PowerPC fastpaths for mutex subsystem

[Andrew Morton]

Joel Schopp <jschopp at austin.ibm.com> wrote:
>
> This is the second pass at optimizing the fastpath for the new mutex subsystem 
>  on PowerPC.  I think it is ready to be included in the series with the other 
>  mutex patches now.  Tested on a 4 core (2 SMT threads/core) Power5 machine with 
>  gcc 3.3.2.
> 
>  Test results from synchro-test.ko:
> 
>  All tests run for default 5 seconds
>  Threads semaphores  mutexes     mutexes+attached
>  1       63,465,364  58,404,630  62,109,571
>  4       58,424,282  35,541,297  37,820,794
>  8       40,731,668  35,541,297  40,281,768
>  16      38,372,769  37,256,298  41,751,764
>  32      38,406,895  36,933,675  38,731,571
>  64      37,232,017  36,222,480  40,766,379
> 

Doens't this mean that the sped-up mutexes are still slower than semaphores?