News:

Masm32 SDK description, downloads and other helpful links
Message to All Guests
NB: Posting URL's See here: Posted URL Change

Main Menu

Managing data structures: lists

Started by NoCforMe, January 31, 2024, 03:03:00 PM

Previous topic - Next topic

sinsi

Just let Windows worry about the heap?
🍺🍺🍺

NoCforMe

I think by that you mean "just waste it", as I described above, is that right?

That's certainly a valid choice. You can always allocate more if you run out of space.
Assembly language programming should be fun. That's why I do it.

sinsi

New string - HeapAlloc
Delete string - HeapFree
Resize string - HeapReAlloc

Windows uses a low-fragmentation heap, so it will take care of coalescing free blocks etc.
The downside is that there are a few bytes of overhead, so many small allocations might not work so well.
🍺🍺🍺

jj2007

Quote from: sinsi on February 06, 2024, 08:28:37 PMmany small allocations might not work so well

The Windows heap is optimised for many small allocations (there is VirtualAlloc for bigger ones, although it never makes a difference, speed-wise). Still, there is overhead that slows things down.

There are two faster options:
1. the stack (by default, up to around 800kBytes, more if you use linker option /stack:nnnnh)
2. a ring buffer; used in MB e.g. for Print Str$(number1), "<>", Str$(number2)

Both have the advantage that no *Free is needed.

Speed-wise, all three options are most of the time sufficient: you rarely use HeapAlloc in a speed-critical innermost loop. Tests with StackBuffer() show that it has an advantage over HeapAlloc for about 0-500kBytes; above that value, HeapAlloc is equally fast or slightly faster (YMMV).

NoCforMe

I have successfully incorporated my LM (list management) module into my editor, the latest version of which is available in this post.
Assembly language programming should be fun. That's why I do it.

NoCforMe

Quote from: jj2007 on February 06, 2024, 08:56:43 PMThe Windows heap is optimised for many small allocations (there is VirtualAlloc for bigger ones, although it never makes a difference, speed-wise). Still, there is overhead that slows things down.

There are two faster options:
1. the stack (by default, up to around 800kBytes, more if you use linker option /stack:nnnnh)
2. a ring buffer; used in MB e.g. for Print Str$(number1), "<>", Str$(number2)
I just don't get it.

What is it with your obsession with speed here? To me, that's the last thing I'd be looking at when it comes to memory allocation. What I would want out of a memory allocator is
  • Efficiency of allocation; minimal memory wasted
  • Reliability and good error handling
Assembly language programming should be fun. That's why I do it.

jj2007

Quote from: NoCforMe on February 16, 2024, 08:48:57 AMWhat is it with your obsession with speed here?

Quote from: sinsi on February 06, 2024, 08:28:37 PMmany small allocations might not work so well

I wouldn't accuse Sinsi of being obsessed with speed, but his only argument against using the heap was, indeed, speed. Do you get it now?

Quote from: NoCforMe on February 16, 2024, 08:48:57 AMEfficiency of allocation; minimal memory wasted

Why are you so obsessed with memory? I have 16GB of it.

NoCforMe

Quote from: jj2007 on February 16, 2024, 11:02:24 AM
Quote from: NoCforMe on February 16, 2024, 08:48:57 AMEfficiency of allocation; minimal memory wasted
Why are you so obsessed with memory? I have 16GB of it.
Sure; I've got 5Gb, which ain't chicken feed either.

However, someone else who has some gigantic memory-intensive application would be quite concerned with efficient usage of memory. Not me, but others. Speed, probably not so much.
Assembly language programming should be fun. That's why I do it.

jj2007

Somebody with a "gigantic memory-intensive application" would be extremely concerned with speed, for obvious reasons.

sinsi

Actually, my point was about memory - if you allocate lots of 4-byte blocks, there is at least an 8-byte overhead associated with it, so every 4 bytes you allocate actually uses 12. Of course, this is an extreme case :biggrin:
🍺🍺🍺

jj2007

Quote from: sinsi on February 16, 2024, 01:02:52 PMif you allocate lots of 4-byte blocks, there is at least an 8-byte overhead

Point taken :thumbsup: