SPEC CPU 2006 runner

Simulating a cache idea is only useful if you run it on realistic programs. SPEC CPU 2006 is the classic benchmark suite for that: gcc, bzip2, mcf, perlbench, soplex, lbm, omnetpp and more. Running it on gem5 means juggling binaries, input files, cache sizes and dozens of command-line flags, so I wrapped it all in scripts.

What’s in the repo

• run_Spec2006_Simple_Modified.sh: runs the benchmarks one by one on a single core, in gem5’s syscall-emulation mode with a TimingSimpleCPU. Cache geometry (L1I/L1D/L2 sizes, associativity, line size), the instruction limit and fast-forward distance are variables at the top, so a whole experiment is one edit away.
• run_Spec2006_Simple_Multicore.sh: a 4-core version that runs four benchmarks at once (for example perlbench + bzip2 + mcf + soplex) on a detailed ARM CPU model. It fast-forwards 4 billion instructions to skip start-up and then measures 4 billion more, against the four-partition L2 from my research branch.

It started from a SPEC2000 script shared in the gem5 community and grew into my own.

Small things that make long runs bearable

A single benchmark can simulate for hours, so the runner:

• shows a spinner with elapsed time while gem5 runs in the background,
• writes each benchmark’s stats file and console output to its own log,
• prints coloured Started / Finished lines so you can glance at a terminal and know where it is.

runProgran log_403.gcc.txt gcc gcc/ref/input/200.in

Status

Finished. Paths point at my machine, so treat it as a template: set GEM5_ROOT, BINARY_DIR and INPUT_DATA_DIR to yours. SPEC itself is licensed and isn’t included.