From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Tue, 31 Jul 2018 14:15:16 +0000 (+0000)
Subject: AMDGPU: Don't handle FP16_TO_FP in isCanonicalized
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=55c7db8b2ead45d01b6060e9768121a5d78ff576;p=llvm

AMDGPU: Don't handle FP16_TO_FP in isCanonicalized

This needs more special handling to do correctly.
Fixes test in subsequent commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338381 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/SIISelLowering.cpp b/lib/Target/AMDGPU/SIISelLowering.cpp
index acde638a006..64d63e43855 100644
--- a/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -6800,10 +6800,6 @@ static bool isCanonicalized(SelectionDAG &DAG, SDValue Op,
     return Op.getOperand(0).getValueType().getScalarType() != MVT::f16 ||
            ST->hasFP16Denormals();
 
-  case ISD::FP16_TO_FP:
-  case ISD::FP_TO_FP16:
-    return ST->hasFP16Denormals();
-
   // It can/will be lowered or combined as a bit operation.
   // Need to check their input recursively to handle.
   case ISD::FNEG: