Reduction failure on MacOSX with Intel CPU as OpenCL device

[fpvidal]

With Intel CPU devices on Mac OS X: Doesn't properly support OpenCL __local memory when the work-group size is greater than one.

Below are the results of test_reduce:

/usr/local/share/compute/bin/test->./test_reduce
Running 12 test cases...
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:48: last checkpoint
-- Repeat-fill case
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:48: last checkpoint
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:48: last checkpoint
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:112: last checkpoint
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:112: last checkpoint
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:112: last checkpoint
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:112: last checkpoint
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:112: last checkpoint
/usr/local/share/compute/include/boost/compute/command_queue.hpp:1040: fatal error in "boost::compute::event boost::compute::command_queue::enqueue_nd_range_kernel(const boost::compute::kernel &, size_t, const size_t *, const size_t *, const size_t *, const boost::compute::wait_list &)": std::exception: Invalid Work Group Size
/usr/local/share/compute/test/test_reduce.cpp:112: last checkpoint

*** 10 failures detected in test suite "TestReduce"

NB: Issue opened to make sure we don't lose track of the problem

[jszuppe]

I think it's known bug in Mac OS OpenCL drivers. I know a few algorithms need special implementation for CPU without usage of local memory because of this bug.
However, in this case serial_reduce should be used to perform reduction and it does not use local memory, so everything should be fine. I don't have Mac OS X, so I can't reproduce the problem.

[jszuppe]

I think I located the bug. #491 should fix it. Try develop branch after it's merged.

[kylelutz]

This should now be fixed in develop. Please re-open if you still encounter errors. Thanks!