ncnn/tests/test_convolution_oom.cpp

// Copyright 2024 Tencent
// SPDX-License-Identifier: BSD-3-Clause

#include "testutil.h"

static int test_convolution_oom(int w, int h, int c, int outch, int kernel, int dilation, int stride, int pad, int bias)
{
    ncnn::Mat a = RandomMat(w, h, c);

    ncnn::ParamDict pd;
    pd.set(0, outch);
    pd.set(1, kernel);
    pd.set(2, dilation);
    pd.set(3, stride);
    pd.set(4, pad);
    pd.set(5, bias);
    pd.set(6, outch * c * kernel * kernel);

    int activation_type = RAND() % 7; // 0 1 2 3 4 5 6
    ncnn::Mat activation_params(2);
    activation_params[0] = (activation_type == 6) ? RandomFloat(0, 1) : RandomFloat(-1, 0); // alpha
    activation_params[1] = RandomFloat(0, 1);                                               // beta
    pd.set(9, activation_type);
    pd.set(10, activation_params);

    std::vector<ncnn::Mat> weights(bias ? 2 : 1);
    weights[0] = RandomMat(outch * c * kernel * kernel);
    if (bias)
        weights[1] = RandomMat(outch);

    int ret = test_layer_oom("Convolution", pd, weights, a);
    if (ret != 0)
    {
        fprintf(stderr, "test_convolution_oom failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d act=%d actparams=[%f,%f]\n", w, h, c, outch, kernel, dilation, stride, pad, bias, activation_type, activation_params[0], activation_params[1]);
        return ret;
    }

    return ret;
}

static int test_convolution_0()
{
    return 0
           || test_convolution_oom(9, 7, 31, 63, 1, 1, 1, 0, 1)
           || test_convolution_oom(9, 7, 31, 63, 3, 1, 1, 1, 1);
}

#if NCNN_INT8
static int test_convolution_oom_int8(int w, int h, int c, int outch, int kernel, int dilation, int stride, int pad, int bias, bool requant = false, bool test_winograd23 = false)
{
    ncnn::Mat a = RandomMat(w, h, c);

    ncnn::ParamDict pd;
    pd.set(0, outch);
    pd.set(1, kernel);
    pd.set(2, dilation);
    pd.set(3, stride);
    pd.set(4, pad);
    pd.set(5, bias);
    pd.set(6, outch * c * kernel * kernel);
    pd.set(8, requant ? 101 : 1); // int8_scale_term

    int activation_type = RAND() % 7; // 0 1 2 3 4 5 6
    ncnn::Mat activation_params(2);
    activation_params[0] = (activation_type == 6) ? RandomFloat(0, 1) : RandomFloat(-1, 0); // alpha
    activation_params[1] = RandomFloat(0, 1);                                               // beta
    pd.set(9, activation_type);
    pd.set(10, activation_params);

    std::vector<ncnn::Mat> weights(bias ? 5 : 4);
    weights[0] = RandomMat(outch * c * kernel * kernel);

    ncnn::Mat weight_scales = scales_mat(weights[0], outch, c * kernel * kernel, c * kernel * kernel);
    ncnn::Mat input_scales = scales_mat(a, 1, w * h * c, a.cstep);
    ncnn::Mat top_scales = requant ? scales_mat(a, 1, w * h * c, a.cstep) : ncnn::Mat();

    if (kernel == 3 && dilation == 1 && stride == 1)
    {
        // test for 6bit quant
        for (int i = 0; i < weight_scales.w; i++)
            weight_scales[i] = weight_scales[i] / 4.f;
    }

    if (bias)
    {
        weights[1] = RandomMat(outch);
        weights[2] = weight_scales;
        weights[3] = input_scales;
        weights[4] = top_scales;
    }
    else
    {
        weights[1] = weight_scales;
        weights[2] = input_scales;
        weights[3] = top_scales;
    }

    int flag = TEST_LAYER_DISABLE_GPU_TESTING;
    int ret = test_layer_oom("Convolution", pd, weights, a, flag);
    if (ret != 0)
    {
        fprintf(stderr, "test_convolution_oom_int8 failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d requant=%d act=%d actparams=[%f,%f]\n", w, h, c, outch, kernel, dilation, stride, pad, bias, requant, activation_type, activation_params[0], activation_params[1]);
        return ret;
    }

    if (test_winograd23)
    {
        ncnn::Option opt;
        opt.num_threads = 1;
        opt.use_packing_layout = false;
        opt.use_fp16_packed = false;
        opt.use_fp16_storage = false;
        opt.use_fp16_arithmetic = false;
        opt.use_bf16_packed = false;
        opt.use_bf16_storage = false;
        opt.use_sgemm_convolution = false;
        opt.use_winograd_convolution = true;
        opt.use_winograd23_convolution = true;
        opt.use_winograd43_convolution = false;
        opt.use_winograd63_convolution = false;

        ret = test_layer_oom_opt("Convolution", pd, weights, opt, a, flag);
        if (ret != 0)
        {
            fprintf(stderr, "test_convolution_oom_int8 failed w=%d h=%d c=%d outch=%d kernel=%d dilation=%d stride=%d pad=%d bias=%d requant=%d winograd=23 act=%d actparams=[%f,%f]\n", w, h, c, outch, kernel, dilation, stride, pad, bias, requant, activation_type, activation_params[0], activation_params[1]);
            return ret;
        }
    }

    return ret;
}

static int test_convolution_1()
{
    return 0
           || test_convolution_oom_int8(9, 7, 31, 32, 1, 1, 1, 0, 1)
           || test_convolution_oom_int8(9, 7, 31, 32, 3, 1, 1, 1, 1, false, true)
           || test_convolution_oom_int8(9, 7, 31, 63, 1, 1, 1, 0, 1)
           || test_convolution_oom_int8(9, 7, 31, 63, 3, 1, 1, 1, 1)
           || test_convolution_oom_int8(3, 1, 32769, 3, 1, 1, 1, 0, 1);
}

static int test_convolution_2()
{
    return 0
           || test_convolution_oom_int8(9, 7, 31, 32, 1, 1, 1, 0, 1, true)
           || test_convolution_oom_int8(9, 7, 31, 32, 3, 1, 1, 1, 1, true)
           || test_convolution_oom_int8(9, 7, 31, 63, 1, 1, 1, 0, 1, true)
           || test_convolution_oom_int8(9, 7, 31, 63, 3, 1, 1, 1, 1, true);
}
#endif // NCNN_INT8

int main()
{
    SRAND(7767517);

#if NCNN_INT8
    return test_convolution_0() || test_convolution_1() || test_convolution_2();
#else
    return test_convolution_0();
#endif
}