🚀 go-pugleaf

In one of the gnu.gcc groups there was mention that -fstrength-reduce (which
used to speed programs up quite a bit, although results varied by program)
did not do very good optimization.  I checked this out, and in fact
-fstrength-reduce (using gcc 2.1) really sucks.

I used dhrystone 1.1 to see what kind of difference it might make.  First,
I tried the old gcc (v 1.36, the one that comes with the system).
$OPT is a variable that turns on ALL optimization for cc.  $OPT2 turns
on ALL optimization (except loop unrolling) for gcc (v 2.1).  $LOOP2 turns
on loop unrolling.

gorgatron% cc $OPT dry.c
a.out
gorgatron% a.out
Dhrystone time for 500000 passes = 16
This machine benchmarks at 30581 dhrystones/second
gorgatron% gcc $OPT2 dry.c
gorgatron% a.out
Dhrystone time for 500000 passes = 13
This machine benchmarks at 36945 dhrystones/second

So, clearly gcc 2.1 generates better code.  Now let's try some stuff with
-fno-strength-reduce.

gorgatron% gcc -O -O2 -fno-strength-reduce dry.c
agorgatron% a.out
Dhrystone time for 500000 passes = 13
This machine benchmarks at 37831 dhrystones/second
gorgatron% gcc $OPT2 -fno-strength-reduce dry.c  
gorgatron% a.out
Dhrystone time for 500000 passes = 13
This machine benchmarks at 38071 dhrystones/second
gorgatron% gcc $LOOP2 $OPT2 -fno-strength-reduce dry.c
gorgatron% a.out
Dhrystone time for 500000 passes = 11
This machine benchmarks at 43041 dhrystones/second

43041! Not bad.  So turning strength reduction off seems to be a performance
win (at least for this program).

David Koski