[Spice-devel] fail to test virtio-gpu with heaven benchmark4.0

[Marc-André Lureau]

Hi

----- Original Message -----
> 
> 
> On 2016年08月29日 17:03, Marc-André Lureau wrote:
> > hi
> >
> > ----- Original Message -----
> >> Hi all,
> >>
> >> I tried to use virtio-gpu to do some test, the glmark2 works, but when I
> >> ran heaven benchmark in Guest,
> >>
> >> the application could not be ran. I have collected the output message.
> >> pls see attachment file.
> >>
> >> I built my environment according to the requirement in
> >> https://copr.fedorainfracloud.org/coprs/kraxel/virgl/，
> > It should no longer be necessary, everything is in fedora 24
> >
> > See also:
> > http://blog.wikichoon.com/2016/05/spice-openglvirgl-acceleration-on.html
> Hi Marc, thanks, I saw that this use spice to display VM, it means
> that virtio-GPU already support remote 3D  accelerator? if true,
> it's great.
> 
> >

No, it's only local atm. There is some experimental work on-going for remoting with spice though.

> >> both guest/host kernel version: 4.6
> >>
> >> I used the latest qemu forked from master branch
> >> (e00da552a0dc82f4ec7896281eada7201e69f1db),
> >>
> >> and run qemu with:
> >>
> >> ~/data/linux-src/qemu$ ./x86_64-softmmu/qemu-system-x86_64 -vga virtio
> >> -display gtk,gl=on --enable-kvm -m 4000 -smp 4,sockets=1,cores=4 -hda
> >> /home/chenfan/imgs/fedora23.qcow2  -cpu host
> >>
> >> when running the heaven benchmark application the qemu has some output
> >> as following, I don't know how to fix that,
> >>
> >> do you have some hints? thanks in advance :)
> >>
> >>
> >> the qemu output messages:
> >> (qemu-system-x86_64:4628): Gdk-WARNING **:
> >> gdk_gl_context_set_required_version - GL context versions less than 3.2
> >> are not supported.
> >> gl_version 33 - core profile enabled
> >> vrend_compile_shader: context error reported 3 "Xorg" Illegal shader 0
> >> shader failed to compile
> >> 0:23(34): error: could not implicitly convert operands to arithmetic
> >> operator
> >> 0:23(34): error: operands to arithmetic operators must be numeric
> >> 0:23(17): error: no matching function for call to
> >> `uintBitsToFloat(error)'; candidates are:
> >> 0:23(17): error:    vec4 uintBitsToFloat(uvec4)
> >> 0:23(17): error:    float uintBitsToFloat(uint)
> >> 0:23(17): error:    vec2 uintBitsToFloat(uvec2)
> >> 0:23(17): error:    vec3 uintBitsToFloat(uvec3)
> >> 0:23(17): error:    vec4 uintBitsToFloat(uvec4)
> >> 0:23(12): error: cannot construct `vec4' from a non-numeric data type
> >> (snip)
> >> vrend_report_buffer_error: context error reported 3 "Xorg" Illegal
> >> command buffer 149816321
> > This is a virglrenderer bug, translating a shader. You could try with
> > virglrenderer from git, and if it still fails, report a bug to the virgl
> > mailing list.
> >
> > Regarding benchmarks, those aren't so great if you compare them with the
> > host today (especially "artificial" benchmarks like glmark2), I would say
> > it's between 30-50% of host, so there is room for improvements.
> >
> > thanks
> >
> yes, I used the latest virglrender from git. so
> 
> Cc:virglrenderer-devel at lists.freedesktop.org
> <mailto:virglrenderer-devel at lists.freedesktop.org>

thanks