[Intel-gfx] sna: Experimental support for write-combining mmaps (wc-mmap)

[Sedat Dilek]

On Fri, Nov 7, 2014 at 12:14 PM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> On Fri, Nov 07, 2014 at 11:59:26AM +0100, Sedat Dilek wrote:
>> On Fri, Nov 7, 2014 at 11:43 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
>> > On Fri, Nov 07, 2014 at 11:35:01AM +0100, Sedat Dilek wrote:
>> >> On Fri, Nov 7, 2014 at 11:29 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
>> >> > On Fri, Nov 07, 2014 at 11:08:41AM +0100, Sedat Dilek wrote:
>> >> >> Hi,
>> >> >>
>> >> >> I am playing with wc-mmap [1]... (and sent a trivial patch [0]).
>> >> >>
>> >> >> Inspecting my logs...
>> >> >>
>> >> >> $ egrep 'wc|mmap'
>> >> >> logs/build-and-install-log_XF86-Video-Intel-v2-99-916-137-g0aa2edb_WCMMAP_llvm-3-4-2.txt
>> >> >>   Experimental support: TearFree mmap(wc)
>> >> >>
>> >> >> $ egrep -i 'wc|mmap|write|comb' /var/log/Xorg.0.log
>> >> >> [ EMPTY ]
>> >> >>
>> >> >> TearFree [1] can be set via an option in xorg.conf.
>> >> >> Doing a very quick look into the sources, such a flag/option is not recognized.
>> >> >>
>> >> >> How do I see if wc-mmap is enabled/active on my desktop?
>> >> >
>> >> > Er, hopefully you don't notice anything except it speeds up certain
>> >> > operations (quite considerably) :). You have to enable the DBG output to
>> >> > see it in action, as it is quite a lowlevel feature (i.e. how we
>> >> > interact with the kernel) and should not add any user visible features.
>> >> > However, it does also require the kernel patch for it to be enabled.
>> >>
>> >> Can you point me to that kernel-patch?
>> >> Is Linux v3.18-rc2+ good enough or do I need d-i-l-n / d-i-n?
>> >
>>
>> Hmm... against which trees are those patches?
>>
>> > http://patchwork.freedesktop.org/patch/35049/
>> "drm/i915: Broaden application of set-domain(GTT)"
>>
>> ...is not applicable against d-i-f or d-i-n.
>
> Try
> http://cgit.freedesktop.org/~ickle/linux-2.6/commit/?h=nightly&id=2b03060c7585df31e6e90fef91af4064769bf394

Here some numbers...

$ diff -uprN benchmarks/x11perf-aa10text_no-wcmmap.txt
benchmarks/x11perf-aa10text_wcmmap.txt
--- benchmarks/x11perf-aa10text_no-wcmmap.txt   2014-11-07
12:57:49.932264840 +0100
+++ benchmarks/x11perf-aa10text_wcmmap.txt      2014-11-07
12:53:17.028273399 +0100
@@ -2,16 +2,16 @@ $ x11perf -aa10text
 x11perf - X11 performance program, version 1.2
 The X.Org Foundation server version 11501000 on :0
 from fambox
-Fri Nov  7 12:56:29 2014
+Fri Nov  7 12:51:37 2014

-Sync time adjustment is 0.0814 msecs.
+Sync time adjustment is 0.0817 msecs.

-24000000 reps @   0.0002 msec (4420000.0/sec): Char in 80-char aa
line (Charter 10)
-24000000 reps @   0.0002 msec (4420000.0/sec): Char in 80-char aa
line (Charter 10)
-24000000 reps @   0.0002 msec (4380000.0/sec): Char in 80-char aa
line (Charter 10)
-24000000 reps @   0.0002 msec (4410000.0/sec): Char in 80-char aa
line (Charter 10)
-24000000 reps @   0.0002 msec (4410000.0/sec): Char in 80-char aa
line (Charter 10)
-120000000 trep @   0.0002 msec (4410000.0/sec): Char in 80-char aa
line (Charter 10)
+24000000 reps @   0.0002 msec (4490000.0/sec): Char in 80-char aa
line (Charter 10)
+24000000 reps @   0.0002 msec (4490000.0/sec): Char in 80-char aa
line (Charter 10)
+24000000 reps @   0.0002 msec (4480000.0/sec): Char in 80-char aa
line (Charter 10)
+24000000 reps @   0.0002 msec (4490000.0/sec): Char in 80-char aa
line (Charter 10)
+24000000 reps @   0.0002 msec (4490000.0/sec): Char in 80-char aa
line (Charter 10)
+120000000 trep @   0.0002 msec (4490000.0/sec): Char in 80-char aa
line (Charter 10)


 -dileks // 07-NOV-2014

- Sedat -