Re: Fastcode Challenge Status 19

Hi Pierre etc.

> In my opinion the solution (for the people who insist on 16-byte
alignment)
> is to make benchmarks that perform very poorly with non-aligned memory
> blocks. If there really is a tangible advantage to 16-byte alignment then
> this will be reflected in the scores.

Then we need some MM's in 2-3 versions - 4, 8, 16 byte alignment otherwise
identical.

> Regards,
> Pierre

Regards
Dennis


.

Relevant Pages

  • Re: Fastcode Challenge Status 19
    ... > is to make benchmarks that perform very poorly with non-aligned memory ... If there really is a tangible advantage to 16-byte alignment then ... Regards ...
    (borland.public.delphi.language.basm)
  • Re: upcoming kerneloops.org item: get_page_from_freelist
    ... with SLUB + slub debug, that gave back non-aligned memory, causing ... Yes, but AFAICT, SLUB_DEBUG doesn't respect the given alignment without ... use objsize rather than size when calculating alignment). ...
    (Linux-Kernel)
  • Re: Alignment and format in listbox columns
    ... Regards, ... Claus ... > Alignment is set when you create the column header. ... > | how to control alignment for multiple columns. ...
    (microsoft.public.excel.programming)
  • Re: More on garbage collection
    ... >That's news to me. ... >> need to do either of these for physical memory rather than just ... >alignment, but I'm not sure what you would consider as a solution. ... Regards, ...
    (sci.crypt)
  • Re: Aligning Dynamic Arrays
    ... I am only asking for changes to dynamic arrays, ... > I am not convinced that 16-byte alignment is a good idea with very small ... > Our own benchmarks show that 8-byte alignment is faster than 16-byte ... >> This way we can get 16 byte aligned dynamic arrays, ...
    (borland.public.delphi.language.basm)