Changeset 243


Ignore:
Timestamp:
Dec 19, 2008, 7:49:09 AM (10 years ago)
Author:
cameron
Message:

P2S under RefA/RefB/IDISA-A/B

Location:
docs/ASPLOS09
Files:
2 edited

Legend:

Unmodified
Added
Removed
  • docs/ASPLOS09/asplos094-cameron.tex

    r242 r243  
    897897of versions of the \verb#simd<8>::mergeh# and \verb#simd<8>::mergel#
    898898operations that are available with each of the SSE and Altivec instruction
    899 sets.  These algorithms take 72 operations to perform the
    900 inverse transposition of 8 parallel registers of bit stream
    901 data into 8 serial registers of byte stream data. 
     899sets.  To perform the full inverse transform of 8 parallel
     900registers of bit stream data into 8 serial registers of byte stream data,
     901a RefA implementation requires 120 operations, while a RefB
     902implementation reduces this to 72.
    902903
    903904\begin{figure}[tbh]
     
    934935\end{figure}
    935936
    936 An algorithm employing only 24 operations using the
    937 inductive doubling instruction set architecture is relatively
     937An algorithm employing only 24 operations using IDISA-A/B is relatively
    938938straightforward.. In stage 1, parallel registers for individual bit streams
    939939are first merged with bit-level interleaving
     
    963963parallel bit stream form can then each be used at will in
    964964character stream applications.
    965 
    966 
    967965
    968966
Note: See TracChangeset for help on using the changeset viewer.