Prev: What makes a trivial operator trivial ?
Next: May empty entire math departments
From: Chip Eastham on 1 Feb 2010 09:43
On Jan 31, 1:23 pm, Jeffrey Walton <noloa...(a)gmail.com> wrote:
> Hi All,
>
> I'm looking at some code that is seg faulting while attempting to
> perform a divide using a 64 bit dividend and divisor. The code is
> written without a native 64 bit word, so:
>
> struct MY_INT_64
> {
>     int32 high;  // signed
>     uint32 low;  // unsigned
>
> }
>
> The code is generalized using Knuth's algorithm from AOCP (Volume 2,
> p. 273). I think the best way to proceed is to use a canned
> implementation specialized for 32 bit word sizes rather than vet out
> the problems with the generalized implementation.
>
> Does anyone have a C/C++ implementation for 64 bit division using two
> 32 bit registers? I only need the quotient, but I'll take something
> that computes both q and r.
>
> Thanks,
> JW

I'm a bit confused about the requirements.
A C implementation might tell something about
the operand sizes, but "using two 32 bit
registers" is not something one would get
out of a high-level language. Is it your
desire to have a C (or C++) subroutine that
uses inline assembler?

If I understand your struct MY_INT_64, you
are trying to implement your non-native
64 bit integer as a signed operand. If the
divisor and dividend are both to be of this
type, I think more than two 32 bit registers
are needed since the dividend itself would
take up that much room. [If I remember the
Knuth material you are talking about, there
is a proposed algorithm for refining from
32-bit to 64-bit quotients. Is that what
you have in mind?]

regards, chip
From: Michael Press on 2 Feb 2010 16:29
In article
<6b91375f-78b1-440f-9ccb-d3212f9a7cbe(a)30g2000vbj.googlegroups.com>,
Jeffrey Walton <noloader(a)gmail.com> wrote:

> Hi All,
>
> I'm looking at some code that is seg faulting while attempting to
> perform a divide using a 64 bit dividend and divisor. The code is
> written without a native 64 bit word, so:
>
> struct MY_INT_64
> {
> int32 high; // signed
> uint32 low; // unsigned
> }
>
> The code is generalized using Knuth's algorithm from AOCP (Volume 2,
> p. 273). I think the best way to proceed is to use a canned
> implementation specialized for 32 bit word sizes rather than vet out
> the problems with the generalized implementation.
>
> Does anyone have a C/C++ implementation for 64 bit division using two
> 32 bit registers? I only need the quotient, but I'll take something
> that computes both q and r.

Look for lldiv() in stdlib.
It returns the remainder and quotient.

--
Michael Press
From: Tim Little on 2 Feb 2010 19:26
On 2010-02-01, Jeffrey Walton <noloader(a)gmail.com> wrote:
> I believe 'long long' makes an assumption that the size is 64 bits,

According to the current C standard, "long long int" is guaranteed to
have at least 64 value bits, and hence suffices.


- Tim
First  |  Prev  | 
Pages: 1 2
Prev: What makes a trivial operator trivial ?
Next: May empty entire math departments