https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78687
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Status|UNCONFIRMED |NEW
Last reconfirmed| |2016-12-06
CC| |jamborm at gcc dot gnu.org,
| |rguenth at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Trunk rejects it with (might be worth another bug?)
t.C: In function ‘ref_proxy<qual_option, inplace_ref<qual_option> >
make_object_4(ref_proxy<option_2, variant_ref<option_1, option_2> >&&)’:
t.C:469:39: error: conversion from ‘inplace_ref<qual_option>’ to
‘ref_proxy<qual_option, inplace_ref<qual_option> >’ is ambiguous
return make_object_3(std::move(a0));
^
t.C:394:8: note: candidate: constexpr
inplace_ref<qual_option>::inplace_ref(const inplace_ref<qual_option>&)
struct inplace_ref
^~~~~~~~~~~
t.C:380:17: note: inherited here
using Base::Base;
^~~~
t.C:394:8: note: candidate: constexpr
inplace_ref<qual_option>::inplace_ref(inplace_ref<qual_option>&&)
struct inplace_ref
^~~~~~~~~~~
t.C:380:17: note: inherited here
using Base::Base;
^~~~
t.C:387:5: note: candidate: ref_proxy<T, Base>::ref_proxy(Base) [with T =
qual_option; Base = inplace_ref<qual_option>]
ref_proxy(Base ptr)
^~~~~~~~~
Confirmed with GCC 6.2 - we have quite many aggregate temporaries not optimized
away:
__attribute__((noinline))
ref_proxy<qual_option, inplace_ref<qual_option> > f() ()
{
struct qual_optionD.34988 D.41115;
struct ref_proxyD.34990 D.41114;
struct qual_optionD.34988 innerD.41113;
struct ref_proxyD.34990 typeD.41111;
struct ref_proxyD.34990 typeD.41112;
struct ref_proxyD.36874 D.41109;
struct ref_proxyD.36874 ptrD.41110;
struct inplace_refD.37561 D.41107;
struct inplace_refD.37561 ptrD.41106;
<bb 2>:
MEM[(struct ref_proxyD.36874 *)&D.41109 + 40B] = 2;
ptrD.41110 = D.41109;
MEM[(struct &)&D.41114] ={v} {CLOBBER};
MEM[(struct variantD.35001 *)&ptrD.41110 + 40B] = 2;
MEM[(struct variant_refD.34986 *)&D.41114].inner_storage_D.35021 =
ptrD.41110.D.36914.inner_storage_D.35021;
MEM[(struct ref_proxyD.34990 *)&D.41114 + 40B] = 2;
typeD.41111 = D.41114;
MEM[(struct ref_proxyD.34990 *)&typeD.41111 + 40B] = 2;
typeD.41112 = typeD.41111;
MEM[(struct &)&D.41115] ={v} {CLOBBER};
MEM[(struct ref_proxyD.34990 *)&typeD.41112 + 40B] = 2;
D.41115.type_D.36509 = typeD.41112;
MEM[(struct qual_optionD.34988 *)&D.41115 + 40B] = 2;
MEM[(struct qual_optionD.34988 *)&D.41115 + 48B] = 0;
innerD.41113 = D.41115;
MEM[(struct &)&D.41107] ={v} {CLOBBER};
MEM[(struct qual_optionD.34988 *)&innerD.41113 + 48B] = 0;
D.41107.inner_D.37568 = innerD.41113;
D.41115 ={v} {CLOBBER};
D.41114 ={v} {CLOBBER};
MEM[(struct inplace_refD.37561 *)&D.41107 + 48B] = 0;
ptrD.41106 = D.41107;
MEM[(struct &)&<retval>] ={v} {CLOBBER};
MEM[(struct inplace_refD.37561 *)&ptrD.41106 + 48B] = 0;
<retval>.D.37651 = ptrD.41106;
return <retval>;
}
it usually is SRAs job to do this but it seems to fail in this case:
<bb 2>:
MEM[(struct &)&D.37695] ={v} {CLOBBER};
SR.31_24 = 2;
MEM[(struct ref_proxyD.36874 *)&D.41109 + 40B] = SR.31_24;
SR.34_26 = SR.31_24;
MEM[(struct ref_proxyD.36874 *)&D.41109 + 40B] = SR.34_26;
ptrD.41110 = D.41109;
^^^ aggregate copy not scalarized
ptr$D36914$inner_storage_$_storage$_which_28 = SR.34_26;
MEM[(struct &)&D.41114] ={v} {CLOBBER};
MEM[(struct variantD.35001 *)&ptrD.41110 + 40B] =
ptr$D36914$inner_storage_$_storage$_which_28;
in fact ptrD.41110 is not scalarized at all. Looks like the init from D.41109
makes us fail here.
Martin?
