srielau opened a new pull request, #59480: URL: https://github.com/apache/spark/pull/59480
### What changes were proposed in this pull request? This PR is stacked on #58410. The incremental commit adds a generic IEEE 754 BID-k path on top of the specialized `Bid64` / `Bid128` kernels. - `FormatParams` describes packed widths `k = 32, 64, ..., 256` (`p`, `w`, `t`, `emax`, `bias`). - `BidCodec` pack/unpacks the combination field into little-endian `long[]` words. - `BidK` / `BidKRaw` implement add, sub, mul, div, sqrt, FMA, quantize, and convert using `DecNum` and `UIntLimbs` (up to 4 longs). - Decimal96 is the proving format, including convert to BID64/128. - Decimal160 through Decimal256 use the same arithmetic path. - Transcendentals for `k <= 128` convert through `Bid128` and the binary128 DPML engine. `k > 128` libm throws until a binary256 engine is funded. `Bid64` and `Bid128` remain the hot-path kernels. No Spark SQL consumer is added. ### Why are the changes needed? IEEE 754 decimal interchange is parameterized by packed width `k`, not only 64 and 128. Spark's decimal precision goes to 38, which is above decimal64/128's 16/34 digits. Decimal96 (25 digits) and the wider formats let LibBID cover that range without inventing a non-IEEE encoding. ### Does this PR introduce _any_ user-facing change? No. LibBID remains an unpublished internal module with no Spark execution path. ### How was this patch tested? - `BidKTest`: IEEE format table, codec round-trips vs `Bid64`/`Bid128`, Decimal96 convert, add/mul/div/sqrt property tests vs `BigDecimal` for k=96 and 160-256, five rounding modes, SQL adapters, DPML `exp` on Decimal96, and unsupported libm for k=160. - `libbid/testOnly org.bidfp.LibraryTests` (29 tests, including Intel vectors and `bidK`). - `./dev/lint-java` (checkstyle passed). ### Was this patch authored or co-authored using generative AI tooling? Generated-by: Cursor Grok 4.6 -- 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]
