Re: How does compare gcc to VS C++ ?

From: John-Paul Stewart (jpstewart_at_binaryfoundry.ca)
Date: 08/12/04 

Date: Thu, 12 Aug 2004 11:27:13 -0400

Jason Bowen wrote:
> John-Paul Stewart wrote:
>
>> Jason Bowen wrote:
>>
>>> John-Paul Stewart wrote:
>>>>
>>>> Then explain to me how you can be certain the observed performance
>>>> difference can be attributed solely to code generation.
>>>
>>> It's easy. A compiler can base it's code generation off of a target.
>>> The knowledge of the target of the compiler builder can vary. A
>>> compiler can be smart enough to know how to order instructions better
>>> to enhance cache access or minimize latency between instructions
>>> based on how long it takes a particular instruction to complete.
>>
>> So you're not talking about an *observed* performance difference based
>> on running the generated code. You're talking about a *calculated*
>> performance difference based on reading the generated code and
>> calculating latencies, etc. Correct?
>
> I was never comparing an executable running under Windows to one running
> under Linux. I was talking about the difference in code generated by a
> couple of compilers. I'm thinking in terms of flat x86 instructions.

That's all well and good if all you care about is theoretical
performance, but real-word performance is influenced by far more factors
than this one. And that's been my point all along.