This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Draft "Unsafe fp optimizations" project description.
- To: gcc at gcc dot gnu dot org, trt at cs dot duke dot edu
- Subject: Re: Draft "Unsafe fp optimizations" project description.
- From: dewar at gnat dot com
- Date: Sun, 5 Aug 2001 21:08:22 -0400 (EDT)
<<But it could benefit from a subsetted option "-fthumb-nose-at-ieee"
which doesn't worry about -0 or Inf or NaN or denorms.
>>
I actually think this can reasonably be the default, there is nothing in
C which requires -0 semantics??? Then do this stuff only in -mieee mode.
Your list of transformations is interesting
FREQ TYPE
2818 x/C => x*(1/C)
477 x relop x => const
349 x+0 => x
158 x*0 => 0
157 x-0 => x
126 -(x-y) => y-x
61 0-x => -x
35 -x + y -> y-x
19 x - -y => x+y
8 0/x => 0
3 x-x => 0
Some of these seem to be value preserving transformations. All of them
are value preserving except for minus zero. I wonder if it is really
appropriate to worry so much about minus zeroes in default mode, it
seems wrong to me.
For the x/C => x*(1/C), it would be interesting to know the distribution
of constants to know how many are idempotent transformations (or does
gcc always transform x/2.0 to x*0.5 [this latter transformation should
definition be done by default].