This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: pre egcs-1.1 testing and Linux 2.1.x
Jamie Lokier wrote:
>David S. Miller wrote:
>> This person gave up because Kenner was so hard to work with, hopefully
>> now he'll resubmit his changes to the egcs people and they will go in,
>> and everyone will get what they want and be happy.
>Hmm. I also have some patches (code generation and C++ bug fixes,
>better ix86 code for 64-bit arithmetic, __builtin_constant_p in inline
>functions). These patches are also about 2 years old. And I gave up
>trying to integrate them also due to arguments with Richard Kenner.
>Perhaps he was having a hard time back then?
Yep. About two and three years ago (the arguments extended over
a period longer than a year :-). Not to say that Kenners arguments
weren't unreasonable at times, but the cases where he seemed to be
nitpicking were too often.
>BTW, I have used -mregparm=1 throughout a 250,000 line C++ program and
>only encountered one code generation bug, with an indirect function
>call.
Yes. "Indirect function calls", back then the only known case which
causes enough register pressure on the x86 to panic the register allocator
of GCC.
>Having used -mregparm=1 extensively, I suggest it is actually very
>reliable. -mregparm=2, =3 etc. perhaps not.
The reliability should be the same, it's the same indirect function
calls that mess things up. Even regparm=5 works just fine.
>I used -mregparm=1 because it generally produced the smallest code. I
>would have used -mrtd too, if __attribute__((regparm(0),cdecl)) worked
>(it didn't).
I'm surprised. This has worked at one time, I'm almost certain.
--
Sincerely, srb@cuci.nl
Stephen R. van den Berg (AKA BuGless).
This signature third word omitted, yet is comprehensible.