> On 29 Sep 2025, at 16:29, Alfie Richards <[email protected]> wrote:
>
> This looks good to me, but I cannot approve.
>
> Is it easy to add a test case? It would be great to have more coverage of FMV
> in LTO to catch similar issues.
>
Yes. I might provide that case in a few days.
And there is a good news: After another patch to prevent virtual functions from
auto generating FMV (one of my research project), I've successfully built
SPECCPU (intrate, fprate) x (2006, 2017) including all hot functions to have
FMV (even with Fortran benchmarks), without any issues. At least all hot
functions in a generic CPU benchmark is being covered with LTO.
Thanks,
Yangyu Chen
> Thanks,
> Alfie
>
> Sent from Outlook for iOS
> From: Yangyu Chen <[email protected]>
> Sent: Sunday, September 28, 2025 5:14:57 PM
> To: [email protected] <[email protected]>
> Cc: Richard Sandiford <[email protected]>; Alice Carlotti
> <[email protected]>; Michael Meissner <[email protected]>; Alfie
> Richards <[email protected]>; Jason Merrill <[email protected]>; Jeff Law
> <[email protected]>; Kito Cheng <[email protected]>; David Edelsohn
> <[email protected]>
> Subject: Re: [PATCH -fixes] fmv: Check DECL_VIRTUAL_P for virtual functions
> for LTO handling Ping.
>
> On 31/7/2025 23:46, Yangyu Chen wrote:
> > When building with LTO, DECL_VINDEX is not always set for virtual
> > functions, which allows virtual functions to be incorrectly treated as
> > non-virtual and then being multi-versioned, and causes errors sometimes.
> >
> > This patch addresses the issue by ensuring that we also check
> > DECL_VIRTUAL_P in addition to DECL_VINDEX when handling virtual
> > functions during multiversioning.
> >
> > gcc/ChangeLog:
> >
> > * config/aarch64/aarch64.cc
> > (aarch64_generate_version_dispatcher_body): Check DECL_VIRTUAL_P for
> > virtual functions for LTO handling.
> > * config/i386/i386-features.cc
> > (ix86_generate_version_dispatcher_body): Ditto.
> > * config/riscv/riscv.cc (riscv_generate_version_dispatcher_body):
> > Ditto.
> > * config/rs6000/rs6000.cc (rs6000_generate_version_dispatcher_body):
> > Ditto.
> >
> > Signed-off-by: Yangyu Chen <[email protected]>
> > ---
> > This should also be backported to GCC stable branches.
> > ---
> > gcc/config/aarch64/aarch64.cc | 2 +-
> > gcc/config/i386/i386-features.cc | 2 +-
> > gcc/config/riscv/riscv.cc | 2 +-
> > gcc/config/rs6000/rs6000.cc | 2 +-
> > 4 files changed, 4 insertions(+), 4 deletions(-)
> >
> > diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
> > index fff8d9da49d..fd77e24ce12 100644
> > --- a/gcc/config/aarch64/aarch64.cc
> > +++ b/gcc/config/aarch64/aarch64.cc
> > @@ -20763,7 +20763,7 @@ aarch64_generate_version_dispatcher_body (void
> > *node_p)
> > not. This happens for methods in derived classes that override
> > virtual methods in base classes but are not explicitly marked as
> > virtual. */
> > - if (DECL_VINDEX (versn->decl))
> > + if (DECL_VIRTUAL_P (versn->decl) || DECL_VINDEX (versn->decl))
> > sorry ("virtual function multiversioning not supported");
> >
> > fn_ver_vec.safe_push (versn->decl);
> > diff --git a/gcc/config/i386/i386-features.cc
> > b/gcc/config/i386/i386-features.cc
> > index 31f3ee2ef17..043e86ece8e 100644
> > --- a/gcc/config/i386/i386-features.cc
> > +++ b/gcc/config/i386/i386-features.cc
> > @@ -4421,7 +4421,7 @@ ix86_generate_version_dispatcher_body (void *node_p)
> > not. This happens for methods in derived classes that override
> > virtual methods in base classes but are not explicitly marked as
> > virtual. */
> > - if (DECL_VINDEX (versn->decl))
> > + if (DECL_VIRTUAL_P (versn->decl) || DECL_VINDEX (versn->decl))
> > sorry ("virtual function multiversioning not supported");
> >
> > fn_ver_vec.safe_push (versn->decl);
> > diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc
> > index a0657323f65..90f2cbce68e 100644
> > --- a/gcc/config/riscv/riscv.cc
> > +++ b/gcc/config/riscv/riscv.cc
> > @@ -13747,7 +13747,7 @@ riscv_generate_version_dispatcher_body (void
> > *node_p)
> > not. This happens for methods in derived classes that override
> > virtual methods in base classes but are not explicitly marked as
> > virtual. */
> > - if (DECL_VINDEX (versn->decl))
> > + if (DECL_VIRTUAL_P (versn->decl) || DECL_VINDEX (versn->decl))
> > sorry ("virtual function multiversioning not supported");
> >
> > fn_ver_vec.safe_push (versn->decl);
> > diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc
> > index 12dbde2bc63..0c13d524124 100644
> > --- a/gcc/config/rs6000/rs6000.cc
> > +++ b/gcc/config/rs6000/rs6000.cc
> > @@ -25656,7 +25656,7 @@ rs6000_generate_version_dispatcher_body (void
> > *node_p)
> > not. This happens for methods in derived classes that override
> > virtual methods in base classes but are not explicitly marked as
> > virtual. */
> > - if (DECL_VINDEX (version->decl))
> > + if (DECL_VIRTUAL_P (version->decl) || DECL_VINDEX (version->decl))
> > sorry ("Virtual function multiversioning not supported");
> >
> > fn_ver_vec.safe_push (version->decl);
>
