double oclMorphFilter::runSubtractKernel(
const oclImage2DHolder& sourceA,
const oclImage2DHolder& sourceB,
oclImage2DHolder& output)
{
cl_int err;
err = kernelSubtract.setArg(0, sourceA.img);
err = kernelSubtract.setArg(1, sourceB.img);
err |= kernelSubtract.setArg(2, output.img);
if(!oclContext::oclError("Error while setting kernel arguments", err))
return 0.0;
cl::NDRange offset(computeOffset(0, 0));
cl::NDRange gridDim(computeGlobal(0, 0));
cl::Event evt;
err = ctx->commandQueue().enqueueNDRangeKernel(
kernelSubtract, offset, gridDim, ctx->workgroupSize(),
nullptr, &evt);
evt.wait();
oclContext::oclError("Error while executing kernel over ND range!", err);
return oclContext::oclElapsedEvent(evt);
}
double oclMorphFilter::runMorphologyKernel(
cl::Kernel* kernel,
const oclImage2DHolder& source,
oclImage2DHolder& output)
{
cl_int err;
err = kernel->setArg(0, source.img);
err |= kernel->setArg(1, output.img);
err |= kernel->setArg(2, structuringElement.buf);
err |= kernel->setArg(3, structuringElement.size / static_cast<int>(sizeof(cl_int2)));
if(!oclContext::oclError("Error while setting kernel arguments", err))
return 0.0;
cl::NDRange offset(computeOffset(0, 0));
cl::NDRange gridDim(computeGlobal(0, 0));
cl::Event evt;
err = ctx->commandQueue().enqueueNDRangeKernel(
*kernel, offset, gridDim, ctx->workgroupSize(),
nullptr, &evt);
evt.wait();
oclContext::oclError("Error while executing kernel over ND range!", err);
return oclContext::oclElapsedEvent(evt);
}
bool PRDF::perform(OptCTX & oc)
{
START_TIMER_AFTER();
m_ru->checkValidAndRecompute(&oc, PASS_RPO, PASS_UNDEF);
List<IRBB*> * bbl = m_ru->get_bb_list();
if (bbl->get_elem_count() == 0) {
return false;
}
List<IR const*> lst;
C<IRBB*> * ct;
for (bbl->get_head(&ct); ct != bbl->end(); ct = bbl->get_next(ct)) {
IRBB * bb = ct->val();
ASSERT0(bb);
computeLocal(bb, lst);
}
computeGlobal();
//dump();
END_TIMER_AFTER(get_pass_name());
return false;
}
