| Another reason for this is that it requires assembler patch.

The GNU 'as' bug triggered by my code has been fixed in June 2004.
So, yes, you need a patch if you are using an older 'as'.

You'll be amazed for how long bugs can stick around:-) Work around them...

| One of unwritten OpenSSL design rule is to make things | work assuming the least about target environment.
| Can you make 64-bit lea operations optional to break dependency from
| assembler patch?


Hmm I am going to see what is possible.

Another thing to consider in the same line of minimal assumption. Explicitly "prototype" mnemonics, e.g. not xor %rax,%rax, but xorq %rax,%rax. Yes, GNU assembler works with "unprototyped" mnemonics, but Solaris assembler is known to be allergic to them. Well, we haven't seen Solaris x86_64 assembler yet, but it's plausible to assume that it won't be any different in this respect from x86.


Another unwritten rule is don't get fixated on particular u-architecture, Opteron in this case. Consider other implementations too, EM64T in this case. Favor *all-round* performance!

Ok. So, here is my question to the OpenSSL community: what algorithm
would you like to see optimized for AMD64 ? AES, SHA-1, Blowfish, RC5 ?

I might have an opportunity to play with AES some day this year... Blowfish is perfectly comfortable in tiny IA-32 register bank and compiler-generated code was observed to be as fast as assembler implementation on x86. Meaning that it will be even more comfortable in x86_64 register bank and compiler has all chances to do decent job.
Who uses RC5? A.
______________________________________________________________________
OpenSSL Project http://www.openssl.org
Development Mailing List [EMAIL PROTECTED]
Automated List Manager [EMAIL PROTECTED]

Reply via email to