Port-sparc archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: OpenSSL build error on 8.2



On 18 May 2020 11:39:24 BST, Erik Fair <fair%netbsd.org@localhost> wrote:
It would be interesting to performance test the gcc -mcpu and -mtune sparc specific options on ssh and see if it’s better - IIRC, there are some instruction additions to supersparc and hypersparc (both of which are documented options in gcc(1)) which might help.

Of course, you give up being able to use those binaries on earlier sparc processors, e.g. sun4 and sun4c systems, but if you’re not sharing /usr across NFS (or something) with such systems, or don’t have any, that may not be a problem for you.

The standard NetBSD distribution for a given architecture is usually least common denominator so that it boots & runs on everything, but that won’t take advantage of more powerful features of a given architecture’s “later” revision processors. The older a system, the more important to optimize for speed is likely to be.

Erik


Apart from losing 80-bit Extended F.P., the main difference for SPARCv8 over the original SPARCv7 is the single-instruction 32-bit integer multiply, over the prior multiple-instruction method using the special-purpose Y register. HyperSPARC also had (limited) compare-and-branch fusion.

The former (still multi-cycle, but freeing up some decoding bandwidth) might help a little bit, but don't get over excited.

-- Mike

Home | Main Index | Thread Index | Old Index