_pack8to4 = 0;

    winograd_padding = 0;
    winograd_crop = 0;
    pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input = 0;
    pipeline_convolution_pack4_3x3s1d1_winograd23_gemm = 0;
    pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output = 0;

    pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input = 0;
    pipeline_convolution_pack8_3x3s1d1_winograd23_gemm = 0;
    pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output = 0;

    innerproduct = 0;
}

int Convolution_vulkan::create_pipeline(const Option& opt)
{
    const Mat& shape = bottom_shapes.empty() ? Mat() : bottom_shapes[0];
    const Mat& out_shape = top_shapes.empty() ? Mat() : top_shapes[0];

    const int maxk = kernel_w * kernel_h;
    int num_input = weight_data_size / maxk / num_output;

    // fc
    if (kernel_w == 1 && kernel_h == 1)
    {
        innerproduct = ncnn::create_layer(ncnn::LayerType::InnerProduct);
        innerproduct->vkdev = vkdev;

        innerproduct->bottom_shapes.resize(1);
        innerproduct->bottom_shapes[0] = shape;
        innerproduct->top_shapes.resize(1);
        innerproduct->top_shapes[0] = out_shape;

        ncnn::ParamDict pd;
        pd.set(0, num_output);
        pd.set(1, bias_term);
        pd.set(2, weight_data_size);// TODO int8

        innerproduct->load_param(pd);

        ncnn::Mat weights[2];
        weights[0] = weight_data;
        weights[1] = bias_data;
        ncnn::ModelBinFromMatArray mb(weights);

        innerproduct->load_model(mb);

        innerproduct->create_pipeline(opt);

        if (shape.dims == 1 && shape.w == num_input)
        {
            return 0;
        }
    }

    // the shape after padding
    Mat shape_bordered;
    if (shape.dims != 0)
    {
        if (pad_left > 0 || pad_right > 0 || pad_top > 0 || pad_bottom > 0)
        {
            shape_bordered = Mat(shape.w + pad_left + pad_right, shape.h + pad_top + pad_bottom, shape.c, (void*)0);
        }
        else if ((pad_left == -233 && pad_right == -233 && pad_top == -233 && pad_bottom == -233)
            || (pad_left == -234 && pad_right == -234 && pad_top == -234 && pad_bottom == -234))
        {
            const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1;
            const int kernel_extent_h = dilation_h * (kernel_h - 1) + 1;

            int wpad = kernel_extent_w + (shape.w - 1) / stride_w * stride_w - shape.w;
            int hpad = kernel_extent_h + (shape.h - 1) / stride_h * stride_h - shape.h;
            if (wpad > 0 || hpad > 0)
            {
                shape_bordered = Mat(shape.w + wpad, shape.h + hpad, shape.c, (void*)0);
            }
        }
        else
        {
            shape_bordered = shape;
        }
    }

    {
        padding = ncnn::create_layer(ncnn::LayerType::Padding);
        padding->vkdev = vkdev;

        padding->bottom_shapes.resize(1);
        padding->bottom_shapes[0] = shape;
        padding->top_shapes.resize(1);
        padding->top_shapes[0] = shape_bordered;

        ncnn::ParamDict pd;
        pd.set(0, pad_top);
        pd.set(1, pad_bottom);
        pd.set(2, pad_left);
        pd.set(3, pad_right);
        pd.set(4, 0);
        pd.set(5, pad_value);

        padding->load_param(pd);

        padding->create_pipeline(opt);
    }

    int elempack = opt.use_shader_pack8 && num_input % 8 == 0 ? 8 : num_input % 4 == 0 ? 4 : 1;
    int out_elempack = opt.use_shader_pack8 && num_output % 8 == 0 ? 8 : num_output % 4 == 0 ? 4 : 1;

    size_t elemsize;
    size_t out_elemsize;
    if (opt.use_image_storage && opt.use_fp16_storage)
    {
        elemsize = elempack * 2u;
        out_elemsize = out_elempack * 2u;
    }
    else if (opt.use_image_storage && opt.use_fp16_packed)
    {
        elemsize = elempack == 1 ? 4u : elempack * 2u;
        out_elemsize = out_elempack == 1 ? 4u : out_elempack * 2u;
    }
    else if (opt.use_image_storage)
    {
        elemsize = elempack * 4u;
        out_elemsize = out_elempack * 4u;
    }
    else if (opt.use_fp16_storage)
    {
        elemsize = elempack * 2u;
        out_elemsize = out_elempack * 2u;
    }
    else if (opt.use_fp16_packed)
    {
        elemsize = elempack == 1 ? 4u : elempack * 2u;
        out_elemsize = out_elempack == 1 ? 4u : out_elempack * 2u;
    }
    else
    {
        elemsize = elempack * 4u;
        out_elemsize = out_elempack * 4u;
    }

    Mat shape_bordered_packed;
    if (shape_bordered.dims == 3) shape_bordered_packed = Mat(shape_bordered.w, shape_bordered.h, num_input / elempack, (void*)0, elemsize, elempack);

    Mat out_shape_packed;
    if (out_shape.dims == 3) out_shape_packed = Mat(out_shape.w, out_shape.h, num_output / out_elempack, (void*)0, out_elemsize, out_elempack);

    std::vector<vk_specialization_type> specializations(10 + 10);
    specializations[0].i = kernel_w;
    specializations[1].i = kernel_h;
    specializations[2].i = dilation_w;
    specializations[3].i = dilation_h;
    specializations[4].i = stride_w;
    specializations[5].i = stride_h;
    specializations[6].i = bias_term;
    specializations[7].i = activation_type;
    specializations[8].f = activation_params.w >= 1 ? activation_params[0] : 0.f;
    specializations[9].f = activation_params.w == 2 ? activation_params[1] : 0.f;
    specializations[10 + 0].i = shape_bordered_packed.dims;
    specializations[10 + 1].i = shape_bordered_packed.w;
    specializations[10 + 2].i = shape_bordered_packed.h;
    specializations[10 + 3].i = shape_bordered_packed.c;
    specializations[10 + 4].i = shape_bordered_packed.cstep;
    specializations[10 + 5].i = out_shape_packed.dims;
    specializations[10 + 6].i = out_shape_packed.w;
    specializations[10 + 7].i = out_shape_packed.h;
    specializations[10 + 8].i = out_shape_packed.c;
    specializations[10 + 9].i = out_shape_packed.cstep;

    Mat local_size_xyz(8, 8, std::min(4, num_output / out_elempack), (void*)0);
    if (out_shape_packed.dims != 0)
    {
        local_size_xyz.w = std::min(8, out_shape_packed.w);
        local_size_xyz.h = std::min(8, out_shape_packed.h);
        local_size_xyz.c = std::min(4, out_shape_packed.c);
    }

    bool is_conv1x1s1d1 = kernel_w == 1 && kernel_h == 1 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1;
    bool is_conv3x3s1d1 = kernel_w == 3 && kernel_h == 3 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1;

    // pack1
    if (elempack == 1 && out_elempack == 1)
    {
        if (is_conv1x1s1d1)
        {
            pipeline_convolution_1x1s1d1 = new Pipeline(vkdev);
            if (opt.use_image_storage)
            {
            Mat local_size_xyz_local(4, 4, std::min(4, num_output / out_elempack), (void*)0);
            if (out_shape_packed.dims != 0)
            {
                local_size_xyz_local.w = std::max(1, std::min(4, (out_shape_packed.w + 1) / 2));
                local_size_xyz_local.h = std::max(1, std::min(4, (out_shape_packed.h + 1) / 2));
                local_size_xyz_local.c = std::min(4, out_shape_packed.c);
            }
            pipeline_convolution_1x1s1d1->set_optimal_local_size_xyz(local_size_xyz_local);
            }
            else
            {
            pipeline_convolution_1x1s1d1->set_local_size_xyz(8, 1, std::min(8, num_output));
            }
            pipeline_convolution_1x1s1d1->create(LayerShaderType::convolution_1x1s1d1, opt, specializations);
        }
        else
        {
            pipeline_convolution = new Pipeline(vkdev);
            pipeline_convolution->set_optimal_local_size_xyz(local_size_xyz);
            pipeline_convolution->create(LayerShaderType::convolution, opt, specializations);
        }
    }

    // pack4
    if (elempack == 4 && out_elempack == 4)
    {
        if (is_conv1x1s1d1)
        {
            pipeline_convolution_pack4_1x1s1d1 = new Pipeline(vkdev);
            if (opt.use_image_storage)
            {
            Mat local_size_xyz_local(4, 4, std::min(4, num_output / out_elempack), (void*)0);
            if (out_shape_packed.dims != 0)
            {
                local_size_xyz_local.w = std::max(1, std::min(4, (out_shape_packed.w + 1) / 2));
                local_size_xyz_local.h = std::max(1, std::min(4, (out_shape_packed.h + 1) / 2));
                local_size_xyz_local.c = std::min(4, out_shape_packed.c);
            }
            pipeline_convolution_pack4_1x1s1d1->set_optimal_local_size_xyz(local_size_xyz_local);
            }
            else
            {
            pipeline_convolution_pack4_1x1s1d1->set_local_size_xyz(8, 1, std::min(8, num_output / 4));
            }
            pipeline_convolution_pack4_1x1s1d1->create(LayerShaderType::convolution_pack4_1x1s1d1, opt, specializations);
        }
        else if (is_conv3x3s1d1 && num_input >= 16 && num_output >= 16)
        {
            // winograd23
            int block_x = 0;
            int block_y = 0;
            Mat shape_winograd_bordered;
            Mat shape_winograd_input_transformed;
            Mat shape_winograd_gemm;
            Mat shape_winograd_out_bordered;
            if (out_shape.dims != 0)
            {
                int outw_bordered = (out_shape.w + 1) / 2 * 2;
                int outh_bordered = (out_shape.h + 1) / 2 * 2;

                int w_bordered = outw_bordered + 2;
                int h_bordered = outh_bordered + 2;

                block_x = outw_bordered / 2;
                block_y = outh_bordered / 2;

                shape_winograd_bordered = Mat(w_bordered, h_bordered, shape.c, (void*)0);
                shape_winograd_input_transformed = Mat(16, block_x * block_y, shape.c, (void*)0);
                shape_winograd_gemm = Mat(16, block_x * block_y, out_shape.c, (void*)0);
                shape_winograd_out_bordered = Mat(outw_bordered, outh_bordered, out_shape.c, (void*)0);
            }

            Mat shape_winograd_bordered_packed;
            if (shape_winograd_bordered.dims == 3) shape_winograd_bordered_packed = Mat(shape_winograd_bordered.w, shape_winograd_bordered.h, shape_winograd_bordered.c / elempack, (void*)0, elemsize, elempack);

            Mat shape_winograd_input_transformed_packed;
            if (shape_winograd_input_transformed.dims == 3) shape_winograd_input_transformed_packed = Mat(shape_winograd_input_transformed.w, shape_winograd_input_transformed.h, shape_winograd_input_transformed.c / elempack, (void*)0, elemsize, elempack);

            Mat shape_winograd_gemm_packed;
            if (shape_winograd_gemm.dims == 3) shape_winograd_gemm_packed = Mat(shape_winograd_gemm.w, shape_winograd_gemm.h, shape_winograd_gemm.c / out_elempack, (void*)0, out_elemsize, out_elempack);

            Mat shape_winograd_out_bordered_packed;
            if (shape_winograd_out_bordered.dims == 3) shape_winograd_out_bordered_packed = Mat(shape_winograd_out_bordered.w, shape_winograd_out_bordered.h, shape_winograd_out_bordered.c / out_elempack, (void*)0, out_elemsize, out_elempack);

            {
                winograd_padding = ncnn::create_layer(ncnn::LayerType::Padding);
                winograd_padding->vkdev = vkdev;

                winograd_padding->bottom_shapes.resize(1);
                winograd_padding->bottom_shapes[0] = shape_bordered;
                winograd_padding->top_shapes.resize(1);
                winograd_padding->top_shapes[0] = shape_winograd_bordered;

                ncnn::ParamDict pd;
                pd.set(0, -233);
                pd.set(1, -233);
                pd.set(2, -233);
                pd.set(3, -233);
                pd.set(4, 0);
                pd.set(5, 0.f);

                winograd_padding->load_param(pd);

                winograd_padding->create_pipeline(opt);
            }

            {
                winograd_crop = ncnn::create_layer(ncnn::LayerType::Crop);
                winograd_crop->vkdev = vkdev;

                winograd_crop->bottom_shapes.resize(1);
                winograd_crop->bottom_shapes[0] = shape_winograd_out_bordered;
                winograd_crop->top_shapes.resize(1);
                winograd_crop->top_shapes[0] = out_shape;

                ncnn::ParamDict pd;
                pd.set(0, -233);
                pd.set(1, -233);
                pd.set(2, -233);
                pd.set(3, 0);
                pd.set(4, 0);
                pd.set(5, 0);

                winograd_crop->load_param(pd);

                winograd_crop->create_pipeline(opt);
            }

            {
                std::vector<vk_specialization_type> specializations(0 + 7);
                specializations[0 + 0].i = shape_winograd_bordered_packed.w;
                specializations[0 + 1].i = shape_winograd_bordered_packed.h;
                specializations[0 + 2].i = shape_winograd_bordered_packed.c;
                specializations[0 + 3].i = shape_winograd_bordered_packed.cstep;
                specializations[0 + 4].i = shape_winograd_input_transformed_packed.cstep;
                specializations[0 + 5].i = block_x;
                specializations[0 + 6].i = block_y;

                pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input = new Pipeline(vkdev);
                pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input->set_local_size_xyz(8, 8, 1);
                pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input->create(LayerShaderType::convolution_pack4_3x3s1d1_winograd23_transform_input, opt, specializations);
            }

            {
                std::vector<vk_specialization_type> specializations(0 + 5);
                specializations[0 + 0].i = shape_winograd_input_transformed_packed.c;
                specializations[0 + 1].i = shape_winograd_input_transformed_packed.cstep;
                specializations[0 + 2].i = shape_winograd_gemm_packed.h;
                specializations[0 + 3].i = shape_winograd_gemm_packed.c;
                specializations[0 + 4].i = shape_winograd_gemm_packed.cstep;

                pipeline_convolution_pack4_3x3s1d1_winograd23_gemm = new Pipeline(vkdev);
                pipeline_convolution_pack4_3x3s1d1_winograd23_gemm->set_local_size_xyz(4, 4, std::min(4, num_output / 4));
                pipeline_convolution_pack4_3x3s1d1_winograd23_gemm->create(LayerShaderType::convolution_pack4_3x3s1d1_winograd23_gemm, opt, specializations);
            }

            {
                std::vector<vk_specialization_type> specializations(4 + 7);
                specializations[0].i = bias_term;
                specializations[1].i = activation_type;
                specializations[2].f = activation_params.w >= 1 ? activation_params[0] : 0.f;
                specializations[3].f = activation_params.w == 2 ? activation_params[1] : 0.f;
                specializations[4 + 0].i = shape_winograd_gemm_packed.c;
                specializations[4 + 1].i = shape_winograd_gemm_packed.cstep;
                specializations[4 + 2].i = block_x;
                specializations[4 + 3].i = block_y;
                specializations[4 + 4].i = shape_winograd_out_bordered_packed.w;
                specializations[4 + 5].i = shape_winograd_out_bordered_packed.h;
                specializations[4 + 6].i = shape_winograd_out_bordered_packed.cstep;

                pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output = new Pipeline(vkdev);
                pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output->set_local_size_xyz(8, 8, 1);
                pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output->create(LayerShaderType::convolution_pack4_3x3s1d1_winograd23_transform_output, opt, specializations);
            }
        }
        else
        {
            pipeline_convolution_pack4 = new Pipeline(vkdev);
            pipeline_convolution_pack4->set_optimal_local_size_xyz(local_size_xyz);
            pipeline_convolution_pack4->create(LayerShaderType::convolution_pack4, opt, specializations);
        }
    }

    // pack1to4
    if (elempack == 1 && out_elempack == 4)
    {
        pipeline_convolution_pack1to4 = new Pipeline(vkdev);
        pipeline_convolution_pack1to4->set_optimal_local_size_xyz(local_size_xyz);
        pipeline_convolution_pack1to4->create(LayerShaderType::convolution_pack1to4, opt, specializations);
    }

    // pack4to1
    if (elempack == 4 && out_elempack == 1)
    {
        pipeline_convolution_pack4to1 = new Pipeline(vkdev);
        pipeline_convolution_pack4to1->set_optimal_local_size_xyz(local_size_xyz);
        pipeline_convolution_pack4to1->create(LayerShaderType::convolution_pack4to1, opt, specializations);
    }

    // pack8
    if (elempack == 8 && out_elempack == 8)
    {
        if (is_conv1x1s1d1)
        {
            pipeline_convolution_pack8_1x1s1d1 = new Pipeline(vkdev);
            if (opt.use_image_storage)
            {
            Mat local_size_xyz_local(4, 4, std::min(4, num_output / out_elempack), (void*)0);
            if (out_shape_packed.dims != 0)
            {
                local_size_xyz_local.w = std::max(1, std::min(4, (out_shape_packed.w + 1) / 2));
                local_size_xyz_local.h = std::max(1, std::min(4, (out_shape_packed.h + 1) / 2));
                local_size_xyz_local.c = std::min(4, out_shape_packed.c);
            }
            pipeline_convolution_pack8_1x1s1d1->set_optimal_local_size_xyz(local_size_xyz_local);
            }
            else
            {
            pipeline_convolution_pack8_1x1s1d1->set_local_size_xyz(8, 1, std::min(8, num_output / 8));
            }
            pipeline_convolution_pack8_1x1s1d1->create(LayerShaderType::convolution_pack8_1x1s1d1, opt, specializations);
        }
        else if (is_conv3x3s1d1 && num_input >= 16 && num_output >= 16)
        {
            // winograd23
            int block_x = 0;
            int block_y = 0;
            Mat shape_winograd_bordered;
            Mat shape_winograd_input_transformed;
            Mat shape_winograd_gemm;
            Mat shape_winograd_out_bordered;
            if (out_shape.dims != 0)
            {
                int outw_bordered = (out_shape.w + 1) / 2 * 2;
                int outh_bordered = (out_shape.h + 1) / 2 * 2;

                int w_bordered = outw_bordered + 2;
                int h_bordered = outh_bordered + 2;

                block_x = outw_bordered / 2;
                block_y = outh_bordered / 2;

                shape_winograd_bordered = Mat(w_bordered, h_bordered, shape.c, (void*)0);
                shape_winograd_input_transformed = Mat(16, block_x * block_y, shape.c, (void*)0);
                shape_winograd_gemm = Mat(16, block_x * block_y, out_shape.c, (void*)0);
                shape_winograd_out_bordered = Mat(outw_bordered, outh_bordered, out_shape.c, (void*)0);
            }

            Mat shape_winograd_bordered_packed;
            if (shape_winograd_bordered.dims == 3) shape_winograd_bordered_packed = Mat(shape_winograd_bordered.w, shape_winograd_bordered.h, shape_winograd_bordered.c / elempack, (void*)0, elemsize, elempack);

            Mat shape_winograd_input_transformed_packed;
            if (shape_winograd_input_transformed.dims == 3) shape_winograd_input_transformed_packed = Mat(shape_winograd_input_transformed.w, shape_winograd_input_transformed.h, shape_winograd_input_transformed.c / elempack, (void*)0, elemsize, elempack);

            Mat shape_winograd_gemm_packed;
            if (shape_winograd_gemm.dims == 3) shape_winograd_gemm_packed = Mat(shape_winograd_gemm.w, shape_winograd_gemm.h, shape_winograd_gemm.c / out_elempack, (void*)0, out_elemsize, out_elempack);

            Mat shape_winograd_out_bordered_packed;
            if (shape_winograd_out_bordered.dims == 3) shape_winograd_out_bordered_packed = Mat(shape_winograd_out_bordered.w, shape_winograd_out_bordered.h, shape_winograd_out_bordered.c / out_elempack, (void*)0, out_elemsize, out_elempack);

            {
                winograd_padding = ncnn::create_layer(ncnn::LayerType::Padding);
                winograd_padding->vkdev = vkdev;

                winograd_padding->bottom_shapes.resize(1);
                winograd_padding->bottom_shapes[0] = shape_bordered;
                winograd_padding->top_shapes.resize(1);
                winograd_padding->top_shapes[0] = shape_winograd_bordered;

                ncnn::ParamDict pd;
                pd.set(0, -233);
                pd.set(1, -233);
                pd.set(2, -233);
                pd.set(3, -233);
                pd.set(4, 0);
                pd.set(5, 0.f);

                winograd_padding->load_param(pd);

                winograd_padding->create_pipeline(opt);
            }

            {
                winograd_crop = ncnn::create_layer(ncnn::LayerType::Crop);
                winograd_crop->vkdev = vkdev;

                winograd_crop->bottom_shapes.resize(1);
                winograd_crop->bottom_shapes[0] = shape_winograd_out_bordered;
                winograd_crop->top_shapes.resize(1);
                winograd_crop->top_shapes[0] = out_shape;

                ncnn::ParamDict pd;
                pd.set(0, -233);
                pd.set(1, -233);
                pd.set(2, -233);
                pd.set(3, 0);
                pd.set(4, 0);
                pd.set(5, 0);

                winograd_crop->load_param(pd);

                winograd_crop->create_pipeline(opt);
            }

            {
                std::vector<vk_specialization_type> specializations(0 + 7);
                specializations[0 + 0].i = shape_winograd_bordered_packed.w;
                specializations[0 + 1].i = shape_winograd_bordered_packed.h;
                specializations[0 + 2].i = shape_winograd_bordered_packed.c;
                specializations[0 + 3].i = shape_winograd_bordered_packed.cstep;
                specializations[0 + 4].i = shape_winograd_input_transformed_packed.cstep;
                specializations[0 + 5].i = block_x;
                specializations[0 + 6].i = block_y;

                pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input = new Pipeline(vkdev);
                pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input->set_local_size_xyz(8, 8, 1);
                pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input->create(LayerShaderType::convolution_pack8_3x3s1d1_winograd23_transform_input, opt, specializations);
            }

            {
                std::vector<vk_specialization_type> specializations(0 + 5);
                specializations[0 + 0].i = shape_winograd_input_transformed_packed.c;
                specializations[0 + 1].i = shape_winograd_input_transformed_packed.cstep;
                specializations[0 + 2].i = shape_winograd_gemm_packed.h;
                specializations[0 + 3].i = shape_winograd_gemm_packed.c;
                specializations[0 + 4].i = shape_winograd_gemm_packed.cstep;

                pipeline_convolution_pack8_3x3s1d1_winograd23_gemm = new Pipeline(vkdev);
                pipeline_convolution_pack8_3x3s1d1_winograd23_gemm->set_local_size_xyz(4, 4, std::min(4, num_output / 8));
                pipeline_convolution_pack8_3x3s1d1_winograd23_gemm->create(LayerShaderType::convolution_pack8_3x3s1d1_winograd23_gemm, opt, specializations);
            }

            {
                std::vector<vk_specialization_type> specializations(4 + 7);
                specializations[0].i = bias_term;
                specializations[1].i = activation_type;
                specializations[2].f = activation_params.w >= 1 ? activation_params[0] : 0.f;
                specializations[3].f = activation_params.w == 2 ? activation_params[1] : 0.f;
                specializations[4 + 0].i = shape_winograd_gemm_packed.c;
                specializations[4 + 1].i = shape_winograd_gemm_packed.cstep;
                specializations[4 + 2].i = block_x;
                specializations[4 + 3].i = block_y;
                specializations[4 + 4].i = shape_winograd_out_bordered_packed.w;
                specializations[4 + 5].i = shape_winograd_out_bordered_packed.h;
                specializations[4 + 6].i = shape_winograd_out_bordered_packed.cstep;

                pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output = new Pipeline(vkdev);
                pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output->set_local_size_xyz(8, 8, 1);
                pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output->create(LayerShaderType::convolution_pack8_3x3s1d1_winograd23_transform_output, opt, specializations);
            }
        }
        else
        {
            pipeline_convolution_pack8 = new Pipeline(vkdev);
            pipeline_convolution_pack8->set_optimal_local_size_xyz(local_size_xyz);
            pipeline_convolution_pack8->create(LayerShaderType::convolution_pack8, opt, specializations);
        }
    }

    // pack1to8
    if (elempack == 1 && out_elempack == 8)
    {
        pipeline_convolution_pack1to8 = new Pipeline(vkdev);
        pipeline_convolution_pack1to8->set_optimal_local_size_xyz(local_size_xyz);
        pipeline_convolution_pack1to8->create(LayerShaderType::convolution_pack1to8, opt, specializations);
    }

    // pack4to8
    if (elempack == 4 && out_elempack == 8)
    {
        pipeline_convolution_pack4to8 = new Pipeline(vkdev);
        pipeline_convolution_pack4to8->set_optimal_local_size_xyz(local_size_xyz);
        pipeline_convolution_pack4to8->create(LayerShaderType::convolution_pack4to8, opt, specializations);
    }

    // pack8to4
    if (elempack == 8 && out_elempack == 4)
    {
        pipeline_convolution_pack8to4 = new Pipeline(vkdev);
        pipeline_convolution_pack8to4->set_optimal_local_size_xyz(local_size_xyz);
        pipeline_convolution_pack8to4->create(LayerShaderType::convolution_pack8to4, opt, specializations);
    }

    // pack8to1
    if (elempack == 8 && out_elempack == 1)
    {
        pipeline_convolution_pack8to1 = new Pipeline(vkdev);
        pipeline_convolution_pack8to1->set_optimal_local_size_xyz(local_size_xyz);
        pipeline_convolution_pack8to1->create(LayerShaderType::convolution_pack8to1, opt, specializations);
    }

    return 0;
}

int Convolution_vulkan::destroy_pipeline(const Option& opt)
{
    if (padding)
    {
        padding->destroy_pipeline(opt);
        delete padding;
        padding = 0;
    }

    delete pipeline_convolution;
    pipeline_convolution = 0;

    delete pipeline_convolution_1x1s1d1;
    pipeline_convolution_1x1s1d1 = 0;

    delete pipeline_convolution_pack4;
    pipeline_convolution_pack4 = 0;

    delete pipeline_convolution_pack4_1x1s1d1;
    pipeline_convolution_pack4_1x1s1d1 = 0;

    delete pipeline_convolution_pack1to4;
    pipeline_convolution_pack1to4 = 0;

    delete pipeline_convolution_pack4to1;
    pipeline_convolution_pack4to1 = 0;

    delete pipeline_convolution_pack8;
    pipeline_convolution_pack8 = 0;

    delete pipeline_convolution_pack8_1x1s1d1;
    pipeline_convolution_pack8_1x1s1d1 = 0;

    delete pipeline_convolution_pack1to8;
    pipeline_convolution_pack1to8 = 0;

    delete pipeline_convolution_pack4to8;
    pipeline_convolution_pack4to8 = 0;

    delete pipeline_convolution_pack8to4;
    pipeline_convolution_pack8to4 = 0;

    delete pipeline_convolution_pack8to1;
    pipeline_convolution_pack8to1 = 0;

    if (winograd_padding)
    {
        winograd_padding->destroy_pipeline(opt);
        delete winograd_padding;
        winograd_padding = 0;
    }

    if (winograd_crop)
    {
        winograd_crop->destroy_pipeline(opt);
        delete winograd_crop;
        winograd_crop = 0;
    }

    delete pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input;
    delete pipeline_convolution_pack4_3x3s1d1_winograd23_gemm;
    delete pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output;
    pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input = 0;
    pipeline_convolution_pack4_3x3s1d1_winograd23_gemm = 0;
    pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output = 0;

    delete pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input;
    delete pipeline_convolution_pack8_3x3s1d1_winograd23_gemm;
    delete pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output;
    pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input = 0;
    pipeline_convolution_pack8_3x3s1d1_winograd23_gemm = 0;
    pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output = 0;

    // fc
    if (innerproduct)
    {
        innerproduct->destroy_pipeline(opt);
        delete innerproduct;
        innerproduct = 0;
    }

    return 0;
}

int Convolution_vulkan::upload_model(VkTransfer& cmd, const Option& opt)
{
    if (padding)
    {
        padding->upload_model(cmd, opt);
    }

    if (winograd_padding)
    {
        winograd_padding->upload_model(cmd, opt);
    }

    if (winograd_crop)
    {
        winograd_crop->upload_model(cmd, opt);
    }

    const int maxk = kernel_w * kernel_h;
    int num_input = weight_data_size / maxk / num_output;

    int elempack = opt.use_shader_pack8 && num_input % 8 == 0 ? 8 : num_input % 4 == 0 ? 4 : 1;
    int out_elempack = opt.use_shader_pack8 && num_output % 8 == 0 ? 8 : num_output % 4 == 0 ? 4 : 1;

    // src = kw-kh-inch-outch
    // dst = pa-pb-kw-kh-inch/pa-outch/pb
    Mat weight_data_packed;
    {
        Mat weight_data_r2 = weight_data.reshape(maxk, num_input, num_output);

        weight_data_packed.create(maxk, num_input/elempack, num_output/out_elempack, (size_t)4*elempack*out_elempack, elempack*out_elempack);

        for (int q=0; q+(out_elempack-1)<num_output; q+=out_elempack)
        {
            Mat g0 = weight_data_packed.channel(q/out_elempack);

            for (int p=0; p+(elempack-1)<num_input; p+=elempack)
            {
                float* g00 = g0.row(p/elempack);

                for (int k=0; k<maxk; k++)
                {

                    for (int i=0; i<out_elempack; i++)
                    {
                        const Mat k0 = weight_data_r2.channel(q+i);

                        for (int j=0; j<elempack; j++)
                        {
                            const float* k00 = k0.row(p+j);

                            g00[0] = k00[k];

                            g00++;
                        }
                    }
                }
            }
        }
    }

    if (opt.use_image_storage)
    {
        cmd.record_upload(weight_data_packed, weight_data_gpu_image, opt);
    }
    else
    {
        cmd.record_upload(weight_data_packed, weight_data_gpu, opt);
    }

    bool is_conv3x3s1d1 = kernel_w == 3 && kernel_h == 3 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1;

    // pack4
    if (elempack == 4 && out_elempack == 4)
    {
        if (is_conv3x3s1d1 && num_input >= 16 && num_output >= 16)
        {
            // winograd23 transform kernel
            Mat weight_data_tm;
            weight_data_tm.create(4*4, num_input, num_output);

            // G
            const float ktm[4][3] = {
                {   1.0f,     0.0f,     0.0f},
                { 1.0f/2,   1.0f/2,   1.0f/2},
                { 1.0f/2,  -1.0f/2,   1.0f/2},
                {   0.0f,     0.0f,     1.0f}
            };

            #pragma omp parallel for
            for (int p = 0; p<num_output; p++)
            {
                for (int q = 0; q<num_input; q++)
                {
                    const float* kernel0 = (const float*)weight_data + p*num_input * 9 + q * 9;
                    float* kernel_tm0 = weight_data_tm.channel(p).row(q);

                    // transform kernel
                    const float* k0 = kernel0;
                    const float* k1 = kernel0 + 3;
                    const float* k2 = kernel0 + 6;

                    // h
                    float tmp[4][3];
                    for (int i=0; i<4; i++)
                    {
                        tmp[i][0] = k0[0] * ktm[i][0] + k0[1] * ktm[i][1] + k0[2] * ktm[i][2];
                        tmp[i][1] = k1[0] * ktm[i][0] + k1[1] * ktm[i][1] + k1[2] * ktm[i][2];
                        tmp[i][2] = k2[0] * ktm[i][0] + k2[1] * ktm[i][1] + k2[2] * ktm[i][2];
                    }

                    // U
                    for (int j=0; j<4; j++)
                    {
                        float* tmpp = &tmp[j][0];

                        for (int i=0; i<4; i++)
                        {
                            kernel_tm0[j*4 + i] = tmpp[0] * ktm[i][0] + tmpp[1] * ktm[i][1] + tmpp[2] * ktm[i][2];
                        }
                    }
                }
            }

            // src = 16-inch-outch
            // dst = 4a-4b-16-inch/4a-outch/4b
            Mat weight_data_pack4_tm;
            {
                weight_data_pack4_tm.create(16, num_input/4, num_output/4, (size_t)4*16, 16);

                for (int q=0; q+3<num_output; q+=4)
                {
                    const Mat k0 = weight_data_tm.channel(q);
                    const Mat k1 = weight_data_tm.channel(q+1);
                    const Mat k2 = weight_data_tm.channel(q+2);
                    const Mat k3 = weight_data_tm.channel(q+3);

                    Mat g0 = weight_data_pack4_tm.channel(q/4);

                    for (int p=0; p+3<num_input; p+=4)
                    {
                        const float* k00 = k0.row(p);
                        const float* k01 = k0.row(p+1);
                        const float* k02 = k0.row(p+2);
                        const float* k03 = k0.row(p+3);

                        const float* k10 = k1.row(p);
                        const float* k11 = k1.row(p+1);
                        const float* k12 = k1.row(p+2);
                        const float* k13 = k1.row(p+3);

                        const float* k20 = k2.row(p);
                        const float* k21 = k2.row(p+1);
                        const float* k22 = k2.row(p+2);
                        const float* k23 = k2.row(p+3);

                        const float* k30 = k3.row(p);
                        const float* k31 = k3.row(p+1);
                        const float* k32 = k3.row(p+2);
                        const float* k33 = k3.row(p+3);

                        float* g00 = g0.row(p/4);

                        for (int k=0; k<16; k++)
                        {
                            g00[0] = k00[k];
                            g00[1] = k01[k];
                            g00[2] = k02[k];
                            g00[3] = k03[k];

                            g00[4] = k10[k];
                            g00[5] = k11[k];
                            g00[6] = k12[k];
                            g00[7] = k13[k];

                            g00[8] = k20[k];
                            g00[9] = k21[k];
                            g00[10] = k22[k];
                            g00[11] = k23[k];

                            g00[12] = k30[k];
                            g00[13] = k31[k];
                            g00[14] = k32[k];
                            g00[15] = k33[k];

                            g00 += 16;
                        }
                    }
                }
            }

            if (opt.use_image_storage)
            {
                cmd.record_upload(weight_data_pack4_tm, weight_data_gpu_pack4_tm_image, opt);
            }
            else
            {
                cmd.record_upload(weight_data_pack4_tm, weight_data_gpu_pack4_tm, opt);
            }
        }
    }

    // pack8
    if (elempack == 8 && out_elempack == 8)
    {
        if (is_conv3x3s1d1 && num_input >= 16 && num_output >= 16)
        {
            // winograd23 transform kernel
            Mat weight_data_tm;
            weight_data_tm.create(4*4, num_input, num_output);

            // G
            const float ktm[4][3] = {
                {   1.0f,     0.0f,     0.0f},
                { 1.0f/2,   1.0f/2,   1.0f/2},
                { 1.0f/2,  -1.0f/2,   1.0f/2},
                {   0.0f,     0.0f,     1.0f}
            };

            #pragma omp parallel for
            for (int p = 0; p<num_output; p++)
            {
                for (int q = 0; q<num_input; q++)
                {
                    const float* kernel0 = (const float*)weight_data + p*num_input * 9 + q * 9;
                    float* kernel_tm0 = weight_data_tm.channel(p).row(q);

                    // transform kernel
                    const float* k0 = kernel0;
                    const float* k1 = kernel0 + 3;
                    const float* k2 = kernel0 + 6;

                    // h
                    float tmp[4][3];
                    for (int i=0; i<4; i++)
                    {
                        tmp[i][0] = k0[0] * ktm[i][0] + k0[1] * ktm[i][1] + k0[2] * ktm[i][2];
                        tmp[i][1] = k1[0] * ktm[i][0] + k1[1] * ktm[i][1] + k1[2] * ktm[i][2];
                        tmp[i][2] = k2[0] * ktm[i][0] + k2[1] * ktm[i][1] + k2[2] * ktm[i][2];
                    }

                    // U
                    for (int j=0; j<4; j++)
                    {
                        float* tmpp = &tmp[j][0];

                        for (int i=0; i<4; i++)
                        {
                            kernel_tm0[j*4 + i] = tmpp[0] * ktm[i][0] + tmpp[1] * ktm[i][1] + tmpp[2] * ktm[i][2];
                        }
                    }
                }
            }

            // src = 64-inch-outch
            // dst = 8a-8b-16-inch/8a-outch/8b
            Mat weight_data_pack8_tm;
            {
                weight_data_pack8_tm.create(16, num_input/8, num_output/8, (size_t)4*64, 64);

                for (int q=0; q+7<num_output; q+=8)
                {
                    Mat g0 = weight_data_pack8_tm.channel(q/8);

                    for (int p=0; p+7<num_input; p+=8)
                    {
                        float* g00 = g0.row(p/8);

                        for (int k=0; k<16; k++)
                        {
                            for (int i=0; i<8; i++)
                            {
                                const Mat k0 = weight_data_tm.channel(q+i);

                                for (int j=0; j<8; j++)
                                {
                                    const float* k00 = k0.row(p+j);

                                    g00[0] = k00[k];

                                    g00++;
                                }
                            }
                        }
                    }
                }
            }

            if (opt.use_image_storage)
            {
                cmd.record_upload(weight_data_pack8_tm, weight_data_gpu_pack8_tm_image, opt);
            }
            else
            {
                cmd.record_upload(weight_data_pack8_tm, weight_data_gpu_pack8_tm, opt);
            }
        }
    }

    if (bias_term)
    {
        Mat bias_data_packed;
        convert_packing(bias_data, bias_data_packed, out_elempack);

        if (opt.use_image_storage)
        {
            cmd.record_upload(bias_data_packed, bias_data_gpu_image, opt);
        }
        else
        {
            cmd.record_upload(bias_data_packed, bias_data_gpu, opt);
        }
    }

    if (innerproduct)
    {
        innerproduct->upload_model(cmd, opt);
    }

    return 0;
}

int Convolution_vulkan::forward(const VkMat& bottom_blob, VkMat& top_blob, VkCompute& cmd, const Option& opt) const
{
    int w = bottom_blob.w;
    int h = bottom_blob.h;
    int channels = bottom_blob.c;
    size_t elemsize = bottom_blob.elemsize;
    int elempack = bottom_blob.elempack;

    // flattened blob, implement as InnerProduct
    if (bottom_blob.dims == 1 && kernel_w == 1 && kernel_h == 1)
    {
        int num_input = weight_data_size / num_output;
        if (bottom_blob.w * bottom_blob.elempack == num_input)
        {
            return innerproduct->forward(bottom_blob, top_blob, cmd, opt);
        }
    }

    const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1;
    const int kernel_extent_h = dilation_h * (kernel_h - 1) + 1;

    VkMat bottom_blob_bordered = bottom_blob;
    if (pad_left > 0 || pad_right > 0 || pad_top > 0 || pad_bottom > 0)
    {
        Option opt_pad = opt;
        opt_pad.blob_vkallocator = opt.workspace_vkallocator;

        padding->forward(bottom_blob, bottom_blob_bordered, cmd, opt_pad);
    }
    else if (pad_left == -233 && pad_right == -233 && pad_top == -233 && pad_bottom == -233)
    {
        int wpad = kernel_extent_w + (w - 1) / stride_w * stride_w - w;
        int hpad = kernel_extent_h + (h - 1) / stride_h * stride_h - h;
        if (wpad > 0 || hpad > 0)
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = hpad / 2;
            padding_params[1] = hpad - hpad / 2;
            padding_params[2] = wpad / 2;
            padding_params[3] = wpad - wpad / 2;

            std::vector<VkMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkMat> padding_outputs(1);
            padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }
    }
    else if (pad_left == -234 && pad_right == -234 && pad_top == -234 && pad_bottom == -234)
    {
        int wpad = kernel_extent_w + (w - 1) / stride_w * stride_w - w;
        int hpad = kernel_extent_h + (h - 1) / stride_h * stride_h - h;
        if (wpad > 0 || hpad > 0)
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = hpad - hpad / 2;
            padding_params[1] = hpad / 2;
            padding_params[2] = wpad - wpad / 2;
            padding_params[3] = wpad / 2;

            std::vector<VkMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkMat> padding_outputs(1);
            padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }
    }

    w = bottom_blob_bordered.w;
    h = bottom_blob_bordered.h;

    int outw = (w - kernel_extent_w) / stride_w + 1;
    int outh = (h - kernel_extent_h) / stride_h + 1;
    int out_elempack = opt.use_shader_pack8 && num_output % 8 == 0 ? 8 : num_output % 4 == 0 ? 4 : 1;
    size_t out_elemsize = elemsize / elempack * out_elempack;

    if (opt.use_fp16_packed && !opt.use_fp16_storage)
    {
        if (out_elempack == 8) out_elemsize = 8*2u;
        if (out_elempack == 4) out_elemsize = 4*2u;
        if (out_elempack == 1) out_elemsize = 4u;
    }

    bool is_conv3x3s1d1 = kernel_w == 3 && kernel_h == 3 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1;

    if (elempack == 4 && out_elempack == 4 && is_conv3x3s1d1 && channels * elempack >= 16 && num_output >= 16)
    {
        // winograd23
        int outw_bordered = (outw + 1) / 2 * 2;
        int outh_bordered = (outh + 1) / 2 * 2;

        int w_bordered = outw_bordered + 2;
        int h_bordered = outh_bordered + 2;

        int block_x = outw_bordered / 2;
        int block_y = outh_bordered / 2;

        // pad to 2n+2
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = 0;
            padding_params[1] = h_bordered - bottom_blob_bordered.h;
            padding_params[2] = 0;
            padding_params[3] = w_bordered - bottom_blob_bordered.w;

            std::vector<VkMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob_bordered;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkMat> padding_outputs(1);
            winograd_padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }

        // transform input
        VkMat bottom_tm_blob;
        {
            bottom_tm_blob.create(16, block_x * block_y, channels, elemsize, elempack, opt.workspace_vkallocator);
            if (bottom_tm_blob.empty())
                return -100;

            std::vector<VkMat> bindings(2);
            bindings[0] = bottom_blob_bordered;
            bindings[1] = bottom_tm_blob;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = bottom_blob_bordered.w;
            constants[1].i = bottom_blob_bordered.h;
            constants[2].i = bottom_blob_bordered.c;
            constants[3].i = bottom_blob_bordered.cstep;
            constants[4].i = bottom_tm_blob.cstep;
            constants[5].i = block_x;
            constants[6].i = block_y;

            VkMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = bottom_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input, bindings, constants, dispatcher);
        }

        // gemm
        VkMat top_tm_blob;
        {
            top_tm_blob.create(16, block_x * block_y, num_output / out_elempack, out_elemsize, out_elempack, opt.workspace_vkallocator);
            if (top_tm_blob.empty())
                return -100;

            std::vector<VkMat> bindings(3);
            bindings[0] = bottom_tm_blob;
            bindings[1] = top_tm_blob;
            bindings[2] = weight_data_gpu_pack4_tm;

            std::vector<vk_constant_type> constants(5);
            constants[0].i = bottom_tm_blob.c;
            constants[1].i = bottom_tm_blob.cstep;
            constants[2].i = top_tm_blob.h;
            constants[3].i = top_tm_blob.c;
            constants[4].i = top_tm_blob.cstep;

            VkMat dispatcher;
            dispatcher.w = top_tm_blob.w;
            dispatcher.h = (top_tm_blob.h + 3) / 4;
            dispatcher.c = top_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack4_3x3s1d1_winograd23_gemm, bindings, constants, dispatcher);
        }

        // transform output
        VkMat top_blob_bordered;
        {
            top_blob_bordered.create(outw_bordered, outh_bordered, num_output / out_elempack, out_elemsize, out_elempack, opt.blob_vkallocator);
            if (top_blob_bordered.empty())
                return -100;

            std::vector<VkMat> bindings(3);
            bindings[0] = top_tm_blob;
            bindings[1] = top_blob_bordered;
            bindings[2] = bias_data_gpu;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = top_tm_blob.c;
            constants[1].i = top_tm_blob.cstep;
            constants[2].i = block_x;
            constants[3].i = block_y;
            constants[4].i = top_blob_bordered.w;
            constants[5].i = top_blob_bordered.h;
            constants[6].i = top_blob_bordered.cstep;

            VkMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = top_blob_bordered.c;

            cmd.record_pipeline(pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output, bindings, constants, dispatcher);
        }

        // crop top_blob
        {
            VkMat crop_param_blob(6, (size_t)4u, 1, opt.staging_vkallocator);
            int* crop_params = crop_param_blob.mapped();

            crop_params[0] = 0;
            crop_params[1] = 0;
            crop_params[2] = 0;
            crop_params[3] = outw;
            crop_params[4] = outh;
            crop_params[5] = num_output;

            std::vector<VkMat> crop_inputs(2);
            crop_inputs[0] = top_blob_bordered;
            crop_inputs[1] = crop_param_blob;

            std::vector<VkMat> crop_outputs(1);
            winograd_crop->forward(crop_inputs, crop_outputs, cmd, opt);
            top_blob = crop_outputs[0];
        }

        return 0;
    }
    if (elempack == 8 && out_elempack == 8 && is_conv3x3s1d1 && channels * elempack >= 16 && num_output >= 16)
    {
        // winograd23
        int outw_bordered = (outw + 1) / 2 * 2;
        int outh_bordered = (outh + 1) / 2 * 2;

        int w_bordered = outw_bordered + 2;
        int h_bordered = outh_bordered + 2;

        int block_x = outw_bordered / 2;
        int block_y = outh_bordered / 2;

        // pad to 2n+2
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = 0;
            padding_params[1] = h_bordered - bottom_blob_bordered.h;
            padding_params[2] = 0;
            padding_params[3] = w_bordered - bottom_blob_bordered.w;

            std::vector<VkMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob_bordered;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkMat> padding_outputs(1);
            winograd_padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }

        // transform input
        VkMat bottom_tm_blob;
        {
            bottom_tm_blob.create(16, block_x * block_y, channels, elemsize, elempack, opt.workspace_vkallocator);
            if (bottom_tm_blob.empty())
                return -100;

            std::vector<VkMat> bindings(2);
            bindings[0] = bottom_blob_bordered;
            bindings[1] = bottom_tm_blob;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = bottom_blob_bordered.w;
            constants[1].i = bottom_blob_bordered.h;
            constants[2].i = bottom_blob_bordered.c;
            constants[3].i = bottom_blob_bordered.cstep;
            constants[4].i = bottom_tm_blob.cstep;
            constants[5].i = block_x;
            constants[6].i = block_y;

            VkMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = bottom_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input, bindings, constants, dispatcher);
        }

        // gemm
        VkMat top_tm_blob;
        {
            top_tm_blob.create(16, block_x * block_y, num_output / out_elempack, out_elemsize, out_elempack, opt.workspace_vkallocator);
            if (top_tm_blob.empty())
                return -100;

            std::vector<VkMat> bindings(3);
            bindings[0] = bottom_tm_blob;
            bindings[1] = top_tm_blob;
            bindings[2] = weight_data_gpu_pack8_tm;

            std::vector<vk_constant_type> constants(5);
            constants[0].i = bottom_tm_blob.c;
            constants[1].i = bottom_tm_blob.cstep;
            constants[2].i = top_tm_blob.h;
            constants[3].i = top_tm_blob.c;
            constants[4].i = top_tm_blob.cstep;

            VkMat dispatcher;
            dispatcher.w = top_tm_blob.w;
            dispatcher.h = (top_tm_blob.h + 3) / 4;
            dispatcher.c = top_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack8_3x3s1d1_winograd23_gemm, bindings, constants, dispatcher);
        }

        // transform output
        VkMat top_blob_bordered;
        {
            top_blob_bordered.create(outw_bordered, outh_bordered, num_output / out_elempack, out_elemsize, out_elempack, opt.blob_vkallocator);
            if (top_blob_bordered.empty())
                return -100;

            std::vector<VkMat> bindings(3);
            bindings[0] = top_tm_blob;
            bindings[1] = top_blob_bordered;
            bindings[2] = bias_data_gpu;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = top_tm_blob.c;
            constants[1].i = top_tm_blob.cstep;
            constants[2].i = block_x;
            constants[3].i = block_y;
            constants[4].i = top_blob_bordered.w;
            constants[5].i = top_blob_bordered.h;
            constants[6].i = top_blob_bordered.cstep;

            VkMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = top_blob_bordered.c;

            cmd.record_pipeline(pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output, bindings, constants, dispatcher);
        }

        // crop top_blob
        {
            VkMat crop_param_blob(6, (size_t)4u, 1, opt.staging_vkallocator);
            int* crop_params = crop_param_blob.mapped();

            crop_params[0] = 0;
            crop_params[1] = 0;
            crop_params[2] = 0;
            crop_params[3] = outw;
            crop_params[4] = outh;
            crop_params[5] = num_output;

            std::vector<VkMat> crop_inputs(2);
            crop_inputs[0] = top_blob_bordered;
            crop_inputs[1] = crop_param_blob;

            std::vector<VkMat> crop_outputs(1);
            winograd_crop->forward(crop_inputs, crop_outputs, cmd, opt);
            top_blob = crop_outputs[0];
        }

        return 0;
    }

    top_blob.create(outw, outh, num_output / out_elempack, out_elemsize, out_elempack, opt.blob_vkallocator);
    if (top_blob.empty())
        return -100;

    std::vector<VkMat> bindings(4);
    bindings[0] = bottom_blob_bordered;
    bindings[1] = top_blob;
    bindings[2] = weight_data_gpu;
    bindings[3] = bias_data_gpu;

    std::vector<vk_constant_type> constants(10);
    constants[0].i = bottom_blob_bordered.dims;
    constants[1].i = bottom_blob_bordered.w;
    constants[2].i = bottom_blob_bordered.h;
    constants[3].i = bottom_blob_bordered.c;
    constants[4].i = bottom_blob_bordered.cstep;
    constants[5].i = top_blob.dims;
    constants[6].i = top_blob.w;
    constants[7].i = top_blob.h;
    constants[8].i = top_blob.c;
    constants[9].i = top_blob.cstep;

    // record
    if (elempack == 1 && out_elempack == 1 && kernel_w == 1 && kernel_h == 1 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1)
    {
        VkMat dispatcher;
        dispatcher.w = top_blob.cstep / 4;
        dispatcher.h = 1;
        dispatcher.c = top_blob.c;

        cmd.record_pipeline(pipeline_convolution_1x1s1d1, bindings, constants, dispatcher);
    }
    else if (elempack == 4 && out_elempack == 4 && kernel_w == 1 && kernel_h == 1 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1)
    {
        VkMat dispatcher;
        dispatcher.w = (top_blob.w * top_blob.h + 3) / 4;
        dispatcher.h = 1;
        dispatcher.c = top_blob.c;

        cmd.record_pipeline(pipeline_convolution_pack4_1x1s1d1, bindings, constants, dispatcher);
    }
    else if (elempack == 8 && out_elempack == 8 && kernel_w == 1 && kernel_h == 1 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1)
    {
        VkMat dispatcher;
        dispatcher.w = (top_blob.w * top_blob.h + 3) / 4;
        dispatcher.h = 1;
        dispatcher.c = top_blob.c;

        cmd.record_pipeline(pipeline_convolution_pack8_1x1s1d1, bindings, constants, dispatcher);
    }
    else
    {
        const Pipeline* pipeline = 0;
        if (elempack == 1 && out_elempack == 1)
        {
            pipeline = pipeline_convolution;
        }
        else if (elempack == 4 && out_elempack == 4)
        {
            pipeline = pipeline_convolution_pack4;
        }
        else if (elempack == 1 && out_elempack == 4)
        {
            pipeline = pipeline_convolution_pack1to4;
        }
        else if (elempack == 4 && out_elempack == 1)
        {
            pipeline = pipeline_convolution_pack4to1;
        }
        else if (elempack == 8 && out_elempack == 8)
        {
            pipeline = pipeline_convolution_pack8;
        }
        else if (elempack == 1 && out_elempack == 8)
        {
            pipeline = pipeline_convolution_pack1to8;
        }
        else if (elempack == 4 && out_elempack == 8)
        {
            pipeline = pipeline_convolution_pack4to8;
        }
        else if (elempack == 8 && out_elempack == 4)
        {
            pipeline = pipeline_convolution_pack8to4;
        }
        else if (elempack == 8 && out_elempack == 1)
        {
            pipeline = pipeline_convolution_pack8to1;
        }

        cmd.record_pipeline(pipeline, bindings, constants, top_blob);
    }

    return 0;
}

int Convolution_vulkan::forward(const VkImageMat& bottom_blob, VkImageMat& top_blob, VkCompute& cmd, const Option& opt) const
{
    int w = bottom_blob.w;
    int h = bottom_blob.h;
    int channels = bottom_blob.c;
    size_t elemsize = bottom_blob.elemsize;
    int elempack = bottom_blob.elempack;

    // flattened blob, implement as InnerProduct
    if (bottom_blob.dims == 1 && kernel_w == 1 && kernel_h == 1)
    {
        int num_input = weight_data_size / num_output;
        if (bottom_blob.w * bottom_blob.elempack == num_input)
        {
            return innerproduct->forward(bottom_blob, top_blob, cmd, opt);
        }
    }

    const int kernel_extent_w = dilation_w * (kernel_w - 1) + 1;
    const int kernel_extent_h = dilation_h * (kernel_h - 1) + 1;

    VkImageMat bottom_blob_bordered = bottom_blob;
    if (pad_left > 0 || pad_right > 0 || pad_top > 0 || pad_bottom > 0)
    {
        Option opt_pad = opt;
        opt_pad.blob_vkallocator = opt.workspace_vkallocator;

        padding->forward(bottom_blob, bottom_blob_bordered, cmd, opt_pad);
    }
    else if (pad_left == -233 && pad_right == -233 && pad_top == -233 && pad_bottom == -233)
    {
        int wpad = kernel_extent_w + (w - 1) / stride_w * stride_w - w;
        int hpad = kernel_extent_h + (h - 1) / stride_h * stride_h - h;
        if (wpad > 0 || hpad > 0)
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkImageMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = hpad / 2;
            padding_params[1] = hpad - hpad / 2;
            padding_params[2] = wpad / 2;
            padding_params[3] = wpad - wpad / 2;

            std::vector<VkImageMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkImageMat> padding_outputs(1);
            padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }
    }
    else if (pad_left == -234 && pad_right == -234 && pad_top == -234 && pad_bottom == -234)
    {
        int wpad = kernel_extent_w + (w - 1) / stride_w * stride_w - w;
        int hpad = kernel_extent_h + (h - 1) / stride_h * stride_h - h;
        if (wpad > 0 || hpad > 0)
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkImageMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = hpad - hpad / 2;
            padding_params[1] = hpad / 2;
            padding_params[2] = wpad - wpad / 2;
            padding_params[3] = wpad / 2;

            std::vector<VkImageMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkImageMat> padding_outputs(1);
            padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }
    }

    w = bottom_blob_bordered.w;
    h = bottom_blob_bordered.h;

    int outw = (w - kernel_extent_w) / stride_w + 1;
    int outh = (h - kernel_extent_h) / stride_h + 1;
    int out_elempack = opt.use_shader_pack8 && num_output % 8 == 0 ? 8 : num_output % 4 == 0 ? 4 : 1;
    size_t out_elemsize = elemsize / elempack * out_elempack;

    if (opt.use_fp16_packed && !opt.use_fp16_storage)
    {
        if (out_elempack == 8) out_elemsize = 8*2u;
        if (out_elempack == 4) out_elemsize = 4*2u;
        if (out_elempack == 1) out_elemsize = 4u;
    }

    bool is_conv3x3s1d1 = kernel_w == 3 && kernel_h == 3 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1;

    if (elempack == 4 && out_elempack == 4 && is_conv3x3s1d1 && channels * elempack >= 16 && num_output >= 16)
    {
        // winograd23
        int outw_bordered = (outw + 1) / 2 * 2;
        int outh_bordered = (outh + 1) / 2 * 2;

        int w_bordered = outw_bordered + 2;
        int h_bordered = outh_bordered + 2;

        int block_x = outw_bordered / 2;
        int block_y = outh_bordered / 2;

        // pad to 2n+2
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkImageMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = 0;
            padding_params[1] = h_bordered - bottom_blob_bordered.h;
            padding_params[2] = 0;
            padding_params[3] = w_bordered - bottom_blob_bordered.w;

            std::vector<VkImageMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob_bordered;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkImageMat> padding_outputs(1);
            winograd_padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }

        // transform input
        VkImageMat bottom_tm_blob;
        {
            bottom_tm_blob.create(16, block_x * block_y, channels, elemsize, elempack, opt.workspace_vkallocator);
            if (bottom_tm_blob.empty())
                return -100;

            std::vector<VkImageMat> bindings(2);
            bindings[0] = bottom_blob_bordered;
            bindings[1] = bottom_tm_blob;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = bottom_blob_bordered.w;
            constants[1].i = bottom_blob_bordered.h;
            constants[2].i = bottom_blob_bordered.c;
            constants[3].i = 0;//bottom_blob_bordered.cstep;
            constants[4].i = 0;//bottom_tm_blob.cstep;
            constants[5].i = block_x;
            constants[6].i = block_y;

            VkImageMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = bottom_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack4_3x3s1d1_winograd23_transform_input, bindings, constants, dispatcher);
        }

        // gemm
        VkImageMat top_tm_blob;
        {
            top_tm_blob.create(16, block_x * block_y, num_output / out_elempack, out_elemsize, out_elempack, opt.workspace_vkallocator);
            if (top_tm_blob.empty())
                return -100;

            std::vector<VkImageMat> bindings(3);
            bindings[0] = bottom_tm_blob;
            bindings[1] = top_tm_blob;
            bindings[2] = weight_data_gpu_pack4_tm_image;

            std::vector<vk_constant_type> constants(5);
            constants[0].i = bottom_tm_blob.c;
            constants[1].i = 0;//bottom_tm_blob.cstep;
            constants[2].i = top_tm_blob.h;
            constants[3].i = top_tm_blob.c;
            constants[4].i = 0;//top_tm_blob.cstep;

            VkImageMat dispatcher;
            dispatcher.w = top_tm_blob.w;
            dispatcher.h = (top_tm_blob.h + 3) / 4;
            dispatcher.c = top_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack4_3x3s1d1_winograd23_gemm, bindings, constants, dispatcher);
        }

        // transform output
        VkImageMat top_blob_bordered;
        {
            top_blob_bordered.create(outw_bordered, outh_bordered, num_output / out_elempack, out_elemsize, out_elempack, opt.blob_vkallocator);
            if (top_blob_bordered.empty())
                return -100;

            std::vector<VkImageMat> bindings(3);
            bindings[0] = top_tm_blob;
            bindings[1] = top_blob_bordered;
            bindings[2] = bias_data_gpu_image;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = top_tm_blob.c;
            constants[1].i = 0;//top_tm_blob.cstep;
            constants[2].i = block_x;
            constants[3].i = block_y;
            constants[4].i = top_blob_bordered.w;
            constants[5].i = top_blob_bordered.h;
            constants[6].i = 0;//top_blob_bordered.cstep;

            VkImageMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = top_blob_bordered.c;

            cmd.record_pipeline(pipeline_convolution_pack4_3x3s1d1_winograd23_transform_output, bindings, constants, dispatcher);
        }

        // crop top_blob
        {
            VkImageMat crop_param_blob(6, (size_t)4u, 1, opt.staging_vkallocator);
            int* crop_params = crop_param_blob.mapped();

            crop_params[0] = 0;
            crop_params[1] = 0;
            crop_params[2] = 0;
            crop_params[3] = outw;
            crop_params[4] = outh;
            crop_params[5] = num_output;

            std::vector<VkImageMat> crop_inputs(2);
            crop_inputs[0] = top_blob_bordered;
            crop_inputs[1] = crop_param_blob;

            std::vector<VkImageMat> crop_outputs(1);
            winograd_crop->forward(crop_inputs, crop_outputs, cmd, opt);
            top_blob = crop_outputs[0];
        }

        return 0;
    }
    if (elempack == 8 && out_elempack == 8 && is_conv3x3s1d1 && channels * elempack >= 16 && num_output >= 16)
    {
        // winograd23
        int outw_bordered = (outw + 1) / 2 * 2;
        int outh_bordered = (outh + 1) / 2 * 2;

        int w_bordered = outw_bordered + 2;
        int h_bordered = outh_bordered + 2;

        int block_x = outw_bordered / 2;
        int block_y = outh_bordered / 2;

        // pad to 2n+2
        {
            Option opt_pad = opt;
            opt_pad.blob_vkallocator = opt.workspace_vkallocator;

            VkImageMat padding_param_blob(4, (size_t)4u, 1, opt.staging_vkallocator);
            int* padding_params = padding_param_blob.mapped();

            padding_params[0] = 0;
            padding_params[1] = h_bordered - bottom_blob_bordered.h;
            padding_params[2] = 0;
            padding_params[3] = w_bordered - bottom_blob_bordered.w;

            std::vector<VkImageMat> padding_inputs(2);
            padding_inputs[0] = bottom_blob_bordered;
            padding_inputs[1] = padding_param_blob;

            std::vector<VkImageMat> padding_outputs(1);
            winograd_padding->forward(padding_inputs, padding_outputs, cmd, opt_pad);
            bottom_blob_bordered = padding_outputs[0];
        }

        // transform input
        VkImageMat bottom_tm_blob;
        {
            bottom_tm_blob.create(16, block_x * block_y, channels, elemsize, elempack, opt.workspace_vkallocator);
            if (bottom_tm_blob.empty())
                return -100;

            std::vector<VkImageMat> bindings(2);
            bindings[0] = bottom_blob_bordered;
            bindings[1] = bottom_tm_blob;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = bottom_blob_bordered.w;
            constants[1].i = bottom_blob_bordered.h;
            constants[2].i = bottom_blob_bordered.c;
            constants[3].i = 0;//bottom_blob_bordered.cstep;
            constants[4].i = 0;//bottom_tm_blob.cstep;
            constants[5].i = block_x;
            constants[6].i = block_y;

            VkImageMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = bottom_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack8_3x3s1d1_winograd23_transform_input, bindings, constants, dispatcher);
        }

        // gemm
        VkImageMat top_tm_blob;
        {
            top_tm_blob.create(16, block_x * block_y, num_output / out_elempack, out_elemsize, out_elempack, opt.workspace_vkallocator);
            if (top_tm_blob.empty())
                return -100;

            std::vector<VkImageMat> bindings(3);
            bindings[0] = bottom_tm_blob;
            bindings[1] = top_tm_blob;
            bindings[2] = weight_data_gpu_pack8_tm_image;

            std::vector<vk_constant_type> constants(5);
            constants[0].i = bottom_tm_blob.c;
            constants[1].i = 0;//bottom_tm_blob.cstep;
            constants[2].i = top_tm_blob.h;
            constants[3].i = top_tm_blob.c;
            constants[4].i = 0;//top_tm_blob.cstep;

            VkImageMat dispatcher;
            dispatcher.w = top_tm_blob.w;
            dispatcher.h = (top_tm_blob.h + 3) / 4;
            dispatcher.c = top_tm_blob.c;

            cmd.record_pipeline(pipeline_convolution_pack8_3x3s1d1_winograd23_gemm, bindings, constants, dispatcher);
        }

        // transform output
        VkImageMat top_blob_bordered;
        {
            top_blob_bordered.create(outw_bordered, outh_bordered, num_output / out_elempack, out_elemsize, out_elempack, opt.blob_vkallocator);
            if (top_blob_bordered.empty())
                return -100;

            std::vector<VkImageMat> bindings(3);
            bindings[0] = top_tm_blob;
            bindings[1] = top_blob_bordered;
            bindings[2] = bias_data_gpu_image;

            std::vector<vk_constant_type> constants(7);
            constants[0].i = top_tm_blob.c;
            constants[1].i = 0;//top_tm_blob.cstep;
            constants[2].i = block_x;
            constants[3].i = block_y;
            constants[4].i = top_blob_bordered.w;
            constants[5].i = top_blob_bordered.h;
            constants[6].i = 0;//top_blob_bordered.cstep;

            VkImageMat dispatcher;
            dispatcher.w = block_x;
            dispatcher.h = block_y;
            dispatcher.c = top_blob_bordered.c;

            cmd.record_pipeline(pipeline_convolution_pack8_3x3s1d1_winograd23_transform_output, bindings, constants, dispatcher);
        }

        // crop top_blob
        {
            VkImageMat crop_param_blob(6, (size_t)4u, 1, opt.staging_vkallocator);
            int* crop_params = crop_param_blob.mapped();

            crop_params[0] = 0;
            crop_params[1] = 0;
            crop_params[2] = 0;
            crop_params[3] = outw;
            crop_params[4] = outh;
            crop_params[5] = num_output;

            std::vector<VkImageMat> crop_inputs(2);
            crop_inputs[0] = top_blob_bordered;
            crop_inputs[1] = crop_param_blob;

            std::vector<VkImageMat> crop_outputs(1);
            winograd_crop->forward(crop_inputs, crop_outputs, cmd, opt);
            top_blob = crop_outputs[0];
        }

        return 0;
    }

    top_blob.create(outw, outh, num_output / out_elempack, out_elemsize, out_elempack, opt.blob_vkallocator);
    if (top_blob.empty())
        return -100;

    std::vector<VkImageMat> bindings(4);
    bindings[0] = bottom_blob_bordered;
    bindings[1] = top_blob;
    bindings[2] = weight_data_gpu_image;
    bindings[3] = bias_data_gpu_image;

    std::vector<vk_constant_type> constants(10);
    constants[0].i = bottom_blob_bordered.dims;
    constants[1].i = bottom_blob_bordered.w;
    constants[2].i = bottom_blob_bordered.h;
    constants[3].i = bottom_blob_bordered.c;
    constants[4].i = 0;//bottom_blob_bordered.cstep;
    constants[5].i = top_blob.dims;
    constants[6].i = top_blob.w;
    constants[7].i = top_blob.h;
    constants[8].i = top_blob.c;
    constants[9].i = 0;//top_blob.cstep;

    // record
    if (elempack == 1 && out_elempack == 1 && kernel_w == 1 && kernel_h == 1 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1)
    {
        VkImageMat dispatcher;
        dispatcher.w = (top_blob.w + 1) / 2;
        dispatcher.h = (top_blob.h + 1) / 2;
        dispatcher.c = top_blob.c;

        cmd.record_pipeline(pipeline_convolution_1x1s1d1, bindings, constants, dispatcher);
    }
    else if (elempack == 4 && out_elempack == 4 && kernel_w == 1 && kernel_h == 1 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1)
    {
        VkImageMat dispatcher;
        dispatcher.w = (top_blob.w + 1) / 2;
        dispatcher.h = (top_blob.h + 1) / 2;
        dispatcher.c = top_blob.c;

        cmd.record_pipeline(pipeline_convolution_pack4_1x1s1d1, bindings, constants, dispatcher);
    }
    else if (elempack == 8 && out_elempack == 8 && kernel_w == 1 && kernel_h == 1 && stride_w == 1 && stride_h == 1 && dilation_w == 1 && dilation_h == 1)
    {
        VkImageMat dispatcher;
        dispatcher.w = (top_blob.w + 1) / 2;
        dispatcher.h = (top_blob.h + 1) / 2;
        dispatcher.c = top_blob.c;

        cmd.record_pipeline(pipeline_convolution_pack8_1x1s1d1, bindings, constants, dispatcher);
    }
    else
    {
        const Pipeline* pipeline = 0;
        if (elempack == 1 && out_elempack == 1)
        {
            pipeline = pipeline_convolution;
        }
        else if (elempack == 4 && out_elempack == 4)
        {
            pipeline = pipeline_convolution_pack4;
        }
        else if (elempack == 1 && out_elempack == 4)
        {
            pipeline = pipeline_convolution_pack1to4;
        }
        else if (elempack == 4 && out_elempack == 1)
        {
            pipeline = pipeline_convolution_pack4to1;
        }
        else if (elempack == 8 && out_elempack == 8)
        {
            pipeline = pipeline_convolution_pack8;
        }
        else if (elempack == 1 && out_elempack == 8)
        {
            pipeline = pipeline_convolution_pack1to8;
        }
        else if (elempack == 4 && out_elempack == 8)
        {
            pipeline = pipeline_convolution_pack4to8;
        }
        else if (elempack == 8 && out_elempack == 4)
        {
            pipeline = pipeline_convolution_pack8to4;
        }
        else if (elempack == 8 && out_elempack == 1)
        {
            pipeline = pipeline_convolution_pack8to1;
        }

        cmd.record_pipeline(pipeline, bindings, constants, top_blob);
    }

    return 0;
}

} // namespace ncnn