[prev in list] [next in list] [prev in thread] [next in thread] 

List:       john-users
Subject:    Re: [john-users] Breaking UNIX crypt() on the PlayStation 3
From:       "Project: NANIYA" <geek4civic () gmail ! com>
Date:       2008-10-19 8:27:06
Message-ID: 36404eab0810190127y5bcc91ebmed6950a1b28f9959 () mail ! gmail ! com
[Download RAW message or body]

Alexander,

Konnichiwa. 1st post here for me.

2008/10/17 Solar Designer <solar@openwall.com>:

> Isn't the PPU capable of issuing more instructions per cycle (than the
> SPUs)?  If not more AltiVec instructions, then maybe more PowerPC
> instructions intermixed with AltiVec ones?  BTW, this additional
> optimization - using native 64-bit instructions intermixed with vector
> ones - might work on all of: SSE2, AltiVec, SPU.

Intemixing VMX and ALU has a problem.

VMX load and store always spend 1 or 2 GPRs, and they have
no "constant displacement" addressings.
eg) LVX V7,R0,R1 # load from (R0+R1), when R0 is always 0
VMX does not have "load from nnn(Rn)."

Thus, VMX sequences make pressure to GPR or ALU OP.
And now, we have achieved more efficient sequences on SPU (and PPU VMX :))
it is harder to interleave 64-bit GPR sequences into (shorter) VMX sequences.

I think it would be better for PPU to let run simulataneously with two threads,
VMX and ALU. But still I have interests to the former strategy.


Btw, do you need faster implementation of DES?
(on x86, x64 and other chips)


--
NAKAMURA, Takumi

-- 
To unsubscribe, e-mail john-users-unsubscribe@lists.openwall.com and reply
to the automated confirmation request that will be sent to you.

[prev in list] [next in list] [prev in thread] [next in thread]