? (0xe2.0x9a.0x9b@gmail.com) on 6/24/10 wrote:
---------------------------
>1. My impression is that the Linux programs typically found in a Linux distribution
>have approximately the *same* performance regardless of whether they are compiled
>with GCC 3.4, 4.0, 4.1, 4.2, 4.3, 4.4 or 4.5. As another example, when I switched
>my application from GCC 4.4 to GCC 4.5 I did *not* notice any significant performance
>boost. Is your experience different?

It depends on the application and on also on your target architecture.

>2. As far as my experience goes, LTO:
>
>- does not bring about the expected results (assembler code)

I've did only some testing, for performance I would have expected better results but for code size I've noticed some very nice improvements I expected to be there.

>- is buggy

True, like pretty much every other GCC with a version number ending with a 0. It's supposed to be that way because even though the test suite is huge and exhaustive they cannot test all possible scenarios on all combinations of target architectures, operating system and option flags.

>Bottom line: Not counting the support for new C/C++ features, how exactly is GCC 4.5 better than GCC 3.4 ?

Well, if you look at GCC development features and architecture support *are* the most important things that have been added to GCC, utmost performance is important for only a very limited amount of people. That said FP codes on PowerPC processors are going to be *much* faster on GCC 4.5 than 3.4 for example. If performance hasn't improved in your particular combination of code and target architecture that doesn't mean there wasn't any improvements overall.