The GitHub Actions job "Lint" on tvm.git/main has failed. Run started by GitHub user tqchen (triggered by tqchen).
Head commit for run: dbb1157bc8802a0069b3a99dc405ab17a01bfa91 / Hongyi Jin <[email protected]> [FIX][IR] Complete lazy subscript realization (#20251) ## Motivation Since #20246, expression subscription returns a lazy `SubscriptProxy` until a consumer realizes the expression. Most parser expression boundaries already honor that contract, but PTX predicate operand coercion bypasses the generic realization path. As a result, code such as `setp(..., buffer[0], ...)` dispatches the proxy itself and rejects it as an invalid operand. The delayed realization also means source metadata must travel with the proxy. Without that, realizing `A[0]` creates a `TensorLoad` without the exact subscription span, so downstream diagnostics such as TIRx racecheck and GDN cannot attribute the access to its source location. ## Summary - realize lazy subscript operands used directly by PTX predicate inputs and destinations - preserve parser source spans through `SubscriptProxy` realization - propagate the span through tuple, Relax, and TIRx subscription callbacks This completes the lazy-subscript contract introduced by #20246 and fixes downstream regressions exposed by the TIRx migration in #20247. ## Testing - `cmake --build build -j 16` - `python -m pytest -q tests/python/tvmscript/test_tvmscript_parser_source.py::test_parser_attaches_span_to_nested_tensor_load tests/python/tvmscript/test_tvmscript_parser_source.py::test_parser_attaches_span_to_direct_call tests/python/tirx/codegen/test_ptx_dialect.py::test_ptx_comparison_selection_dispatch` - downstream wiki-kernel integration: 5 passed - downstream full `tirx_tools` gate: 2933 passed Report URL: https://github.com/apache/tvm/actions/runs/33504060165 With regards, GitHub Actions via GitBox --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
