[kaffe] armeb (big-endian) linux JIT test error

[David Tiller]

FYI, This may be similar to the problem I reported this morning wrt jit & xscale doubles getting corrupted when no lower bits are on. My guess is its caused by a sneaky optimization gone wrong, but that's just my opinion.


-----Original Message-----
From:	Dalibor Topic [mailto:robilad at kaffe.org]
Sent:	Thu 2/3/2005 11:04 AM
To:	Dwayne Fontenot <dwayne.fontenot at att.net>
Cc:	Helmer Krämer; Guilhem Lavaux; Timothy Stack; Kiyo Inaba; kaffe at kaffe.org
Subject:	Re: [kaffe] armeb (big-endian) linux JIT test error
Dwayne Fontenot <dwayne.fontenot at att.net> wrote:
> Hi,

Hi Dwayne,

> make check fails on internal test with:
> 
>   field: test__float_nan_to_int = 0x00000000
> asked to translate = TypeConversion.float_nan_to_int(()I)
> successfully verified = TypeConversion.float_nan_to_int(()I)
> pc = 0, npc = 3
> @0:     getstatic 43
> 8:      maxStack = 1, maxLocal = 0, maxTemp = 15
> 20:     move_label_const 0, ?
> 24:     ld 0, [0]
> 2c:     call [0]
> 3c:     move_int_const 1, 1381752 (0x151578)
> 40:     load_float 0, [1]
> pc = 3, npc = 4
> @3:     f2i
> 44:     cvt_float_int r2, f0
> pc = 4, npc = 5
> @4:     ireturn
> 48:     returnarg_int R0,2
> 4c:     branch type 0 at 0x48 from 0x50
> 4c:     maxStack = 1, maxLocal = 0, maxTemp = 15
> L2:
> Translated TypeConversion.float_nan_to_int()I (static) 0x15e560
> Failure for test__float_nan_to_int, got: 7fffffff nan
>   expected: 00000000 0.000000
> FAIL: jitBasic
> ================================
> 1 of 1 tests failed
> Please report to kaffe at kaffe.org
> ================================
> 
> Note that when built in interp mode, all tests (including regression) pass.

That would indeed indicate that something goes bad in the jit on arm(eb) 
regarding the conversion. I've CC:ed the local resident jit(3) gurus for 
pointers how to debug such failures in internal tests for newcomers to 
the kaffe jits. It would make a nice addition to FAQ.debugging :)

I'm particularly interested in how to go about ficing such issues, as 
I'm getting a rather early fasilure myself on powerpc-linux with jit3, 
which I'd like to nail down, if possible, so I'm looking forward to a 
lecture on the internals and how to detect and fix breakage there.

cheers,
dalibor topic

_______________________________________________
kaffe mailing list
kaffe at kaffe.org
http://kaffe.org/cgi-bin/mailman/listinfo/kaffe