How to Use the Cache Replacement Competition Kit

Setting Up the Simulation Infrastructure:

1. Subscribe to the replacement competition mailing list by sending an empty email to crc-1-subscribe AT yahoogroups.com. You do not need to have a Yahoo email account - you only need to subscribe to the mailing list. When prompted, type in a dummy message to the moderator.

This mailing list will be used for any announcments regarding the competition, any simulation infrastructure updates, and participants may post questions or report problems.

2. Download the replacement competition kit file from here and save it in your home directory.

3. unpack the kit:

 	tar -xzvf CRC_DISTRIB.tgz
  	cd CRC 

4. Set the variable 'PIN_DIR' in pinkit/CONFIG/makefile.gnu.config to the absolute path for pinkit/pin-2.7-31933-gcc.3.4.6-ia32_intel64-linux/ in the kit. When you edit the makefile.gnu.config file, modify the fourth line to:

	PIN_DIR ?= $FULL_PATH_TO_CRC_KIT/CRC/pinkit/pin-2.7-31933-gcc.3.4.6-ia32_intel64-linux

5.Edit Makefile.competition and modify the fifth line `default: CMPsim32' depending whether you are on a 32-bit (IA-32, i686) or a 64-bit (x86_64) system. To get that configuration, type "uname -a" in Linux.

• For 32-bit machines, modify to:

  	default: CMPsim32

• For 64-bit machines, modify to:

  	default: CMPsim64

6. To compile the simulator, make sure you are in the CRC/ directory and that your gcc compiler path is set up correctly, then type:

	make

	make CMPsim32

and for 64-bit machines, type:

	make CMPsim64

Generating Traces to Simulate:

To generate a trace for any single-threaded application, do the following (from the CRC/ directory):

	mkdir traces
	pinkit/pin-2.7-31933-gcc.3.4.6-ia32_intel64-linux/pin -t ./bin/CMPsim.gentrace.32 -threads 1 -fwd [fast forward instructions in millions] -icount [traced instructions in millions] -o traces/[outputfile] -- [application] [application args]

OR for 64-bit machines:

	pinkit/pin-2.7-31933-gcc.3.4.6-ia32_intel64-linux/pin -t ./bin/CMPsim.gentrace.64 -threads 1 -fwd [fast forward instructions in millions] -icount [traced instructions in millions] -o [outputfile] -- [application] [application args]

Example:

	pinkit/pin-2.7-31933-gcc.3.4.6-ia32_intel64-linux/pin -t ./bin/CMPsim.gentrace.64 -threads 1 -o traces/ls.out -- /bin/ls
 OR:
	pinkit/pin-2.7-31933-gcc.3.4.6-ia32_intel64-linux/pin -t ./bin/CMPsim.gentrace.64 -threads 1 -o traces/cat.out -- cat README 

Output of Trace Generator (from the first example above):

• ls.out.gz: summary of instructions traced
• ls.out.trace.gz: binary of the trace file collected
• ls.out.dep.gz: ASCII representation of the instruction dependency information for out-or-order simulation

To collect traces for a snippet of an application, use the -fwd and -icount options to the pin tool.

	-fwd :   e.g. -fwd 100 will fast-forward 100 million instructions (note: arg is in millions)
	-icount :  e.g. -icount 10 will trace 10 million instructions of the application

For example, to generate a trace of "ls -al" that fast-forwards for 1 million instructions and traces another 1 million instructions:

	pinkit/pin-2.7-31933-gcc.3.4.6-ia32_intel64-linux/pin -t ./bin/CMPsim.gentrace.64 -threads 1 -fwd 1 -icount 1 -o traces/lsal.out -- /bin/ls -al

(Hint: When evaluating the replacement techniques, the organizers will use an undistributed trace list. A significant part of that trace list will be based on SPEC2006 benchmarks. When generating these traces, you should warm them up for 40 billion instructions and generate a trace for 100 million instructions, i.e., -fwd 40000 -icount 100)

Generating Multi-Program Traces:

To run 4-core traces, you need to get single-threaded traces first. To construct a multiprogram trace for the 4-core configuration, you need to create a text file containing the full path for four such single threaded traces. For example, the following file contains two ls traces and two cat traces:

	cat > traces/mix_ls_cat.mix
	YOUR_HOME_DIR/CRC/traces/ls.out.trace.gz
	YOUR_HOME_DIR/CRC/traces/cat.out.trace.gz
	YOUR_HOME_DIR/CRC/traces/ls.out.trace.gz
	YOUR_HOME_DIR/CRC/traces/cat.out.trace.gz
	CTRL-D

Running the Simulator:

To run the simulator on the generated trace, do the following (from the CRC/ directory):

	mkdir runs
	cd runs/

The command format is:

	../bin/CMPsim.usetrace.32 -threads [# threads] (-t [trace file name] | -mix [trace mix file name]) -icount [number of instructions in millions] -o [output file name] -cache [cache configuration] -LLCrepl [Replacement policy number]

OR for 64-bit machines:

	../bin/CMPsim.usetrace.64 -threads [# threads] (-t [trace file name] | -mix [trace mix file name]) -icount [number of instructions in millions] -o [output file name] -cache [cache configuration] -LLCrepl [Replacement policy number]

• # threads: 1 (single-core simulation) or 4 (four-core simulation)
• trace file name: path to one of the generated single-threaded traces
• trace mix file name: A trace file mix generated by including four single-threaded traces in the same file (see above)
• cache configuration: UL3:Cache_Size_In_KB:Line_Size:Associativity
• Replacement policy number: 0 for LRU, 1 for random, 2 for your contest algorithm

For example, to run a single-core simulation using the default cache configuration and the LRU replacement algorithm:

	../bin/CMPsim.usetrace.64 -threads 1 -t ../traces/ls.out.trace.gz -o ls.stats -cache UL3:1024:64:16 -LLCrepl 0

You can examine statistics from this run by opening ls.stats.gz

To run a four-core simulation using the default cache configuration and the LRU replacement algorithm:

	../bin/CMPsim.usetrace.64 -threads 4 -mix ../traces/mix_ls_cat.mix -cache UL3:4096:64:16 -autorewind 1 -icount 1 -o mix_ls_cat.stats -LLCrepl 0

Note that "autorewind 1" is necessary to ensure that a fast application continues to give contention in the presence of slow applications (should the trace run out of instructions to provide)

You can examine statistics from this simulation by opening mix_ls_cat.stats.gz

Writing Your Own Replacement Algorithm:

In your CRC/src/LLCsim/ directory, there are two files you can modify: replacement_state.cpp and replacement_state.h.

Only these two files should be submitted with your contest submission. All other files in that directory should not be modified.