MegEngine/dnn/test/fallback/elemwise.cpp

/**
 * \file dnn/test/fallback/elemwise.cpp
 * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
 *
 * Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
 *
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
 * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 */

#include "test/fallback/fixture.h"

#include <ctime>
#include "test/common/checker.h"
#include "test/common/elemwise.h"
#include "test/common/task_record_check.h"
#include "test/common/tensor.h"

using namespace megdnn;
using namespace test;

template <typename tag>
class FALLBACK_ELEMWISE : public FALLBACK {};
TYPED_TEST_CASE(FALLBACK_ELEMWISE, elemwise::test_types);
TYPED_TEST(FALLBACK_ELEMWISE, run) {
    elemwise::run_test<TypeParam>(this->handle());
}
TEST_F(FALLBACK, ELEMWISE_RECORD) {
    TaskRecordChecker<Elemwise> checker{1};
    checker.set_param({Elemwise::Mode::ADD});
    checker.set_dtype(0, dtype::Float32());
    checker.set_dtype(1, dtype::Float32());
    checker.set_dtype(2, dtype::Float32());
    UniformIntRNG rng{-100, 100};
    checker.set_rng(0, &rng);
    checker.set_rng(1, &rng);
    checker.set_rng(2, &rng);
    checker.execs({{10, 10, 32}, {10, 10, 32}, {}});
}
#if MEGDNN_WITH_BENCHMARK
TEST_F(FALLBACK, BENCHMARK_ELEMWISE) {
    auto naive_handle = create_cpu_handle(2);
    auto run = [&](const TensorShape& shp0, const TensorShape& shp1) {
        TensorShape shpo;
        Elemwise::deduce_shape({shp0, shp1}, shpo);
        Tensor<> op0(handle(), {shp0, dtype::Float32()}),
                op1(handle(), {shp1, dtype::Float32()}),
                out(handle(), {shpo, dtype::Float32()});
        auto opr_cur = handle()->create_operator<Elemwise>();
        auto opr_naive = naive_handle->create_operator<Elemwise>();
        opr_cur->param() = {Elemwise::Mode::ADD};
        opr_naive->param() = {Elemwise::Mode::ADD};

        auto timeit = [&](Elemwise* opr) {
            opr->exec({op0.tensornd(), op1.tensornd()}, out.tensornd());
            auto start = clock();
            opr->exec({op0.tensornd(), op1.tensornd()}, out.tensornd());
            auto stop = clock();
            return (stop - start) * 1e3 / CLOCKS_PER_SEC;
        };
        auto t0 = timeit(opr_cur.get()), t1 = timeit(opr_naive.get());
        double tot_size_gb_ms =
                (op0.layout().span().dist_byte() + op1.layout().span().dist_byte() +
                 out.layout().span().dist_byte()) /
                1024.0 / 1024.0 / 1024.0 * 1e3;
        printf("%15s+%-15s: fallback=%7.3fms,%5.2fGiB/s "
               "naive=%7.3fms,%5.2fGiB/s\n",
               shp0.to_string().c_str(), shp1.to_string().c_str(), t0,
               tot_size_gb_ms / t0, t1, tot_size_gb_ms / t1);
    };
    // contig
    run({1024, 1024, 32}, {1024, 1024, 32});
    // bcast 101
    run({1024, 1024, 32}, {1, 1024, 1});
    // bcast 01
    run({4096 * 4, 1024}, {4096 * 4, 1});
    // bcast 10
    run({4096 * 4, 1024}, {1, 1024});

    // non-contig, fallback to naive
    run({1024, 1024, 32}, {1024, 1, 32});
}
#endif

// vim: syntax=cpp.doxygen