JIT support for m68k/{linux,netbsd}

Kiyo Inaba inaba at src.ricoh.co.jp
Sat Mar 27 20:39:20 PST 1999


Well, JIT support for m68/{linux,netbsd} comes close to be completed.
I summarize what you can get from the latest snap.

m68k/linux/jit:
	I only tested with '--with-staticlib --with-staticvm --without-x'.
	Or actually, I usually do it on the x86/linux box.
	DoublePrint, SoTimeout, BeanBug, ProcessClassInst, SoInterrupt fail.
	But on my machine, the test reports more problems, when I just do
	'make check'. When I do 'make TESTS=foobar.java check' some of them
	works fine. Since I am not so familiar with m68k/linux I don't know
	what's wrong, but I guess swap problem or kernel bug.

m68k/linux/intrp:
	You have to specify '--with-engine=intrp' explicitly.
	There should be no problem if you test each of them individually,
	but 'run everything by one command' on my machine reports several
	errors.

m68k/netbsd1/jit:
	The most funny thing happens here.
	Test reports it can not compile 'TestFloatDouble.java' with

	../../../kaffe-snap/test/regression/TestFloatDouble.java:23: floating point number too large
			float f1 = 500.005f;

	../../../kaffe-snap/test/regression/TestFloatDouble.java:24: floating point number too large
			float f2 = 400.004f;
				   ^
	2 errors

	And it happens with Klasses.jar revision 1.81 or higher. I don't
	say it's bug in Klasses.jar (because this works fine for intrp
	or other platforms) and of course it's bug in jit code for m68k.

	But I have to point out which class files in Klasses.jar makes
	this problem before debug.

	Except for this, OverFlow and ProcessClassInst fail.

m68k/netbsd1/intrp:
	OverFlow fails.
	Since OverFlow does not fail for linux box, it should be netbsd
	specific or Sun3 specific (which I am not so sure).

These are the issues for next weekend :-)
# And also, '-O2' support for jit is needed.

Kiyo


More information about the kaffe mailing list