[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAOLP8p6yCZ+nP=a3KQOFnaJXbuv+vPuamMOo3fxWWW+wHEwLvw@mail.gmail.com>
Date: Wed, 26 Feb 2014 12:22:09 -0500
From: Bill Cox <waywardgeek@...il.com>
To: discussions@...sword-hashing.net
Subject: Re: [PHC] multiply-hardening (Re: NoelKDF ready for submission)
I'm going to reintegrate multiplication back into the memory hashing
threads and eliminate the multiplication hardening thread. At least
for Haswell, a single scalar multiply and XOR seem to run nicely in
parallel with AVX2 memory hashing, even at L1 hashing speeds.
I'm going to make an option for between 0 and 8 multiplications per
256-bits of memory hashing. 0 would be useful for applications where
multipliers are very slow or not available and have to be emulated, or
if the CPU has no multiple instruction issue capability, meaning any
multiplies add directly to user's runtime. 1 seems like a good match
for AVX2 running in L1 cache, and I'm guessing 2 will be good for SSE.
For hashing into external memory, up to 8 seems reasonable.
Bill
Powered by blists - more mailing lists