Support showing the global bus events metrics

[ryoqun]

hi, thanks for maintaining this crate. i found this useful. :)

Callgrind supports collecting global bus events. This is useful for measuring how much atomic instructions are issued for given algorithm:

       --collect-bus=<no|yes> [default: no]
           This specifies whether the number of global bus events executed should be collected. The event type "Ge" is used for these events.

However, iai-callgrind doesn't support this? Ideally, I'd like to see this output when it's given with opt-in cli flag like --collect-bus:

test_lib_bench_readme_example_fibonacci::bench_fibonacci_group::bench_fibonacci short:10
  Instructions:                1733|N/A             (*********)
  L1 Hits:                     2359|N/A             (*********)
  L2 Hits:                        0|N/A             (*********)
  RAM Hits:                       2|N/A             (*********)
  Total read+write:            2361|N/A             (*********)
  Estimated Cycles:            2429|N/A             (*********)
  Global Bus Events:            223|N/A             (*********)  # <- NEW 

[gamma0987]

Thanks for your interest in iai-callgrind.

There are different possibilities in place to achieve collecting global bus events (and the other optional callgrind events):

You can switch on --collect-bus using the LibraryBenchmarkConfig (library benchmarks) or BinaryBenchmarkConfig (binary benchmarks). For example:

main!(
    config = LibraryBenchmarkConfig::default()
        .raw_callgrind_args(["--collect-bus=yes"]);
...
);

The global bus events are printed in the Ge column in the output. Or, using the cli with environment variables

IAI_CALLGRIND_ARGS="--collect-bus=yes" cargo bench

with arguments

cargo bench -- --callgrind-args="--collect-bus=yes"

[ryoqun]

thanks for the reply!

There are different possibilities in place to achieve collecting global bus events (and the other optional callgrind events):

You can switch on --collect-bus using the LibraryBenchmarkConfig (library benchmarks) or BinaryBenchmarkConfig (binary benchmarks). For example:

main!(
    config = LibraryBenchmarkConfig::default()
        .raw_callgrind_args(["--collect-bus=yes"]);
...
);

Well, this worked nice!

The global bus events are printed in the Ge column in the output.

yeah, i saw like this:

  Instructions:                 729|729             (No change)
  L1 Hits:                     1093|1096            (-0.27372%) [-1.00274x]
  L2 Hits:                        0|0               (No change)
  RAM Hits:                      61|58              (+5.17241%) [+1.05172x]
  Total read+write:            1154|1154            (No change)
  Estimated Cycles:            3228|3126            (+3.26296%) [+1.03263x]
  Ge:                             8|N/A             (*********)

Or, using the cli with environment variables

IAI_CALLGRIND_ARGS="--collect-bus=yes" cargo bench

however, this didn't work for some reason.

with arguments

cargo bench -- --callgrind-args="--collect-bus=yes"

and this didn't too.

[gamma0987]

Oops. Sorry, the environment variable changed to IAI_CALLGRIND_CALLGRIND_ARGS="--collect-bus=yes" cargo bench.

For cargo bench -- --callgrind-args="--collect-bus=yes" to work you may need to specify the benchmark with --bench BENCHMARK, for example cargo bench --bench my_iai_benchmark -- --callgrind-args="--collect-bus=yes" or add the following to your Cargo.toml

[lib]
bench = false

and if you have binaries

[[bin]]
name = "my-binary"
path = "src/bin/my-binary.rs"
bench = false

[gamma0987]

You can always use cargo bench --bench my_iai_benchmark -- --help to list the recognized environment variables and command line arguments.

[gamma0987]

Is it working now?