Re: Inline assembler reference

Concerning the DOS club...... well, I would rather join Linux one if MSFT
takes away the possibility of writing platform-specific code (according to
Intel manuals, Pentium D seems to be quite interesting platform for
system-level developers). In fact, I am not 100% sure MSFT will stick to
its
current position on this issue - there is a good chance that they will be
pressed hard to remove the restrictions
thay want to introduce.

I really don't want to get too involved in this. I, too, am greatly
dissappointed that there is no more inlining of 64-bit assembly instructions
in the 64-bit c/c++ compilers. I had a project that was highly optimized
that way and I would have to rewrite from scratch for the 64-bit compile and
just isn't that important to me. The compiler intrinsics that we are
supposed to use in its stead don't produce the same code I would by hand and
yes, I do know how to beat the compiler with float/SSE* instructions. Using
MASM64 and C++ to link the ml64 output into my build only cases an extra
CALL instruction most of the time when inlining it could avoid those
precious clock cycles in the crtitical loops.

Some would say that I'm clock cycle obsessed. I'm actually not. But I
don't come of the opinion that just because all new machines have a Gig of
ram I need to write an application that consumes half of it just to perform
some menial action. I write my own code the way I want and for my own
reasons, and that includes using assembly language here and there.
Sometimes I'll write the entire Windows application in assembly, just
because I can. That's my choice. However, MS has limited my choice by
making it difficult to do what I want to do. Compiler intrinsics they
provide really are also hard to type in and hurt on my hands.

Anyway, get used to it. Its not just Microsoft. The Intel and GCC
64-compilers don't support inlining of 64-bit code, either. In fact, no
Intel compatible 64-compiler at the moment supports inlining of assembly
instructions. The vendors can brainwash us to thinks its because there's
not enough demand all they want, and the people who feel its an unimportant
feature will agree and cheer. The fact is, people who use inlining of
assembly langauge *usually* know what they're doing and have a good reason
for it and are annoyed that the feature has been removed and likely will
never be added again. It is the compiler vendors that are just too lazy to
add the feature, really. Fortunately, with GCC we can add the feature
ourselves (and since the Intel compiler boasts compatibility with GCC pehaps
they'd follow suit) but I have never been able to make sense of the GCC code
and so it won't be me that makes the change. I don't use GCC anyway.

They days of being in control of your own code are over, we must now welcome
the days of the compiler limiting us and doing our jobs for us. Not saying
thats bad, really, it isn't. But to completely undermine my own abilities
and make it as difficult as possible to perform optimizations that
outperform the compiler's own output as a bit too extreme, IMO. The
workarounds add overhead and the point is to remove overhead, and increase
performance. I shouldn't have to sacrifice. But, using any Microsoft tool,
I always have to sacrifice. I'm a fan and arduant supporter of MS, but lets
face it, more and more since the last 7 years or so, increasingly we have to
conform to MS rather then them conforming to the customer.


Thanks,
Shawn


.

Relevant Pages

  • Re: Richard Stallman is responsible for the shrinking economy
    ... than fixing problems if and when the spirit moves them as per Gcc. ... Intel wants to co-develop the GCC ... Three Intel employees are to contribute to the development of the GNU ... the development of its Intel Compiler Collection." ...
    (comp.arch.embedded)
  • Re: __cxa_throw () from libcxa.so.5 intel lib on LINUX
    ... Everything works absolutely fine with gcc. ... I have successfully managed to build/link the producton intel ... server aborts with below back trace. ... so a forum that deals with Intel's compiler would be even ...
    (comp.lang.c)
  • Re: Richard Stallman is responsible for the shrinking economy
    ... Intel wants to co-develop the GCC ... Three Intel employees are to contribute to the development of the GNU ... the development of its Intel Compiler Collection." ...
    (comp.arch.embedded)
  • Re: Inlining can be _very_bad...
    ... (as I recall many discussions about inlining yes or no...). ... The only fault is to post this issue here instead of the gcc Bugzilla. ... funtions in tcp or raid that are inlined expecting to be faster ... But most times that's already only up to the compiler - e.g. current gcc ...
    (Linux-Kernel)
  • Re: code portability
    ... I see massive gcc and imake requirements. ... GCC is hardly your average OSS project (and Intel C is hardly standard ... the intel compiler complies with C99. ...
    (comp.lang.c)