[Bug string/25131] memcpy perfomance problem with ARM 32 A9be due to high cache-misses

["adhemerval.zanella at linaro dot org" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=25131

Adhemerval Zanella <adhemerval.zanella at linaro dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |adhemerval.zanella at linaro dot o
                   |                            |rg

--- Comment #7 from Adhemerval Zanella <adhemerval.zanella at linaro dot org> ---
(In reply to Florian Weimer from comment #6)
> The usual way we deal with this is to use some mechanism to probe for the
> platform in question and use an IFUNC handler to switch to the optimized
> version for this platform only.
> 
> I'm not sure if 32-bit Arm platform variants can be identified easily at the
> CPU level or from the auxiliary vector. It may be necessary to do some
> kernel work first.

Another option is to try to improve the generic armv7 implementation, the main
challenge is to create a strategy that does not regress on other chip
implementation.

One option is to check if !USE_VFP code path (that does use PLD instruction)
yields any performance gain and check against at least some different chip
implementations.

-- 
You are receiving this mail because:
You are on the CC list for the bug.