Class AxpyPerformanceTest

src/tests/performance/perf-axpy.cpp:59–118

namespace clMath {

template <typename ElemType> class AxpyPerformanceTest : public PerformanceTest
{
public:
    virtual ~AxpyPerformanceTest();

    virtual int prepare(void);
    virtual nano_time_t etalonPerfSingle(void);
    virtual nano_time_t clblasPerfSingle(void);

    static void runInstance(BlasFunction fn, TestParams *params)
    {
        AxpyPerformanceTest<ElemType> perfCase(fn, params);
        int ret = 0;
        int opFactor;
        BlasBase *base;

        base = clMath::BlasBase::getInstance();

        opFactor =1;

        if (((fn == FN_DAXPY) || (fn == FN_ZAXPY)) &&
            !base->isDevSupportDoublePrecision()) {

            std::cerr << ">> WARNING: The target device doesn't support native "
                         "double precision floating point arithmetic" <<
                         std::endl << ">> Test skipped" << std::endl;
            return;
        }

        if (!perfCase.areResourcesSufficient(params)) {
            std::cerr << ">> RESOURCE CHECK: Skip due to unsufficient resources" <<
                        std::endl;
			return;
        }
        else {
            ret = perfCase.run(opFactor);
        }

        ASSERT_GE(ret, 0) << "Fatal error: can not allocate resources or "
                             "perform an OpenCL request!" << endl;
        EXPECT_EQ(0, ret) << "The OpenCL version is slower in the case" << endl;
    }

private:
    AxpyPerformanceTest(BlasFunction fn, TestParams *params);

    bool areResourcesSufficient(TestParams *params);

    TestParams params_;
    ElemType alpha_;
    ElemType *X_;
    ElemType *Y_;
    ElemType *blasX_;
    ElemType *blasY_;
    cl_mem mobjX_;
    cl_mem mobjY_;
    size_t  lengthX;
    size_t  lengthY;