[PATCH] Lower FNEG ( FABS (x) ) -> FNABS (x) [X86 codegen] PR20578

[Demikhovsky, Elena]

Ok. You are right. I have no more comments.

-  Elena


-----Original Message-----
From: Sanjay Patel [mailto:spatel at rotateright.com] 
Sent: Saturday, September 27, 2014 02:26
To: spatel at rotateright.com; chandlerc at gmail.com; nrotem at apple.com; hfinkel at anl.gov; Andrea_DiBiagio at sn.scee.net; Demikhovsky, Elena
Cc: llvm-commits at cs.uiuc.edu
Subject: Re: [PATCH] Lower FNEG ( FABS (x) ) -> FNABS (x) [X86 codegen] PR20578

================
Comment at: lib/Target/X86/X86ISelLowering.cpp:12344
@@ +12343,3 @@
+  if (IsFABS)
+    for (SDNode *User : Op->uses())
+      if (User->getOpcode() == ISD::FNEG)
----------------
spatel wrote:
> delena wrote:
> > I think this check is not necessary because the graph always comes from bottom.
> Thanks for looking at this, Elena.
> 
> Without this check, all of the vector tests in the proposed test file fail because we process the FNEG first. I don't know why this is different between scalar and vector, but it is.
Sorry - that should have been process the *FABS* first.

http://reviews.llvm.org/D5201


---------------------------------------------------------------------
Intel Israel (74) Limited

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.