8/16 element vectors unsupported about clspv HOT 1 CLOSED

Clspv targets Vulkan. I don't know of any Vulkan implementations that allow vectors larger than 4 elements. So this is a deliberate restriction.

Before someone makes a comment on that, I did notice it was a documented restriction. Do people find issues useful to track lifting these restrictions?

:-)

I think it's useful to track if the output of Clspv could run on some Vulkan implementation, even allowing for vendor extensions. Otherwise I think it's best to mark as Invalid. I just made up this rule. What do you think?

Even with such a rule, I think we should consider changes that enable development of experimental features in a Vulkan implementation. For example, someone may want to extend Mesa's Vulkan implementation and finds something like OpenCL C to be the best shader programming model. I think it's reasonable to allow that, perhaps under a command line flag explicitly opting into the experimental behaviour.

from clspv.