On Mon, 23 Aug 2010, Raymond Toy wrote:
< On 8/23/10 7:17 PM, Raymond Toy wrote:
< > On 8/23/10 5:47 PM, Leo Butler wrote:
< >> Sorry to restart this thread, but I rebuilt 5.22.1 today with cmucl 20a
< >> from the Debian testing repository and the result is a nightmare of
< >> errors. I've posted the complete output of the build process here:
< >>
< >> http://pastebin.com/zvpxqiCr
< >>
< >> The relevant errors start around line 4500 and all of them seem to
< >> be related to (b)floats and transcendental functions.
< > I'll look into this. However, I do know that 20a had a few bugs in its
< > SSE2 support, which I didn't catch until later. More recent snapshots
< > have these issues fixed, but, of course, that doesn't help if you're
< > using 20a. It might be possible to supply a patch for these issues.
<
< I can reproduce these with 20a and SSE2. With x87, there are no errors
< at all.
<
< I couldn't figure out how to configure maxima to use cmucl with x87.
< (Perhaps configure should have a way to specify Lisp-specific flags?)
< What I ended up doing it basically cd src; make CMUCL_NAME="<cmucl path>
< -fpu x87", where cmucl path is the value of CMUCL_NAME in src/Makefile.
< This is not the best way to do this, but it does show that the issues
< are due to sse2. Once maxima.core is built, there's no need to specify
< -fpu x87 anymore. The core will use x87.
Thanks, Ray. I've filed an addendum to a bug report on debian's cmucl
20a. If I understand another bug report and the dev's reply, it was a
policy decision to ship only sse2.
I should also note that 5.22.1 built with sbcl v 1:1.0.40.0-2 fails
test 383 in rtest16.
Leo
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.