I am not by any kind a GPU driver programmer,...yet, but I think that DX front-end(HLSL,runtime, etc.) is written by m$ and is closed source and in the end it spits something which can easily converted to TGSI, if not the same. Maybe someone from VMware can explain better.
Originally Posted by smitty3268
I recall reading something to the effect that with this conversion, GL ASM could be optimized better (though still not as well as GLSL). Do I remember right?
Here's a picture showing interactions between shader IRs and drivers in Mesa: