Function TEST_F

dnn/test/aarch64/convolution.cpp:12–60

using namespace test;
#if __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
TEST_F(AARCH64, CONVOLUTION_BACKWARD_DATA_FP16) {
    Checker<ConvolutionBackwardData> checker(handle());
    using Param = ConvolutionBackwardData::Param;
    Param param;
    auto run = [&](size_t n, size_t ic, size_t oh, size_t ow, size_t oc, size_t fh,
                   size_t fw, size_t stride, size_t padding, size_t group = 1) {
        param.pad_h = param.pad_w = padding;
        param.stride_h = param.stride_w = stride;

        TensorLayout diff = TensorLayout{{n, oc * group, oh, ow}, dtype::Float16()};
        TensorLayout grad;
        TensorLayout filter;
        if (group == 1) {
            param.sparse = Param::Sparse::DENSE;
            filter = {{oc, ic, fh, fw}, dtype::Float16()};
        } else {
            param.sparse = Param::Sparse::GROUP;
            filter = {{group, oc, ic, fh, fw}, dtype::Float16()};
        }
        // TensorLayout grad;
        {
            auto opr = handle()->create_operator<ConvolutionBackwardData>();
            opr->param() = param;
            opr->deduce_layout(filter, diff, grad);
        }
        NormalRNG rng(10.f);
        checker.set_param(param)
                .set_dtype(0, dtype::Float16())
                .set_dtype(1, dtype::Float16())
                .set_dtype(2, dtype::Float16())
                .set_rng(0, &rng)
                .set_rng(1, &rng)
                .set_epsilon(1e-2)
                .set_before_exec_callback(
                        AlgoChecker<ConvolutionBackwardData>("DeconvMatmul"));
        checker.exec(TensorLayoutArray{filter, diff, grad});
    };

    for (auto mode : {Param::Mode::CONVOLUTION, Param::Mode::CROSS_CORRELATION}) {
        param.mode = mode;
        run(4, 3, 10, 13, 5, 1, 1, 1, 0, 1);
        run(4, 3, 10, 45, 2, 1, 1, 1, 0, 4);
        run(2, 3, 9, 12, 2, 4, 6, 1, 0, 1);
        run(3, 4, 17, 32, 2, 3, 2, 5, 4, 4);
        run(5, 5, 24, 43, 11, 9, 3, 3, 12, 2);
        run(2, 3, 20, 33, 3, 5, 7, 4, 15, 2);
        run(4, 4, 6, 7, 9, 3, 2, 2, 1, 3);
    }
}

#if MEGDNN_WITH_BENCHMARK
TEST_F(AARCH64, BENCHMARK_MATRIX_MUL_QUICK_FP16) {