model-writtenrun 6232 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Marek Szuba proposed native RISC-V spinlocks using the existing ARM compiler-intrinsic implementation and reported successful PostgreSQL 13.3 testing on physical rv64gc hardware; support for other RISC-V widths remained untested #1.
- Andres Freund endorsed the change and proposed backpatching, which Tom Lane supported because enabling spinlocks should not disrupt configurations that explicitly disabled them #3 #4.
- Tom Lane reviewed the patch and raised architecture-specific atomic support, while Andres Freund explained that generic compiler intrinsics already cover the architecture; they agreed further optimization could wait #5 #6 #7 #8 #9.
- Tom Lane committed the patch from #1 as c32fcac56a and backpatched it to supported branches, separating RISC-V from ARM into its own equivalent code block to allow future architecture-specific additions #9.
- Daniel Gustafsson closed the CommitFest entry as committed, and Christoph Berg subsequently reported successful testing with the change cherry-picked onto PostgreSQL 13.4 in Debian's RISC-V port #10 #12.
- Thomas Munro later proposed declaring 64-bit single-copy atomicity for RISC-V once supporting architectural documentation could be found, and reported that FreeBSD 13 testing worked except for LLVM JIT instruction-selection failures #13.
- Andres Freund requested dumped JIT bitcode, and Thomas Munro supplied a minimal failing query and diagnostic files, identifying an unrecognized generic CPU name #14 #15.
- Thomas Munro experimentally forced the CPU name to generic-rv64, which exposed an unimplemented relocation error and a stack trace entering SystemZ relocation handling; the thread records no completed JIT fix #16.
- In November 2023, Thomas Munro revived the unresolved single-copy atomicity question for RISC-V and s390x, and Christoph Berg emphasized RISC-V's growing Debian importance; spinlock support remained landed, with atomicity declarations and the earlier JIT problems unresolved in this thread #17 #18.