Q. intel_gpu_top report interpretation

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hello.

I'm a user of Intel graphics driver and looking for information about
the output of "intel_gpu_top" command.
I hope I can ask this kind of question on this ML.
If I shouldn't, I deeply apologize you for my posting.

Before posting this message, I tried to read the description
about this ML.
But the site "http://www.intellinuxgraphics.org"; linked from
http://lists.freedesktop.org/mailman/listinfo/intel-gfx has been 
unreachable with "Host is down".
I also read that some documentations about Intel GPU architecture
are available on the site, but it is down.
I would appreciate it very much if anybody could direct me
to the right place.

My questions are the followings.

Q1.
What are the meanings of the task names listed in output from
intel_gpu_top like GAM, GAFS, SF, VS, URBM, etc. ?
Is there any more detailed documentation about the tool than
"man intel_gpu_top"?

The followings are similar questions posted on this ML,
which brought me here with searching in mesa-3d ML archive.
I read them through, but I was not able to find the answer.
intel_gpu_top decode..
http://lists.freedesktop.org/archives/intel-gfx/2010-October/008312.html
intel_gpu_top profiles
http://lists.freedesktop.org/archives/intel-gfx/2009-October/004707.html

Q2.
With the attached snapshot, is the GPU quite busy?
Is it hitting the ceiling of its potential performance?

Task GAM is 98% busy, but the number "render busy" shows only 31%.

http://dri.freedesktop.org/wiki/IntelPerformanceTuning
tells that "ring idle" is the most important number,
but I don't see it in the report.

Q3.
If the GPU is quite busy, what operation makes GAM so busy?
My application does 2D image composition using a lot of texture images.

Q4.
When I run "intel_gpu_top" on Core i5 3550S Ivy Bridge
for a few minutes with the application,
the whole system hangs up (freezes).

"intel_gpu_top" also prints strange negative values occasionally for
"render space", "bitstream space" and "blitter space".

When I run "intel_gpu_top" alone (0% busy), I don't have those problems.

When I run "intel_gpu_top" on Core i7 Sandy Bridge
with the same application, "intel_gpu_top" reports lower "busy values"
(as expected with i7) and I don't have those problems.

Are the problems (freezing & strange negative values) triggered
by the busy GPU?

A snapshot of intel_gpu_top output.
(unnecessary white spaces are removed for better visibility)
----------------------------------------
     render busy:  31%: ???????        render space: 1437518475/4096
  bitstream busy:   0%:             bitstream space: 0/4096
    blitter busy:  11%: ???           blitter space: 117/4096

            task  percent busy
             GAM:  98%: ????????????????????    vert fetch: 0 (0/sthe ec)
            GAFS:  73%: ???????????????         prim fetch: 0 (0/sec)
              SF:  63%: ?????????????        VS invocations: 0 (0/sec)
              VS:  63%: ?????????????        GS invocations: 0 (0/sec)
            URBM:  63%: ?????????????             GS prims: 0 (0/sec)
            GAFM:  62%: ?????????????        CL invocations: 0 (0/sec)
              VF:  62%: ?????????????             CL prims: 0 (0/sec)
              CL:  62%: ?????????????        PS invocations: 0 (0/sec)
             SVG:  62%: ?????????????        PS depth pass: 0 (0/sec)
              DS:  62%: ?????????????
              HS:  62%: ?????????????
             TDG:  62%: ?????????????
              GS:  62%: ?????????????
             SOL:  62%: ?????????????
              TE:  62%: ?????????????
             TSG:  62%: ?????????????
             VFE:  62%: ?????????????

My environment
------------------------
CPU & chipset:  i5 3550S Ivy Bridge 3.00GHz (GPU HD2500)
OS : Cent OS 6 Linux kernel 2.6.32-279 SMP x86_64
intel-gpu-tools  : yum version 2.20.16 Release: 1.el6 Arch: x86_64
mesa-dri-drivers : yum version 8.0.3   Release: 3.el6 Arch: x86_64


I'll appreciate it very much if anybody could help me with any of
my questions.

Thank you.

----------------
Masaki NISHIKAWA, Sony Corp.


[Index of Archives]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]
  Powered by Linux