|different results with different f77 optimizers firstname.lastname@example.org (Tom Keyes) (1995-09-24)|
|Re: different results with different f77 optimizers ah739@cleveland.Freenet.Edu (1995-09-29)|
|Re: different results with different f77 optimizers email@example.com (1995-09-30)|
|Re: different results with different f77 optimizers firstname.lastname@example.org (1995-10-05)|
|Re: different results with different f77 optimizers email@example.com (1995-10-06)|
|From:||firstname.lastname@example.org (Cliff Click)|
|Date:||Thu, 5 Oct 1995 21:57:32 GMT|
ah739@cleveland.Freenet.Edu (Leslie J. Somos) writes:
> See CMG Transactions (CMG=Computer Measurement Group)
> number 52, Spring 1986, article "Do Fortran Compilers Really Optimize?"
> by Dr. David S. Lindsay.
> He compares optimized and unoptimized versions of Fortran code
> specifically coded to be optimizable, on commercial compilers.
> Quote from his Conclusion:
> "1. Compiler optimization of generated code is haphazard at best.
> 2. Compiler optimization if _not_ a mature technology, at least
> not if IBM's VS FORTRAN and DEC's VAX/VMS FORTRAN are representative.
> 3. Trade, technical, and academic sources have no basis for their
> dogma about how all optimizing compilers do so-and-so.
> 4. Empirical tests are _badly_ needed to verify vendors' claims of
> their compilers' generation of optimized code.
> 5. Empirical tests are probably also the only way to improve
> optimization technology.
> 6. Academics should have their students run tests, not just
> learn techniques."
> (The article is from 1986, I have no idea if this address is stale.)
> [I also wonder if the optimization biz has advanced any in the past decade.
> I sure hope so. -John]
Well, I've certainly seen a lot of easy and old optimizations missed
or done poorly by a number of compilers. Alas, optimizing compilers
are big, and big software moves slowly. Nonetheless, the state of
optimizers _has_ improved in the last decade. I no longer use the
'register' keyword - compilers almost always get key values into
registers. Many compilers do some memory heirarchy transformations -
a must for matrix multiply and big SpecFP numbers.
I think the trend is that new optimizations show up, but improved
versions of old ones are very slow to appear. Most folks do some kind
of local value numbering; many fewer do global value numbering. Most
folks fold constants, very few do global constant propagation. Most
folks do induction variable elimination on addressing expressions in
loops; fewer use the more generic algorithm that handles constant
multiplies in loops (even multiply by 4! This crops up when you are
doing multi-dimensional arrays in C).
P.S. My "test case compilers" include gcc 2.6.x, an older IBM xlc, an
older Sun cc, HP's cc, and Motorola's cc - not an exhaustive list fer
sure, and a little out of date. Anyone want to post a summary of more
Cliff Click Compiler Researcher & Designer
RISC Software, Motorola PowerPC Compilers
email@example.com (512) 891-7240
Return to the
Search the comp.compilers archives again.