Re: [PATCH] Perf stat: Fix the ratio comments of cache miss-events

From: Qi Liu
Date: Mon Feb 17 2020 - 03:08:09 EST



This modification seems fine for me, thanks for your work!

On 2020/2/13 16:55, Lin Feng wrote:
> Perf stat displays miss ratio of L1-dcache, L1-icache, dTLB cache,
> iTLB cache and LL-cache, while the comments for them seem a bit
> misleading. Take L1-dcache for example, its miss ratio
> is caculated as "L1-dcache-load-misses/L1-dcache-loads". So "of all
> L1-dcache hits" is unsuitable to describe it, and "of all L1-dcache
> accesses/references" seems better.
>
> 285,132,521 cache-misses # 53.696 % of all cache refs [83.34%]
> 531,015,219 cache-references [83.20%]
> 220,465,183 LLC-load-misses # 72.33% of all LL-cache miss [83.44%]
> ^^^^
> 304,787,745 LLC-loads [66.60%]
>
> There is an old patch here https://lkml.org/lkml/2019/11/16/37, but
> seems not upstreamed yet. This patch follows suggestions by Arnaldo.
> But one difference is using "refs" to follow the convention because
> there is comment also for cache-misses event which uses
> "of all cache refs".
> The comments of L1-icache, dTLB cache, iTLB cache and LL-cache are
> fixed in the same way.
>
> P.S. Liu and Andi if you don't mind this patch adds your Reviewed-by
> and Signed-off-by, thanks.
>
> Reviewed-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> Signed-off-by: Qi Liu <liuqi115@xxxxxxxxxxxxx>
> Signed-off-by: Lin Feng <linf@xxxxxxxxxx>
> ---
> tools/perf/util/stat-shadow.c | 20 ++++++++++----------
> 1 file changed, 10 insertions(+), 10 deletions(-)
>
> diff --git a/tools/perf/util/stat-shadow.c b/tools/perf/util/stat-shadow.c
> index 2c41d47f6f83..070e9749e934 100644
> --- a/tools/perf/util/stat-shadow.c
> +++ b/tools/perf/util/stat-shadow.c
> @@ -506,7 +506,7 @@ static void print_l1_dcache_misses(struct perf_stat_config *config,
>
> color = get_ratio_color(GRC_CACHE_MISSES, ratio);
>
> - out->print_metric(config, out->ctx, color, "%7.2f%%", "of all L1-dcache hits", ratio);
> + out->print_metric(config, out->ctx, color, "%7.2f%%", "of all L1-dcache refs", ratio);
> }
>
> static void print_l1_icache_misses(struct perf_stat_config *config,
> @@ -527,7 +527,7 @@ static void print_l1_icache_misses(struct perf_stat_config *config,
> ratio = avg / total * 100.0;
>
> color = get_ratio_color(GRC_CACHE_MISSES, ratio);
> - out->print_metric(config, out->ctx, color, "%7.2f%%", "of all L1-icache hits", ratio);
> + out->print_metric(config, out->ctx, color, "%7.2f%%", "of all L1-icache refs", ratio);
> }
>
> static void print_dtlb_cache_misses(struct perf_stat_config *config,
> @@ -547,7 +547,7 @@ static void print_dtlb_cache_misses(struct perf_stat_config *config,
> ratio = avg / total * 100.0;
>
> color = get_ratio_color(GRC_CACHE_MISSES, ratio);
> - out->print_metric(config, out->ctx, color, "%7.2f%%", "of all dTLB cache hits", ratio);
> + out->print_metric(config, out->ctx, color, "%7.2f%%", "of all dTLB cache refs", ratio);
> }
>
> static void print_itlb_cache_misses(struct perf_stat_config *config,
> @@ -567,7 +567,7 @@ static void print_itlb_cache_misses(struct perf_stat_config *config,
> ratio = avg / total * 100.0;
>
> color = get_ratio_color(GRC_CACHE_MISSES, ratio);
> - out->print_metric(config, out->ctx, color, "%7.2f%%", "of all iTLB cache hits", ratio);
> + out->print_metric(config, out->ctx, color, "%7.2f%%", "of all iTLB cache refs", ratio);
> }
>
> static void print_ll_cache_misses(struct perf_stat_config *config,
> @@ -587,7 +587,7 @@ static void print_ll_cache_misses(struct perf_stat_config *config,
> ratio = avg / total * 100.0;
>
> color = get_ratio_color(GRC_CACHE_MISSES, ratio);
> - out->print_metric(config, out->ctx, color, "%7.2f%%", "of all LL-cache hits", ratio);
> + out->print_metric(config, out->ctx, color, "%7.2f%%", "of all LL-cache refs", ratio);
> }
>
> /*
> @@ -872,7 +872,7 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
> if (runtime_stat_n(st, STAT_L1_DCACHE, ctx, cpu) != 0)
> print_l1_dcache_misses(config, cpu, evsel, avg, out, st);
> else
> - print_metric(config, ctxp, NULL, NULL, "of all L1-dcache hits", 0);
> + print_metric(config, ctxp, NULL, NULL, "of all L1-dcache refs", 0);
> } else if (
> evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
> evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_L1I |
> @@ -882,7 +882,7 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
> if (runtime_stat_n(st, STAT_L1_ICACHE, ctx, cpu) != 0)
> print_l1_icache_misses(config, cpu, evsel, avg, out, st);
> else
> - print_metric(config, ctxp, NULL, NULL, "of all L1-icache hits", 0);
> + print_metric(config, ctxp, NULL, NULL, "of all L1-icache refs", 0);
> } else if (
> evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
> evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_DTLB |
> @@ -892,7 +892,7 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
> if (runtime_stat_n(st, STAT_DTLB_CACHE, ctx, cpu) != 0)
> print_dtlb_cache_misses(config, cpu, evsel, avg, out, st);
> else
> - print_metric(config, ctxp, NULL, NULL, "of all dTLB cache hits", 0);
> + print_metric(config, ctxp, NULL, NULL, "of all dTLB cache refs", 0);
> } else if (
> evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
> evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_ITLB |
> @@ -902,7 +902,7 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
> if (runtime_stat_n(st, STAT_ITLB_CACHE, ctx, cpu) != 0)
> print_itlb_cache_misses(config, cpu, evsel, avg, out, st);
> else
> - print_metric(config, ctxp, NULL, NULL, "of all iTLB cache hits", 0);
> + print_metric(config, ctxp, NULL, NULL, "of all iTLB cache refs", 0);
> } else if (
> evsel->core.attr.type == PERF_TYPE_HW_CACHE &&
> evsel->core.attr.config == ( PERF_COUNT_HW_CACHE_LL |
> @@ -912,7 +912,7 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,
> if (runtime_stat_n(st, STAT_LL_CACHE, ctx, cpu) != 0)
> print_ll_cache_misses(config, cpu, evsel, avg, out, st);
> else
> - print_metric(config, ctxp, NULL, NULL, "of all LL-cache hits", 0);
> + print_metric(config, ctxp, NULL, NULL, "of all LL-cache refs", 0);
> } else if (perf_evsel__match(evsel, HARDWARE, HW_CACHE_MISSES)) {
> total = runtime_stat_avg(st, STAT_CACHEREFS, ctx, cpu);
>
>