Пример #1
0
TEST(testlocal, reduceviascratch_multipleworkgroups_ints) {
    EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
    CLKernel *kernel = cl->buildKernel("testlocal.cl", "reduceViaScratch_multipleworkgroups_ints");
    int workgroupSize = min(512, cl->getMaxWorkgroupSize());
    const int numWorkgroups = workgroupSize;
    const int N = workgroupSize * numWorkgroups;
    cout << "numworkgroups " << numWorkgroups << " workgroupsize " << workgroupSize << " N " << N << endl;
    int *myarray = new int[N];
    int sumViaCpu = 0;
    int localSumViaCpu = 0;
    int localSumViaCpu2 = 0;
    int *localSumsViaCpu = new int[numWorkgroups];
    memset(localSumsViaCpu, 0, sizeof(int)*numWorkgroups);
    for(int i = 0; i < N; i++) {
        myarray[i] = ((i + 7) * 3) % 50;
        sumViaCpu += myarray[i];
        if(i < workgroupSize) {
            localSumViaCpu += myarray[i];
        }
        if(i >= workgroupSize && i < workgroupSize * 2) {
            localSumViaCpu2 += myarray[i];
        }
        int workgroupId = i / workgroupSize;
        localSumsViaCpu[workgroupId] += myarray[i];
    }
    ASSERT_EQ(localSumViaCpu, localSumsViaCpu[0]);
    ASSERT_EQ(localSumViaCpu2, localSumsViaCpu[1]);
    ASSERT_NE(myarray[0], sumViaCpu);

//    Timer timer;

    CLWrapper *a1wrapper = cl->wrap(N, myarray);
    a1wrapper->copyToDevice();
    int *a2 = new int[numWorkgroups];
    CLWrapper *a2wrapper = cl->wrap(numWorkgroups, a2);
    kernel->in(a1wrapper);
    kernel->out(a2wrapper);
    kernel->localInts(workgroupSize);
    kernel->run_1d(N, workgroupSize);

    int finalSum;
    kernel->in(a2wrapper);
    kernel->out(1, &finalSum);
    kernel->localInts(workgroupSize);
    kernel->run_1d(numWorkgroups, workgroupSize);
//    timer.timeCheck("finished 2-way reduce");

    EXPECT_EQ(sumViaCpu, finalSum);

    delete a1wrapper;
    delete a2wrapper;
    delete[] a2;
    delete[]myarray;
    delete kernel;
    delete cl;
}
Пример #2
0
TEST(SLOW_testlocal, selfdot_3levels_withscratch) {
    EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
    CLKernel *kernel = cl->buildKernel("testlocal.cl", "selfdot_ints_withscratch");
    int workgroupSize = min(512, cl->getMaxWorkgroupSize());
    const int numWorkgroups = workgroupSize;
    const int level3size = numWorkgroups / 4;
    const int N = workgroupSize * numWorkgroups * level3size;
    cout << "numworkgroups " << numWorkgroups << " workgroupsize " << workgroupSize << " N " << N << endl;
    int *myarray = new int[N];
    for(int i = 0; i < N; i++) {
        myarray[i] = ((i + 7) * 3) % 5;
    }

//    Timer timer;

    CLWrapper *a1wrapper = cl->wrap(N, myarray);
    a1wrapper->copyToDevice();
//    timer.timeCheck("copied array to device");
    int *a2 = new int[numWorkgroups*level3size];
    CLWrapper *a2wrapper = cl->wrap(numWorkgroups * level3size, a2);
    kernel->in(a1wrapper);
    kernel->out(a2wrapper);
    kernel->localInts(workgroupSize);
    kernel->localInts(workgroupSize);
    kernel->run_1d(N, workgroupSize);
    cl->finish();

    int *a3 = new int[numWorkgroups];
    CLWrapper *a3wrapper = cl->wrap(level3size, a3);
    kernel->in(a2wrapper);
    kernel->out(a3wrapper);
    kernel->localInts(workgroupSize);
    kernel->localInts(workgroupSize);
    kernel->run_1d(workgroupSize * level3size, workgroupSize);
    cl->finish();

    int finalSum;
    kernel->in(a3wrapper);
    kernel->out(1, &finalSum);
    kernel->localInts(level3size);
    kernel->localInts(workgroupSize);
    kernel->run_1d(level3size, level3size);
//    timer.timeCheck("finished 3-level reduce");

    EXPECT_EQ(-1306309159, finalSum);

    delete a1wrapper;
    delete a2wrapper;
    delete a3wrapper;
    delete[] a3;
    delete[] a2;
    delete[]myarray;
    delete kernel;
    delete cl;
}