Re: Tuning John
Solar Designer <solar@...
2006-05-02 12:09:31 GMT
On Tue, May 02, 2006 at 11:04:25AM +0200, Ami Schwartzman wrote:
> 1) What is the performance gain (if any) by moving john on linux-x86-mmx
> to linux-64-x86-mmx for Intel processors?
There should be no performance difference between the linux-x86-mmx and
linux-x86-64-mmx targets. The only difference is that the former is
for 32-bit Linux/x86 distributions, whereas the latter is for 64-bit
Linux/x86-64 ones that have 32-bit compatibility libraries installed.
Both will use MMX for effective 64-bitness in exactly the same way.
The target which actually produces different code is linux-x86-64. It
uses native x86-64 instructions instead of MMX. In practice, the
performance is similar to that of MMX - although it may be 20% worse or
10% better for different hash types, CPUs, and gcc versions.
> 2) Why does the CYGWIN and DJGPP versions do not calculate virtual c/s
> in the benchmark,
For Cygwin builds, that's because virtual (processor) times are not
available on Windows 9x and I did not bother implementing a (trivial)
check for 9x vs. NT.
For DJGPP builds, that's because processor times are supposed to match
real times (DOS is not a multi-tasking OS), and when they don't it is
not certain what host OS is used to emulate DOS and how it might report
processor times to DOS applications (if at all). Not that I would
bother implementing this even if it were supported. The DOS build is
for plain DOS.
> and in general how does John calculate virtual c/s?