Re: [PATCH v4 5/6] AutoFDO: Enable machine function split optimization for AutoFDO
From: Masahiro Yamada
Date: Wed Oct 23 2024 - 02:50:36 EST
On Tue, Oct 22, 2024 at 8:28 AM Rong Xu <xur@xxxxxxxxxx> wrote:
>
> On Sun, Oct 20, 2024 at 8:18 PM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
> >
> > On Tue, Oct 15, 2024 at 6:33 AM Rong Xu <xur@xxxxxxxxxx> wrote:
> > >
> > > Enable the machine function split optimization for AutoFDO in Clang.
> > >
> > > Machine function split (MFS) is a pass in the Clang compiler that
> > > splits a function into hot and cold parts. The linker groups all
> > > cold blocks across functions together. This decreases hot code
> > > fragmentation and improves iCache and iTLB utilization.
> > >
> > > MFS requires a profile so this is enabled only for the AutoFDO builds.
> > >
> > > Co-developed-by: Han Shen <shenhan@xxxxxxxxxx>
> > > Signed-off-by: Han Shen <shenhan@xxxxxxxxxx>
> > > Signed-off-by: Rong Xu <xur@xxxxxxxxxx>
> > > Suggested-by: Sriraman Tallam <tmsriram@xxxxxxxxxx>
> > > Suggested-by: Krzysztof Pszeniczny <kpszeniczny@xxxxxxxxxx>
> > > ---
> > > include/asm-generic/vmlinux.lds.h | 6 ++++++
> > > scripts/Makefile.autofdo | 2 ++
> > > 2 files changed, 8 insertions(+)
> > >
> > > diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
> > > index ace617d1af9b..20e46c0917db 100644
> > > --- a/include/asm-generic/vmlinux.lds.h
> > > +++ b/include/asm-generic/vmlinux.lds.h
> > > @@ -565,9 +565,14 @@ defined(CONFIG_AUTOFDO_CLANG)
> > > __unlikely_text_start = .; \
> > > *(.text.unlikely .text.unlikely.*) \
> > > __unlikely_text_end = .;
> > > +#define TEXT_SPLIT \
> > > + __split_text_start = .; \
> > > + *(.text.split .text.split.[0-9a-zA-Z_]*) \
> > > + __split_text_end = .;
> > > #else
> > > #define TEXT_HOT *(.text.hot .text.hot.*)
> > > #define TEXT_UNLIKELY *(.text.unlikely .text.unlikely.*)
> > > +#define TEXT_SPLIT
> > > #endif
> >
> >
> > Why conditional?
>
> The condition is to ensure that we don't change the default kernel
> build by any means.
> The new code will introduce a few new symbols.
Same.
Adding two __split_text_start and __split_text_end markers
do not affect anything. It just increases the kallsyms table slightly.
You can do it unconditionally.
>
> >
> >
> > Where are __unlikely_text_start and __unlikely_text_end used?
>
> These new symbols are currently unreferenced within the kernel source tree.
> However, they provide a valuable means of identifying hot and cold
> sections of text,
> and how large they are. I think they are useful information.
Should be explained in the commit description.
--
Best Regards
Masahiro Yamada