Forums before death by AOL, social media and spammers... "We can't have nice things"
|    comp.lang.asm.x86    |    Ahh, the lost art of x86 assembly    |    4,675 messages    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
|    Message 3,957 of 4,675    |
|    Rick C. Hodgin to Bart    |
|    Re: I'm looking for a mathematical libra    |
|    20 Sep 19 16:13:17    |
      From: rick.c.hodgin@gmail.com              On 9/20/2019 3:48 PM, Bart wrote:       > On 19/09/2019 13:41, Rick C. Hodgin wrote:       >> For a long time I've wanted to re-purpose the x87 opcodes to have       >> a different kind of math library. I've wanted to add arbitrary-       >> precision support in lieu of 80-bit reals. Use real hardware for       >> 32-bit and 64-bit, but when using 80-bit loads and operations on       >> 80-bit loaded values, use GMP and MPFR. Haven't gotten around to       >> it yet. Too many other things to do first.       >       > Arbitrary-precision floating point is going to be much slower than even       > emulation of 8087.              I often tell the people I work with when it comes to the algorithms       we're writing and why we're writing them that way, "We don't build       bridges for sunny days. We build bridges for hurricanes and 200 mph       winds and houses and trees slamming into them in the flood waters."       The same goes for our software algorithms.              Arbitrary precision emulation via hardware mechanisms would be a       nice feature in some cases. GMP and MPFR are pretty fast, even       for arbitrary precision. They're just many orders of magnitude       slower than hardware FPU ops, but they also have the benefit of       carrying their work out to far more significant digits.              It all depends on what you need.                     > But in either case, overheads of using a function call are       > going to be insignificant.              Agreed.              > So I can't see the need to re-use 8087 opcodes. In any case, those are       > designed to work with a small stack of 80-bit floating point values (in that       > the operands will often be implicit). Arbitrary precision numbers would have       > a different representation, and need to be put elsewhere, and the results       > also need to be placed elsewhere              Absolutely.              --       Rick C. Hodgin              --- SoupGate-Win32 v1.05        * Origin: you cannot sedate... all the things you hate (1:229/2)    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
(c) 1994, bbs@darkrealms.ca