Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 21 additions & 0 deletions llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -2529,6 +2529,23 @@ static SDValue foldAddSubBoolOfMaskedVal(SDNode *N, SelectionDAG &DAG) {
return DAG.getNode(IsAdd ? ISD::SUB : ISD::ADD, DL, VT, C1, LowBit);
}

// Attempt to form avgceilu(A, B) from (A | B) - ((A ^ B) >> 1)
static SDValue combineFixedwidthToAVGCEILU(SDNode *N, SelectionDAG &DAG) {
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SDValue N0 = N->getOperand(0);
EVT VT = N0.getValueType();
SDLoc DL(N);
if (TLI.isOperationLegal(ISD::AVGCEILU, VT)) {
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

AVGFLOORU -> AVGCEILU

This shares quite a lot with #84903 and could be combined into the same function once that is submitted.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

AVGFLOORU -> AVGCEILU

Updated.

This shares quite a lot with #84903 and could be combined into the same function once that is submitted

Sure.

SDValue A, B;
if (sd_match(N, m_Sub(m_Or(m_Value(A), m_Value(B)),
m_Srl(m_Xor(m_Deferred(A), m_Deferred(B)),
m_SpecificInt(1))))) {
return DAG.getNode(ISD::AVGCEILU, DL, VT, A, B);
}
}
return SDValue();
}

/// Try to fold a 'not' shifted sign-bit with add/sub with constant operand into
/// a shift and add with a different constant.
static SDValue foldAddSubOfSignBit(SDNode *N, SelectionDAG &DAG) {
Expand Down Expand Up @@ -3849,6 +3866,10 @@ SDValue DAGCombiner::visitSUB(SDNode *N) {
if (SDValue V = foldAddSubOfSignBit(N, DAG))
return V;

// Try to match AVGCEILU fixedwidth pattern
if (SDValue V = combineFixedwidthToAVGCEILU(N, DAG))
return V;

if (SDValue V = foldAddSubMasked1(false, N0, N1, DAG, SDLoc(N)))
return V;

Expand Down
14 changes: 11 additions & 3 deletions llvm/test/CodeGen/AArch64/hadd-combine.ll
Original file line number Diff line number Diff line change
Expand Up @@ -329,9 +329,17 @@ define <8 x i16> @hadds_i_undef(<8 x i16> %t, <8 x i16> %src1) {
ret <8 x i16> %result
}




define <8 x i16> @sub_fixedwidth_v4i32(<8 x i16> %a0, <8 x i16> %a1) {
; CHECK-LABEL: sub_fixedwidth_v4i32:
; CHECK: // %bb.0:
; CHECK-NEXT: urhadd v0.8h, v0.8h, v1.8h
; CHECK-NEXT: ret
%or = or <8 x i16> %a0, %a1
%xor = xor <8 x i16> %a0, %a1
%srl = lshr <8 x i16> %xor, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
%res = sub <8 x i16> %or, %srl
ret <8 x i16> %res
}

define <8 x i16> @rhaddu_base(<8 x i16> %src1, <8 x i16> %src2) {
; CHECK-LABEL: rhaddu_base:
Expand Down