Function TEST_F

dnn/test/armv7/convolution.cpp:13–57

#if MEGDNN_WITH_BENCHMARK
TEST_F(ARMV7, BENCHMARK_CONVOLUTION_STRIDE2) {
    using Param = param::Convolution;
    auto run = [&](const TensorShapeArray& shapes, Param param) {
        Benchmarker<Convolution> benchmarker_float(handle());
        size_t RUN = 100;
        auto tfloat = benchmarker_float.set_display(false)
                              .set_times(RUN)
                              .set_param(param)
                              .exec(shapes);
        size_t IC = shapes[1][1];
        size_t FH = shapes[1][2];
        size_t FW = shapes[1][3];
        TensorLayout dst_layout;
        auto opr = handle()->create_operator<Convolution>();
        opr->param() = param;
        opr->deduce_layout(
                {shapes[0], dtype::Float32()}, {shapes[1], dtype::Float32()},
                dst_layout);
        printf("flops: %.3f mflops\n", (IC * dst_layout.total_nr_elems() * FH * FW *
                                        2) / (tfloat / RUN * 1000));
    };

    auto profile = [&](size_t oc, size_t ic, size_t w, size_t h, size_t kernel,
                       size_t stride) {
        Param param;
        param.stride_h = stride;
        param.stride_w = stride;
        param.pad_h = kernel / 2;
        param.pad_w = kernel / 2;
        printf("oc: %zd ic: %zd w: %zd h: %zd stride: %zd kernel_size: %zd\n", oc, ic,
               w, h, stride, kernel);

        run({{1, ic, h, w}, {oc, ic, kernel, kernel}, {}}, param);
    };

    for (size_t kernel : {2, 3, 5, 7}) {
        for (size_t ic : {3, 6, 12, 24}) {
            for (size_t oc : {3, 6, 12, 24}) {
                for (size_t size : {4, 7, 8, 14, 16, 17, 28, 32, 34, 64, 112}) {
                    profile(oc, ic, size, size, kernel, 2);
                }
            }
        }
    }
}
#endif

TEST_F(ARMV7, BENCHMARK_CONVOLUTION_1X1) {