ndeng@EULER.BERKELEY.EDU (03/03/90)
I have a question about which language is the "best" for scientific computation (read: number crunching), especially for execution speed. A majority of people use FORTRAN, not because this language is good or flexible, but due to historical reasons. Some others use C or Pascal, which offer better system interaction and graphics support. There is also a group of people who still use BASIC or ALGOL. I wonder if anybody has used two or more different languages to solve an identical problem and compared the results for speed/accuracy differences. Do the C/Pascal/BASIC version of LINPACK or Livermore Bunchmarks exist? How about the speed of these bunchmarks when compared with FORTRAN version? I think this would be a topic of general interest and would like to see discussions. I apologize if this topic has been discussed before. (In that case, would somebody be kind enough to give me some hints for whereabout of the discussion archives?) And also, please no flames, I am not trying to compare apples and oranges. If I get enough responses by e-mail, I'll post a summary. N. Deng ndeng@euler.berkeley.edu ----------------------------- Standard Disclaimer Applies.
yahoo@unix.cis.pitt.edu (Kenneth L Moore) (03/05/90)
In article <9003021955.AA28774@euler.Berkeley.EDU> ndeng@EULER.BERKELEY.EDU writes:
=>I have a question about which language is the "best" for scientific
=>computation (read: number crunching), especially for execution speed.
=>N. Deng
=>ndeng@euler.berkeley.edu
=>Standard Disclaimer Applies.
Here is a posting that I just made to sci.math.num-analysis. It
was a summary of mail I got in response to posting the question,
"What language do YOU use for numerical analysis?"
In the case of FORTRAN and C (the major languages mentioned), I tried to
include a summary of what people thought were the advantages of their
choice.
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
1 APL2
1 Assembler
1 ALGOL
1 BASIC
1 Common Lisp
1 Common Lisp Object System
13 FORTRAN
15 C
4 C++
1 Bliss
1 Forth
4 Pascal
1 Modula-2
1 S
FORTRAN - complex numbers, arrays at 1, existing debugged code such as
Eispak & Linpac, bounds checking, supercomputing, portability,
speed of execution, does not promote float to double, compact code,
double double precision, the way arrays are handled, intrinsic
functions such as 2**3.2 MAX and Min, each variable does not
have to be declared
C - systems interface, algorithm translation (structures),
portability, allows for development of a user friendly front end,
compact code, speed of development, speed of execution,
readability, dynamic memory allocation, flexible.
--
I don't yell and I don't tell and I'm grateful as hell: Benny Hillbright@Data-IO.COM (Walter Bright) (03/06/90)
In article <9003021955.AA28774@euler.Berkeley.EDU> ndeng@EULER.BERKELEY.EDU writes:
<I have a question about which language is the "best" for scientific
<computation (read: number crunching), especially for execution speed.
I hate to say it, but Fortran is still the best for the following reasons:
1. It's been around the longest, so the floating point routines are
very mature and reliable. Lots of people rely on this stability, so
if an implementation is weak in this regard, it's quickly corrected.
C is not as well used for number crunching, and so the libraries are
less well developed and less used.
2. Fortran has built-in support for complex data. Yeah, I know, you can
fake it in C++. But nothing matches the convenience of being able to
say:
z = a + b*i
Hardwiring a type into the compiler produces better results than
a user-defined type.
3. Fortran has variably-dimensioned arrays. Doing this in C is a bit
klunky.
4. Fortran compilers tend to optimize better than C compilers, because
a) they've been around longer and b) have less aliasing problems.
There's a group, the NCEG, trying to address these deficiencies in C.
(BTW, Fortran is a miserable language for doing about anything else!)