The compiled xlib-shim.so can be used to intercept and prevent SDL
from calling XSync, which causes latency issues on some video
hardware.
To use specify the full path to the file in the LD_PRELOAD
environment variable, like so:
LD_PRELOAD=/full/path/xlib-shim.so ./looking-glass
This fixes a race condition which causes the mouse ringbuffer to
overflow. It also corrects out of order message index IDs due to
multiple threads sending messages asyncronously.
This partially resolves lag issues on hosts running the amdgpu driver.
If mouse caputure is enable the lag issue returns, this is because SDL
calls `XSync` in `X11_WarpMouse` and `X11_WarpMouseGlobal`, if these
calls are removed all input lag issues dissapear.
This issue has been reported to SDL as the calls to `XSync` are not
required per the xlib documentation.
the client now will look for a configuration file in the following
locations by default.
* /etc/looking-glass.conf
* ~/.looking-glass.conf
All configuration files are loaded and may override values specified by
any prior configuration files loaded.
Sample Config:
global:
{
fullScreen=false;
showFPS=true;
x=0;
y=0;
w=800;
h=600;
}
OpenGL:
{
mipmap="false";
}
Since we do not use IRQs anymore we can use the ivshmem-plain device
which doesn't need the ivshmem-server. The QEMU arguments now should be
as follows:
-device ivshmem-plain,memdev=ivshmem
-object memory-backend-file,id=ivshmem,share=on,mem-path=/dev/shm/looking-glass,size=32M
Obviously adjusting the memory size as required. It is suggested that
the shared memory file be created before the guest is started with the
appropriate permissions, for example:
touch /dev/shm/looking-glass
chown user:kvm /dev/shm/looking-glass
chmod 660 /dev/shm/looking-glass
The prior patch to correct the mouse loop resolves the CPU load issue
with the atomic locking method. SDL mutexes are still available if
desired but full mutex locking is far slower then fast spinlocks
This is the first of two commits that completely turn the rendering code
on it's head. This change set decouples the guest's capture rate from
the host's render rate for both cursor and frame updates. This helps
prevent the host application from stalling when waiting for frame draws
when all it want's to do is send cursor updates.
* Breaks OpenGL-Basic for now
This feature was previously hardcoded enabled, it is now optional as
some hardware pipelines stall with the excessive flushes resulting in a
jumpy cursor. Default is disabled and may be re-enabled with
`-o opengl:splitmouse=1` or `-o opengl-basic:splitmouse=1` depending on
the renderer selected.
This is based on bf8da7fd7e4397fe9931e82e52753806f1eee7e0 by @kiljacken
Per issue #11 on some cards this method performs much better then the
persistant buffering used in the standard OpenGL renderer.
Closes#11
