These options are defined for Nvidia PTX:
Generate code for 32-bit or 64-bit ABI.
Link in code for a __main kernel. This is for stand-alone instead of offloading execution.
Apply partitioned execution optimizations. This is the default when any level of optimization is selected.
Generate code that does not use
directly for stack storage. Instead, a per-warp stack pointer is
maintained explicitly. This enables variable-length stack allocation (with
variable-length arrays or
alloca), and when global memory is used for
underlying storage, makes it possible to access automatic variables from other
threads, or with atomic instructions. This code generation variant is used
for OpenMP offloading, but the option is exposed on its own for the purpose
of testing the compiler; to generate code suitable for linking into programs
using OpenMP offloading, use option -mgomp.
Switch to code generation variant that allows to execute all threads in each
warp, while maintaining memory state and side effects as if only one thread
in each warp was active outside of OpenMP SIMD regions. All atomic operations
and calls to runtime (malloc, free, vprintf) are conditionally executed (iff
current lane index equals the master lane index), and the register being
assigned is copied via a shuffle instruction from the master lane. Outside of
SIMD regions lane 0 is the master; inside, each thread sees itself as the
master. Shared memory array
int __nvptx_uni stores all-zeros or
all-ones bitmasks for each warp, indicating current mode (0 outside of SIMD
regions). Each thread can bitwise-and the bitmask at position
with current lane index to compute the master lane index.
Generate code for use in OpenMP offloading: enables -msoft-stack and -muniform-simt options, and selects corresponding multilib variant.