[plt-scheme] JIT and flonums

This is fixed in SVN, finally.

Matthew

At Fri, 4 Jan 2008 07:36:29 -0700, "Doug Williams" wrote:
> Any idea on the second problem/feature - subcollections not working (or not
> available) in PLaneT?
> 
> Doug
> 
> On Jan 4, 2008 7:27 AM, Matthew Flatt <mflatt at cs.utah.edu> wrote:
> 
> > The JIT flonum problems have been fixed, and the JIT's flonum support
> > is re-enabled in SVN.
> >
> > There's a good chance that your plots will work, now. The JIT bug was
> > leaving the FP stack in a bad state, which mostly affected FP code
> > *not* generated by the JIT (e.g., the FP code in the plot library).
> >
> > Matthew
> >
> > At Thu, 3 Jan 2008 20:46:39 -0700, Matthew Flatt wrote:
> > > We've discovered today some problems with the JIT's new flonum handling.
> > > That may be the problem with the plots.
> > >
> > > It's not quite fixed, yet, so I've disabled the JIT's flonum code for
> > > now in SVN.
> > >
> > > Matthew
> > >
> > > At Thu, 3 Jan 2008 17:33:34 -0700, "Doug Williams" wrote:
> > > > The problem does seem to be in PLoT.  When I run demo-1.ss (in
> > C:\Program
> > > > Files\PLT-FULL-3.99.0.8\collects\plot\demos on my machine), it
> > generally has
> > > > an empty plot.  Occasionally it will show the plot - maybe 1 time in
> > 10 or
> > > > so, the other times it is just the axes and labels.  [Don't be
> > confused by
> > > > the image in the source file, which also prints out when evaluated.  I
> > just
> > > > comment it out or delete it.  Then just hit the run button repeatedly.
> > > > Also, it has to be run in the Pretty Big language because of
> > build-list.]
> > > >
> > > > I also checked it with some of the science collections tests (for
> > example,
> > > > uniform-int-histogram-int-example.ss in the example subdirectory of
> > the
> > > > science collection, which doesn't have the subcollection reference
> > > > problem).  It exhibits the same behavior.
> > > >
> > > > For the PLaneT subcollection problem, the exact error message I get
> > is:
> > > > bernoulli-example.ss:1:9: planet: bad module path in: (planet "
> > bernoulli.ss"
> > > > ("williams" "science.plt") "random-distributions").  But, (require
> > (planet "
> > > > science.ss" ("williams" "science.plt"))) works fine.
> > > >
> > > > But there is a noticeable speedup in the overall test execution.
> > > >
> > > > On Jan 3, 2008 3:25 PM, Doug Williams <m.douglas.williams at gmail.com>
> > wrote:
> > > >
> > > > > I tried running the science collection examples in 3.99.  Most of
> > the
> > > > > plots were just empty, so I need to see if it's a problem with the
> > PLoT
> > > > > package or with my code.  [They didn't error, just gave empty
> > plots.]  For
> > > > > the ones that did run and which are computationally intensive (with
> > > flonums)
> > > > > - like Chebyshev evaluation and (especially) the ordinary
> > differential
> > > > > equation solvers - they 'seemed' much faster.  But, it's anecdotal -
> > I just
> > > > > eyeballed it.  I have to track down my uses of set-cdr!, etc in the
> > > > > simulation collection before I try any of my larger examples there.
> > > > >
> > > > > One interesting problem came up.  Has the use of subcollections in
> > PLaneT
> > > > > changed with 3.99?  For example, (require (planet "bernoulli.ss"
> > > > > ("williams" "science.plt") "random-distributions")) says it can't
> > find the
> > > > > file, but it works fine in V372 (and before).
> > > > >
> > > > > Doug
> > > > >
> > > > >
> > > > >
> > > > > On Dec 28, 2007 3:55 PM, Noel Welsh <noelwelsh at gmail.com> wrote:
> > > > >
> > > > > > Joyous day!  Should my code run in 3.99 I'll do some benchmarking.
> > > > > >
> > > > > > N.
> > > > > >
> > > > > > On Dec 28, 2007 10:52 PM, Matthew Flatt < mflatt at cs.utah.edu>
> > wrote:
> > > > > > > The latest in SVN (v3.99.0.7) includes new JIT support on x86
> > for
> > > > > > > inlining flonum (i.e., inexact real) arithmetic. I expect to
> > fill in
> > > > > > > x86_64 and PPC support next week.
> > > > > > >
> > > > > > > This change makes inexact-heavy microbenchmarks run 2-3 times as
> > fast.
> > > > > > > It might be useful for real programs, too, but I don't seem to
> > have
> > > > > > any
> > > > > > > interesting floating-point applications to try out.
> > > > > > >
> > > > > > > Matthew
> > > > > > _________________________________________________
> > > > > >  For list-related administrative tasks:
> > > > > >  http://list.cs.brown.edu/mailman/listinfo/plt-scheme
> > > > > >
> > > > >
> > > > >
> > > _________________________________________________
> > >   For list-related administrative tasks:
> > >   http://list.cs.brown.edu/mailman/listinfo/plt-scheme
> >