🚀 go-pugleaf

Has anyone managed to get tcsh running under MINIX/532?  Before I waste
a lot of time on this, I thought I'd check.

Also, what is the latest version of gcc on MINIX/532?  Is G++ or Objective-C
running yet?  How about the libraries?

Finally, while perusing the gcc sources I noticed a couple of interesting
things that I thought I'd ask some questions about-

1) To pop 4 or 8 bytes of the stack, gcc emits "cmpqd 0, tos" and "cmpd tos,tos"
   instructions.  I wrote a rudimentary benchmark in assembly to compare the
   performance against "adjspd -4" and "adjspd -8".  The "ADJSPD" instructions
   were faster (though longer).  Is this true in general for the 32532?  Did
   they pick these sequences to save space, or do they actually run faster on
   pre-532 chips, or were they concerned about cache space?  

2) For values larger than 4 or 8 bytes, they use ADJSPB (<64 bytes), ADJSPW
    (<8192 bytes), or ADJSPD (all others).  I seem to recall reading that there
   is a bug in the 32000 family such that adjspb and adjspw only affect the 
   L.O. byte or word of the SP register.  Does this problem still exist on the
   32532?  In any case, the ADJSPD instruction is faster than ADJSPB or
   ADJSPW, even if it is longer.  Shouldn't these instructions get switched
   to ADJSPD?

   BTW, the 32532 timing chart claims CMP tos,tos requires t(eu)+t(au) =
   2+(2+2) = 6 cycles to execute while ADJSPD requires only 4.  This is
   consistent with my measured timings.  CMPQD 0,tos, OTOH, should run at the
   same speed (approximately) which I did not find to be the case with my
   simple benchmark.

3) I presume (by looking at object code) that gcc trashes R0-R2 and preserves
   the other registers?

*** Randy Hyde O-)