Re: Creating fastest Mul31

• Subject: Re: Creating fastest Mul31
• From: Ian Ollmann <email@hidden>
• Date: Thu, 17 Nov 2005 09:37:20 -0800
• Delivered-to: email@hidden
• Delivered-to: email@hidden

On G5, you can probably do this faster with mulld.

Ian

On Nov 17, 2005, at 7:18 AM, Rustam Muginov wrote:

Thank you, i'd try it and check if performance would change.

--
Sincerely,
	Rustam Muginov

On Nov 17, 2005, at 3:54 PM, Jonas Maebe wrote:

On 17 nov 2005, at 12:01, Jonas Maebe wrote:

You can replace the "slwi" and "or" at the end with a rlwimi (if you put the result of the mullhw immediately in result instead of in hi_prod), but I doubt it will make that much of a difference (majority of the time is spent multiplying). Something like (untested)

mullhw result, x, y
rlwimi result, lo_prod, 1, 0, 30

Still not tested, but definitely more correct than the above:

    mullw lo_prod, x, y
    srwi result, lo_prod, 31

    mulhw hi_prod, x, y
    rlwimi result, hi_prod, 1, 0, 30

_______________________________________________
Do not post admin requests to the list. They will be ignored.
PerfOptimization-dev mailing list (PerfOptimization- email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/perfoptimization-dev/iano% 40apple.com

This email sent to email@hidden

_______________________________________________
Do not post admin requests to the list. They will be ignored.
PerfOptimization-dev mailing list      (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/perfoptimization-dev/email@hidden

• Prev by Date: Re: Creating fastest Mul31
• Next by Date: Fastest way to find max element in vector
• Previous by thread: Re: Creating fastest Mul31
• Next by thread: Fastest way to find max element in vector
• Index(es):
  • Date
  • Thread