src/treelearner/ocl/histogram256.cl (3 lines):
	- line 149: // TODO: try to avoid bank conflict here
	- line 720: // FIXME: 2-way bank conflict
	- line 738: // FIXME: how to do the above on AMD GPUs??


src/treelearner/ocl/histogram16.cl (1 line):
	- line 717: // FIXME: how to do the above on AMD GPUs??


src/treelearner/gpu_tree_learner.cpp (1 line):
	- line 246: // allocate memory for all features (FIXME: 4 GB barrier on some devices, need to split to multiple buffers)


src/treelearner/ocl/histogram64.cl (1 line):
	- line 693: // FIXME: how to do the above on AMD GPUs??