Bob Lees wrote:
>>From memory, I think that EABI systems are supposed to have binary
> compatibility with non-eabi binaries, but I could never make that work.
The kernel can be told do support OABI binaries, even when it's compiled
using the EABI.
If the userland (i.e. libc) is OABI, then most of the same old
applications should run no problem (still as slow as before).
If you compile a test application statically with the eabi, then it
should show a drastic speed improvement.
I would just suggest with using a fully eabi userland, with
-mfloat-abi=softfp
>
> Out of the box the cross compiler builds float to use the kernel floating
> point emulator which is supposed to be and at the last count was somewhat
> faster than the non-eabi emulator.
>
> If you want to use soft-float then you need a cross-compiler which has been
> built for soft-float support which is NOT the default
>
> AFAIK the current Marvell chip on the 7800 does not have native floating
> point
> support and as you point out Maverick Crunch is Cirrus specific.
>
> I was using arm-2005q3-2-arm-none-linux-gnueabi-i686-pc-linux-gnu.tar.bz2 to
> build my eabi binaries for hard float (kernel emulation).
>
> Hope this helps
>
> Bob
>
> On Wednesday 06 February 2008 02:54:18 delapluie wrote:
>> all right, I just discovered this thread-
>>
>> http://tech.groups.yahoo.com/group/ts-7000/message/9683
>>
>> which seems to indicate that the 7800 ships with an OABI kernel. But
>> when I try readelf -h /bin/ls on my two day old TS-7800, I get this-
>>
>> ELF Header:
>> ...
>> Flags: 0x2, has entry point, GNU EABI
>>
>> Which sure suggests it is EABI to me. Is it possible that I will get
>> better performance than the native compiler if I can get one of the
>> EABI cross compilers to work?
>>
>> -mike
>>
>> --- In "delapluie" <> wrote:
>>> I recently got a TS-7800, in the hope that because of the EABI support
>>> and the 500 MHz processor, it would perform much better at floating
>>> point math than the 7260 I was testing before.
>>>
>>> I was unsuccessful in getting the crosstools to work from Ubuntu.
>>> When I ftp'd my executable to the 7800, and did a "chmod +x test", I
>>> keep getting-"-bash: ./test: No such file or directory" If anyone has
>>> any thoughts on this problem, I'd appreciate it.
>>>
>>> But I'm really eager to determine if the 7800 can solve my performance
>>> problem, so I ftp'd my source code over, and compiled it natively on
>>> the 7800. No compiler options (do I need any?)
>>>
>>> Strangely, benchmarking 1000 iterations of my algorithm, I got almost
>>> exactly the same performance as with the 7260- about 54ms per
>>> iteration. I expected that even if EABI wasn't doing it's job, I
>>> would at least get a 2.5 times speed up from the clock rate. My test
>>> code reads a line (maybe 100 bytes) from a data file in the NAND flash
>>> once per iteration; otherwise, I can't think of anything outside the
>>> processor that takes any time.
>>>
>>> Has anyone else benchmarked floating point on the 7800, and gotten
>>> better results? Any idea what I might be doing wrong? Thanks,
>>>
>>> -Mike
>
>
>
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/ts-7000/
<*> Your email settings:
Individual Email | Traditional
<*> To change settings online go to:
http://groups.yahoo.com/group/ts-7000/join
(Yahoo! ID required)
<*> To change settings via email:
<*> To unsubscribe from this group, send an email to:
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|