Does a VLIW command count as 1 instruction?

[Luminescent]

I believe a VLIW command is a single instruction containing an array of sub-instructions packed into it (simd, scalar, etc.). I just wanted to verify if this was the case. This would mean that the floating-point op per clock performance of the pixel program processor in the NV30 is not only limited to 1 op, but an array of scalar/vector ops per clock (ala R300). This is very important in dictating the real value of the touted long and flexible shading routines inherent with CineFX.

 

[RussSchultz]

It depends on the system implementation.

 

[Luminescent]

Well a pixel processing block must be something like the Emotion engine albeit some branching/scheduling flexibility and only 1 VU (like VU0, controlled by some sort of command processor). So I was wondering, if the pixel processor pipeline was issued an alu command for the pixel program processor, which issued a VLIW to its internal units, would it be considered 1 single instruction?

 

[Basic]

Who's counting, and why?

 

[Dio]

It's a buzzword, and frankly, all that you need to attach it and be completely correct by the strict meaning of the acronym is that the Instruction Word is Very Long. And for pixel/vertex shaders, it is. You can count the bits yourself

 

[Luminescent]

Very funny guys. I know I make something out of nothing, and I did know the acronym meant very long instruction word. Anyways, thankyou for the replys.