<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
  <head>
    <meta content="text/html; charset=ISO-8859-1"
      http-equiv="Content-Type">
  </head>
  <body bgcolor="#ffffff" text="#000000">
    On 4/10/2011 18:35, Mariano Martinez Peck wrote:
    <blockquote
      cite="mid:BANLkTinncrF6O2S2pFoT7VpG=Jgb_QFgkg@mail.gmail.com"
      type="cite">Sorry for arriving late to the thread. I have a simiar
      problem now trying to compile Cog in a Windows box with MinGW. <br>
      <br>
      The problem is described in another thread: <a
        moz-do-not-send="true"
href="http://forum.world.st/Cannot-generate-VM-in-Windows-with-CMake-and-Git-tp3434566p3434566.html">http://forum.world.st/Cannot-generate-VM-in-Windows-with-CMake-and-Git-tp3434566p3434566.html</a><br>
      <br>
      So...if we remove "-mno-fused-madd"&nbsp; from windows confs like
      CogMsWindowsConfig, then Corquet wouldn't behave correctly ?&nbsp; but
      if we keep them, it is difficult to compile in some platforms.
      So...I wonder, I am the only one trying to compile this on
      Windows?&nbsp; Who do the Croquet guys do to compile with this flag in
      Windows?<br>
    </blockquote>
    <br>
    gcc 3.x accepts -mno-fused-madd fine.<br>
    <br>
    Cheers,<br>
    &nbsp; - Andreas<br>
    <br>
    <blockquote
      cite="mid:BANLkTinncrF6O2S2pFoT7VpG=Jgb_QFgkg@mail.gmail.com"
      type="cite">
      <br>
      Thanks<br>
      <br>
      Mariano<br>
      <br>
      <div class="gmail_quote">On Tue, Jan 25, 2011 at 9:05 PM, Nicolas
        Cellier <span dir="ltr">&lt;<a moz-do-not-send="true"
            href="mailto:nicolas.cellier.aka.nice@gmail.com">nicolas.cellier.aka.nice@gmail.com</a>&gt;</span>
        wrote:<br>
        <blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt
          0.8ex; border-left: 1px solid rgb(204, 204, 204);
          padding-left: 1ex;"><br>
          2011/1/25 Andreas Raab &lt;<a moz-do-not-send="true"
            href="mailto:andreas.raab@gmx.de">andreas.raab@gmx.de</a>&gt;:<br>
          <div class="im">&gt;<br>
            &gt; On 1/25/2011 11:38 AM, Igor Stasenko wrote:<br>
            &gt;&gt;&gt;<br>
            &gt;&gt;&gt; Squeak uses highly unportable flag
            "-mno-fused-madd".<br>
            &gt;&gt;&gt; It isn't present in many systems, even those
            using GCC,<br>
            &gt;&gt;&gt; since many systems use older compilers.<br>
            &gt;&gt;&gt;<br>
            &gt;&gt;&gt; Also, why do you build fdlibm? It is old stuff
            that works not so well,<br>
            &gt;&gt;&gt; we have better libm. Is there a way to use our
            libm?<br>
            &gt;&gt;&gt;<br>
            &gt;&gt; I asked &nbsp;same question few weeks ago.<br>
            &gt;&gt; Check mailing list archive for discussion.<br>
            &gt;&gt; In short: differrent libm implementations work
            differently and some<br>
            &gt;&gt; have bad support of IEEE standard.<br>
            &gt;<br>
            &gt; Actually, that's not quite the point. The issue is that
            Croquet requires<br>
            &gt; bit-identical computations including floating point.
            For FPU computations,<br>
            &gt; the use of -mno-fused-madd avoids the use of the fused
            multiply-add<br>
            &gt; operation by compilers which support it which would
            generate different<br>
            &gt; results from compilers not using fused madd.<br>
            &gt;<br>
            &gt; The usage of fdlibm is similar. As Nicolas has pointed
            out elsewhere, fdlibm<br>
            &gt; is in some cases actually inferior of the platform
            libms (one might say<br>
            &gt; outright broken) but the requirement for the usage in
            Croquet isn't really<br>
            &gt; whether it's "correct" or "good". The requirement is
            bit-identical results<br>
            &gt; across all platforms. The results can be wrong as long
            as they are<br>
            &gt; consistently wrong. But they mustn't be different.<br>
            &gt;<br>
            <br>
          </div>
          Yes, I was disappointed by exp(1), but fdlibm sin(pi+epsilon)
          is far<br>
          far superior to i86 hardwired answer (i86 approximation of pi
          is well<br>
          known to be poor).<br>
          <font color="#888888"><br>
            Nicolas<br>
          </font>
          <div>
            <div class="h5"><br>
              &gt; Having said that, for your regular Squeak VM (i.e.,
              not requiring<br>
              &gt; bit-identical floating point results) there really
              isn't a requirement to<br>
              &gt; use either -mno-fused-madd or fdlibm.<br>
              &gt;<br>
              &gt; Cheers,<br>
              &gt; &nbsp;- Andreas<br>
              &gt;<br>
            </div>
          </div>
        </blockquote>
      </div>
      <br>
      <br clear="all">
      <br>
      -- <br>
      Mariano<br>
      <a moz-do-not-send="true" href="http://marianopeck.wordpress.com"
        target="_blank">http://marianopeck.wordpress.com</a><br>
      <br>
    </blockquote>
  </body>
</html>