NetBSD-1.4.2 kernel is slower than 1.4 and 1.4.1

[Inaki Saez <jisaez%sfe.indra.es@localhost>]

I have tested netbsd 1.4.2 generic kernels for hp300 and amiga and
compared the performance. I have found that 1.4.2 is slower than
the previous 1.4.1 and 1.4 kernels.

- First  reconfiguring zsh-3.1.6 (./configure) I found that on hp300
  architecture the total time increments from 2:36 minutes to 3:15.
  Which is 25% slower. The same test on amiga gives a result of the
  same proportion, 25% or 35% more time.

- Then I have run as a speed test program "bytebench-3.6" package
  and found that the performance loss is specially significant in
  Execl Throughput Test, Shell Script Test and Dc: sqrt(2) to 99 decimal places
test

  In hp300 Execl drops 70% :
  NetBSD-1.4-generic:   Execl Throughput Test                         6.4 lps  
(10 secs, 6 samples)
  NetBSD-1.4.1:         Execl Throughput Test                         6.5 lps  
(10 secs, 6 samples)
  NetBSD-1.4.2-generic: Execl Throughput Test                         3.8 lps  
(10 secs, 6 samples)

  In amiga Execl drops 170%:
  NetBSD-1.4.1:         Execl Throughput Test                         5.1 lps  
(9 secs, 6 samples)
  NetBSD-1.4.2:         Execl Throughput Test                         2.0 lps  
(10 secs, 6 samples)
  NetBSD-1.4.2-generic: Execl Throughput Test                         1.9 lps  
(10 secs, 6 samples)

  In hp300 Dc: sqrt(2) to 99 decimal places test drops 58%:
  NetBSD-1.4-generic:   Dc: sqrt(2) to 99 decimal places            310.0 lpm  
(60 secs, 6 samples)
  NetBSD-1.4.1:         Dc: sqrt(2) to 99 decimal places            312.5 lpm  
(60 secs, 6 samples)
  NetBSD-1.4.1:         Dc: sqrt(2) to 99 decimal places            197.9 lpm  
(60 secs, 6 samples)

  In amiga Dc: sqrt(2) to 99 decimal places test drops 115%:
  NetBSD-1.4.1:         Dc: sqrt(2) to 99 decimal places            224.3 lpm  
(60 secs, 6 samples)
  NetBSD-1.4.2:         Dc: sqrt(2) to 99 decimal places            104.2 lpm  
(60 secs, 6 samples)
  NetBSD-1.4.2-generic: Dc: sqrt(2) to 99 decimal places             99.6 lpm  
(60 secs, 6 samples)



Something must be wrong with this version. The problem may be in other
architectures
but I cannot check it.


Note: The only thing that I have changed is the kernel I boot with, the rest is
the
      same system. The version not marked as -generic are compiled by me.