TEST( testCopyBlock, basic ) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
float a[] = { 1,2,3,4,
5,6,7,8,
9,10,11,12 };
float b[10];
memset(b, 0, sizeof(float)*10);
CLKernel *kernel = makeBasicKernel( cl );
kernel->in( 12, a )->out( 6, b )->in( ( 3<<10)|4)->in( (0<<10)|1)->in((2<<10)|3);
kernel->localFloats( 2 * 3 );
kernel->run_1d(12,4);
// kernel->run_1d(12,12);
cl->finish();
float expected[] = { 2,3,4,
6,7,8 };
for( int i = 0; i < 2; i++ ) {
for( int j = 0; j < 3; j++ ) {
cout << b[i*3+j] << " ";
EXPECT_EQ( expected[i*3+j], b[i*3+j] );
}
cout << endl;
}
cout << endl;
for( int i = 6; i < 10; i++ ) {
cout << b[i] << " ";
EXPECT_EQ( 0, b[i] );
}
cout << endl;
cout << endl;
kernel->in( 12, a )->out( 6, b )->in( ( 3<<10)|4)->in( (1<<10)|0)->in((2<<10)|3);
kernel->localFloats( 2 * 3 );
// kernel->run_1d(12,4);
kernel->run_1d(12,4);
cl->finish();
float expected2[] = { 5,6,7,
9,10,11 };
for( int i = 0; i < 2; i++ ) {
for( int j = 0; j < 3; j++ ) {
cout << b[i*3+j] << " ";
EXPECT_EQ( expected2[i*3+j], b[i*3+j] );
}
cout << endl;
}
cout << endl;
for( int i = 6; i < 10; i++ ) {
cout << b[i] << " ";
EXPECT_EQ( 0, b[i] );
}
cout << endl;
cout << endl;
delete kernel;
delete cl;
}
TEST( testCopyLocal, basic ) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
float a[] = { 1,2,3,4,
5,6,7,8,
9,10,11,12 };
float b[16];
memset(b, 0, sizeof(float)*16);
CLKernel *kernel = makeKernel( cl );
kernel->in( 12, a )->out( 16, b )->in( 12 );
kernel->localFloats( 12 );
kernel->run_1d(12,12);
cl->finish();
for( int i = 0; i < 3; i++ ) {
for( int j = 0; j < 4; j++ ) {
cout << b[i*4+j] << " ";
EXPECT_EQ( i * 4 + j + 1, b[i*4+j] );
}
cout << endl;
}
cout << endl;
for( int i = 12; i < 16; i++ ) {
cout << b[i] << " ";
EXPECT_EQ( 0, b[i] );
}
cout << endl;
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, uselocal) {
EasyCL *cl = EasyCL::createForFirstGpuOtherwiseCpu();
CLKernel *kernel = cl->buildKernel("testlocal.cl", "useLocal");
int workgroupSize = 64;
float *myarray = new float[workgroupSize];
kernel->in(workgroupSize);
kernel->inout(workgroupSize, myarray);
kernel->localFloats(workgroupSize);
kernel->run_1d(workgroupSize, workgroupSize);
delete[]myarray;
delete kernel;
delete cl;
}
