perf vendor events arm64: Fix incorrect CPU_CYCLE in metrics expr

[PATCH] perf vendor events arm64: Fix incorrect CPU_CYCLE in metrics expr

Posted by Yangyu Chen 10 months, 1 week ago

Some existing metrics for Neoverse N3 and V3 expressions use CPU_CYCLE
to represent the number of cycles, but this is incorrect. The correct
event to use is CPU_CYCLES.

I encountered this issue while working on a patch to add pmu events for
Cortex A720 and A520 by reusing the existing patch for Neoverse N3 and
V3 by James Clark [1] and my check script [2] reported this issue.

[1] https://lore.kernel.org/lkml/20250122163504.2061472-1-james.clark@linaro.org/
[2] https://github.com/cyyself/arm-pmu-check

Signed-off-by: Yangyu Chen <cyy@cyyself.name>
---
 .../perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json | 6 +++---
 .../perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json b/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json
index 1f7c9536cb88..eb3a35f244e7 100644
--- a/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json
+++ b/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json
@@ -169,7 +169,7 @@
     },
     {
         "MetricName": "fp_ops_per_cycle",
-        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLE",
+        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLES",
         "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by any instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
         "MetricGroup": "FP_Arithmetic_Intensity",
         "ScaleUnit": "1operations per cycle"
@@ -383,7 +383,7 @@
     },
     {
         "MetricName": "nonsve_fp_ops_per_cycle",
-        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLE",
+        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLES",
         "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by an instruction that is not an SVE instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
         "MetricGroup": "FP_Arithmetic_Intensity",
         "ScaleUnit": "1operations per cycle"
@@ -421,7 +421,7 @@
     },
     {
         "MetricName": "sve_fp_ops_per_cycle",
-        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLE",
+        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLES",
         "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by SVE instructions. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
         "MetricGroup": "FP_Arithmetic_Intensity",
         "ScaleUnit": "1operations per cycle"
diff --git a/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json b/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json
index d022ae25c864..4a671f55eaf3 100644
--- a/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json
+++ b/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json
@@ -169,7 +169,7 @@
     },
     {
         "MetricName": "fp_ops_per_cycle",
-        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLE",
+        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLES",
         "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by any instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
         "MetricGroup": "FP_Arithmetic_Intensity",
         "ScaleUnit": "1operations per cycle"
@@ -383,7 +383,7 @@
     },
     {
         "MetricName": "nonsve_fp_ops_per_cycle",
-        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLE",
+        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLES",
         "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by an instruction that is not an SVE instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
         "MetricGroup": "FP_Arithmetic_Intensity",
         "ScaleUnit": "1operations per cycle"
@@ -421,7 +421,7 @@
     },
     {
         "MetricName": "sve_fp_ops_per_cycle",
-        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLE",
+        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLES",
         "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by SVE instructions. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
         "MetricGroup": "FP_Arithmetic_Intensity",
         "ScaleUnit": "1operations per cycle"
-- 
2.47.2

Re: [PATCH] perf vendor events arm64: Fix incorrect CPU_CYCLE in metrics expr

Posted by Namhyung Kim 10 months ago

On Thu, 13 Feb 2025 16:44:09 +0800, Yangyu Chen wrote:
> Some existing metrics for Neoverse N3 and V3 expressions use CPU_CYCLE
> to represent the number of cycles, but this is incorrect. The correct
> event to use is CPU_CYCLES.
> 
> I encountered this issue while working on a patch to add pmu events for
> Cortex A720 and A520 by reusing the existing patch for Neoverse N3 and
> V3 by James Clark [1] and my check script [2] reported this issue.
> 
> [...]
Applied to perf-tools-next, thanks!

Best regards,
Namhyung

Re: [PATCH] perf vendor events arm64: Fix incorrect CPU_CYCLE in metrics expr

Posted by James Clark 10 months, 1 week ago


On 13/02/2025 8:44 am, Yangyu Chen wrote:
> Some existing metrics for Neoverse N3 and V3 expressions use CPU_CYCLE
> to represent the number of cycles, but this is incorrect. The correct
> event to use is CPU_CYCLES.
> 
> I encountered this issue while working on a patch to add pmu events for
> Cortex A720 and A520 by reusing the existing patch for Neoverse N3 and
> V3 by James Clark [1] and my check script [2] reported this issue.
> 
> [1] https://lore.kernel.org/lkml/20250122163504.2061472-1-james.clark@linaro.org/
> [2] https://github.com/cyyself/arm-pmu-check
> 
> Signed-off-by: Yangyu Chen <cyy@cyyself.name>

Thanks for the fix. I'll report this issue on the source data.

Reviewed-by: James Clark <james.clark@linaro.org>

> ---
>   .../perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json | 6 +++---
>   .../perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json | 6 +++---
>   2 files changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json b/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json
> index 1f7c9536cb88..eb3a35f244e7 100644
> --- a/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json
> +++ b/tools/perf/pmu-events/arch/arm64/arm/neoverse-n3/metrics.json
> @@ -169,7 +169,7 @@
>       },
>       {
>           "MetricName": "fp_ops_per_cycle",
> -        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLE",
> +        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLES",
>           "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by any instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
>           "MetricGroup": "FP_Arithmetic_Intensity",
>           "ScaleUnit": "1operations per cycle"
> @@ -383,7 +383,7 @@
>       },
>       {
>           "MetricName": "nonsve_fp_ops_per_cycle",
> -        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLE",
> +        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLES",
>           "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by an instruction that is not an SVE instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
>           "MetricGroup": "FP_Arithmetic_Intensity",
>           "ScaleUnit": "1operations per cycle"
> @@ -421,7 +421,7 @@
>       },
>       {
>           "MetricName": "sve_fp_ops_per_cycle",
> -        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLE",
> +        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLES",
>           "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by SVE instructions. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
>           "MetricGroup": "FP_Arithmetic_Intensity",
>           "ScaleUnit": "1operations per cycle"
> diff --git a/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json b/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json
> index d022ae25c864..4a671f55eaf3 100644
> --- a/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json
> +++ b/tools/perf/pmu-events/arch/arm64/arm/neoverse-v3/metrics.json
> @@ -169,7 +169,7 @@
>       },
>       {
>           "MetricName": "fp_ops_per_cycle",
> -        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLE",
> +        "MetricExpr": "(FP_SCALE_OPS_SPEC + FP_FIXED_OPS_SPEC) / CPU_CYCLES",
>           "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by any instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
>           "MetricGroup": "FP_Arithmetic_Intensity",
>           "ScaleUnit": "1operations per cycle"
> @@ -383,7 +383,7 @@
>       },
>       {
>           "MetricName": "nonsve_fp_ops_per_cycle",
> -        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLE",
> +        "MetricExpr": "FP_FIXED_OPS_SPEC / CPU_CYCLES",
>           "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by an instruction that is not an SVE instruction. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
>           "MetricGroup": "FP_Arithmetic_Intensity",
>           "ScaleUnit": "1operations per cycle"
> @@ -421,7 +421,7 @@
>       },
>       {
>           "MetricName": "sve_fp_ops_per_cycle",
> -        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLE",
> +        "MetricExpr": "FP_SCALE_OPS_SPEC / CPU_CYCLES",
>           "BriefDescription": "This metric measures floating point operations per cycle in any precision performed by SVE instructions. Operations are counted by computation and by vector lanes, fused computations such as multiply-add count as twice per vector lane for example.",
>           "MetricGroup": "FP_Arithmetic_Intensity",
>           "ScaleUnit": "1operations per cycle"