From 82c3dc63474b3e25bf70db31b2e1d71903485b81 Mon Sep 17 00:00:00 2001
From: Chris Lattner <sabre@nondot.org>
Date: Sun, 14 Mar 2010 23:03:31 +0000
Subject: [PATCH] tidy up and expound more on how half-float works.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@98524 91177308-0d34-0410-b5e6-96231b3b80d8
---
docs/LangRef.html | 35 +++++++++++++++++++++--------------
1 file changed, 21 insertions(+), 14 deletions(-)
diff --git a/docs/LangRef.html b/docs/LangRef.html
index a867ee01c53..9d2d226ddc6 100644
--- a/docs/LangRef.html
+++ b/docs/LangRef.html
@@ -257,8 +257,8 @@
</li>
<li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
<ol>
- <li><a href="#int_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
- <li><a href="#int_from_fp16">'<tt>llvm.conver.from.fp16</tt>' Intrinsic</a></li>
+ <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
+ <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
</ol>
</li>
<li><a href="#int_debugger">Debugger intrinsics</a></li>
@@ -6619,16 +6619,23 @@ LLVM</a>.</p>
<div class="doc_text">
-<p>Half precision floating point is storage-only format. That is why the values
- in such format should be promoted to single precision format before any
- operations. LLVM provides intrinsics for conversions to single precision and
- back.</p>
-
+<p>Half precision floating point is storage-only format. This means that it is
+ an encoding (typically for density in memory) but typically does not support
+ computation in the format.</p>
+
+<p>This means that you code must first load the half-precision floating point
+ value as an i16, then convert it to float with <a
+ href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
+ Computation can then be performed on the float value (including extending to
+ double etc). When the value is ready to be stored back to memory, this is
+ done by converting to float if needed, then converting to i16 with
+ <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
+ storing as an i16 value.</p>
</div>
<!-- _______________________________________________________________________ -->
<div class="doc_subsubsection">
- <a name="int_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a>
+ <a name="int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a>
</div>
<div class="doc_text">
@@ -6650,8 +6657,8 @@ LLVM</a>.</p>
<h5>Semantics:</h5>
<p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
a conversion from single precision floating point format to half precision
- floating point format. Since the format is storage only the return value is
- just an <tt>i16</tt> which contains the converted number.</p>
+ floating point format. The return value is just an <tt>i16</tt> which
+ contains the converted number.</p>
<h5>Examples:</h5>
<pre>
@@ -6663,7 +6670,7 @@ LLVM</a>.</p>
<!-- _______________________________________________________________________ -->
<div class="doc_subsubsection">
- <a name="int_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a>
+ <a name="int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a>
</div>
<div class="doc_text">
@@ -6684,9 +6691,9 @@ LLVM</a>.</p>
<h5>Semantics:</h5>
<p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
- conversion from half single precision floating point format to signle
- precision floating point format. Since the format is storage only the
- argument is represented by an <tt>i16</tt> value.</p>
+ conversion from half single precision floating point format to signgle
+ precision floating point format. The input half-float value is represented by
+ an <tt>i16</tt> value.</p>
<h5>Examples:</h5>
<pre>