Without a shadow, these emojis were nearly unreadable on the fixed chat
background. These hand-drawn replacements are detailed enough to be
clearly understood against all kinds of backgrounds.
This allows using opencl as an alternative to opengl compute shaders on
macos, which does not support compute shaders. Now, macos can finally
use the extended draw distance feature of the gpu plugin.
This also includes code for using opencl with Windows and Linux if we
want to enable that in the future. A copy of the existing compute
shaders have been checked in and ported to opencl, keeping support for
opengl compute shaders on Windows and Linux.
Co-authored-by: Paul Norton <napkinorton@gmail.com>
Add 4 dose Extended Antifires and Anti-venoms to the herblore skill calculator. These potions are able and often are made using 4 doses at a time.
Super Antifires and Anti-venom+'s can not be made using 3 doses at a time, so were fixed to display the right potion dose (4). The xp for these potions was already correct.
This adjusts the draw callbacks api to take in the color of the full
screen overlay the client would have rendered. This is primarily used in
godwars, underwater, darkmeyer, etc. Having them rasterized on the cpu
is slow, especially with gpu on since we additionally have to compute
the alpha per-pixel.
While nothing passed to comp_unordered uses orientation, the uniform
block is not bound in this shader, and rotate() accesses sinCosTable. So
I'm not sure why this has ever worked.
Having all 3 sampling branches feed into one colorblind() call was
breaking some versions of Intel's drivers whenever the xbr sampling code
was merely present. Changing each branch to individually call
colorblind() seems to fix it.
Prior to 8c00f6da88 the interpolation of fragment
shader inputs appears to not have been accounting for perspective. This broke
texturing due to the texture ids being interpolated for perspective which was
fixed in 47e0ac8032. The hsl values similarly
require not accounting for perspective, so the noperspective interpolation
qualifier has been added. It is unclear to me why the geometry shader removal
would affect the interpolation like this since it was emitting vertices with
the correct z values.
The face sorting still requires the toScreen function to match exactly
the same order that would be computed on CPU, but for the projection for
rendering we can simplify it by replacing it with a projection matrix.
This also drops the requirement of having a geometry shader to remove
faces too close to or behind the camera. Upon closer inspection it looks
like the client can draw faces which have at least one vertex which is
>= 50 from the camera, so this now more accurately reflects the software
renderer.
Co-authored-by: Adam <Adam@sigterm.info>
