RFC: g++ aliasing (was: Re: g++ and aliasing bools)

[Paolo Carlini]

Dan Nicolaescu wrote:

> If you can run some C++ benchmarks please do tell if this patch makes
> a difference.

Hi Dan, my compliments!

For sure it makes a /v/e/r/y /b/i/g difference for "Haney Speed"!

These are the numbers (-O2 -funroll-loops -fomit-frame-pointer, on aPII-400):

Timing Complex matrix operations for n = 200
1...2...
Language     Answer Check            Time     Iter Time
O-O C++      (-1.06982e+07,-2.82516e+08): 13.68    0.5472         <=======
Hand-coded C (-1.06982e+07,-2.82516e+08): 11.08    0.4432
Fortran: Not available

Timing Real matrix operations for n = 200
1...2...
Language     Answer Check   Time    Iter Time
O-O C++        (2.41602e+06):   14.02    0.1402           <=======
Hand-coded C   (2.41602e+06):   15.36    0.1536
Fortran: Not available

Timing Vector operations for n = 200
1...2...
Language     Answer Check   Time     Iter Time
O-O C++      (0.000150793): 4.66    0.0001864
Hand-coded C (0.000150793): 3.6    0.000144
Fortran: Not available


VS:
---

Timing Complex matrix operations for n = 200
1...2...
Language     Answer Check            Time     Iter Time
O-O C++      (-1.06982e+07,-2.82516e+08): 20.37    0.8148        <=======
Hand-coded C (-1.06982e+07,-2.82516e+08): 11.54    0.4616
Fortran: Not available

Timing Real matrix operations for n = 200
1...2...
Language     Answer Check   Time    Iter Time
O-O C++        (2.41602e+06):   29.07    0.2907         <=======
Hand-coded C   (2.41602e+06):   15.33    0.1533
Fortran: Not available

Timing Vector operations for n = 200
1...2...
Language     Answer Check   Time     Iter Time
O-O C++      (0.000150793): 4.58    0.0001832
Hand-coded C (0.000150793): 3.63    0.0001452
Fortran: Not available


Ciao,
Paolo.