Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming > #2553 > unrolled thread

integer Vs floating point efficiency?

Back to article view | Back to comp.programming

Contents

  integer Vs floating point efficiency? Rui Maciel <rui.maciel@gmail.com> - 2012-12-01 12:39 +0000
    integer Vs floating point efficiency? Rui Maciel <rui.maciel@gmail.com> - 2012-12-01 12:41 +0000
      Re: integer Vs floating point efficiency? Herman Rubin <hrubin@skew.stat.purdue.edu> - 2012-12-02 23:10 +0000
        Re: integer Vs floating point efficiency? Robin Vowels <robin.vowels@gmail.com> - 2012-12-02 16:36 -0800
      Re: integer Vs floating point efficiency? Rui Maciel <rui.maciel@gmail.com> - 2012-12-03 10:03 +0000
    Re: integer Vs floating point efficiency? BGB <cr88192@hotmail.com> - 2012-12-01 16:49 -0600
    Re: integer Vs floating point efficiency? "Chris Uppal" <chris.uppal@metagnostic.REMOVE-THIS.org> - 2012-12-03 10:22 +0000

#2553 — integer Vs floating point efficiency?

Is there any information on the relative efficiency of modern CPUs with 
regards to algebraic operations on integer and floating point data types?  

If there is, where can I get my hands on it?


Thanks in advance,
Rui Maciel

[toc] | [next] | [standalone]

#2554

Gratuitous cross-posting to extend this query to s.m.num-analysis.

Rui Maciel wrote:

> Is there any information on the relative efficiency of modern CPUs with
> regards to algebraic operations on integer and floating point data types?
> 
> If there is, where can I get my hands on it?
> 
> 
> Thanks in advance,
> Rui Maciel

[toc] | [prev] | [next] | [standalone]

#2568

On 2012-12-01, Rui Maciel <rui.maciel@gmail.com> wrote:
> Gratuitous cross-posting to extend this query to s.m.num-analysis.

> Rui Maciel wrote:

>> Is there any information on the relative efficiency of modern CPUs with
>> regards to algebraic operations on integer and floating point data types?

>> If there is, where can I get my hands on it?


>> Thanks in advance,
>> Rui Maciel

It depends on the operations.  Unless the operations are integer
operations, there is always conversion of integer to float.  This
conversion is not that efficient; it typically requires putting
the integer in an integer register, shifting if necessary, placing
an exponent in that register, putting it back in memory, reading
it into a floating point register, and subtracting what the result
would have been if it were 0.  Even if memory references were not
needed, and shifing not needed, and the same registers could be
used for integers and floats, there are at least two operations
required if the needed constants were already in memory.  This
is converting "single precision" integers into "double precision"
floats.

Getting the integer part of a float is even worse if not hardware.




-- 
This address is for information only.  I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@stat.purdue.edu         Phone: (765)494-6054   FAX: (765)494-0558

[toc] | [prev] | [next] | [standalone]

#2571

On Dec 3, 10:10 am, Herman Rubin <hru...@skew.stat.purdue.edu> wrote:
> On 2012-12-01, Rui Maciel <rui.mac...@gmail.com> wrote:
>
> > Gratuitous cross-posting to extend this query to s.m.num-analysis.
> > Rui Maciel wrote:
> >> Is there any information on the relative efficiency of modern CPUs with
> >> regards to algebraic operations on integer and floating point data types?
> >> If there is, where can I get my hands on it?
> >> Thanks in advance,
> >> Rui Maciel
>
> It depends on the operations.  Unless the operations are integer
> operations, there is always conversion of integer to float.  This
> conversion is not that efficient; it typically requires putting
> the integer in an integer register, shifting if necessary, placing
> an exponent in that register, putting it back in memory, reading
> it into a floating point register, and subtracting what the result
> would have been if it were 0.  Even if memory references were not
> needed, and shifing not needed, and the same registers could be
> used for integers and floats, there are at least two operations
> required if the needed constants were already in memory.  This
> is converting "single precision" integers into "double precision"
> floats.
>
> Getting the integer part of a float is even worse if not hardware.

On Intel, instructions are provided to convert directly from integer
to float, and vice-versa, and without referencing memory..

But the OP wanted to know the speed differences between performing
operations on integers compared to performing operations on floats
(and not the conversions between integers an floats).

[toc] | [prev] | [next] | [standalone]

#2577

Gordon Sande wrote:

> On 2012-12-01 08:41:21 -0400, Rui Maciel said:
>
> Instruction timings etc are to be found in the processor's data sheets.
> Ask the manufacturer of one. The web sites will have copies or request
> pages.

I've followed your advice, and I'm in the process of digesting AMD and 
Intel's docs that cover this subject.  If anyone is interested, the 
documents are available at:

http://support.amd.com/us/psearch/Pages/psearch.aspx?type=2.1%3b2.12&product=2.1.2&contentType=Tech%20Doc%20Processor
http://www.intel.com/content/www/us/en/processors/architectures-software-
developer-manuals.html


Unfortunately, the information isn't compiled or easily accessible.  To make 
matters worse, I've stumbled on a claim that Intel's docs have been 
"incomplete and full of errors", which isn't very reassuring.  

Nevertheless, some info on this subject has been compiled by third parties, 
such as the following article:

http://www.agner.org/optimize/instruction_tables.pdf


As should be expected, latency varies with the make and model of the 
processor, and instruction latency alone doesn't say everything there is to 
say (or, possibly, anything) about the inherent efficiency of integer Vs 
floating point data types. In addition, there are more than a way to skin 
some numerical cats with the AMD64 platform, which adds to the complexity of 
this issue.

Nevertheless, from what I gathered so far it appears that, in general, 
integer operations do present a small, but possibly entirely meaningless, 
efficiency advantage.


Rui Maciel

[toc] | [prev] | [next] | [standalone]

#2558

On 12/1/2012 6:39 AM, Rui Maciel wrote:
> Is there any information on the relative efficiency of modern CPUs with
> regards to algebraic operations on integer and floating point data types?
>
> If there is, where can I get my hands on it?
>

informal results are that integer is slightly faster, especially when 
working with integer data (because integer<->float conversions are 
expensive).

the difference at this point is "modest" though, as both seem to be more 
at the whim of "how quickly the CPU can perform operations".


the main advantage of integer though is that it is possible to more 
precisely control round-off behavior (allowing for more determinism), 
which, combined with it not suffering from conversion-performance 
issues, makes integer math better suited to things like signal 
processing and A/V codecs (which are typically, mostly, integer and 
fixed-point math).

there are downsides to fixed-point though (vs floats), namely that it is 
more awkward than floating-point math (since manual handling of rounding 
and lots of shifts are needed), if used naively may be actually slower 
than floats, and typically has a more limited value-range (the number of 
bits above and below the decimal point may become an important factor in 
the operation of an algorithm).


>
> Thanks in advance,
> Rui Maciel
>

[toc] | [prev] | [next] | [standalone]

#2578

Rui Maciel wrote:
> Is there any information on the relative efficiency of modern CPUs with
> regards to algebraic operations on integer and floating point data types?

A very valuable resource, comprehensive, and apparently expert (not being an 
expert myself, I can't really tell):

    http://www.agner.org/optimize/

Not sure how well he addresses your specific question (think he does but I 
can't remember for certain), but a great source of understanding of these 
highly non-trivial architectures.

    -- chris 

[toc] | [prev] | [standalone]

Back to top | Article view | comp.programming

csiph-web