* [PATCH 0/3] Reduce length of the eagerfpu path during x86 context switches
@ 2014-08-06 12:55 Mel Gorman
0 siblings, 0 replies; 4+ messages in thread
From: Mel Gorman @ 2014-08-06 12:55 UTC (permalink / raw)
To: H Peter Anvin, Suresh Siddha; +Cc: Linux-X86, LKML, Mel Gorman
Eager FPU switching is used on CPUs that support xsave on the grounds
that CPUs that support it can optimise the switch with xsaveopt and xrstor
instead of serialising by updating cr0.TS which has serialising semantics.
The path for eagerfpu is fatter than it needs to be because it still
maintains the fpu_counter for lazy FPU switches even though the information
is never used. This patch splits the paths optimises the eagerfpu path a
little. The benefit is marginal, it was just noticed when looking at why
integer-only workloads were spending time saving/restoring FPU states.
arch/x86/include/asm/fpu-internal.h | 46 +++++++++++++++++++++++++++++--------
arch/x86/kernel/process_32.c | 2 +-
arch/x86/kernel/process_64.c | 2 +-
3 files changed, 38 insertions(+), 12 deletions(-)
--
1.8.4.5
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH 0/3] Reduce length of the eagerfpu path during x86 context switches
2014-08-27 16:03 ` Mel Gorman
@ 2014-08-27 17:03 ` H. Peter Anvin
0 siblings, 0 replies; 4+ messages in thread
From: H. Peter Anvin @ 2014-08-27 17:03 UTC (permalink / raw)
To: Mel Gorman, Suresh Siddha; +Cc: Mike Galbraith, Linux-X86, LKML
On 08/27/2014 09:03 AM, Mel Gorman wrote:
> On Wed, Aug 06, 2014 at 01:55:45PM +0100, Mel Gorman wrote:
>> Eager FPU switching is used on CPUs that support xsave on the grounds
>> that CPUs that support it can optimise the switch with xsaveopt and xrstor
>> instead of serialising by updating cr0.TS which has serialising semantics.
>>
>> The path for eagerfpu is fatter than it needs to be because it still
>> maintains the fpu_counter for lazy FPU switches even though the information
>> is never used. This patch splits the paths optimises the eagerfpu path a
>> little. The benefit is marginal, it was just noticed when looking at why
>> integer-only workloads were spending time saving/restoring FPU states.
>>
>
> This was initially sent when it would collide with the merge window
> which was stupid timing. Nothing has actually changed since but I wonder
> if anyone had a chance to take a look at this patches?
>
Looking at it now.
-hpa
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH 0/3] Reduce length of the eagerfpu path during x86 context switches
2014-08-06 12:55 Mel Gorman
@ 2014-08-27 16:03 ` Mel Gorman
2014-08-27 17:03 ` H. Peter Anvin
0 siblings, 1 reply; 4+ messages in thread
From: Mel Gorman @ 2014-08-27 16:03 UTC (permalink / raw)
To: H Peter Anvin, Suresh Siddha; +Cc: Mike Galbraith, Linux-X86, LKML
On Wed, Aug 06, 2014 at 01:55:45PM +0100, Mel Gorman wrote:
> Eager FPU switching is used on CPUs that support xsave on the grounds
> that CPUs that support it can optimise the switch with xsaveopt and xrstor
> instead of serialising by updating cr0.TS which has serialising semantics.
>
> The path for eagerfpu is fatter than it needs to be because it still
> maintains the fpu_counter for lazy FPU switches even though the information
> is never used. This patch splits the paths optimises the eagerfpu path a
> little. The benefit is marginal, it was just noticed when looking at why
> integer-only workloads were spending time saving/restoring FPU states.
>
This was initially sent when it would collide with the merge window
which was stupid timing. Nothing has actually changed since but I wonder
if anyone had a chance to take a look at this patches?
Thanks
--
Mel Gorman
SUSE Labs
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH 0/3] Reduce length of the eagerfpu path during x86 context switches
@ 2014-08-06 12:55 Mel Gorman
2014-08-27 16:03 ` Mel Gorman
0 siblings, 1 reply; 4+ messages in thread
From: Mel Gorman @ 2014-08-06 12:55 UTC (permalink / raw)
To: H Peter Anvin, Suresh Siddha; +Cc: Mike Galbraith, Linux-X86, LKML, Mel Gorman
Eager FPU switching is used on CPUs that support xsave on the grounds
that CPUs that support it can optimise the switch with xsaveopt and xrstor
instead of serialising by updating cr0.TS which has serialising semantics.
The path for eagerfpu is fatter than it needs to be because it still
maintains the fpu_counter for lazy FPU switches even though the information
is never used. This patch splits the paths optimises the eagerfpu path a
little. The benefit is marginal, it was just noticed when looking at why
integer-only workloads were spending time saving/restoring FPU states.
arch/x86/include/asm/fpu-internal.h | 46 +++++++++++++++++++++++++++++--------
arch/x86/kernel/process_32.c | 2 +-
arch/x86/kernel/process_64.c | 2 +-
3 files changed, 38 insertions(+), 12 deletions(-)
--
1.8.4.5
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2014-08-27 17:03 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-08-06 12:55 [PATCH 0/3] Reduce length of the eagerfpu path during x86 context switches Mel Gorman
2014-08-06 12:55 Mel Gorman
2014-08-27 16:03 ` Mel Gorman
2014-08-27 17:03 ` H. Peter Anvin
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.