Repository navigation
riscv: allow single-precision float ABI (lp64f / ilp32f) - #349
Merged
Merged
Conversation
The header rejected __riscv_float_abi_single with an #error, but the F and D extensions share the same FCSR. The hard-float inline implementations below the guard use only csrr/csrw/csrc on fcsr and fflags, both of which are present whenever F is present. So lp64f / ilp32f can use the same code path as the double ABI. Closes JuliaMath#262. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The lp64f / ilp32f ABI used to be rejected by openlibm_fenv_riscv.h with an #error. Add a compile-only cross job that rebuilds the static library with -mabi=lp64f so the removed #error stays removed. A full lp64f shared-lib link or qemu run is impossible because Ubuntu's gcc-riscv64-linux-gnu ships only the lp64d sysroot; the static archive needs no libc link, so it still compiles riscv64/fenv.c (and every other source) under the single-float ABI. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ViralBShah
force-pushed
the
fix-262-riscv-lp64f
branch
from
June 22, 2026 23:33
4ed0692 to
ef48409
Compare
maleadt
added a commit
to JuliaLang/julia
that referenced
this pull request
Sep 3, 2026
Picks up JuliaMath/openlibm#360, which stops the riscv64 port from exporting its fenv functions. Those symbols shadowed glibc's through the dynamic linker while carrying FreeBSD's pre-shifted `FE_*` encodings, so `fesetround` rejected every constant a caller had compiled against the system header and left the rounding mode untouched. Julia's own rounding path stopped going through that symbol in `6081d7ceb7`, but anything else linked against the bundled library still did. The release also carries three riscv64 fixes that never reached the 0.8.7 build: JuliaMath/openlibm#324 (`feholdexcept` always returned -1), and JuliaMath/openlibm#330 and JuliaMath/openlibm#349 for the lp64f ABI. Assisted-by: Claude Code (Opus 5)
social4hyq
pushed a commit
to social4hyq/homebrew-core
that referenced
this pull request
Sep 20, 2026
openlibm 0.8.8 Created-by: HarmonybrewBot Commit-by: HarmonybrewBot Merged-by: HarmonybrewBot Description: Created by `brew bump` --- Created with `brew bump-formula-pr`.<details> <summary>release notes</summary> <pre>## What's Changed * Update riscv64 fenv.c initialization to fix warnings by @ViralBShah in JuliaMath/openlibm#330 * Musl compatibility fixes for s390 and powerpc by @sertonix in JuliaMath/openlibm#331 * Fix cmake error for 4.x by @HIllya51 in JuliaMath/openlibm#333 * Bump actions/checkout from 4 to 5 by @dependabot[bot] in JuliaMath/openlibm#335 * Bump actions/upload-artifact from 4 to 5 by @dependabot[bot] in JuliaMath/openlibm#336 * Bump actions/checkout from 5 to 6 by @dependabot[bot] in JuliaMath/openlibm#338 * Bump actions/upload-artifact from 5 to 6 by @dependabot[bot] in JuliaMath/openlibm#339 * Bump actions/upload-artifact from 6 to 7 by @dependabot[bot] in JuliaMath/openlibm#341 * Bump codecov/codecov-action from 5 to 6 by @dependabot[bot] in JuliaMath/openlibm#342 * ci: add native ARM (Linux + Windows) and Intel macOS runners by @ViralBShah in JuliaMath/openlibm#345 * ci: workflow hygiene (concurrency, timeouts, loongarch fold, mips64, fork-safe codecov) by @ViralBShah in JuliaMath/openlibm#346 * riscv: Fix feholdexcept() by @AlekseyZhmulin in JuliaMath/openlibm#324 * CI: add -Werror strict lane and an auto-discovered regression-test harness by @ViralBShah in JuliaMath/openlibm#350 * ci: speed up Windows jobs by trimming the msys2 install by @ViralBShah in JuliaMath/openlibm#354 * ci: make codecov coverage status informational by @ViralBShah in JuliaMath/openlibm#358 * riscv: allow single-precision float ABI (lp64f / ilp32f) by @ViralBShah in JuliaMath/openlibm#349 * Fix powl() returning NaN instead of +0 on extreme underflow (#334) by @ViralBShah in JuliaMath/openlibm#351 * Make powl() thread-safe (#222) by @ViralBShah in JuliaMath/openlibm#355 * Bump codecov/codecov-action from 6 to 7 by @dependabot[bot] in JuliaMath/openlibm#359 * riscv64: Keep hard-float fenv functions private by @maleadt in JuliaMath/openlibm#360 ## New Contributors * @sertonix made their first contribution in JuliaMath/openlibm#331 * @HIllya51 made their first contribution in JuliaMath/openlibm#333 * @AlekseyZhmulin made their first contribution in JuliaMath/openlibm#324 **Full Changelog**: https://github.com/JuliaMath/openlibm/compare/v0.8.7...v0.8.8</pre> <p>View the full release notes at <a href="https://github.com/JuliaMath/openlibm/releases/tag/v0.8.8">https://github.com/JuliaMath/openlibm/releases/tag/v0.8.8</a>.</p> </details> <hr> See merge request: Harmonybrew/homebrew-core!18388
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fixes #262. The RISC-V fenv header rejected
__riscv_float_abi_single(lp64f, ilp32f) with an `#error`, but the rejection isn't warranted: the F and D extensions share the same `fcsr` / `fflags` CSRs. The hard-float inline implementations below the guard use only `csrr`/`csrw`/`csrc` on those CSRs, all of which are present whenever F is present.So lp64f / ilp32f can use the existing hard-float code path unchanged.
```diff
-#if !defined(__riscv_float_abi_soft) && !defined(__riscv_float_abi_double)
-#if defined(__riscv_float_abi_single)
-#error single precision floating point ABI not supported
-#else
+#if !defined(__riscv_float_abi_soft) && !defined(__riscv_float_abi_double) && \
#error compiler did not set soft/hard float macros
#endif
-#endif
```
Test plan
doubleABI users; lp64f users no longer hit the `#error`).🤖 Generated with Claude Code