Re: AMD64 and EM64T instruction sets.

From: Dennis (marianndkc_at_home3.gvdnet.dk)
Date: 02/25/05 

Date: Fri, 25 Feb 2005 11:14:20 +0100

Hi Per

> Hmm - according to the second one there, old-style x87 instructions are
not
> supported in 64-bit mode. That's news to me. I guess it's all SSE now,
then.

I cannot belive that. SSE has no support for extended precision or Sin, Cos,
Log etc. Then it is impossible to develop scientific applications.

> That means more work for the compiler porters than would otherwise have
been
> the case, but I guess the good news is that then they'll have an
opportunity
> to improve the quality of the output code.

Quality of calculations will decrease, because of less precision. Speed can
increase, because of less precision and possibility of running two single
precision SSE calculations in parallel in the extended precision pipeline.
Code will be of same quality.

> - Per

Regards
Dennis

Relevant Pages

  • Re: AMD64 and EM64T instruction sets.
    ... SSE has no support for extended precision or Sin, ... recent SIMD instructions. ...
    (borland.public.delphi.language.basm)
  • Re: double to int conversion yields strange results
    ... giving me two different values for the int. ... the question raised is not about the precision of the ... is done in extended precision. ... It is the conversion to 'double' before the conversion to 'int' that ...
    (comp.lang.c)
  • Re: surprising behaviour of .EQ.
    ... The only problem that those "4 other compilers" use x87 instead of SSE ... for floating point calculations. ... extra precision that, for many numerical methods, would give the users the ...
    (comp.lang.fortran)
  • Re: SP on a 64-bit CPU and DP on a 32-bit CPU
    ... The SSE floating point unit, ... >> you're using double precision and 40 less for single precision reals. ... > which initialized x87 precision mode to 53 bits. ... > the range of standard double. ...
    (comp.lang.fortran)
  • Re: Who uses clapack?
    ... Use of traditional BLAS interposes a mandatory memory save ... > the sqrtin extended precision, but it is not forced to truncate ... > do a simple dot product to compute the vector norm, ...
    (sci.math.num-analysis)