TEST(testqueues, defaultqueue) {
cout << "start" << endl;
if(!EasyCL::isOpenCLAvailable()) {
cout << "opencl library not found" << endl;
exit(-1);
}
cout << "found opencl library" << endl;
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernelFromString(getKernel(), "test", "");
float in[5];
float out[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
out[i] = 0;
}
kernel->input(5, in);
kernel->output(5, out);
size_t global = 5;
size_t local = 5;
kernel->run(1, &global, &local);
cl->finish();
assertEquals(out[0] , 7);
assertEquals(out[1] , 10);
assertEquals(out[2] , 13);
assertEquals(out[3] , 16);
assertEquals(out[4] , 19);
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
out[i] = 0;
}
cout << "cl->queue " << (void *)cl->queue << endl;
cout << "*cl->queue " << (void *)*cl->queue << endl;
cout << "cl->default_queue->queue" << (void *)cl->default_queue->queue << endl;
kernel->input(5, in);
kernel->output(5, out);
kernel->run(cl->default_queue, 1, &global, &local);
cl->finish();
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;
}
TEST(testfloatwrapperconst, main) {
if(!EasyCL::isOpenCLAvailable()) {
cout << "opencl library not found" << endl;
exit(-1);
}
cout << "found opencl library" << endl;
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernelFromString(getKernel(), "test", "");
float in[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
}
float out[5];
CLWrapper *inwrapper = cl->wrap(5, (float const *)in);
CLWrapper *outwrapper = cl->wrap(5, out);
inwrapper->copyToDevice();
kernel->input(inwrapper);
kernel->output(outwrapper);
kernel->run_1d(5, 5);
outwrapper->copyToHost();
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( 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(testqueues, main) {
cout << "start" << endl;
if(!EasyCL::isOpenCLAvailable()) {
cout << "opencl library not found" << endl;
exit(-1);
}
cout << "found opencl library" << endl;
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLQueue *queue2 = cl->newQueue();
CLQueue *queue3 = cl->newQueue();
CLKernel *kernel = cl->buildKernelFromString(getKernel(), "test", "");
// CLKernel *kernel2 = cl->buildKernelFromString(queue, getKernel(), "test", "");
float in[5];
float out[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
out[i] = 0;
}
kernel->input(5, in);
kernel->output(5, out);
size_t global = 5;
size_t local = 5;
kernel->run(1, &global, &local);
assertEquals(out[0] , 7);
assertEquals(out[1] , 10);
assertEquals(out[2] , 13);
assertEquals(out[3] , 16);
assertEquals(out[4] , 19);
// in[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
out[i] = 0;
}
// float out[5];
kernel->input(5, in);
kernel->output(5, out);
// size_t global = 5;
// size_t local = 5;
kernel->run(queue2, 1, &global, &local);
assertEquals(out[0] , 7);
assertEquals(out[1] , 10);
assertEquals(out[2] , 13);
assertEquals(out[3] , 16);
assertEquals(out[4] , 19);
// in[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
out[i] = 0;
}
// float out[5];
kernel->input(5, in);
kernel->output(5, out);
// size_t global = 5;
// size_t local = 5;
kernel->run(queue3, 1, &global, &local);
assertEquals(out[0] , 7);
assertEquals(out[1] , 10);
assertEquals(out[2] , 13);
assertEquals(out[3] , 16);
assertEquals(out[4] , 19);
// in[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
out[i] = 0;
}
// float out[5];
kernel->input(5, in);
kernel->output(5, out);
// size_t global = 5;
// size_t local = 5;
kernel->run(1, &global, &local);
assertEquals(out[0] , 7);
assertEquals(out[1] , 10);
assertEquals(out[2] , 13);
assertEquals(out[3] , 16);
assertEquals(out[4] , 19);
// in[5];
for(int i = 0; i < 5; i++) {
in[i] = i * 3;
out[i] = 0;
}
// float out[5];
kernel->input(5, in);
kernel->output(5, out);
// size_t global = 5;
// size_t local = 5;
kernel->run(queue2, 1, &global, &local);
assertEquals(out[0] , 7);
assertEquals(out[1] , 10);
assertEquals(out[2] , 13);
assertEquals(out[3] , 16);
assertEquals(out[4] , 19);
delete queue2;
delete queue3;
cout << "tests completed ok" << endl;
}
