Thank you for all the answers.. now the subject is clearer. And I thought that I knew something about comp. architecture... Thanks!<br><br>Ivan<br><br><div><span class="gmail_quote">2006/11/22, Richard Walsh <<a href="mailto:rbw@ahpcrc.org">
rbw@ahpcrc.org</a>>:</span><blockquote class="gmail_quote" style="margin-top: 0; margin-right: 0; margin-bottom: 0; margin-left: 0; margin-left: 0.80ex; border-left-color: #cccccc; border-left-width: 1px; border-left-style: solid; padding-left: 1ex">
Ivan Paganini wrote:<br>> Hello everybody at beowulf. Sorry about the _really_ newbie question,<br>> but after doing some tests and researching a little, a question arose<br>> when fooling around with amd64 (more precisely, an amd64 Athlon 4200
<br>> X2) and gcc and sun studio 11. The architecture has 64 bits integer<br>> registers and 128 bits floating point registers, but my test programs<br>> in C just gave me the same precision that I got with an old athlon
<br>> 2400 xp (32bits), that is, long double go only to 1x10^ 4961, even<br>> with the -m64 flag. I always imagined that I would get the double<br>> precision without the long double declaration (or, maybe, 40bits
<br>> precision). What am I missing here? Is the compiler (gcc 4.1, sun<br>> studio express 11), the operating system (ubuntu 64bits edgy), or just<br>> an error in my logic?<br>><br>Ivan,<br><br>A common confusion ... x86_64 changes nothing about the precision of
<br>floats or doubles in<br>C or Fortran. They remain the same format as does the basic inner<br>workings of the functional<br>units. The 64-ness refers to the added memory addressibility provided<br>by the larger address
<br>registers. This potential itself is not fully utilized as I believe<br>only 40-bits are used (the socket<br>F series may have bumped this up to 48-bits).<br><br>So computed epsilons on on either platform should be the same. Long
<br>doubles, which would<br>need to be supported in software, would be need to get more.<br><br>rbw<br><br>PS In his note, Mark Hahn makes reference to the different extended<br>precisions inside<br> the processor of the SIMD floating-point units versus the
<br>old-style floating point.<br> Results are truncated to 64-bits when stored to memory, but a<br>path through the SIMD<br> units could produce slightly different results in "vectorized"<br>code from the non-vector
<br> old-style functional unit.<br><br>><br>> --<br>> -----------------------------------------------------------<br>> Ivan S. P. Marin<br>> Laboratório de Física Computacional<br>> <a href="http://lfc.ifsc.usp.br">
lfc.ifsc.usp.br</a> <<a href="http://lfc.ifsc.usp.br">http://lfc.ifsc.usp.br</a>><br>> Instituto de Física de São Carlos - USP<br>> ----------------------------------------------------------<br>> ------------------------------------------------------------------------
<br>><br>> _______________________________________________<br>> Beowulf mailing list, <a href="mailto:Beowulf@beowulf.org">Beowulf@beowulf.org</a><br>> To change your subscription (digest mode or unsubscribe) visit
<a href="http://www.beowulf.org/mailman/listinfo/beowulf">http://www.beowulf.org/mailman/listinfo/beowulf</a><br>><br><br><br>--<br><br>Richard B. Walsh<br><br>"The world is given to me only once, not one existing and one
<br> perceived. The subject and object are but one."<br><br>Erwin Schroedinger<br><br>Project Manager<br>Network Computing Services, Inc.<br>Army High Performance Computing Research Center (AHPCRC)<br><a href="mailto:rbw@ahpcrc.org">
rbw@ahpcrc.org</a> | 612.337.3467<br><br>-----------------------------------------------------------------------<br>This message (including any attachments) may contain proprietary or<br>privileged information, the use and disclosure of which is legally
<br>restricted. If you have received this message in error please notify<br>the sender by reply message, do not otherwise distribute it, and delete<br>this message, with all of its contents, from your files.<br>-----------------------------------------------------------------------
<br><br></blockquote></div><br><br clear="all"><br>-- <br>-----------------------------------------------------------<br>Ivan S. P. Marin<br>Laboratório de Física Computacional<br><a href="http://lfc.ifsc.usp.br">lfc.ifsc.usp.br
</a><br>Instituto de Física de São Carlos - USP<br>----------------------------------------------------------