Re: gf_unit 16 A -1 -m SPLIT 16 4 -r ALTMAP - failure

[Loic Dachary <loic@dachary.org>]

[-- Attachment #1: Type: text/plain, Size: 2429 bytes --]



On 02/02/2015 22:14, Janne Grunau wrote:
> On 2015-01-10 18:54:43 +0100, Loic Dachary wrote:
>> Hi Kevin & Janne,
>>
>> The test gf_unit 16 A -1 -m SPLIT 16 4 -r ALTMAP - fails on the current gf-complete master. The first commit where it fails is
>>
>> commit 474010a91d35fef5ca7dea77205b6a5c7e68c3e9
>> Author: Janne Grunau <j@jannau.net>
>> Date:   Wed Sep 17 16:10:25 2014 +0200
>>
>>     arm: NEON optimisations for gf_w16
>>
>>     Optimisations for the 4,16 split table region multiplications.
>>
>>     Selected time_tool.sh 16 -A -B results for a 1.7 GHz cortex-a9:
>>     Region Best (MB/s):   532.14   W-Method: 16 -m SPLIT 16 4 -r SIMD -
>>     Region Best (MB/s):   212.34   W-Method: 16 -m SPLIT 16 4 -r NOSIMD -
>>     Region Best (MB/s):   801.36   W-Method: 16 -m SPLIT 16 4 -r SIMD -r ALTMAP -
>>     Region Best (MB/s):    93.20   W-Method: 16 -m SPLIT 16 4 -r NOSIMD -r ALTMAP -
>>     Region Best (MB/s):   273.99   W-Method: 16 -m SPLIT 16 8 -
>>     Region Best (MB/s):   270.81   W-Method: 16 -m SPLIT 8 8 -
>>     Region Best (MB/s):    70.42   W-Method: 16 -m COMPOSITE 2 - -
>>     Region Best (MB/s):   393.54   W-Method: 16 -m COMPOSITE 2 - -r ALTMAP -
>>
>> but the test did exit(0) on error instead of exit(1) and we failed to notice.
>>
>> gf_unit 16 A -1 -m SPLIT 16 4 -r ALTMAP -
>> Args: 16 A -1 -m SPLIT 16 4 -r ALTMAP - / size (bytes): 524428
>> Problem with region multiply (all values in hex):
>>    Target address base: 0x8fd08e.  Word 0x1 of 0x1fee.  Xor: 0
>>    Value: 2
>>    Original source word: d00a
>>    Product word: a000
>>    It should be: b01f
>>
>> Do you have an idea why this happens ? For the record here is the bisect command I used:
> 
> yes, it happens because the arm code is incomplete and bogus. I did the 
> arm64 code first and then ported it to arm. This function uses arm64 
> specific instructions and thus needed a port to arm. The arm version 
> misses several computation steps. I guess I trusted the results of the
> unit tests and didn't looked at gf_unit output on arm.
> 
> I'll send a fix tomorrow.

That's a relief :-)

Thanks !

> 
> Janne
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

-- 
Loïc Dachary, Artisan Logiciel Libre


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]