home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!haven.umd.edu!purdue!mentor.cc.purdue.edu!pop.stat.purdue.edu!hrubin
- From: hrubin@pop.stat.purdue.edu (Herman Rubin)
- Subject: Re: FP-number cache? Unclocked VLSI design.
- Message-ID: <C0Dsuu.7wE@mentor.cc.purdue.edu>
- Summary: This will only make it worse
- Sender: news@mentor.cc.purdue.edu (USENET News)
- Organization: Purdue University Statistics Department
- References: <1993Jan5.085415.19676@klaava.Helsinki.FI>
- Date: Tue, 5 Jan 1993 12:47:17 GMT
- Lines: 37
-
- In article <1993Jan5.085415.19676@klaava.Helsinki.FI> veijalai@klaava.Helsinki.FI (Tony Veijalainen) writes:
-
- >This is my third posting in my cracy thoughts series :-)
-
- >There is common trend towards harwardizing the memory lane by separating
- >instructions and data to separate caches with own bus to other parts of
- >CPU.
-
- >On the other hand FPU-units that appear more and more are in modern big
- >CPU:s are quite far conceptually from other operations. I have
- >suspision that FP-arithmetic tends to cluster quite heavily, and because
- >of traditional efficiency thinking and fixed number arithmetic in
- >business applications some big parts of programs are integer only (not
- >much FP-operations in interupt code for example :-).
-
- >So have somebody researched the havoc FPU-instructions make to general
- >data cache? Is there possible advantage of having FP-number (with
- >separate bus to FPU-register file) and fixnumber caches with advantages
- >outdoing the cost on CPU (like diminiching the general cache size, is
- >this over specialization?).
-
- Those of us who use more sophisticated numerical procedures consider this
- to be essentially the opposite direction to go. It is not the case that
- integers are only used for addresses and indices; they are vitally needed
- for what is commonly called "number-crunching." Fixed-point arithmetic
- has great advantages in many situations, and often is so slow in some of
- the "modern" machines, especially multiplication, that it pays to go to
- the massive headaches of making the floating arithmetic emulate fixed-point
- arithmetic. This is especially bad with the formats in which floats are
- automatically normalized and with the most significant bit missing, and
- where there are even separate registers for integers and floats.
-
- --
- Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
- Phone: (317)494-6054
- hrubin@snap.stat.purdue.edu (Internet, bitnet)
- {purdue,pur-ee}!snap.stat!hrubin(UUCP)
-