On 11/13/2025 8:50 AM, Ian Rogers wrote:
> Use the counter mask (cmask) to see how many cycles an instruction
> takes to retire. Present as a set of ILP metrics.
>
> Signed-off-by: Ian Rogers <irogers@google.com>
> ---
> tools/perf/pmu-events/amd_metrics.py | 37 ++++++++++++++++++++++++++++
> 1 file changed, 37 insertions(+)
>
> diff --git a/tools/perf/pmu-events/amd_metrics.py b/tools/perf/pmu-events/amd_metrics.py
> index 53f7c2198147..9a39e23075b8 100755
> --- a/tools/perf/pmu-events/amd_metrics.py
> +++ b/tools/perf/pmu-events/amd_metrics.py
> @@ -121,6 +121,42 @@ def AmdBr():
> description="breakdown of retired branch instructions")
>
>
> +def AmdIlp() -> MetricGroup:
> + tsc = Event("msr/tsc/")
> + c0 = Event("msr/mperf/")
> + low = tsc - c0
> + inst_ret = Event("ex_ret_instr")
> + inst_ret_c = [Event(f"{inst_ret.name}/cmask={x}/") for x in range(1, 6)]
> + ilp = [d_ratio(max(inst_ret_c[x] - inst_ret_c[x + 1], 0), cycles)
> + for x in range(0, 4)]
> + ilp.append(d_ratio(inst_ret_c[4], cycles))
According to AMD PPRs, "the corresponding PERF_CTR[5:0] register increments by 1,
if the number of events occurring in a clock cycle is greater than or equal to
the CntMask value".
Please correct me if I am getting this wrong but I am interpreting the values above as
inst_ret_c[0] = cycles in which at least 1 instruction retired
inst_ret_c[1] = cycles in which at least 2 instructions retired
...
inst_ret_c[5] = cycles in which at least 6 instructions retired
ilp[0] = fraction of cycles in which exactly 1 instruction retired
ilp[1] = fraction of cycles in which exactly 2 instructions retired
...
ilp[4] = fraction of cycles in which exactly 5 instructions retired
ilp[5] = fraction of cycles in which at least 5 instructions retired
> + ilp0 = 1
> + for x in ilp:
> + ilp0 -= x
> + return MetricGroup("lpm_ilp", [
> + Metric("lpm_ilp_idle", "Lower power cycles as a percentage of all cycles",
> + d_ratio(low, tsc), "100%"),
> + Metric("lpm_ilp_inst_ret_0",
> + "Instructions retired in 0 cycles as a percentage of all cycles",
> + ilp0, "100%"),
If the interpretation above is correct, ilp0 (fraction of cycles in which no instructions
retired) can be computed as 1 - (inst_ret_c[0] / cycles) as inst_ret_c[0] is the count of
cycles in which at least 1 instruction retired.
> + Metric("lpm_ilp_inst_ret_1",
> + "Instructions retired in 1 cycles as a percentage of all cycles",
> + ilp[0], "100%"),
> + Metric("lpm_ilp_inst_ret_2",
> + "Instructions retired in 2 cycles as a percentage of all cycles",
> + ilp[1], "100%"),
> + Metric("lpm_ilp_inst_ret_3",
> + "Instructions retired in 3 cycles as a percentage of all cycles",
> + ilp[2], "100%"),
> + Metric("lpm_ilp_inst_ret_4",
> + "Instructions retired in 4 cycles as a percentage of all cycles",
> + ilp[3], "100%"),
> + Metric("lpm_ilp_inst_ret_5",
> + "Instructions retired in 5 or more cycles as a percentage of all cycles",
> + ilp[4], "100%"),
ilp[5] corresponds to "5 or more".
> + ])
> +
> +
> def AmdDtlb() -> Optional[MetricGroup]:
> global _zen_model
> if _zen_model >= 4:
> @@ -600,6 +636,7 @@ def main() -> None:
>
> all_metrics = MetricGroup("", [
> AmdBr(),
> + AmdIlp(),
> AmdDtlb(),
> AmdItlb(),
> AmdLdSt(),