https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #12 from Uroš Bizjak ---
(In reply to Segher Boessenkool from comment #11)
> Should LRA do this? Shouldn't it be done earlier? Or later, in a peephole
> for example?
I think that combine should do this propagation, if the
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #11 from Segher Boessenkool ---
Should LRA do this? Shouldn't it be done earlier? Or later, in a peephole
for example?
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #10 from Uroš Bizjak ---
(In reply to Segher Boessenkool from comment #9)
> Ah. So you want this optimisation (which is currently done by LRA) to be
> done
> by combine as well; it's not that the resulting assembler code for this
>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #9 from Segher Boessenkool ---
Ah. So you want this optimisation (which is currently done by LRA) to be done
by combine as well; it's not that the resulting assembler code for this
testcase
is worse than what you'd like to see. And
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #8 from Uroš Bizjak ---
(In reply to Segher Boessenkool from comment #7)
> It's not clear to me what you would have liked it to do instead?
The loads from constant memory pools always have REG_EQUAL of a relevant
constant attached
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #7 from Segher Boessenkool ---
Hi Uros,
It's not clear to me what you would have liked it to do instead?
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #6 from Uroš Bizjak ---
Here is a bit simpler testcase:
--cut here--
typedef float __v4sf __attribute__((__vector_size__ (16)));
__v4sf
foo (__v4sf x)
{
return x + (__v4sf){ 2.3f, 2.3f, 2.3f, 2.3f };
}
--cut here--
"cc1 -O2" on
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #5 from Uroš Bizjak ---
(In reply to Segher Boessenkool from comment #4)
> It tries twice, first just the substitution, and then that modified with
> the REG_EQUAL. You know a mem is not often valid in the resulting insn,
> but
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #4 from Segher Boessenkool ---
It tries twice, first just the substitution, and then that modified with
the REG_EQUAL. You know a mem is not often valid in the resulting insn,
but combine doesn't, and that is not the same thing as
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #3 from Uroš Bizjak ---
Another similar problem:
__m128
bar (__m128 x)
{
return x + _mm_set1_ps (2.3f);
}
gcc -O2 -msse2 creates following _combine dump:
--cut here--
Trying 6 -> 7:
6: r85:V4SF=[`*.LC0']
REG_EQUAL
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87678
--- Comment #2 from Segher Boessenkool ---
This is a much more general problem in combine. In general it only tries once,
and it only tries the fully simplified form, including known bit values etc.
11 matches
Mail list logo
