<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:taxo="http://purl.org/rss/1.0/modules/taxonomy/" version="2.0">
  <channel>
    <title>topic How to get llc misses using perf in Software Tuning, Performance Optimization &amp; Platform Monitoring</title>
    <link>https://community.intel.com/t5/Software-Tuning-Performance/How-to-get-llc-misses-using-perf/m-p/1614925#M8391</link>
    <description>&lt;P&gt;In perf manual, I find two metrics to obtain llc misses:&amp;nbsp;&lt;/P&gt;&lt;PRE&gt;&lt;STRONG&gt;PERF_COUNT_HW_CACHE_MISSES&lt;BR /&gt;Cache misses. Usually this indicates Last Level Cache misses;&lt;/STRONG&gt;&lt;/PRE&gt;&lt;PRE&gt;&lt;STRONG&gt;PERF_COUNT_HW_CACHE_LL&lt;/STRONG&gt;
for measuring Last-Level Cache&lt;BR /&gt;&lt;STRONG&gt;PERF_COUNT_HW_CACHE_OP_READ&lt;/STRONG&gt;
for read accesses
&lt;STRONG&gt;PERF_COUNT_HW_CACHE_OP_WRITE&lt;/STRONG&gt;
for write accesses
&lt;STRONG&gt;PERF_COUNT_HW_CACHE_OP_PREFETCH&lt;/STRONG&gt;
for prefetch accesses&lt;/PRE&gt;&lt;P&gt;I use perf to obtain the two metrics(respectively cache-misses and LLC-load-misses), but there seems to be large difference. &amp;nbsp;Which is the right answer? or any other metrics?&lt;/P&gt;&lt;P&gt;&amp;nbsp;&lt;/P&gt;&lt;PRE&gt;$sudo perf stat --timeout 10000 -e instructions,cache-references,cache-misses,LLC-loads,LLC-load-misses,LLC-store,LLC-store-misses -G kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1&lt;BR /&gt;&lt;BR /&gt;Performance counter stats for 'system wide':&lt;BR /&gt;&lt;BR /&gt;957,616,348,042 instructions kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.33%)&lt;BR /&gt;2,385,485,978 cache-references kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.91%)&lt;BR /&gt;637,991,398 cache-misses kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 # 26.745 % of all cache refs (27.91%)&lt;BR /&gt;38,378,430 LLC-loads kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.91%)&lt;BR /&gt;6,031,894 LLC-load-misses kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 # 15.72% of all LL-cache hits (27.79%)&lt;BR /&gt;1,254,043 LLC-store kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.79%)&lt;BR /&gt;135,446 LLC-store-misses kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.82%)&lt;BR /&gt;&lt;BR /&gt;10.004501298 seconds time elapsed&lt;/PRE&gt;&lt;P&gt;This is my cpu&lt;/P&gt;&lt;PRE&gt;Architecture: x86_64&lt;BR /&gt;CPU op-mode(s): 32-bit, 64-bit&lt;BR /&gt;Byte Order: Little Endian&lt;BR /&gt;CPU(s): 104&lt;BR /&gt;On-line CPU(s) list: 0-103&lt;BR /&gt;Thread(s) per core: 2&lt;BR /&gt;Core(s) per socket: 26&lt;BR /&gt;Socket(s): 2&lt;BR /&gt;NUMA node(s): 2&lt;BR /&gt;Vendor ID: GenuineIntel&lt;BR /&gt;CPU family: 6&lt;BR /&gt;Model: 85&lt;BR /&gt;Model name: Intel(R) Xeon(R) Platinum 8269CY CPU @ 2.50GHz&lt;/PRE&gt;&lt;P&gt;&amp;nbsp;&lt;/P&gt;</description>
    <pubDate>Mon, 15 Jul 2024 04:14:56 GMT</pubDate>
    <dc:creator>wanger</dc:creator>
    <dc:date>2024-07-15T04:14:56Z</dc:date>
    <item>
      <title>How to get llc misses using perf</title>
      <link>https://community.intel.com/t5/Software-Tuning-Performance/How-to-get-llc-misses-using-perf/m-p/1614925#M8391</link>
      <description>&lt;P&gt;In perf manual, I find two metrics to obtain llc misses:&amp;nbsp;&lt;/P&gt;&lt;PRE&gt;&lt;STRONG&gt;PERF_COUNT_HW_CACHE_MISSES&lt;BR /&gt;Cache misses. Usually this indicates Last Level Cache misses;&lt;/STRONG&gt;&lt;/PRE&gt;&lt;PRE&gt;&lt;STRONG&gt;PERF_COUNT_HW_CACHE_LL&lt;/STRONG&gt;
for measuring Last-Level Cache&lt;BR /&gt;&lt;STRONG&gt;PERF_COUNT_HW_CACHE_OP_READ&lt;/STRONG&gt;
for read accesses
&lt;STRONG&gt;PERF_COUNT_HW_CACHE_OP_WRITE&lt;/STRONG&gt;
for write accesses
&lt;STRONG&gt;PERF_COUNT_HW_CACHE_OP_PREFETCH&lt;/STRONG&gt;
for prefetch accesses&lt;/PRE&gt;&lt;P&gt;I use perf to obtain the two metrics(respectively cache-misses and LLC-load-misses), but there seems to be large difference. &amp;nbsp;Which is the right answer? or any other metrics?&lt;/P&gt;&lt;P&gt;&amp;nbsp;&lt;/P&gt;&lt;PRE&gt;$sudo perf stat --timeout 10000 -e instructions,cache-references,cache-misses,LLC-loads,LLC-load-misses,LLC-store,LLC-store-misses -G kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1&lt;BR /&gt;&lt;BR /&gt;Performance counter stats for 'system wide':&lt;BR /&gt;&lt;BR /&gt;957,616,348,042 instructions kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.33%)&lt;BR /&gt;2,385,485,978 cache-references kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.91%)&lt;BR /&gt;637,991,398 cache-misses kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 # 26.745 % of all cache refs (27.91%)&lt;BR /&gt;38,378,430 LLC-loads kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.91%)&lt;BR /&gt;6,031,894 LLC-load-misses kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 # 15.72% of all LL-cache hits (27.79%)&lt;BR /&gt;1,254,043 LLC-store kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.79%)&lt;BR /&gt;135,446 LLC-store-misses kubepods/besteffort/podeef13b4a-d2b2-427a-8219-a9af8b327cf1 (27.82%)&lt;BR /&gt;&lt;BR /&gt;10.004501298 seconds time elapsed&lt;/PRE&gt;&lt;P&gt;This is my cpu&lt;/P&gt;&lt;PRE&gt;Architecture: x86_64&lt;BR /&gt;CPU op-mode(s): 32-bit, 64-bit&lt;BR /&gt;Byte Order: Little Endian&lt;BR /&gt;CPU(s): 104&lt;BR /&gt;On-line CPU(s) list: 0-103&lt;BR /&gt;Thread(s) per core: 2&lt;BR /&gt;Core(s) per socket: 26&lt;BR /&gt;Socket(s): 2&lt;BR /&gt;NUMA node(s): 2&lt;BR /&gt;Vendor ID: GenuineIntel&lt;BR /&gt;CPU family: 6&lt;BR /&gt;Model: 85&lt;BR /&gt;Model name: Intel(R) Xeon(R) Platinum 8269CY CPU @ 2.50GHz&lt;/PRE&gt;&lt;P&gt;&amp;nbsp;&lt;/P&gt;</description>
      <pubDate>Mon, 15 Jul 2024 04:14:56 GMT</pubDate>
      <guid>https://community.intel.com/t5/Software-Tuning-Performance/How-to-get-llc-misses-using-perf/m-p/1614925#M8391</guid>
      <dc:creator>wanger</dc:creator>
      <dc:date>2024-07-15T04:14:56Z</dc:date>
    </item>
  </channel>
</rss>

