Vector Microprocessors

Krste Asanović

EECS Department
University of California, Berkeley
Technical Report No. UCB/CSD-98-1014
1998

Most previous research into vector architectures has concentrated on supercomputing applications and small enhancements to existing vector supercomputer implementations. This thesis expands the body of vector research by examining designs appropriate for single-chip full-custom vector microprocessor implementations targeting a much broader range of applications.

I present the design, implementation, and evaluation of T0 (Torrent-0): the first single-chip vector microprocessor. T0 is a compact but highly parallel processor that can sustain over 24 operations per cycle while issuing only a single 32-bit instruction per cycle. T0 demonstrates that vector architectures are well suited to full-custom VLSI implementation and that they perform well on many multimedia and human-machine interface tasks.

The remainder of the thesis contains proposals for future vector microprocessor designs. I show that the most area-efficient vector register file designs have several banks with several ports, rather than many banks with few ports as used by traditional vector supercomputers, or one bank with many ports as used by superscalar microprocessors. To extend the range of vector processing, I propose a vector flag processing model which enables speculative vectorization of "while" loops. To improve the performance of inexpensive vector memory systems, I introduce virtual processor caches, a new form of primary vector cache which can convert some forms of strided and indexed vector accesses into unit-stride bursts. (Spring 1998)

Advisor: John Wawrzynek


BibTeX citation:

@phdthesis{Asanović:CSD-98-1014,
    Author = {Asanović, Krste},
    Title = {Vector Microprocessors},
    School = {EECS Department, University of California, Berkeley},
    Year = {1998},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/1998/6404.html},
    Number = {UCB/CSD-98-1014},
    Abstract = {Most previous research into vector architectures has concentrated on supercomputing applications and small enhancements to existing vector supercomputer implementations. This thesis expands the body of vector research by examining designs appropriate for single-chip full-custom vector microprocessor implementations targeting a much broader range of applications.  <p>  I present the design, implementation, and evaluation of T0 (Torrent-0): the first single-chip vector microprocessor. T0 is a compact but highly parallel processor that can sustain over 24 operations per cycle while issuing only a single 32-bit instruction per cycle. T0 demonstrates that vector architectures are well suited to full-custom VLSI implementation and that they perform well on many multimedia and human-machine interface tasks.  <p>  The remainder of the thesis contains proposals for future vector microprocessor designs. I show that the most area-efficient vector register file designs have several banks with several ports, rather than many banks with few ports as used by traditional vector supercomputers, or one bank with many ports as used by superscalar microprocessors. To extend the range of vector processing, I propose a vector flag processing model which enables speculative vectorization of "while" loops. To improve the performance of inexpensive vector memory systems, I introduce virtual processor caches, a new form of primary vector cache which can convert some forms of strided and indexed vector accesses into unit-stride bursts. (Spring 1998)}
}

EndNote citation:

%0 Thesis
%A Asanović, Krste
%T Vector Microprocessors
%I EECS Department, University of California, Berkeley
%D 1998
%@ UCB/CSD-98-1014
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/1998/6404.html
%F Asanović:CSD-98-1014