[Caml-list] Bigarray access speed

[Caml-list] Bigarray access speed

[Richard Nyberg]

In the following small programs 3) is faster than 1) and 2), which run
equally fast. However, 4 is significantly slower than the rest. If you
change the line "a.{i} <- a.{i} + i" to "a.{i} <- i" the execution time
is halfed but it's still much slower.

Are access to Bigarrays slower when passed to functions? If so, is it
fixable? or is there some workaround?

I stumbled upon this while coding on a school assignment ((not too ;) fast
multiplication of large integers).

1)
let a = Array.make 1000000 0 in
for i = 0 to 999999 do
    a.(i) <- a.(i) + i;
done;;

2)
let a = Array.make 1000000 0 in
let rec loop a i =
  if i <= 999999 then begin
    a.(i) <- a.(i) + i;
    loop a (i + 1)
  end in
loop a 0;;

3)
open Bigarray;;
let a = Array1.create int c_layout 1000000 in
Array1.fill a 0;
for i = 0 to 999999 do
  a.{i} <- a.{i} + i
done;;

4)
open Bigarray;;
let a = Array1.create int c_layout 1000000 in
Array1.fill a 0;
let rec loop a i =
  if i <= 999999 then begin
    a.{i} <- a.{i} + i;
    loop a (i + 1)
  end in
loop a 0;;

	-Richard
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Bigarray access speed

[malc]

On Fri, 16 Aug 2002, Richard Nyberg wrote:

> In the following small programs 3) is faster than 1) and 2), which run
> equally fast. However, 4 is significantly slower than the rest. If you
> change the line "a.{i} <- a.{i} + i" to "a.{i} <- i" the execution time
> is halfed but it's still much slower.
> 
> Are access to Bigarrays slower when passed to functions? If so, is it
> fixable? or is there some workaround?
> 
> I stumbled upon this while coding on a school assignment ((not too ;) fast
> multiplication of large integers).
> 
> 1)
> let a = Array.make 1000000 0 in
> for i = 0 to 999999 do
>     a.(i) <- a.(i) + i;
> done;;
> 
> 2)
> let a = Array.make 1000000 0 in
> let rec loop a i =
>   if i <= 999999 then begin
>     a.(i) <- a.(i) + i;
>     loop a (i + 1)
>   end in
> loop a 0;;
> 
> 3)
> open Bigarray;;
> let a = Array1.create int c_layout 1000000 in
> Array1.fill a 0;
> for i = 0 to 999999 do
>   a.{i} <- a.{i} + i
> done;;
> 
> 4)
> open Bigarray;;
> let a = Array1.create int c_layout 1000000 in
> Array1.fill a 0;
> let rec loop a i =
>   if i <= 999999 then begin
>     a.{i} <- a.{i} + i;
>     loop a (i + 1)
>   end in
> loop a 0;;

http://caml.inria.fr/archives/200110/msg00148.html

An aside(not all facts are cheked):

Bigarrays(of at least floats) can have a slight edge over normals arrays.
To get maximal speed of the inner loops data needs to be naturally 
aligned. OCaml does nothing to enforce it for non-big arrays. Bigarrays on 
the other hand are mmaped(4k on IA32) and you get perfectly aligned data 
for free. I was thinking that maybe Array can be extended with
make[create]_aligned, for speed/space tradeoff.

-- 
mailto:malc@pulsesoft.com

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Bigarray access speed

[Markus Mottl]

On Fri, 16 Aug 2002, malc wrote:
> Bigarrays(of at least floats) can have a slight edge over normals arrays.
> To get maximal speed of the inner loops data needs to be naturally 
> aligned. OCaml does nothing to enforce it for non-big arrays. Bigarrays on 
> the other hand are mmaped(4k on IA32) and you get perfectly aligned data 
> for free. I was thinking that maybe Array can be extended with
> make[create]_aligned, for speed/space tradeoff.

Additionally, it would also be nice to have a specialized "create"
function for (naturally unboxed) float arrays such that they need not
be initialized with a given float value. This may be beneficial for
algorithms that allocate work space whose contents is not necessarily
fully needed but filled on demand.

Regards,
Markus Mottl

-- 
Markus Mottl                                             markus@oefai.at
Austrian Research Institute
for Artificial Intelligence                  http://www.oefai.at/~markus
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

float array alignment; was Re: [Caml-list] Bigarray access speed

[William Chesters]

malc writes:
 > To get maximal speed of the inner loops data needs to be naturally 
 > aligned. OCaml does nothing to enforce it for non-big arrays. Bigarrays on 
 > the other hand are mmaped(4k on IA32) and you get perfectly aligned data 
 > for free. I was thinking that maybe Array can be extended with
 > make[create]_aligned, for speed/space tradeoff.

I did this once to be able to interface with Fortran libs on Sparc32,
and I still have a patch (against ocaml-2.01) lying around.  It was
actually quite thoroughly tested, but it's probably not very tidy.  The
main gotcha iirc was getting output_value/input_value to preserve
alignment :).
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: float array alignment; was Re: [Caml-list] Bigarray access speed

[Xavier Leroy]

malc writes:
> To get maximal speed of the inner loops data needs to be naturally 
> aligned. OCaml does nothing to enforce it for non-big arrays. Bigarrays on 
> the other hand are mmaped(4k on IA32) and you get perfectly aligned data 
> for free. I was thinking that maybe Array can be extended with
> make[create]_aligned, for speed/space tradeoff.

As William Chester said, allocating 8-aligned arrays isn't really
hard, but keeping them 8-aligned across copying collection,
compaction, and structured I/O is quite a pain.

My experiments indicate that the lack of alignment on float arrays 
(or more precisely the fact that they are 4-aligned instead of
8-aligned) has negligible impact on performance for the IA32 (Pentium)
and PowerPC processors, but non-negligible for SPARC and MIPS.
And of course on a 64-bit architecture the problem goes away because
everything in the Caml heap is then 8-aligned.  Since I expect IA32
and PowerPC to remain dominant until we massively switch to 64-bit
processors, there's no urgent need to do something about float array
alignment.

- Xavier Leroy
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: float array alignment; was Re: [Caml-list] Bigarray access speed

[malc]

On Mon, 19 Aug 2002, Xavier Leroy wrote:

> malc writes:
> > To get maximal speed of the inner loops data needs to be naturally 
> > aligned. OCaml does nothing to enforce it for non-big arrays. Bigarrays on 
> > the other hand are mmaped(4k on IA32) and you get perfectly aligned data 
> > for free. I was thinking that maybe Array can be extended with
> > make[create]_aligned, for speed/space tradeoff.
> 
> As William Chester said, allocating 8-aligned arrays isn't really
> hard, but keeping them 8-aligned across copying collection,
> compaction, and structured I/O is quite a pain.
> 
> My experiments indicate that the lack of alignment on float arrays 
> (or more precisely the fact that they are 4-aligned instead of
> 8-aligned) has negligible impact on performance for the IA32 (Pentium)
> and PowerPC processors, but non-negligible for SPARC and MIPS.
> And of course on a 64-bit architecture the problem goes away because
> everything in the Caml heap is then 8-aligned.  Since I expect IA32
> and PowerPC to remain dominant until we massively switch to 64-bit
> processors, there's no urgent need to do something about float array
> alignment.

IA32 is now much bigger family, and unlucky owners of AMD 7th generation 
machines, such as myself, do pay a price for unaligned double precission
float accesses. 

-- 
mailto:malc@pulsesoft.com

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Bigarray access speed

[Richard Nyberg]

> http://caml.inria.fr/archives/200110/msg00148.html

Yes. That message explained it very well :)
Thanks!

	-Richard
-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners

Re: [Caml-list] Bigarray access speed

[Joe HELL]

Bigarray is designed to facilitate operation on large numerical array when 
used with external C functions.

Pure caml use of bigarray is generally slower.

_________________________________________________________________
Chat with friends online, try MSN Messenger: http://messenger.msn.com

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners