Back to home page

DOS ain't dead

Forum index page

Log in | Register

Back to the forum
Board view  Mix view

Compatibility woes / deprecation (Miscellaneous)

posted by marcov(R), 24.02.2009, 11:32

> The Pentium had been advertised as having a 64-bit bus (or whatever), not
> to mention MMX on later models technically being 64-bit. I know that's not
> the same thing, but that's how it was spun in some cases.

Yes, but they fooled nobody serious with that :-)

> > And in general the P4 depended more on "P4 optimized code" than P-III.
> > Athlon was even less picky about how the code was optimized.
>
> Well, they broke some common optimizations used in the past due to
> architectural differences. Of course I think all cpus are like that, even
> AMDs (from my limited reading of the optimization hints in tech docs).

Yes, but the problem was that the penalties were so steep because the P4 with its deep pipelines was so different.

IOW the difference is that you noticed such things inabout general code by normal compilers. Not last cycle work with the intel compiler.

> That's what I meant, home use. Of course, even the first 64-bit Opterons
> were server only, right?

And workstation. But the workstation pendant Athlon64 predates Opteron, so it is not the same.

(SSE/MMX)
> > It isn't, but it shouldn't be overrated. Rule of thumb, it is useless
> > unless you specifically code for it.
>
> I mean, it just feels somewhat useless since it's weird and involves
> rewriting code manually (although this is a perfect example that "compiler
> outperforms humans" isn't always true since most compilers can't vectorize
> worth a damn).

Note that we currently have exactly this same discussions again about cores. The same advocates that said that "better compilers will solve this" are at it again.

Of course it is not entirely the same, but the pure compiler part is IMHO. But the line is blurred that frameworks often can do something (hiding parallelism for the user)

> GCC has "-mfpmath=", which allows 387, sse, or both. But I'm not sure it
> helps (yet?), highly experimental I think.

FPC has {$FPUTYPE SSE2}

> > Where do you get that? One still installs DDR (1,2,3) in pairs for
> optimal
> > performance, and each is 64-bit. So afaik mem bandwidth is still
> 128bit.
> > (unless you use several QPI/HT)
>
> I mean SSE bandwidth (or so I heard).

Ah ok.

> Anyways, I also read somewhere that Core 2 can (optimally) do four
> instructions per clock unlike AMD at max. three. That alone is pretty
> good, so raw clock speed doesn't matter as much as previously (e.g. 486 or
> 586).

Yes, but only if they are parallizable, and 3-4 loads/stores per clock aren't.
Note that both Phenom II and i7 have now 1 cycle L1 caches, which might improve this a bit. (when not read lineairly)

> Maybe not on Windows for make, bash, gcc, etc, but DOS has no issues.

It's more important on Windows true. And Dos has maximal partition limits? How much storage can you use on a single HD on pure dos? 24 partitions of 2GB or so?

> The only reason to not use something old is if the new improves in
> every way, which is hardly typical.

That is not correct. If it works better/easier over the whole line is enough. Otherwise I would always remain stuck with something old because it is "better" on one not terribly important point.

> Pros:
> + small

Not an advantage per se.

> + fast (even can use EMS or XMS for even faster speeds)

For me 32-bit binaires were always faster.

> + 16-bit code (which GCC still lacks)
> + runs on 16-bit cpus
> + all models: tiny through huge
> - 186/286 optimizations at most (useless for 99% of the world)

Not a requirement. Don't have anything in use below XP2000+

> + supports ANSI C and C++ AT&T 2.0

I'm not a C programmer.

> + nice IDE
> + nice help / function reference

I use FPC's IDE. Works fine. Am improving the help, but it is huge.

> Cons:
> - no sources

Less important if it works right. But for runtime parts sources are a non negotiable requirement.

> - DOS only (no cross compiling supported)

Useless :-)

> - no newer C++ features (generics, templates, etc.)

Those would be the only reason to use C++ in the first place.

> - OpenWatcom is better in most ways (but needs 386+ to host)
>
> Besides, there even a DOS extender that works with it
> (Swallow

Pmode works with FPC too. Likewise untested in recent years.

 

Complete thread:

Back to the forum
Board view  Mix view
15192 Postings in 1365 Threads, 250 registered users, 18 users online (0 registered, 18 guests)
DOS ain't dead | Admin contact
RSS Feed
powered by my little forum