-freduce-all-givs differs on P3 and P4

[Scott Robert Ladd]

I made an odd discovery (at least in my mind) whilst exploring evolution 
in different environments... consider these two command lines:

     A) gcc -lm -lrt -march=pentium3 \
         -O3 \
	-o lpbenchA lpbench.c

     B) gcc -lm -lrt -march=pentium3 \
	-O3 -freduce-all-givs \
	-o lpbenchB lpbench.c

"B" runs 23.5% faster than "A" on a Pentium 3, due to the addition of 
-freduce-all-givs. A very nice improvement.

Consider my surprise, then, when I used the following two commands on a 
Pentium 4 system:

     C) gcc -lm -lrt -march=pentium4 \
         -O3 \
	-o lpbenchC lpbench.c

     D) gcc -lm -lrt -march=pentium4 \
	-O3 -freduce-all-givs \
	-o lpbenchD lpbench.c

Both "C" and "D" run in exactly the same time. Exchanging "pentium3" for 
"pentium4" in the above command lines (i.e., running "A" and "B" on the 
P4) does not change the result: for this program, -freduce-all-givs is 
effective on the Pentium 3, but ineffective on the Pentium 4.

I'm not certain if this is a bug or not. If people consider it a bug, 
I'll be more than happy to track down what the compiler is doing -- or I 
can e-mail the lpbench.c source code to anyone interested. On the other 
hand, if the observed behavior is to be expected, I'd love to know why.

I'll also note that on both test systems, -O3 is not necessary; the same 
phenomena can be seen when substituting "-O1 -fstrength-reduce" for 
"-O3", as in:

     E) gcc -lm -lrt -march=pentium4 \
	-O1 -fstrength-reduce -freduce-all-givs \
	-o lpbenchD lpbench.c

"-O1 -fstrength-reduce" generates code that is just as fast as "-O3" on 
this benchmark.

Command "E" may simplify understanding of the curious behavior by 
reducing the number of optimizations (and thus the complexity) involved 
in code generation.

Thanks.

-- 
Scott Robert Ladd
Coyote Gulch Productions (http://www.coyotegulch.com)
Software Invention for High-Performance Computing