On Mar 10, 2021, Hans-Peter Nilsson <h...@bitrange.com> wrote:
> On Wed, 10 Mar 2021, Alexandre Oliva wrote:
>>
>> The test expects shrink-wrapping of the fsqrt call, but that will only
>> occur when there is a usable sqrt insn.
>>
>> Arrange for dejagnu to add the options that enable the sqrt insn, if
>> one is available, and to skip the test otherwise.
>>
>>
>> H-P, this *should* obviate the mmix-specific dg-skip-if.
> Unfortunately it doesn't.
Uhh :-(
>> Would it be
>> easy for you to confirm that this is the case and, if so, drop it?
> About as easy as for anyone (this is a compile-test),
I figured you'd have a recent toolchain around ;-) Thanks!
> FAIL: gcc.dg/cdce3.c scan-tree-dump cdce "cdce3.c:12: [^\n\r]*
> function call is shrink-wrapped into error conditions."
How surprising! My understanding was that dg-add-options <feature>
implicitly implies dg-require-effective-target <feature>. That was how
I'd read et-dg-runtest.
Now I see the _runtime after the check_effective_target_${target} there,
and my understanding is updated, so some pending changes may need
revisiting. Oh my...
> The dump files and assembly file show no obvious clues to me as
> to what is supposed to happen; attached.
The test just should keep on not running on mmix. Without an insn for
sqrt, there's no point in shrink-wrapping, and the code that performs
the optimization is smart enough to realize that, so it just leaves the
code alone.
--
Alexandre Oliva, happy hacker https://FSFLA.org/blogs/lxo/
Free Software Activist GNU Toolchain Engineer
Vim, Vi, Voltei pro Emacs -- GNUlius Caesar
