aboutsummaryrefslogtreecommitdiff
path: root/lib/Target/X86
diff options
context:
space:
mode:
authorDimitry Andric <dim@FreeBSD.org>2017-07-29 21:25:18 +0000
committerDimitry Andric <dim@FreeBSD.org>2017-07-29 21:25:18 +0000
commit3ad6a4b447326bc16c17df65637ca02330b8d090 (patch)
tree568321855815f8ca008258972e27d4a3ea487475 /lib/Target/X86
parent93c91e39b29142dec1d03a30df9f6e757f56c193 (diff)
Notes
Diffstat (limited to 'lib/Target/X86')
-rw-r--r--lib/Target/X86/X86ISelDAGToDAG.cpp5
-rw-r--r--lib/Target/X86/X86ISelLowering.cpp9
2 files changed, 9 insertions, 5 deletions
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 3c4589ab18f6..8f24f98be681 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -1055,7 +1055,10 @@ static bool foldMaskAndShiftToScale(SelectionDAG &DAG, SDValue N,
// Scale the leading zero count down based on the actual size of the value.
// Also scale it down based on the size of the shift.
- MaskLZ -= (64 - X.getSimpleValueType().getSizeInBits()) + ShiftAmt;
+ unsigned ScaleDown = (64 - X.getSimpleValueType().getSizeInBits()) + ShiftAmt;
+ if (MaskLZ < ScaleDown)
+ return true;
+ MaskLZ -= ScaleDown;
// The final check is to ensure that any masked out high bits of X are
// already known to be zero. Otherwise, the mask has a semantic impact
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 44eecd664714..ba8eb8656585 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1672,8 +1672,8 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
// TODO: These control memcmp expansion in CGP and could be raised higher, but
// that needs to benchmarked and balanced with the potential use of vector
- // load/store types (PR33329).
- MaxLoadsPerMemcmp = 4;
+ // load/store types (PR33329, PR33914).
+ MaxLoadsPerMemcmp = 2;
MaxLoadsPerMemcmpOptSize = 2;
// Set loop alignment to 2^ExperimentalPrefLoopAlignment bytes (default: 2^4).
@@ -22022,8 +22022,9 @@ static SDValue LowerScalarImmediateShift(SDValue Op, SelectionDAG &DAG,
return getTargetVShiftByConstNode(X86Opc, dl, VT, R, ShiftAmt, DAG);
// i64 SRA needs to be performed as partial shifts.
- if ((VT == MVT::v2i64 || (Subtarget.hasInt256() && VT == MVT::v4i64)) &&
- Op.getOpcode() == ISD::SRA && !Subtarget.hasXOP())
+ if (((!Subtarget.hasXOP() && VT == MVT::v2i64) ||
+ (Subtarget.hasInt256() && VT == MVT::v4i64)) &&
+ Op.getOpcode() == ISD::SRA)
return ArithmeticShiftRight64(ShiftAmt);
if (VT == MVT::v16i8 ||