|
|
Francois LE COAT wrote:
> The 64-bits addressing is not the main feature of GCC 4.0 that I'm
> looking forward ... I was speaking of refined optimizations such
> as parallelization and vectorization of the PPC code ... It seems
> that GCC 4.0 is a great step in the direction of more efficient
> code generation.
>
> Do you know more than I know myself ?
On the gcc mailing list there have been several notes that code actually is
slower when compiled with gcc 4.0 than with 3.4 due to some optimizations
they had to leave behind when getting their new optimizer framework in.
Supposedly only gcc 4.1 will bring most of those things back.
Besides, I am not exactly very confident with what Apple will ship with Mac
OS X 10.4 - while they claim it will be gcc 4.0, it cannot actually be
because that way only released *today* but Apple is already sending out CDs,
which must have been pressed at least a week ago. Besides, in the past
serious bugs in Apple's custom version of gcc have broken popular C++
libraries such as boost - the current Apple supplied gcc 3.3.x (not the real
gcc) has had an infinite loop allocating memory bug in the *compiler* (not
the generated code) for several months and Apple has not taken any action.
As it stands, I would be more interested to get the 'real' gcc generate AIX
for PowerPC code as that uses the sane and properly designed PEF/CFM
execuatable format* rather than the flawed Mach-O format (see for example
the summary on <http://www.unsanity.org/archives/000044.php> to learn why
Mach-O applications are so slow)...
Thorsten
* IBM developed it along with the first Power processor, the superset of the
instruction set and processor architecture PowerPC processors are based on.
When Apple moved to PowerPC, the first Mac OS was developed using IBM
compilers running on Power-based AIX workstations.
Post a reply to this message
|
|