|Summary:||Missing support for fpsetround()|
|Product:||Base System||Reporter:||Sean Bruno <sbruno>|
|Component:||arm||Assignee:||freebsd-arm mailing list <freebsd-arm>|
|Severity:||Affects Only Me||CC:||Andrew, emaste|
|Bug Depends on:|
Description Sean Bruno 2015-08-07 18:25:37 UTC
TCL ports seem to have a need for fpsetround() on ARM64 that is missing. I'm unsure where this would be added, so I'll just drop these logs here for future review. http://dirty.ysv.freebsd.org/data/11aarch64-default/2015-08-07_14h09m10s/logs/errors/tcl86-8.6.4.log http://dirty.ysv.freebsd.org/data/11aarch64-default/2015-08-07_14h09m10s/logs/errors/tcl85-8.5.18_1.log
Comment 1 Andrew Turner 2015-08-08 11:22:51 UTC
I would like it if we don't need to implement these. I'm planning on a ports exp-run on amd64 to see what breaks when the fp* functions are removed.
Comment 2 Ed Maste 2015-08-10 13:29:49 UTC
From tcl8.4.20/unix/tclUnixInit.c for reference: > #if defined(__FreeBSD__) && defined(__GNUC__) > /* > * Adjust the rounding mode to be more conventional. Note that FreeBSD > * only provides the __fpsetreg() used by the following two for the GNU > * Compiler. When using, say, Intel's icc they break. (Partially based on > * patch in BSD ports system from email@example.com) > */ > > fpsetround(FP_RN); > fpsetmask(0L); > #endif
Comment 3 Ed Maste 2016-06-22 14:37:48 UTC
Should we just add `&& !defined(__aarch64__)` to the conditional for now?
Comment 4 Ed Maste 2016-09-07 13:56:52 UTC
lang/tcl86, failing due to lack of fpsetround, is responsible for 615 skipped ports in the arm64 quarterly ports build http://thunderx1.nyi.freebsd.org/build.html?mastername=110arm64-quarterly&build=421308
Comment 5 Ed Maste 2016-10-28 17:51:11 UTC
TCL has been fixed. Sean are you aware of any other ports affected by this?
Comment 6 Sean Bruno 2016-10-28 18:33:37 UTC
no, I think we're good here. I can't recall any others, but I'll keep my eye out.