lkml.org 
[lkml]   [2000]   [Oct]   [31]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: 2.2.18Pre Lan Performance Rocks!


Ingo Molnar wrote:
>
> On Tue, 31 Oct 2000, Jeff V. Merkey wrote:
>
> > [...] These types of optimizations are possible when people have
> > acccess to Intel Red Cover documents, [...]
>
> optimizing away AGIs has been documented by public Intel PDFs for years:
>
> [...] Since the Pentium processor has two integer pipelines, a register
> used as the base or index component of an effective address calculation
> (in either pipe) causes an additional clock cycle if that register is the
> destination of either instruction from the immediately preceding clock
> cycle. This effect is known as Address Generation Interlock (AGI).
>
> (ditto for the p6 core CPUs), and GCC (IIRC) tries to avoid AGI conflicts
> as much as possible. (and this kind of stuff belongs into the compiler)

Odd. When I profile Linux with EMON, I see tons of them. Anywhere code
does

mov eax, addr
mov [addr], ebx

one will get generated. Even something as simple as:

mov eax, addr
inc eax
dec eax
mov [addr]. ebx

will avoid an AGI (since the other pipeline has now been allowed a few
clocks to fetch the address and load it).

Jeff



>
> Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 12:45    [W:0.111 / U:0.592 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site