Performance regressions in HEAD

[Neil Mitchell]

Hi Simon,

> > The attached charcount.zip shows the optimised cmm and the assembler.
> > As far as I can tell, the optimised cmm for the hotspot function in
> > both cases is identical, and the code should be spending all its time
> > in the hotspot. The names are Main_zdspreludezu942zull85_entry for 6.7
> > and Main_zdspreludezu942zull85_entry for 6.6.1.
>
> Any chance you could send the .hs file too?

It was attached to the first message, I've reattached it to this one too.

>  I tried to repeat it with the asm
> output you gave, I could compile and run the 6.6.1 version ok but not the HEAD
> version.  I replaced the code for Main_zdspreludezu942zull85_entry (and its
> subsidiary code blocks) with the HEAD version and saw no difference in
> performance.
>
> The only difference I can see in the assembly between those two functions is
> that in the HEAD there is a small optimisation where 3 instructions have been
> replaced by one.

These results were all obtained under Windows, which may be necessary
for replication (or not).

Thanks

Neil
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 4.hs
Type: text/x-haskell
Size: 4301 bytes
Desc: not available
Url : http://www.haskell.org/pipermail/cvs-ghc/attachments/20070706/1a9313d1/4.bin