<?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 What happens in DSB/LSD when an instruction is invalidated using CLFLUSH? in Software Tuning, Performance Optimization &amp; Platform Monitoring</title>
    <link>https://community.intel.com/t5/Software-Tuning-Performance/What-happens-in-DSB-LSD-when-an-instruction-is-invalidated-using/m-p/1144784#M6736</link>
    <description>&lt;P&gt;I have a small loop that fit the rules required to use LSD mechanism. The most relevant rule, in this case, is that all the micro-ops are resident in DSB. From the Optimization Manual, it is known that LSD fetches instructions from DSB while it works. It known that LSD ends when a misprediction takes place: "The loop streams from the micro-op queue, with no more fetching, decoding, or reading micro-ops from any of the caches, until a branch misprediction inevitably ends it."&lt;/P&gt;

&lt;P&gt;It is also known that DSB is a sort of "L0 cache": it is virtually included in L1 Icache. The Optimization Manual says:"Instruction cache evictions must also be evicted from the Decoded ICache, which evicts only the necessary lines".&lt;/P&gt;

&lt;P&gt;The question is the following:&lt;/P&gt;

&lt;P&gt;What happens when i run the described loop in a process and i evict one of its intruction from all the caches using CLFLUSH from another process? It's not clear if the micro-ops associated to the instruction are evicted from the DSB and, so the LSD should stop.&lt;/P&gt;

&lt;P&gt;My idea is that the eviction of micro-ops in DSB does not take place and the DSB continues to work.&lt;/P&gt;</description>
    <pubDate>Sat, 21 Oct 2017 08:56:26 GMT</pubDate>
    <dc:creator>Matteo_Fusi</dc:creator>
    <dc:date>2017-10-21T08:56:26Z</dc:date>
    <item>
      <title>What happens in DSB/LSD when an instruction is invalidated using CLFLUSH?</title>
      <link>https://community.intel.com/t5/Software-Tuning-Performance/What-happens-in-DSB-LSD-when-an-instruction-is-invalidated-using/m-p/1144784#M6736</link>
      <description>&lt;P&gt;I have a small loop that fit the rules required to use LSD mechanism. The most relevant rule, in this case, is that all the micro-ops are resident in DSB. From the Optimization Manual, it is known that LSD fetches instructions from DSB while it works. It known that LSD ends when a misprediction takes place: "The loop streams from the micro-op queue, with no more fetching, decoding, or reading micro-ops from any of the caches, until a branch misprediction inevitably ends it."&lt;/P&gt;

&lt;P&gt;It is also known that DSB is a sort of "L0 cache": it is virtually included in L1 Icache. The Optimization Manual says:"Instruction cache evictions must also be evicted from the Decoded ICache, which evicts only the necessary lines".&lt;/P&gt;

&lt;P&gt;The question is the following:&lt;/P&gt;

&lt;P&gt;What happens when i run the described loop in a process and i evict one of its intruction from all the caches using CLFLUSH from another process? It's not clear if the micro-ops associated to the instruction are evicted from the DSB and, so the LSD should stop.&lt;/P&gt;

&lt;P&gt;My idea is that the eviction of micro-ops in DSB does not take place and the DSB continues to work.&lt;/P&gt;</description>
      <pubDate>Sat, 21 Oct 2017 08:56:26 GMT</pubDate>
      <guid>https://community.intel.com/t5/Software-Tuning-Performance/What-happens-in-DSB-LSD-when-an-instruction-is-invalidated-using/m-p/1144784#M6736</guid>
      <dc:creator>Matteo_Fusi</dc:creator>
      <dc:date>2017-10-21T08:56:26Z</dc:date>
    </item>
  </channel>
</rss>

