chapter1-AppendixA-m2-ziavras

Ziavras simdvector processing 2 over the years

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: FPU. 128-bit: IEEE 754 quadruple-precision Some material is adapted from D. Culler & D. Patterson (UCB) (binary128) S. Ziavras SIMD/Vector Processing • • Status reg.: contains # of elements per vector reg. Many elements are loaded simultaneously (low-order memory interleaving): interleaving): Bank 0 bits Example 0 16 Address in module k 1 2 Module in bank 4 3 1-bank example example 2^k words per module Some material is adapted from D. Culler & D. Patterson (UCB) 14 15 31 S. Ziavras SIMD/Vector Processing (2) • Over the years traditional vector computers added – Strided addressing (skip elements) A A+m Access only elements with address A+n*m address A+n*m A+2m Some material is adapted from D. Culler & D. Patterson (UCB) S. Ziavras SIMD/Vector Processing (3) – Gather/scatter addressing (indices of elements to access are stored in another vector register) Gather Indices of vector elements elements Scatter Vector Some material is adapted from D. Culler & D. Patterson (UCB) Vector Unit Ziavras S. SIMD/Vector Processing (4) • • Vector units in processors can be mainly...
View Full Document

Ask a homework question - tutors are online