<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="/rss.xsl.xml"?>
<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/">
<channel>
    <title>Changes in illegal-eviction-assert.mir</title>
    <description></description>
    <language>en</language>
    <copyright>Copyright 2015</copyright>
    <generator>Java</generator><item>
        <title>d4b1be20 - RegAllocGreedy: Fix illegal eviction assert for urgent evictions</title>
        <link>http://172.16.0.5:8080/history/llvm-project-15.0.7/llvm/test/CodeGen/AMDGPU/illegal-eviction-assert.mir#d4b1be20</link>
        <description>RegAllocGreedy: Fix illegal eviction assert for urgent evictionsThe condition in canEvictInterferenceBasedOnCost is slightly differentfrom the assertion in evictInteference.canEvictInterferenceBasedOnCost uses a &lt;= check for the cascade numberfor legality, but the assert was checking for &lt;. For equal cascadenumbers for an urgent eviction, canEvictInterferenceBasedOnCost couldreturn success. The actual eviction would then hit this assert. Avoidever returning true for equivalent cascade numbers.The resulting failed allocation seems a bit off to me. e.g. inillegal-eviction-assert.mir, I wuold assume %0 gets allocated startingat $vgpr0. That was its initial allocation choice, but was laterevicted. In this example no evictions can help improve anything.

            List of files:
            /llvm-project-15.0.7/llvm/test/CodeGen/AMDGPU/illegal-eviction-assert.mir</description>
        <pubDate>Tue, 29 Mar 2022 12:48:21 +0000</pubDate>
        <dc:creator>Matt Arsenault &lt;Matthew.Arsenault@amd.com&gt;</dc:creator>
    </item>
</channel>
</rss>
