For n = 50(10)100 I inverted an identity matrix, a unit upper triangular
matrix, and a unit lower triangular matrix using first HFP and then BFP.
The results, stated as index numbers with BFP=100, are summarized below.
50 60 70 80 90 100 n
101 101 100 103 101 99 I
101 100 103 101 100 100 UUT
103 100 103 102 100 100 ULT
BFP would appear to be very slightly faster, but certainly not in any
interesting way.
BFP is significantly more accurate, in the sense that its use yielded less
(indeed almost no) roundoff error for my examples.
Worth noting is that I did not do these calculation is C or C++, neither of
which is well adapted to intensive matrix algebra. I did them in Enterprise
PL/I 3.8 using double-precision real arithmetic.
In the event this difference is less important than might be guessed. The
current IBM implementations of these three very different languages share the
same code generator.
John Gilmore Ashland, MA 01721-1817 USA
_________________________________________________________________
The New Busy is not the old busy. Search, chat and e-mail from your inbox.
http://www.windowslive.com/campaign/thenewbusy?ocid=PID27925::T:WLMTAGL:ON:WL:en-US:WM_HMP:032010_3
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
