Which is more expensive ? Post: 302131762

Sponsored Content

Top Forums Programming Which is more expensive ? Post 302131762 by Perderabo on Monday 13th of August 2007 03:39:02 PM

08-13-2007

Administrator Emeritus

I would code your first snippet but compile with an optimizer. These days optimizers will unroll loops if unrolling is advantageous. That particular loop is not a real great candidate for unrolling anyway. A better candidate would be:

for(i=0; i<100; i++) A[i]=0;

Most superscalar cpus can execute:
A[i]=0;
A[i+1]=0;
A[i+2]=0;
simultaneously. How deep it can go depends on the cpu and that's why leaving unrolling to an optimizer is a good idea. The optimizer should know the target cpu. But your case involved a system call which is different. You're only saving some loop overhead.

Apparently, if you explicitly unroll a loop when it is not advantageous, most optimizers will not reroll the loop. At least this was the case circa 1998 when my copy of "High Performance Computing" was published. If you have that book, see chapter 8, "Loop Optimizations" and chapter 9, "Understanding Parallelism". This is still a great book and it's not just for Fortran programmers.

Anyway, if you are not in control of which fd's might be open, you need to to loop up to OPEN_MAX closing them. High fd's might have been opened and then setrlimit() called lower to the max fd.

Perderabo

View Public Profile for Perderabo

Find all posts by Perderabo

2 More Discussions You Might Find Interesting

1. Programming

calling pthread_self (on ubuntu), expensive?

Hi all, Is anyone aware of what operations are involved when a call to pthread_self() is made, obtaining the unique thread ID on a Ubuntu system (or even any Linux flavour)? Specifically, to retrieve the thread id, is there any locking required or atomic operations? I'm building an...

2. What is on Your Mind?

Very Expensive Running Shoes

You really should not need one third of the entire US budget to buy a pair of running shoes... even if they are name brand. What have these guys been smoking? It reminds me of the old joke... Customer: At those prices you aren't going to sell many shoes. Salesman: Ah, but all we need to do is...

LEARN ABOUT CENTOS

dlarfx

dlarfx.f(3)							      LAPACK							       dlarfx.f(3)

NAME

       dlarfx.f -

SYNOPSIS

   Functions/Subroutines
       subroutine dlarfx (SIDE, M, N, V, TAU, C, LDC, WORK)
	   DLARFX applies an elementary reflector to a general rectangular matrix, with loop unrolling when the reflector has order <= 10.

Function/Subroutine Documentation
   subroutine dlarfx (characterSIDE, integerM, integerN, double precision, dimension( * )V, double precisionTAU, double precision, dimension( ldc,
       * )C, integerLDC, double precision, dimension( * )WORK)
       DLARFX applies an elementary reflector to a general rectangular matrix, with loop unrolling when the reflector has order <= 10.

       Purpose:

	    DLARFX applies a real elementary reflector H to a real m by n
	    matrix C, from either the left or the right. H is represented in the
	    form

		  H = I - tau * v * v**T

	    where tau is a real scalar and v is a real vector.

	    If tau = 0, then H is taken to be the unit matrix

	    This version uses inline code if H has order < 11.

       Parameters:
	   SIDE

		     SIDE is CHARACTER*1
		     = 'L': form  H * C
		     = 'R': form  C * H

	   M

		     M is INTEGER
		     The number of rows of the matrix C.

	   N

		     N is INTEGER
		     The number of columns of the matrix C.

	   V

		     V is DOUBLE PRECISION array, dimension (M) if SIDE = 'L'
						or (N) if SIDE = 'R'
		     The vector v in the representation of H.

	   TAU

		     TAU is DOUBLE PRECISION
		     The value tau in the representation of H.

	   C

		     C is DOUBLE PRECISION array, dimension (LDC,N)
		     On entry, the m by n matrix C.
		     On exit, C is overwritten by the matrix H * C if SIDE = 'L',
		     or C * H if SIDE = 'R'.

	   LDC

		     LDC is INTEGER
		     The leading dimension of the array C. LDA >= (1,M).

	   WORK

		     WORK is DOUBLE PRECISION array, dimension
				 (N) if SIDE = 'L'
				 or (M) if SIDE = 'R'
		     WORK is not referenced if H has order < 11.

       Author:
	   Univ. of Tennessee

	   Univ. of California Berkeley

	   Univ. of Colorado Denver

	   NAG Ltd.

       Date:
	   September 2012

       Definition at line 121 of file dlarfx.f.

Author
       Generated automatically by Doxygen for LAPACK from the source code.

Version 3.4.2							  Tue Sep 25 2012						       dlarfx.f(3)