Re: [PATCH] ac7 Athlon-SMP

[Alan Cox]

> Athlon docs recommend inlining for less than 25 machine instructions.
> 
> $ objdump -d vmlinux | grep memcpy -- | wc
> 	430	3709	28419
> 
> so inlining them would cost 412*110 or 45k.
> 
> Who is the author/maintainer of the Athlon code? I don't see any reference in the source
> or the kernel docs.

I am. Take a K7 and some benchmarking tools like lmbench and give it a whirl.
Im pretty sure you want to inline small copies. Note that we have a suprisingly
high proportion of short copies and that those we want to inline via rep movs

The inline code you want is something like

		in_irq ?
		jmp #1
		len < 256
		jmp #1
		jmp out_of_line_copy
	#1
		[existing x86 rep based copy]