<div dir="auto"><div>We need to extend the CS ioctl to allow submitting 2 command buffers at the same time.</div><div dir="auto"><br></div><div dir="auto">Marek<br><br><div class="gmail_quote" dir="auto"><div dir="ltr" class="gmail_attr">On Mon, Feb 25, 2019, 10:06 PM Dieter Nützel <<a href="mailto:Dieter@nuetzel-hh.de">Dieter@nuetzel-hh.de</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hello Marek,<br>
<br>
you wrote with your series sent:<br>
<br>
[-]<br>
Trivial benchmarks such as glxgears can expect 20% decrease<br>
in performance due to the added cost of the SDMA CS ioctl that wasn't<br>
there before.<br>
[-]<br>
<br>
Any ideas to speed this up, again?<br>
glmark2 went from 9766 (best ever) down to 7455 (all with NIR).<br>
Or are micro benchmarks not worth more effort?<br>
<br>
Dieter<br>
<br>
SDMA<br>
=======================================================<br>
     glmark2 2017.07<br>
=======================================================<br>
     OpenGL Information<br>
     GL_VENDOR:     X.Org<br>
     GL_RENDERER:   Radeon RX 580 Series (POLARIS10, DRM 3.30.0, <br>
5.0.0-rc1-1.g7262353-default+, LLVM 9.0.0)<br>
     GL_VERSION:    4.5 (Compatibility Profile) Mesa 19.1.0-devel <br>
(git-a9b32aaa16)<br>
=======================================================<br>
[build] use-vbo=false: FPS: 3694 FrameTime: 0.271 ms<br>
[build] use-vbo=true: FPS: 9341 FrameTime: 0.107 ms<br>
[texture] texture-filter=nearest: FPS: 9140 FrameTime: 0.109 ms<br>
[texture] texture-filter=linear: FPS: 9163 FrameTime: 0.109 ms<br>
[texture] texture-filter=mipmap: FPS: 9161 FrameTime: 0.109 ms<br>
[shading] shading=gouraud: FPS: 9234 FrameTime: 0.108 ms<br>
[shading] shading=blinn-phong-inf: FPS: 9255 FrameTime: 0.108 ms<br>
[shading] shading=phong: FPS: 9226 FrameTime: 0.108 ms<br>
[shading] shading=cel: FPS: 9310 FrameTime: 0.107 ms<br>
[bump] bump-render=high-poly: FPS: 9298 FrameTime: 0.108 ms<br>
[bump] bump-render=normals: FPS: 9121 FrameTime: 0.110 ms<br>
[bump] bump-render=height: FPS: 9120 FrameTime: 0.110 ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 9858 FrameTime: 0.101 ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 9854 FrameTime: <br>
0.101 ms<br>
[pulsar] light=false:quads=5:texture=false: FPS: 8468 FrameTime: 0.118 <br>
ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[desktop] blur-radius=5:effect=blur:passes=1:separable=true:windows=4: <br>
FPS: 5181 FrameTime: 0.193 ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[desktop] effect=shadow:windows=4: FPS: 5374 FrameTime: 0.186 ms<br>
[buffer] <br>
columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: <br>
FPS: 824 FrameTime: 1.214 ms<br>
[buffer] <br>
columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: <br>
FPS: 1114 FrameTime: 0.898 ms<br>
[buffer] <br>
columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: <br>
FPS: 899 FrameTime: 1.112 ms<br>
[ideas] speed=duration: FPS: 3485 FrameTime: 0.287 ms<br>
[jellyfish] <default>: FPS: 7992 FrameTime: 0.125 ms<br>
[terrain] <default>: FPS: 1796 FrameTime: 0.557 ms<br>
[shadow] <default>: FPS: 7350 FrameTime: 0.136 ms<br>
[refract] <default>: FPS: 3595 FrameTime: 0.278 ms<br>
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 9401 FrameTime: <br>
0.106 ms<br>
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 9413 FrameTime: <br>
0.106 ms<br>
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 9417 FrameTime: <br>
0.106 ms<br>
[function] fragment-complexity=low:fragment-steps=5: FPS: 9365 <br>
FrameTime: 0.107 ms<br>
[function] fragment-complexity=medium:fragment-steps=5: FPS: 9451 <br>
FrameTime: 0.106 ms<br>
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 9300 <br>
FrameTime: 0.108 ms<br>
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 9440 <br>
FrameTime: 0.106 ms<br>
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 9392 <br>
FrameTime: 0.106 ms<br>
=======================================================<br>
                                   glmark2 Score: 7455<br>
=======================================================<br>
<br>
<br>
Before<br>
=======================================================<br>
     glmark2 2017.07<br>
=======================================================<br>
     OpenGL Information<br>
     GL_VENDOR:     X.Org<br>
     GL_RENDERER:   Radeon RX 580 Series (POLARIS10, DRM 3.27.0, <br>
4.20.0-rc3-1.g7262353-default+, LLVM 8.0.0)<br>
     GL_VERSION:    4.5 (Compatibility Profile) Mesa 19.0.0-devel <br>
(git-c49b3df3cb)<br>
=======================================================<br>
[build] use-vbo=false: FPS: 3373 FrameTime: 0.296 ms<br>
[build] use-vbo=true: FPS: 13121 FrameTime: 0.076 ms<br>
[texture] texture-filter=nearest: FPS: 12172 FrameTime: 0.082 ms<br>
[texture] texture-filter=linear: FPS: 12557 FrameTime: 0.080 ms<br>
[texture] texture-filter=mipmap: FPS: 12228 FrameTime: 0.082 ms<br>
[shading] shading=gouraud: FPS: 12536 FrameTime: 0.080 ms<br>
[shading] shading=blinn-phong-inf: FPS: 12782 FrameTime: 0.078 ms<br>
[shading] shading=phong: FPS: 12619 FrameTime: 0.079 ms<br>
[shading] shading=cel: FPS: 12735 FrameTime: 0.079 ms<br>
[bump] bump-render=high-poly: FPS: 11412 FrameTime: 0.088 ms<br>
[bump] bump-render=normals: FPS: 12467 FrameTime: 0.080 ms<br>
[bump] bump-render=height: FPS: 12422 FrameTime: 0.081 ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 13252 FrameTime: 0.075 ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 11468 FrameTime: <br>
0.087 ms<br>
[pulsar] light=false:quads=5:texture=false: FPS: 11385 FrameTime: 0.088 <br>
ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[desktop] blur-radius=5:effect=blur:passes=1:separable=true:windows=4: <br>
FPS: 5609 FrameTime: 0.178 ms<br>
libpng warning: iCCP: known incorrect sRGB profile<br>
[desktop] effect=shadow:windows=4: FPS: 5944 FrameTime: 0.168 ms<br>
[buffer] <br>
columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: <br>
FPS: 837 FrameTime: 1.195 ms<br>
[buffer] <br>
columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: <br>
FPS: 1153 FrameTime: 0.867 ms<br>
[buffer] <br>
columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: <br>
FPS: 925 FrameTime: 1.081 ms<br>
[ideas] speed=duration: FPS: 3042 FrameTime: 0.329 ms<br>
[jellyfish] <default>: FPS: 10209 FrameTime: 0.098 ms<br>
[terrain] <default>: FPS: 1791 FrameTime: 0.558 ms<br>
[shadow] <default>: FPS: 9248 FrameTime: 0.108 ms<br>
[refract] <default>: FPS: 3502 FrameTime: 0.286 ms<br>
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 13061 FrameTime: <br>
0.077 ms<br>
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 13225 FrameTime: <br>
0.076 ms<br>
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 12882 FrameTime: <br>
0.078 ms<br>
[function] fragment-complexity=low:fragment-steps=5: FPS: 12806 <br>
FrameTime: 0.078 ms<br>
[function] fragment-complexity=medium:fragment-steps=5: FPS: 12936 <br>
FrameTime: 0.077 ms<br>
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 12954 <br>
FrameTime: 0.077 ms<br>
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: <br>
12788 FrameTime: 0.078 ms<br>
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 12846 <br>
FrameTime: 0.078 ms<br>
=======================================================<br>
                                   glmark2 Score: 9766<br>
=======================================================<br>
</blockquote></div></div></div>