Skip to content

Latest commit

 

History

History
33 lines (24 loc) · 3.85 KB

README_SPEC_INSTALLATION.md

File metadata and controls

33 lines (24 loc) · 3.85 KB
Raw

After installing SPEC-2017 on the disk, for running the SPECint-CPU2017 benchmarks with gem5, you need to create the run-directories for each SPEC benchmark with the following steps (these steps are only needed once for each benchmark):

1. Create configuration file

  • Copy an example configuration to another file, for example, cp config/Example-gcc-linux-x86.cfg config/gem5_se_test.cfg.
  • Open the gem5_se_test.cfg configuration file and change its label. For example, define label gem5_se.
  • We strongly recommend changing the compiler optimization for microarchitecture and setting it to athlon64 to ensure gem5 doesn't encounter unknown instructions. Specifically, we used OPTIMIZE = -g -O3 -march=athlon64 -fno-unsafe-math-optimizations -fno-tree-loop-vectorize.

2. Test run directory generation

  • Source the scripts: source shrc
  • Run the following command: runcpu --config=gem5_se_test --action=runsetup 500.perlbench_r. This command should build a benchmark (if it does not exist), create run directories and copy the executable & inputs to run folder.
    • A helpful resource for understanding this process can be the SPEC2017 installation page, which outlines the steps to install and test SPEC benchmarks, and also this link providing details about runcpu actions.
  • Check $SPEC17_PATH/benchspec/CPU/500.perlbench_r/run/: there should be a run folder (e.g. run_base_refrate_gem5_se-m64.0000) containing the executable and input files (e.g. checkspam.pl) for the benchmark.
  • The exact name of the run-folder might be different based on your OS/CPU/compiler. If so, please follow the steps:
    • Open spec17_benchmarks.py file in gem5/configs/example directory.
    • Update the RUN_DIR_postfix variable as with run-folder (like run_base_refrate_gem5_se-m64.0000) and x86_suffix variable with executable binary postfix (like _r_base.gem5_se-m64).
  • Lastly, you need to make sure gem5 runs with the correct executable names for benchmarks (this can also vary on different SPEC installations). Check the name of the executable in the run-folder. On our system, the binary name is perlbench_r_base.gem5_se-m64. Note the suffix after perlbench (e.g. _r_base.gem5_se-m64).
    • Open spec17_benchmarks.py file in gem5/configs/example directory.
    • Update the x86_suffix variable with the suffix on your system.
  • Verify that the executable runs: cd $SPEC17_PATH/benchspec/CPU/500.perlbench_r/run/run_base_refrate_gem5_se-m64.0000 and ./perlbench_r_base.gem5_se-m64 -I./lib checkspam.pl 2500 5 25 11 150 1 1 1 1

3. Test the Checkpointing Script with SPEC17

  • After these steps,./ckptscript_test.sh perlbench 4 2017 should hopefully work: the script should be able to cd into the correct run-directory for the SPEC-benchmark and run the correct executable name with the right input parameters.
    • You can check the command used in cpts/output/multiprogram_16GBmem_1Mn.SPEC2017.C4/checkpoint_out/perlbench/runscript.log.

4. Generate run directories for all SPEC17 workloads

  • Once perlbench is successfully tested, to generate the run directories and also perform a test run for all the benchmarks, use: runcpu --config=gem5_se_test --action=run --size=ref --copies=1 --noreportable --iterations=1 intrate fprate. You can also use --action=runsetup instead of --action=run to avoid doing a test run and only generating the run directories.
  • Note: After all benchmark run-directories are created, please make sure run-directories of all benchmarks have the same naming format.

After completing these steps, all the SPEC benchmnark run-directories should be set up. Now, you can try the next step using runscript_test.sh in README.md to test gem5 functionality and then run all the experiments.