[m-rev.] for review: pack small args next to local secondary tags

[Zoltan Somogyi]

On Sat, 14 Jul 2018 12:44:14 -0400 (EDT), Julien Fischer <jfischer at opturion.com> wrote:

> On Thu, 12 Jul 2018, Zoltan Somogyi wrote:
> 
> > On Tue, 3 Jul 2018 21:14:22 -0400 (EDT), Julien Fischer <jfischer at opturion.com> wrote:
> >> Specifically, in asm_fast.gc.decldebug.trseg.stseg, the following
> >> fail on my system (x86-64,linux):
> >>
> >>     hard_coded/arith_int16
> >>     hard_coded/arith_int32
> >>     hard_coded/arith_int8
> >>     hard_coded/pack_int32
> >>
> >> The packing one is most likely failing bcause the arithmetic ones are.
> >> (This looks like a regression of the issue fixed in commit b765dae,
> >> but it isn't since the test case for that passes.)
> >
> > The attached diff fixes these failures.
> 
> As of rotd-2018-07-14 the last one of those is still failing for me.
> 
> diff -u pack_int32.exp pack_int32.out
> --- pack_int32.exp      2018-05-16 20:06:06.000000000 -0400
> +++ pack_int32.out      2018-07-14 12:39:16.000000000 -0400
> @@ -1,3 +1,3 @@
> -struct(-2147483648, 2147483647, -32768, 32767)
> +struct(-2147483648, -32768, 990059265, 25)
>   struct(-2147483648, 2147483647, -32768, 32767)
>   struct(-2147483648, 2147483647, -32768, 32767)
> 
> (Same grade and system as above.)

It passed for me in asm_fast.gc.decldebug.stseg, i.e. NOT trseg,
but trailing should have no bearing on this.

Did the failure happen with allow_packing_* enabled or disabled?
And can you confirm that you were generating 64 bit code,
not 32 bit?

Zoltan.