<html>
<head>
<meta content="text/html; charset=windows-1252"
http-equiv="Content-Type">
</head>
<body bgcolor="#FFFFFF" text="#000000">
On 12/01/2014 10:19 AM, Pekka Jääskeläinen wrote:<br>
<blockquote cite="mid:547C8711.3000201@tut.fi" type="cite">Hi,
<br>
<br>
Here's a trivial patch to relax the FP math also
<br>
if -cl-fast-relaxed-math is set.
<br>
</blockquote>
<br>
I think the test should also check for the function target
attribute.<br>
<br>
What about the other flags? I think this patch should handle (maybe
just test?) all of the CL fast math flags instead of just this one (<br>
-cl-no-signed-zeros, -cl-unsafe-math-optimizations,
-cl-finite-math-only, and -cl-fast-relaxed-math)<br>
<br>
<blockquote cite="mid:547C8711.3000201@tut.fi" type="cite">
<br>
This supersedes the previous patch with title
<br>
"if unsafe-fp-math FN attribute is set, enable reduction
<br>
detection with float operations"
<br>
<br>
On 11/28/2014 08:04 PM, Pekka Jääskeläinen wrote:
<br>
<blockquote type="cite">Hi Hal,
<br>
<br>
OK, I didn't know which way was the chosen "future approach" or
<br>
if they both are used for some use cases.
<br>
<br>
I just noticed that when I set -cl-fast-relaxed-math in OpenCL
C,
<br>
it didn't propagate to the FP instructions as 'fast' attributes,
<br>
but the func attribute did get set.
<br>
<br>
So I suppose it should be investigated why the flags are not
<br>
generated to the produced instructions in this case instead
<br>
in case the fnattr is deprecated?
<br>
<br>
On 11/28/2014 07:43 PM, Hal Finkel wrote:
<br>
<blockquote type="cite">Pekka,
<br>
<br>
What's your motivation for doing this? Aside from the fact
that we
<br>
can't yet set fast math flags on function calls (to sqrt,
etc.), I
<br>
think we're pretty close to removing the function attribute
all
<br>
together. The instruction-level flags play much nicer with
LTO, and
<br>
give more precision control to the frontends; it is not clear
to me
<br>
that overriding them with the function-level attribute is
desirable.
<br>
<br>
-Hal
<br>
<br>
----- Original Message -----
<br>
<blockquote type="cite">From: "Pekka Jääskeläinen"
<a class="moz-txt-link-rfc2396E" href="mailto:pekka.jaaskelainen@tut.fi"><pekka.jaaskelainen@tut.fi></a> To: "Commit
<br>
Messages and Patches for LLVM"
<a class="moz-txt-link-rfc2396E" href="mailto:llvm-commits@cs.uiuc.edu"><llvm-commits@cs.uiuc.edu></a> Sent:
<br>
Friday, November 28, 2014 11:13:39 AM Subject: [PATCH]
<br>
[LoopVectorizer] if unsafe-fp-math FN attribute is set,
enable
<br>
reduction detection with float operations
<br>
<br>
Hi,
<br>
<br>
The attached patch checks the function attributes for
<br>
"unsafe-fp-math" to relax FP constraints when detecting
reduction
<br>
operations.
<br>
<br>
-- --Pekka
<br>
<br>
_______________________________________________ llvm-commits
<br>
mailing list <a class="moz-txt-link-abbreviated" href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a>
<br>
<a class="moz-txt-link-freetext" href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a>
<br>
<br>
</blockquote>
<br>
</blockquote>
<br>
<br>
</blockquote>
<br>
<br>
<br>
<fieldset class="mimeAttachmentHeader"></fieldset>
<br>
<pre wrap="">_______________________________________________
cfe-commits mailing list
<a class="moz-txt-link-abbreviated" href="mailto:cfe-commits@cs.uiuc.edu">cfe-commits@cs.uiuc.edu</a>
<a class="moz-txt-link-freetext" href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits">http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits</a>
</pre>
</blockquote>
<br>
</body>
</html>