[kaffe] Re: mipsel/jit3 regression test results
robilad at kaffe.org
Mon Oct 18 08:04:37 PDT 2004
Kevin D. Kissell wrote:
>>The threads about MIPS & jit with Kevin from 2002 (March-June, I
>>believe), 2003, and 2004 with him & Casey also may contain some patches
>>that haven't been applied yet. If you, or Casey are aware of patches
>>that I have let slip under the floor, please point me to URLs, and I'll
>>be very glad to merge them in. There was a lot of intersting commentary
>>on the JIT engine between Kevin, Casey and Tim that was over my head,
>>essentially, as I'm not very familiar with the core JIT engine. But I'd
>>be glad to play patch-bot :)
> Alas, I never submitted any JIT patches as such, because I didn't see any point
> in submitting patches that weren't against the head of the tree (or at least the
> most current "stable" revision), but I couldn't even get the more recent sources
> to build with JIT, and was working mostly from a patched 1.07 tree. That
> version of the JIT actually passed a lot of the regression tests, and could run
> the embedded Caffeinemark. So my "patches" conisted of "I changed X to Y"
> rather than a source diff, which I know is a good deal less useful.
If you by any chance find a copy of those changes somewhere, I guess one
could try to resurrect your changes.
> My recollection is that someone did indeed apply/check in a patch which
> fixed the erroneous allocation of MIPS argument registers by the JIT, but I
> don't think my suggestion to force the use of temp registers, rather than roll
> through spilling and restoring "dead" values, was ever picked up. The
> last time I went at debugging the thing, I was able to step through a certain
> amount of JITed code, walking gdb from JIT buffer to JIT buffer, up to the
> point where it computed an address based on a null pointer and crashed.
> My hypothesis was that there was an assumption in the high-level JIT code
> that a value would have been flushed out to memory, but that in the case
> of the low-level MIPS JIT code, it had not been, leaving the location as
> an uninitialized deathtrap, but that's just an intuition. I'm working on other
> stuff these days, and regret that I've got no time to fire up my MIPS/Linux
> box and dig into this further. I'll just have to track these threads and offer
> whatever advice comes to mind from memory.
No problem, thanks for all your good suggestions and advice. I hope
Mikolaj will be able to fix mips jit using your notes.
More information about the kaffe