1 Jan 2009 01:41
Re: Re: GCC 3.4.6 and Mozilla's jit
Dave Yeo <daveryeo <at> telus.net>
2009-01-01 00:41:00 GMT
2009-01-01 00:41:00 GMT
Hi Paul, On 12/31/08 03:12 pm, Paul Smedley wrote: > > Hi Guys, > > Dave Yeo wrote: >> Maybe related, I have noticed that running the regression tests on >> FFmpeg causes a trap in libc63 if compiled with gcc 3.4.6 or 4.3.2 >> while with 3.3.5 it just outputs bad data. I'm not knowledgeable >> enough to say but it seems the bad data comes from the same place in >> the code as the trap. >> This raises the question, is 3.3.5 actually handling the jit correctly >> or just failing in a more minor way? >> Anyways thanks for looking into this and perhaps Paul will have some >> ideas why the newer compilers are failing here. > > I'd love to be able to provide some input here but I really have no > clues :( Can you point me to the failing ffmpeg test? That may prove a > simpler testcase than Mozilla. Seems the failure has changed with the latest FFmpeg. It was quite a while ago that I tested with 3.4.6 so perhaps the problem was already fixed. Since the last time I ran the regression tests the number of errors has increased from one to three and seem slightly different. 3.4.6 and 3.3.5 both fail the same. Previously with 4.3.2 FFmpeg was trapping in libc. when doing the dv-50 test IIRC. Now crashing in the svq1 test in FFmpeg_g. This could be related to the optimization level. Unluckily FFmpeg doesn't compile with --disable-optimization. (can't find a register)(Continue reading)
RSS Feed