[PATCH] [SDAG] Add SDNodes for umin, umax, smin and smax
Demikhovsky, Elena
elena.demikhovsky at intel.com
Thu May 14 12:02:26 PDT 2015
Hi James,
I thought more about number of values on my way home and concluded that you was right.
You'll never see more than one value on MIN/MAX pattern,
because it comes from "icmp" instruction, that does not work with aggregate types.
- Elena
-----Original Message-----
From: James Molloy [mailto:james.molloy at arm.com]
Sent: Thursday, May 14, 2015 16:53
To: james.molloy at arm.com; resistor at mac.com; t.p.northover at gmail.com; Demikhovsky, Elena
Cc: amara.emerson at arm.com; llvm-commits at cs.uiuc.edu
Subject: Re: [PATCH] [SDAG] Add SDNodes for umin, umax, smin and smax
Hi Elena,
Updated according to comments. I changed the check - we still need to check that all the VTs are the same, otherwise the transformation would be completely incorrect.
The while statement can't be moved inside the if, because the while statement modifies "VT". The aim is to peel off SplitVector operations and suchlike until we get to the bottommost type, which we then check if it's legal or custom.
Cheers,
James
REPOSITORY
rL LLVM
http://reviews.llvm.org/D9746
Files:
include/llvm/CodeGen/ISDOpcodes.h
include/llvm/Target/TargetSelectionDAG.td
lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
lib/CodeGen/TargetLoweringBase.cpp
EMAIL PREFERENCES
http://reviews.llvm.org/settings/panel/emailpreferences/
---------------------------------------------------------------------
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.
More information about the llvm-commits
mailing list