Boyan Li
Boyan Li
Thanks for the interest in cuTile and for the great questions! 1. Pipelining & warp specialization strategies are not exposed in the DSL. Instead, they are automatically chosen and applied...
@irasin That's correct. The compiler decides what hardware-specific optimizations to apply, though we do expose some hints to help you guide the compiler for specific workloads. See https://docs.nvidia.com/cuda/cutile-python/performance.html. Today, cutile-python...
@irasin Yes, `bytecode_buf` contains Tile IR Bytecode.