[0/3] Kbuild: Some fixdep tweaks

Message

Rasmus Villemoes Aug. 15, 2018, 2:27 p.m. UTC

These patches eliminate two (albeit tiny and shortlived) processes
from the cmd_and_fixdep rule, i.e. from every TU being
compiled. Whether the diffstat below is worth it I'll leave to Kbuild
maintainers to decide.

Rasmus Villemoes (3):
  Kbuild: refactor fixdep to use getopt()
  Kbuild: teach fixdep to optionally remove the depfile
  Kbuild: let fixdep do the renaming to .cmd

 scripts/Kbuild.include | 11 ++++------
 scripts/basic/fixdep.c | 56 ++++++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 52 insertions(+), 15 deletions(-)

Comments

Rasmus Villemoes Sept. 26, 2018, 6:58 p.m. UTC | #1

On 15 August 2018 at 16:27, Rasmus Villemoes <[e-mail]> wrote:
> These patches eliminate two (albeit tiny and shortlived) processes
> from the cmd_and_fixdep rule, i.e. from every TU being
> compiled. Whether the diffstat below is worth it I'll leave to Kbuild
> maintainers to decide.

Ping.

Masahiro Yamada Sept. 28, 2018, 2:21 a.m. UTC | #2

Hi Rasmus,


2018年9月27日(木) 3:58 Rasmus Villemoes <[e-mail]>:
>
> On 15 August 2018 at 16:27, Rasmus Villemoes <[e-mail]> wrote:
> > These patches eliminate two (albeit tiny and shortlived) processes
> > from the cmd_and_fixdep rule, i.e. from every TU being
> > compiled. Whether the diffstat below is worth it I'll leave to Kbuild
> > maintainers to decide.
>
> Ping.


Sorry for delay.
As far as I tested, the performance improvement was not noticeable level.

This patch set actually sits on the fence.
I tend to choose not-apply when I cannot make up my mind.


If you have something more to convince me, please let me know.

Rasmus Villemoes Oct. 1, 2018, 12:19 p.m. UTC | #3

On 2018-09-28 04:21, Masahiro Yamada wrote:
> Hi Rasmus,
> 
> This patch set actually sits on the fence.
> I tend to choose not-apply when I cannot make up my mind.
> 
> If you have something more to convince me, please let me know.

No, that's fine. I agree that the overall performance improvement for
whole kernel builds is hard to measure, but that can be said for many
small optimizations. Anyway, as I said, it's up to you, so let's just
drop it.

Thanks,
Rasmus