<div style="font-family:'Segoe UI',Arial,Sans-Serif;font-size:10pt;">
<h1 style="font-size: 150%;font-weight:normal; color:#ff3228;"><a href="https://ci.appveyor.com/project/jrfonseca-fdo/mesa/build/3055" style="color:#ff3228;">Build mesa 3055 failed</a></h1>
<p style="color: #888;">
Commit <a href="">4634cb5921</a> by <a href="mailto:sroland@vmware.com">Roland Scheidegger</a> on 12/21/2016 3:56 AM:
<br />
<span style="font-size: 110%;color:#222;">gallivm: implement aos unpack (to unorm8) for small unorm formats\n\nUsing bit replication. This path now resembles something which might make\nsense. (The logic was mostly copied from llvmpipe fs backend.)\nI am not convinced though it is actually faster than SoA sampling (actually\nI'm quite certain it's always a loss with AVX).\nWith SoA it's just shift/mask/cvt/mul for getting the colors, whereas\nthere's still roughly 3 shifts, 3 or/and per channel for AoS\n(i.e. for SoA it's exactly the same as it would be for a rgba8 format,\nwhereas the extra effort for AoS is significant). The filtering\nmight still be faster (albeit with FMA the instruction count gets down\nquite a bit there on the SoA float filtering path on new cpus). And those\nsmall unorm formats often don't have an alpha channel (which makes things\nworse relatively for AoS path).\n(This also fixes a trivial bug in the llvmpipe fs code this was derived\nfrom, albeit it was only relevant for 4-bit channels.)\n\nReviewed-by: Jose Fonseca <jfonseca@vmware.com></span>
</p>
<p><a href="https://ci.appveyor.com/notifications" style="font-size:85%;color:#999;">Configure your notification preferences</a></p>
</div>