Re: [PATCH v2 3/3] powerpc:selftest update memcmp_64 selftest for VMX implementation

2017-09-25 Thread Simon Guo 
Hi David,
On Mon, Sep 25, 2017 at 09:30:28AM +, David Laight wrote:
> From: wei.guo.si...@gmail.com
> > Sent: 21 September 2017 00:35
> > This patch adjust selftest memcmp_64 so that memcmp selftest can be
> > compiled successfully.
> ...
> >  #define ITERATIONS 1
> > 
> > +#define LARGE_SIZE (5 * 1024)
> > +#define LARGE_ITERATIONS 1000
> ...
> 
> Measuring performance by doing a lot of iterations isn't ideal
> and is pretty pointless.
> Cold cache performance can be more useful.
> Also you don't really want any dynamic branch prediction logic
> tuned to the exact test you keep doing.

I think the (orignal) selftest aims at full coverage of functionality
correctness, since each iteration generates a new data set by random.

Thanks,
- Simon

RE: [PATCH v2 3/3] powerpc:selftest update memcmp_64 selftest for VMX implementation

2017-09-25 Thread David Laight 
From: wei.guo.si...@gmail.com
> Sent: 21 September 2017 00:35
> This patch adjust selftest memcmp_64 so that memcmp selftest can be
> compiled successfully.
...
>  #define ITERATIONS 1
> 
> +#define LARGE_SIZE (5 * 1024)
> +#define LARGE_ITERATIONS 1000
...

Measuring performance by doing a lot of iterations isn't ideal
and is pretty pointless.
Cold cache performance can be more useful.
Also you don't really want any dynamic branch prediction logic
tuned to the exact test you keep doing.

David