jinhongyii opened a new pull request, #20140: URL: https://github.com/apache/tvm/pull/20140
This change removes the remaining replaceable CUDA source helpers needed by TIRx kernels and expresses their semantics through typed IR. ## What changed - add explicit preserve_dst and undefined_dst policies for predicated PTX destinations - add the tcgen05 MXF4 block-scale block32 instruction form - lower SMEM descriptor manipulation and warp uniformization through typed TIR/PTX - classify code after T.device_entry as device-side when binding private PrimFunc helpers - regenerate the T.ptx typing stub ## Validation - PTX dialect: 47 passed, 32 skipped - BindTarget device-helper regression: 1 passed - descriptor lowering regressions: 6 passed - changed-file pre-commit passed - generated tirx.pyi is byte-identical to generator output - downstream affected correctness and fixed-config performance gates passed ## Related Downstream kernel migration: https://github.com/mlc-ai/TIRx-kernels/pull/62 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
