Optimization manual updated again. Now covers Core 2

[agner]

I have updated my manual once again. Now covering everything about the new Intel Core 2 processor including a detailed study of the pipeline and execution units and complete lists of instruction timings.

This time my manual has come before the official manuals from Intel. Their software manuals for the Core 2 are not out yet. Thank you to a friendly person who gave me remote access to a prerelease sample of the Core 2. This enabled me to test almost everything.

The execution core is more powerful than anything we have seen until now. It can do up to three full 128-bit vector calculations per clock cycle. Unfortunately, the instruction fetch and predecode stage has not been expanded enough to keep up with the rest of the pipeline, so this is a serious bottleneck in many situations.

The section on AMD microarchitecture in my manual has also been revised, thanks to help from Andreas Kaiser.

http://www.agner.org/optimize/

[MazeGen]

You work is really unique. Thanks for your hard work, agner :U

[Ghirai]

Very nice, thanks for your work.

Would it be okay to add the asm manual to my mirror (link in my signature)?

[drhowarddrfine]

Thank you very much, Agner! :U

[BogdanOntanu]

Thank you very much ;)

[Vortex]

Agner,

Very nice work :U

[stanhebben]

I found a little error in the optimization manual:

At page 74, you state that the only way to load unaligned data in XMM registers is to use MOVDQU, MOVUPS, MOVUPD or LDDQU. You can use MOVLPD/MOVHPD pairs, which results in faster code.

Stan

[gabor]

Hello agner!

Thanks for your work! It is very nice to see people who do a lot of work in a specific area and are ready to share the gained knowledge! Thank you again!
I had a look on your website. There are interessting topics! This Cultural Selection Theory looks very exciting....

Greets, Gábor

[gwapo]

Great work, thanks!
I thought Mark Larson's "Code Optimization" is already the best optimization document I've got. I guess multiple "the best" documents are always better than having single "the best" document  :U

\
-chris

[agner]

Your link to Mark Larson doesn't work. Where is the document you are referring to?

[PBrennick]

Agner,
The link works just fine.  Please try it again.  The server may have been down when you last tried.

Paul

[KSS]

Agner, thanks you for your work! :U

[theunknownguy]

Agner you rulz ! thanks so much my favorite read for optimization  :dance:

[dedndave]

you guys have woken a 4 year old thread   :P

[clive]

you guys have woken a 4 year old thread   :P

Can we dig up the zombie P4 designers that removed the barrel shifter, and beat them to death again?

-Clive