2005-11-03 Hormozd Gahvari <hormozd@cs.berkeley.edu>
	timing.c, timing.h appear in both spmvbench and bebop_util
	directores. also, including files should be done with directories,
	e.g. #include <matrix_generator/create_rand.h>.

2004-11-30 Mark Hoemmen <mhoemmen@cs.berkeley.edu>

	If you specify an out-of-range command-line option, that value that you
	input is shown as the "default" value when the usage is printed.

2004-11-29 Mark Hoemmen <mhoemmen@cs.berkeley.edu>

 	Calling "./hpcc_interface_tester -f uga.out" causes an error when 
	parsing "block_sizes.csv", but specifying a value for the "-m" 
	command-line option prevents this error.

	Code that will be enabled when build with HPCC is not up-to-date.

	Memory expenditure is much greater than that specified by the HPCC
	parameter, because that parameter only accounts for the memory used by
	the sparse matrix and source and destination vectors.  It doesn't
	account for the set data structure used to construct the random sparse
	matrix, nor does it account for the various other data structures used
	throughout the program.

2004-11-23 Mark Hoemmen <mhoemmen@cs.berkeley.edu>

	Fixed bug in smvm_invoker where the mflop/s rate was inverted.
	Don't worry, it wasn't showing up in results -- the code that 
	we actually have been calling is correct.

2004-11-21 Mark Hoemmen <mhoemmen@cs.berkeley.edu>

	In raw CSV output of benchmark data, mflops field is often negative.  
	It could be that it overflows int.  We should use doubles instead.

	Note (2004-11-23): this requires changing the low-level benchmark as 
	well, so that doubles are used instead of ints.

	2004-11-23:  fixed (was only a problem with output).
