Dalibor Topic wrote:
Dalibor Topic wrote:
Robert Schuster wrote:
Hi,
Robert Schuster schrieb:
I now found a cacheflush implementation in libffi. I adjust this for
kaffe and report back.
I took what was said here[0] as a base for an OABI- & EABI-compliant
implementation of the cache flushing.
It compiles nicely (I can only test the EABI-variant here).
Unfortunately it does not work at runtime. That means I get the same
assertion failure as before:
kaffe-bin: exception.c:100: vmExcept_setJNIFrame: Assertion `fp !=
(JNIFrameAddress)0'
Now I dont know whether those two things are related. The only thing I
know for sure is that I use the same cacheflush implementation for cacao
and there it does its job (and fixes a real bug).
Help! :)
Thanks a lot for the patch, I can reproduce the build failure now, and
I've fixed the build for the interpreter engine on arm linux (3
regression tests fail, which I think has more to do with
fdlibm/classpath, than Kaffe, as there is a patch or two for arm
issues in the classpath bug tracker).
I'll look into fixing the build on arm-linux with the jit engines next
using your patch. I've got aurelien's ARM Debian OABI image running,
and I hope I can get one with EABI running, too.
Ok, I've committed a fix for the build problem. Unfortunately, your
patch still failed to compile on OABI, so I decided to use gcc's cache
flushing code instead.
On to the next problem: currently the jit regression test fails at a
floating point test.
Fixed per just committed patch. This lets the jit regression test pass,
and lets Kaffe run the real regression tests. 26 fails on armel OABI
linux with the jit engine. 0 with the interpreter engine.
The interpreter engine on EABI looks like it will also have 0
regressions. I haven't tried the jit on EABI yet.
cheers,
dalibor topic
_______________________________________________
kaffe mailing list
kaffe@kaffe.org
http://kaffe.org/cgi-bin/mailman/listinfo/kaffe
