tqchen opened a new pull request, #19621: URL: https://github.com/apache/tvm/pull/19621
## Summary Move `include/tvm/runtime/nvtx.h` to `include/tvm/support/cuda/nvtx.h` under namespace `tvm::support`, delete `src/runtime/nvtx.cc`, and inline the two trivial constructor/destructor bodies so the header is header-only. No FFI registrations or static state to relocate; the .cc was a pure RAII wrapper. The CMake `TVM_NVTX_ENABLED=1` gate moves from a per-file `set_source_files_properties` on `nvtx.cc` (now gone) to a target-wide `add_compile_definitions` inside the same `if(USE_CUDA AND USE_NVTX)` block, so every TU that includes the inline header agrees on the definition. ## Files - create: `include/tvm/support/cuda/nvtx.h` - delete: `include/tvm/runtime/nvtx.h`, `src/runtime/nvtx.cc` - modify: `CMakeLists.txt`, `src/runtime/vm/attn_utils.h`, `src/runtime/vm/paged_kv_cache.cc`, `src/runtime/vm/vm.cc` ## Test plan - [x] Default build (`USE_NVTX=OFF`) — full `ninja` succeeds; no undefined references to `NVTXScopedRange`. - [x] `ninja cpptest` — 118/118 passed. - [x] Python smoke: `import tvm; tvm.cuda(0).exist` — True. - [x] `pre-commit run --all-files` — all hooks passed (clang-format reformatted and re-verified clean). -- 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]
