Imagination PowerVR SDK Blog

USSE2 32-bit floats on vector engine?


It’s known, that USSE1 can process 32-bit floats only on scalar engine. However there are some words, that USSE2 can process from 2 to 4 32-bit float at the same time i.e. on vector engine. Can someone clarify this?






Hi, sorry for the lateness of reply.

You are correct, USSE1 is indeed scales, and USSE2 has a vector floating point ALU as part of its general processing hardware.



Thx, Bob. Can you say more about SIMD? Does USSE2 has 128bit wide registers, that can process 4 32bit floats at a time? Or does it processes only 2 floats?


Hi StiX,

Sadly I can’t elaborate further.