On 20 May 2010 03:37, Jason Moxham <[email protected]> wrote:
> On Thursday 20 May 2010 03:11:24 Bill Hart wrote:
>> On 20 May 2010 02:58, Jason Moxham <[email protected]> wrote:
>> > On Thursday 20 May 2010 02:11:11 Bill Hart wrote:
>> >> There's a new Opteron on the block. family 16 model 8. It's an Opteron
>> >>  8439 SE. We don't have it in cpuid.c yet.
>> >
>> > Looks like another K102  , doesn't look like any differences for us ,
>> > although we dont know until we test it.
>>
>> Passes make check once the cpuid is put in. See below for bench. This
>> particular machine is 2.8GHz(owner is in cc). I didn't make tune, but
>> that shouldn't be needed.
>>
>> Index: cpuid.c
>> ===================================================================
>> --- cpuid.c     (revision 2928)
>> +++ cpuid.c     (working copy)
>> @@ -120,7 +120,8 @@
>>         case 16:
>>           if (model == 2) { CPUIS(k10);break;}
>>           if (model == 4) { CPUIS(k102);break;}
>> -         break;
>> +      if (model == 8) { CPUIS(k102);break;}
>> +      break;
>>          }
>>      }
>>    else if (strcmp (vendor_string, "CentaurHauls") == 0)
>>
>> > By the way the difference between what we call the K102 and the K10 , is
>> > the "same" as the difference between early K8 and late K8's , a slight
>> > difference in L1-data cache and store forwarding , which does effect some
>> > functions eg copy,addadd,lshiftcom , some more than others
>> >
>> >> Also we currently have a symbol in the library, mpn_sqr_n and a
>> >> #define in mpir.h to rename it mpn_sqr in line with gmp. But that
>> >> isn't sufficient. It breaks gcc on machines where gcc has been built
>> >> with dynamic libs. On the face of it this shouldn't affect anyone, but
>> >> in fact it will affect lots of people. The only solution I see is to
>> >> rename the library symbol.
>> >
>> > we can just export both names , so we are compatible with the new gmp AND
>> > the old
>>
>> Yeah, I noticed that.
>>
>> Anyhow, good idea, but not in mpir.h. It has to be in the library itself.
>>
> I ment in the library , trying to think how you do it :) , I know how to do it
> in asm , but in C ?

I don't know either, I mean without duplicating the code or incurring
extra overheads. The compiler is going to inline anything efficient
anyway!

>
>> >> Also, GMP defines mpn_copyi and mpn_copyd. On generic x86_64 machines,
>> >> we don't seem to have any assembly for this. There's no fallback
>> >> generic C symbol in the lib.
>> >
>> > we can the fallback C code that a fat build uses , with the new K102
>> > cpuid above there is no current generic x86_64 cpu :)
>>
>> True.
>>
>> >> Bill.
>>
>> AuthenticAMD Family 16 Model 8 Stepping 0
>> Six-Core AMD Opteron(tm) Processor 8439 SE
>> Speed: 0.80 GHz (reported)
>
> powersave on ? , below useless

Probably affects only the first timing. I think it is probably on demand.

>
>>  Category base
>>   Program multiply (weight 1.00)
>>          128       128 => 62415242
>>          512       512 => 14910433
>>         8192      8192 =>   174206
>>       131072    131072 =>     3012
>>      2097152   2097152 =>      108
>>          128       128 => 57065363
>>          512       512 => 13512577
>>         8192      8192 =>   120162
>>       131072    131072 =>     2251
>>      2097152   2097152 =>     74.3
>>        15000     10000 =>    59494
>>        20000     10000 =>    46812
>>        30000     10000 =>    27429
>>     16777216       512 =>      458
>>     16777216    262144 =>     13.7 => 37401,46751
>>   Program divide (weight 1.00)
>>         8192        32 =>  1654949
>>         8192        64 =>  1526756
>>         8192       128 =>   926962
>>         8192      4096 =>   297892
>>       131072     65536 =>     3232
>>      8388608   4194304 =>     11.5
>>         8192      8064 =>  8473345
>>     16777216    262144 =>     4.76 => 31816,39770
>>   Program divexact (weight 1.00)
>>         8192        32 =>  1872460
>>         8192        64 =>  1853921
>>         8192       128 =>   786511
>>         8192      4096 =>   360487
>>       131072     65536 =>     3813
>>      8388608   4194304 =>     11.8
>>         8192      8064 => 14242122
>>     16777216    262144 =>     7.84 => 38603,48254
>>   Program gcd (weight 0.50)
>>          128       128 =>  2284479
>>          512       512 =>   227241
>>         8192      8192 =>     5964
>>       131072    131072 =>      111
>>      1048576   1048576 =>     6.36 =>  4656, 5819
>>   Program gcdext (weight 0.50)
>>          128       128 =>   988477
>>          512       512 =>   175371
>>         8192      8192 =>     4609
>>       131072    131072 =>     68.3
>>      1048576   1048576 =>     4.31 =>  2981, 3726
>>   Program root (weight 0.30)
>>          128         5 =>  1142240
>>          512         3 =>   514009
>>         8192        11 =>   158261
>>       131072         3 =>     4138
>>      1048576         3 =>      190 => 37396,46745
>>   Program fac_ui (weight 0.20)
>>          128       128 =>  1588752
>>         1512      1512 =>    53386
>>        15000     15000 =>     1249
>>      1000010   1000010 =>     4.39
>>      2123456   2123456 =>     1.63 =>   945, 1182 => 18481,23101
>>  Category app
>>   Program rsa (weight 1.00)
>>                    512 =>    24338
>>                   1024 =>     5200
>>                   2048 =>      883 =>  4817, 6021
>>   Program pi (weight 1.00)
>>                  10000 =>      458
>>                 100000 =>     27.0
>>                1000000 =>     1.61 =>  27.1, 33.9
>>   Program bpsw (weight 1.00)
>>                   1024 =>     1138
>>                   4096 =>     32.4
>>                  16384 =>    0.840 =>  31.4, 39.2
>>   Program wagstaff (weight 1.00)
>>                   1024 =>     2626
>>                   4096 =>      105
>>                  16384 =>     3.36 =>  97.5,  122 =>   141,  177 =>  1617,
>> 2021
>
> --
> You received this message because you are subscribed to the Google Groups 
> "mpir-devel" group.
> To post to this group, send email to [email protected].
> To unsubscribe from this group, send email to 
> [email protected].
> For more options, visit this group at 
> http://groups.google.com/group/mpir-devel?hl=en.
>
>

-- 
You received this message because you are subscribed to the Google Groups 
"mpir-devel" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/mpir-devel?hl=en.

Reply via email to