Re: older gcc and pentium
Albert Cahalan (albert@ccs.neu.edu)
Tue, 5 Mar 1996 15:46:55 -0500 (EST)
> On Mon, 4 Mar 1996, Mike Castle wrote:
>> Amazingly enough Kai Schulte said:
>>
>>> Really 486? Don't the -m486 alignments bloat the code unnecessarily?
>>
>> Well, if using more memory results in faster memory accesses, is
>> it unnecessary? The alignments result in faster code accesses.
>> But if you have limited core, perhaps the compact code is more
>> appropriate.
>
> It results in more speed for 486-specific accesses, but not for a 386,
> and not necessarily for a 586, either. Can somebody with "real" hardware
> knowledge help out on this? I really don't know enough about the
> architectures, but i'll take a closer look at the new gcc as soon as I
> have the time, promised ;)
Actual test results for non-kernel code indicate that 386 optimized
code runs about 5% faster than 486 optimized code. This may vary
depending on your cache performance.
Recent kernels can use 486+ opcodes, so there is a tradeoff
here that has not been tested. For the very fastest code
(assuming you do not upgrade your compiler) I'd guess that
you should _compile_ for a 386 but set the #define for 486.