Yes, I believe I commented on that concern in my original email (Perhaps not for this reason, however). Ben pointed out that GVN should be able to clean things up if there are multiple instances of 1/Y.<br><br><div class="gmail_quote">
On Thu, Aug 8, 2013 at 1:21 PM, Shuxin Yang <span dir="ltr"><<a href="mailto:shuxin.llvm@gmail.com" target="_blank">shuxin.llvm@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div bgcolor="#FFFFFF" text="#000000">
I remember why I didn't implement this rule in Instcombine. It add
one instruction. So, <br>
this xform should be driven by a redundancy eliminator if you care
code size. <br><div><div class="h5">
<br>
<div>On 8/8/13 10:13 AM, Shuxin Yang wrote:<br>
</div>
<blockquote type="cite">
I did few transformation in Instruction *InstCombiner::visitFDiv()
in an attempt to remove some divs. <br>
I may miss this case. If you need to implement this rule, it is
better done in Instcombine than in DAG combine. <br>
Doing such xform early expose the redundancy of 1/y, which have
positive impact to neighboring code, <br>
while DAG combine is bit blind. <br>
<br>
You should be very careful, reciprocal is very very very imprecise
transformation. Sometimes you will see big different<br>
with and without this xform. <br>
<br>
<div>On 8/8/13 9:25 AM, Chad Rosier wrote:<br>
</div>
<blockquote type="cite">I would like to transform X/Y -> X*1/Y.
Specifically, I would like to convert:<br>
<br>
define void @t1a(double %a, double %b, double %d) {<br>
entry:<br>
%div = fdiv fast double %a, %d<br>
%div1 = fdiv fast double %b, %d<br>
%call = tail call i32 @foo(double %div, double %div1)<br>
ret void<br>
}<br>
<br>
to:<br>
<br>
define void @t1b(double %a, double %b, double %d) {<br>
entry:<br>
%div = fdiv fast double 1.000000e+00, %d<br>
%mul = fmul fast double %div, %a<br>
%mul1 = fmul fast double %div, %b<br>
%call = tail call i32 @foo(double %mul, double %mul1)<br>
ret void<br>
}<br>
<br>
Is such a transformation best done as a (target-specific) DAG
combine?<br>
<br>
A similar instcombine already exists for the X/C->X*1/C case
(see the CvtFDivConstToReciprocal function in
InstCombineMlDivRem.cpp), but I don't believe the above can be
done as an instcombine as it creates a new instruction (in
addition to replacing the original). Also, I only want to
perform the transformation if there are multiple uses of 1/Y
(like in my test case). Otherwise, the transformation replaces
a fdiv with a fdiv+fmul pair, which I doubt would be profitable.<br>
<br>
FWIW, I'm also pretty sure this combine requires -fast-math.<br>
<br>
Can someone point me in the right direction?<br>
<br>
Thanks,<br>
Chad<br>
<br>
<fieldset></fieldset>
<br>
<pre>_______________________________________________
LLVM Developers mailing list
<a href="mailto:LLVMdev@cs.uiuc.edu" target="_blank">LLVMdev@cs.uiuc.edu</a> <a href="http://llvm.cs.uiuc.edu" target="_blank">http://llvm.cs.uiuc.edu</a>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev</a>
</pre>
</blockquote>
<br>
</blockquote>
<br>
</div></div></div>
<br>_______________________________________________<br>
LLVM Developers mailing list<br>
<a href="mailto:LLVMdev@cs.uiuc.edu">LLVMdev@cs.uiuc.edu</a> <a href="http://llvm.cs.uiuc.edu" target="_blank">http://llvm.cs.uiuc.edu</a><br>
<a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev</a><br>
<br></blockquote></div><br>