-------------------------------------------------------------------------
   Cache Performance of some programs from  proposed SPEC95 Benchmark Suite

	      * Nikhil Sharma
	      * Amit Shukla
	      * Avinash Sodani

--------------------------------------------------------------------------
This file contains the following information.

	o The tables and the corresponding benchmark programs.

	o Description of the various graphs available.
	
	o Instrumenting  the  benchmark     program  executables   for
	  generating traces.

	o Cheetah: the cache simulator used.

	o Versions and Parameters

	o References

======================================================================

1. Files and their contents
   ~~~~~ ~~~ ~~~~~ ~~~~~~~~

	-----------------------------------
	Table #		Benchmark program
	-----------------------------------
	   1		tomcatV	
	   2		su2cor	
	   3		hydro2d	
	   4		mgrid
	   5		swim
	   6		nump
	   7		ora
	   8		appsp
	-----------------------------------



The  files  table[1-8].ascii   and   table[1-8].ps contain  ascii   and
postscript versions respectively of  the miss ratios for the benchmark
programs

----------------------------------------------------------------------

2. The graphs
   ~~~ ~~~~~~

	The graphs show  the  variation of  the  miss ratios  with the
number of references simulated.

	-----------------------------------
	Graph #		Benchmark program
	-----------------------------------
	   1		tomcatV	
	   2		swim
	   3		nump
	   4		ora
	   5		appsp
	-----------------------------------


----------------------------------------------------------------------

3. Instrumenting the executables
   ~~~~~~~~~~~~~ ~~~ ~~~~~~~~~~~

	We used the Quick Program Profiler  and Tracer (QPT) [Larus92]
to  generate  address traces  for all the  programs  for which we have
obtained results.

	QPT is used  to produce a highly compressed  trace  file and a
number of trace regeneration programs.  These programs read the highly
compressed trace file and produce a full program trace.

We ran QPT with original executable file as shown below:


         qpt -t -M -o +19 foo
         gcc -o din din.c foo_sma*.c
         pixtrace foo.qpt <foo.in | din - | cheetah <options>

----------------------------------------------------------------------

4. Cheetah
   ~~~~~~~

	Cheetah is capable  of simulating multiple cache configurations
with a fixed line size and  a varying cache  size. Further details can
be found in [Sugumar91].

----------------------------------------------------------------------

5. Simulation parameters
   ~~~~~~~~~~ ~~~~~~~~~~

	o Compiled on SPARC10s, running SunOS 4.1.3

	o Compilers : Sun Fortran Compiler V1.4
		      Gcc 2.6.0

        o Optimization Level : 2

	o Replacement policy: LRU

	o Number of references used: 1 billion.


----------------------------------------------------------------------

6. References
   ~~~~~~~~~~

[Ball92]    Thomas  Ball  and  James  Larus,  "Optimally Profiling and
	    Tracing  Programs",   Proc.   of   the  19th    Annual ACM
	    SIGPLAN-SIGACT  Symposium  on  Principles  of  Programming 
	    Languages '92


[Sugumar91] Rabin A.  Sugumar  and  Santhosh G.  Abraham,   "Efficient
	    Simulation of Caches using Binomial Trees", CSE TR-111-91,
	    University of Michigan, 1991.

----------------------------------------------------------------------