Brandon S. Allbery KF8NH ha scritto:

On 2009 Mar 26, at 11:39, Manlio Perillo wrote:

...

The execution time and CPU usage is almost the same. However the C++ version requires 305 MB, the GHC version 617 MB.

I wonder how much of that is due to lifting (i.e. laziness).

http://hpaste.org/fastcgi/hpaste.fcgi/view?id=2911 Performances are the same; but I'm not sure if this removes all laziness. I have changed ins t (k,x) = insert k x t to ins t (k,x) = k `seq` x `seq` insert k x t Manlio

Bulat Ziganshin ha scritto:

Hello Manlio,

Thursday, March 26, 2009, 6:39:12 PM, you wrote:

...

The test consists in adding 10000000 elements to an empty map.

+RTS -c -F1.1

then read about garbage collection

It now requires 386 MB of memory, but is 4.7 times slower. So, now memory required is about the same as the C++ version, but how can I optimize memory usage without having to tweak the garbage collector? Thanks Manlio

manlio_perillo:

Bulat Ziganshin ha scritto:

...

Hello Manlio,

Thursday, March 26, 2009, 6:39:12 PM, you wrote:

...

The test consists in adding 10000000 elements to an empty map.

+RTS -c -F1.1

then read about garbage collection

It now requires 386 MB of memory, but is 4.7 times slower.

So, now memory required is about the same as the C++ version, but how can I optimize memory usage without having to tweak the garbage collector?

You'll need similar data structures.

So how does F# IntMap version compares to Haskell's IntMap? On Sat, Apr 4, 2009 at 2:58 PM, Peter Verswyvelen wrote:

On Sat, Apr 4, 2009 at 12:42 PM, Jon Harrop wrote:

...

For comparison:

Haskell hash table: 44s Haskell map: 7s F# hash table: 0.7s

Ouch! That's pretty insane.