Tested the effect of just this particular patch, on BDW GT2, for all GLBenchmark / GfxBench, GpuTest, Unigine demos and SynMark test-cases (3 runs each):

-6% Unigine Valley (fullscreen GL)
 0% GLB 3.0 T-Rex (fullscreen GL)
+2% GLB 2.7 Egypt (windowed GLES)
+4% GLB 2.7 T-Rex (windowed GLES)

On other tests there was no effect that wouldn't have been within their normal variances.


Wendy, there have been mentions of PMA issue having much larger benefits (on HSW) in Valve games and Lightsmark.  Could you mail me information about the performance before and after the indicated Mesa commit in Lightsmark and all the games?

NOTE: You need to do that test with the fixed X intel driver (which came after this commit), so that that issue doesn't distort the numbers.

Test base on Eero's suggestion, got below compare result on BDW GT2 E2:
parents of bad commit	FPS	after of bad commit 	FPS	(Before-After)/Before
doom3 50.1	50.1	doom3 49.5	49.5	1.37%
etqw_1_10 36.5	36.5	etqw_1_10 36.0	36	0.90%
etqw-demo 33.2	33.2	etqw-demo 32.9	32.9	0.78%
lightsmark 124.85	124.85	lightsmark 123.87	123.87	1.67%
openarena 35.9	35.9	openarena 35.3	35.3	2.83%
padman 151.9	151.9	padman 147.6	147.6	-1.16%
smokin-guns 129.5	129.5	smokin-guns 131.0	131	0.52%
urbanterror 96.9	96.9	urbanterror 96.4	96.4	1.58%
warsow01 126.7	126.7	warsow01 124.7	124.7	1.23%
xonotic07 166.31	166.31	xonotic07 164.26	164.26	-0.22%
cs 77.34	77.34	cs 77.51	77.51	-1.04%
tf2 44.17	44.17	tf2 44.63	44.63	-1.33%
hl2 88.55	88.55	hl2 89.73	89.73	0.08%
portal 63.48	63.48	portal 63.43	63.43	1.62%
unigine-heaven_4_0 6.18	6.18	unigine-heaven_4_0 6.08	6.08	8.27%
unigine-vally_1_0 5.44	5.44	unigine-vally_1_0 4.99	4.99	8.27%


Bad commit:
 commit 7423cc891b4d6fcc63bfeb79cc1d711ce81122bd
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Wed Oct 22 08:58:58 2014 -0700

    i965: Implement the PMA stall fix.

Parents of bad commit: 
commit 8ccf54ab098032da4652b314761c04f7724a7277
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Wed Oct 22 08:58:57 2014 -0700

    i965: Add #defines for Broadwell HiZ workarounds in CACHE_MODE_1.

    This patch adds macros needed for the HiZ PMA stall optimization.

Signed-off-by: Kenneth Graunke kenneth@whitecape.org


After bad commit:
commit 6107557f8fa34e0b7191813792be43eaa03aed19
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Wed Oct 22 08:58:59 2014 -0700

    i965: Re-enable Z16 on Gen8+.

    Improves performance in GLBenchmark 2.7 TRex by 3.88889% +/- 0.336383%
    (n=80) at 1280x720 on Broadwell GT3.  Together with the previous patch,
    it improves performance by 5.42738% +/- 0.541971% (n=10) at 1920x1080.

    Note that without the PMA stall fix, this would instead decrease
    performance by 22%.

    v2: Update comment (noticed by Kristian Høgsberg).

Signed-off-by: Kenneth Graunke kenneth@whitecape.org


Fixed other gfx sw driver to below list 2014-11-17 
Libdrm:		(master)libdrm-2.4.58-4-g00847fa48b83a85b0cb882594a12ed1511f780db
 Xserver:		(master)xorg-server-1.16.99.901-3-g63bb5c5ef16edf652179770294dcca4fc07dc992
 Xf86_video_intel:		(master)2.99.916-145-g6c2707d7bbc0ebb422be66618b6f78887c46446e
 Cairo:		(master)121f384c0e231c9c5d9c937b216d342bfc7810a6
 Libva:		(master)f9309a6f44b51bb2c463a6a16d3ccf3edc6e6c7a
 Libva_intel_driver:		(master)8e34fb34ed402811e512f9d41b14345f3795bac5
 Kernel:   (drm-intel-nightly)e49ebf9ed863e9522260ebd7bd0338ef5641c0e6

(In reply to Eero Tamminen from comment #1)
> Wendy, there have been mentions of PMA issue having much larger benefits (on
> HSW) in Valve games and Lightsmark.  Could you mail me information about the
> performance before and after the indicated Mesa commit in Lightsmark and all
> the games?

Eero, I never actually tested PMA on Lightsmark or Valve games...I wouldn't expect it to make a big impact.  It sounds like you're thinking of the shadow_c performance fix, which is different.

I'm okay with reverting the PMA stuff if it's a loss overall.  I'm kind of surprised though...