> I was really surprised by the 15% increases, they sounded impossible on
> any real modern hardware (any reasonable modern graphics hardware tends
> to not need all that much CPU access, and the TLB doesn't matter for the
> graphics card itself which certainly has direct access to all of its own
> memory).
i have clearly described in my mail that i have forced the X server to do
all operations 'unaccelerated', ie. in the framebuffer. So the numbers
made sense. Even if this doesnt happen often on any modern system, it was
an opportunity to test the impact of 4M pages on RL applications.
> I can imagine that large pages made sense for older bitmapped devices
> without hardware acceleration and with faster buses. But I think it's
> an optimization that is no longer very relevant as most of the real
> video activity happens on the graphics hardware these days.
yes, but the workload/impact is analogous to some other (valid) 4M
application.
i think the reason for no performance increase is that 64 TLBs cover 256K
worth of pages, and x11perf even in truecolor mode operates on a window
that has a size comparable to this.
-- mingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/