avr-libc-dev
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [avr-libc-dev] calculating benchmarks


From: Sean D'Epagnier
Subject: Re: [avr-libc-dev] calculating benchmarks
Date: Sat, 21 Feb 2009 07:30:23 -0700

Hi,

I was able to run the script.

The test confirms that the fixed point math routines are much faster
for avrs without a multiplier, and slightly slower for avrs with a
multiplier in most cases.

One problem I had is when executing a function which is an inlined
function in a header file which calls other functions. it gives the
result of '7' cycles which I know to be wrong.

What do you think about integrating this script into the build for
avr-libc somehow?  Then all the library functions could get updated
all the time.   This would also be helpful to show how xmega compares
since it has different cycle counts for many instructions (I know
xmega isn't supported by simulavr yet)

Sean

On 2/20/09, Dmitry K. <address@hidden> wrote:
>> > > Are you able to run this script with the latest simulavr?
>> >
>> > As soon as
>> >
>> > > I try to call either floating or fixed point routines, I am getting
>> > > errors:
>> > >
>> > > rm: cannot remove `core_avr_dump.core': No such file or directory
>
> Hi,
> I have try the latest simulavr: the same error in result.
>
> 1. The latest simulavr is partially broken: the run of first
> test (__addsf3) is failed with 'atmega8' target with message:
>
>    devsupp.c:321: ERROR: reference vdev hasn't been created yet
>
> The 'at90s8515' target is OK.
> (Eric, I will look this in details.  I am interesting in
> simulavr usage).
>
> 2. My script was too poor in extracting of clock number.
> The format of this line is changed in comparison to 0.1.2.1
> version, this leads to wrong result, like 7 cycles.  I have
> correct this (in attach): '[0-9]+ clock cycles'.
>
> Now you can run the latest simulavr with 'at90s8515' target
> only (MCU_LIST), or you can install simulavr-0.1.2.1 with
> the patch (in attach).  The critical is only 1 part of patch:
> where the division by zero is excluded.
>
> Regards,
> Dmitry.
>




reply via email to

[Prev in Thread] Current Thread [Next in Thread]