TEST( SLOW_testintwrapper_huge, testreadwrite ) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testeasycl.cl", "test_stress");
const int N = 1000000;
int *in = new int[N];
for( int i = 0; i < N; i++ ) {
in[i] = i * 3;
}
int *out = new int[N];
CLWrapper *inwrapper = cl->wrap(N, in);
CLWrapper *outwrapper = cl->wrap(N, out);
inwrapper->copyToDevice();
outwrapper->createOnDevice();
kernel->input( inwrapper );
kernel->output( outwrapper );
int globalSize = N;
int workgroupsize = cl->getMaxWorkgroupSize();
globalSize = ( ( globalSize + workgroupsize - 1 ) / workgroupsize ) * workgroupsize;
cout << "globalsize: " << globalSize << " workgroupsize " << workgroupsize << endl;
kernel->run_1d( globalSize, workgroupsize );
outwrapper->copyToHost();
for( int i = 0; i < N; i++ ) {
if( out[i] != 689514 ) {
cout << "out[" << i << "] != 689514: " << out[i] << endl;
exit(-1);
}
}
delete outwrapper;
delete inwrapper;
delete kernel;
delete cl;
}
TEST( SLOW_testintwrapper_huge, testread ) {
Timer timer;
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testeasycl.cl", "test_read");
// const int N = 4500000;
// const int N = (4500000/512)*512;
int N = 100000;
int *out = new int[N];
CLWrapper *outwrapper = cl->wrap(N, out);
kernel->in(3)->in(7);
kernel->output( outwrapper );
int globalSize = N;
int workgroupsize = cl->getMaxWorkgroupSize();
globalSize = ( ( globalSize + workgroupsize - 1 ) / workgroupsize ) * workgroupsize;
cout << "globalsize: " << globalSize << " workgroupsize " << workgroupsize << endl;
timer.timeCheck("before kernel");
kernel->run_1d( globalSize, workgroupsize );
timer.timeCheck("after kernel");
outwrapper->copyToHost();
timer.timeCheck("after copy to host");
for( int i = 0; i < N; i++ ) {
if( out[i] != 4228 ) {
cout << "out[" << i << "] != 4228: " << out[i] << endl;
exit(-1);
}
}
delete outwrapper;
delete kernel;
delete cl;
}
TEST(SLOW_testlocal, selfdot_3levels_withoutscratch) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testlocal.cl", "selfdot_ints_withoutscratch");
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 *second = new int[N];
CLWrapper *secondwrapper = cl->wrap(N, second);
int *a2 = new int[numWorkgroups*level3size];
CLWrapper *a2wrapper = cl->wrap(numWorkgroups * level3size, a2);
kernel->in(a1wrapper);
kernel->out(secondwrapper);
kernel->out(a2wrapper);
kernel->run_1d(N, workgroupSize);
cl->finish();
int *a3 = new int[numWorkgroups];
CLWrapper *a3wrapper = cl->wrap(level3size, a3);
kernel->in(a2wrapper);
kernel->out(secondwrapper);
kernel->out(a3wrapper);
kernel->run_1d(workgroupSize * level3size, workgroupSize);
cl->finish();
int finalSum;
kernel->in(a3wrapper);
kernel->out(secondwrapper);
kernel->out(1, &finalSum);
kernel->run_1d(level3size, level3size);
// timer.timeCheck("finished 3-level reduce");
EXPECT_EQ(-1306309159, finalSum);
delete a1wrapper;
delete a2wrapper;
delete a3wrapper;
delete secondwrapper;
delete[] a3;
delete[] second;
delete[] a2;
delete[]myarray;
delete kernel;
delete cl;
}
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;
}
TEST( testdefines, simple ) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testdefines.cl", "testDefines",
"-D DOUBLE -D SOME_VALUE=5" );
float out[32];
kernel->out( 32, out );
kernel->run_1d( 32, 32 );
EXPECT_EQ( 10, out[3] );
delete kernel;
delete cl;
}
TEST(testlocal, notUselocal) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testlocal.cl", "notUseLocal");
int workgroupSize = 64;
float *myarray = new float[workgroupSize];
kernel->in(workgroupSize);
kernel->inout(workgroupSize, myarray);
kernel->run_1d(workgroupSize, workgroupSize);
delete[]myarray;
delete kernel;
delete cl;
}
TEST(testdirtywrapper, main) {
if(!EasyCL::isOpenCLAvailable()) {
cout << "opencl library not found" << endl;
exit(-1);
}
cout << "found opencl library" << endl;
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testeasycl.cl", "test");
float in[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
}
float out[5];
CLWrapper *inwrapper = cl->wrap(5, in);
CLWrapper *outwrapper = cl->wrap(5, out);
EXPECT_FALSE(inwrapper->isDeviceDirty());
EXPECT_FALSE(outwrapper->isDeviceDirty());
inwrapper->copyToDevice();
EXPECT_FALSE(inwrapper->isDeviceDirty());
EXPECT_FALSE(outwrapper->isDeviceDirty());
kernel->input(inwrapper);
kernel->output(outwrapper);
EXPECT_FALSE(inwrapper->isDeviceDirty());
EXPECT_FALSE(outwrapper->isDeviceDirty());
kernel->run_1d(5, 5);
EXPECT_FALSE(inwrapper->isDeviceDirty());
EXPECT_TRUE(outwrapper->isDeviceDirty());
outwrapper->copyToHost();
EXPECT_FALSE(outwrapper->isDeviceDirty());
assertEquals(out[0] , 7);
assertEquals(out[1] , 10);
assertEquals(out[2] , 13);
assertEquals(out[3] , 16);
assertEquals(out[4] , 19);
cout << "tests completed ok" << endl;
delete inwrapper;
delete outwrapper;
delete kernel;
delete cl;
}
TEST(testlocal, reduceviascratch_multipleworkgroups) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testlocal.cl", "reduceViaScratch_multipleworkgroups");
int workgroupSize = min(512, cl->getMaxWorkgroupSize());
const int numWorkgroups = workgroupSize;
const int N = workgroupSize * numWorkgroups;
float *myarray = new float[N];
float sumViaCpu = 0;
float localSumViaCpu = 0;
for(int i = 0; i < N; i++) {
myarray[i] = ((i + 7) * 3) % 10;
sumViaCpu += myarray[i];
if(i < workgroupSize) {
localSumViaCpu += myarray[i];
}
}
cout << "expected sum, calc'd via cpu, : " << sumViaCpu << endl;
EXPECT_NE(myarray[0], sumViaCpu);
// Timer timer;
CLWrapper *a1wrapper = cl->wrap(N, myarray);
a1wrapper->copyToDevice();
float *a2 = new float[numWorkgroups];
CLWrapper *a2wrapper = cl->wrap(numWorkgroups, a2);
kernel->in(a1wrapper);
kernel->out(a2wrapper);
kernel->localFloats(workgroupSize);
kernel->run_1d(N, workgroupSize);
float finalSum;
kernel->in(a2wrapper);
kernel->out(1, &finalSum);
kernel->localFloats(workgroupSize);
kernel->run_1d(numWorkgroups, workgroupSize);
EXPECT_EQ(sumViaCpu, finalSum);
delete a1wrapper;
delete a2wrapper;
delete[] a2;
delete[]myarray;
delete kernel;
delete cl;
}
TEST(testlocal, localreduce) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testlocal.cl", "reduceViaScratch");
int workgroupSize = min(512, cl->getMaxWorkgroupSize());
float *myarray = new float[workgroupSize];
// Timer timer;
for(int i = 0; i < 2000; i++) {
float sumViaCpu = 0;
for(int i = 0; i < workgroupSize; i++) {
myarray[i] = (i + 7) * 3;
sumViaCpu += myarray[i];
}
EXPECT_NE(myarray[0], sumViaCpu);
kernel->inout(workgroupSize, myarray)->localFloats(workgroupSize);
kernel->run_1d(workgroupSize, workgroupSize);
EXPECT_EQ(myarray[0], sumViaCpu);
}
// timer.timeCheck("after iterations");
delete[]myarray;
delete kernel;
delete cl;
}
int main(int argc, char *argv[])
{
const int test_size = 128;
std::random_device rd;
std::seed_seq s{ rd(), rd(), rd(), rd(), rd(), rd(), rd(), rd() };
std::mt19937 mt(s);
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kern = cl->buildKernel("test.cl", "test");
std::vector<uint32_t> inbuf(test_size*4), outbuf(test_size), compare(test_size);
std::generate(inbuf.begin(), inbuf.end(), mt);
std::cout << "Running CL implementation" << std::endl;
kern->in(inbuf.size(), &inbuf[0]);
kern->out(outbuf.size(), &outbuf[0]);
size_t global_size[] = { test_size };
kern->run(1, global_size, nullptr);
delete kern;
std::cout << "Running local implementation" << std::endl;
for (int i = 0; i < compare.size(); ++i) {
compare[i] = inbuf[i] ^ inbuf[i + 1] ^ inbuf[i + 2] ^ inbuf[i + 3];
}
std::cout << "Comparing CL test with local implementation" << std::endl;
for (int i = 0; i < compare.size(); ++i) {
if (outbuf[i] != compare[i]) {
std::cout << "Error in index " << i << " " << outbuf[i] << " != " << compare[i] << std::endl;
}
}
return 0;
}