This class is mostly a dummy implementation

describes how the contents of a vector vary with the vectorized dimension

Do not change the order of these enums, as the compiler may generate atomic calls based on the int values of these enums

Macros do not handle types with commans in them, so we do things manually

Expands aggregates into primitive function arguments. Aggregate types to expand are expected to be marked using the ByVal attribute.

constructs SSA form for allocas associates every pointer value with the set of allocas it originates from the results of this analysis are used by the

Note: CUevent_st* == cudaEvent_t

Forward declare CUstream_st instead of including cuda_runtime_api.h. It's not possible to include both HIP and CUDA headers since they define conflict

Utility type to generate the set of all function declarations implicitly or explicitly reachable from some initial declaration. NOTE: Must only be us

divergence propagator for reducible CFGs

Replaces extern globals with calls to the SSCP dynamic local memory builtin. This enables support e.g. for CUDA's extern __shared__ syntax.