petermcneeleychromium
petermcneeleychromium
Atomic barrier sum check. Some compliers do not seem to respect the synchronization requirements of the barrier for workgroups sizes on the order of a subgroup. crbug.com/42241359
After a bit of investigation it has been determined that tanh has a max Absolute error of 1e-5 for some devices (nvidia) We discussed polyfilling this function (sinh/cosh) but this...
Some intrinsic implementations of tanh have low precision. This is found across nvidia directx (4070). It may be across all nvidia. This can easily be seen by running the CTS...
### Review of handling of nans for Clamp function. For min,max in wgsl the behavior for nan is called out explicitly This behavior comes from the extended instructions of NMin,...
64 bit atomic operations are impossible (?) to emulate but are desired by some applications like Nanite. There is also some suggestion that algorithms like Scan need key-value 32 bit...
GLSL has 'mediump' precision qualifier to get something equivalent to f16. It is natural to use these for rendering code (gfx pipeline) that does not require full 32 bits However...
Description of wgsl sample_mask 'Sample coverage mask for the current fragment. It contains a bitmask indicating which samples in this fragment are covered by the primitive being rendered.' The interpretation...