Matrix multiplication on PC-s

Preliminary test 96/10/23

(see also: bench_bm_mm_ws.txt)

All this is unchecked (what is f77, does it call g77; where are the libraries, etc.) (f77 seems to use a conversion to C.)

CPU times can be compared to the workstation benchmarks WITHOUT precompiled BLAS libraries.

Table generated by using a dummy tempd_linux.f and using the command "time ".

TABLE 1. Program: p25 (simple do-loops, vector loop inside).
Machine Compiler Opt. level ND N CPU (sec) MFLOPS

Pentium 100 MHz f77 1 400 400 28 4.6
16 MB (Linux) 2 26 5
3 59?

g77 1 25 * 5
2 25 5
3 25 5

* The time overhead for generating the matrices is 0.2 sec. (Obtained by running the program with the matrix multiplication loops deleted. All other loops and calls were retained.)

Machine	Compiler	Opt. level	ND	N	CPU (sec)	MFLOPS
Pentium 100 MHz	f77	1	400	400	28	4.6
16 MB (Linux)		2			26	5
		3			59?
	g77	1			25 *	5
		2			25	5
		3			25	5