Canonicalize barriers: ensure all barriers are in a separate BB containing only the barrier and the terminator, with just one predecessor. This allows

Only for internal use; not part of the public API

for a nested region (top-level loop or nested loop)

create the wi-loops around a kernel or subCFG, LastHeader input should be the load block, ContiguousIdx may be any identifyable value (load from undef

From a string like gfxABC:flag+:flag2- only returns gfxABC

Returns true in case the given basic block ends with a barrier, that is, contains only a branch instruction after a barrier call.

Adapted from: https://stackoverflow.com/a/17387176 Returns the last Win32 error, in string format. Returns an empty string if there is no error.

get the wg size values for the loop bounds

get the work-item state alloca a load reads from (through GEPs..)

Should only be called inside device code