[llvm] [Intrinsics][AArch64] Add intrinsic to mask off aliasing vector lanes (PR #117007)
Sam Tebbs via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 21 01:33:51 PST 2025
https://github.com/SamTebbs33 updated https://github.com/llvm/llvm-project/pull/117007
>From 57f8f2249d11641c080f02f3586f3e21cba97163 Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Fri, 15 Nov 2024 10:24:46 +0000
Subject: [PATCH 01/10] [Intrinsics][AArch64] Add intrinsic to mask off
aliasing vector lanes
It can be unsafe to load a vector from an address and write a vector to
an address if those two addresses have overlapping lanes within a
vectorised loop iteration.
This PR adds an intrinsic designed to create a mask with lanes disabled
if they overlap between the two pointer arguments, so that only safe
lanes are loaded, operated on and stored.
Along with the two pointer parameters, the intrinsic also takes an
immediate that represents the size in bytes of the vector element
types, as well as an immediate i1 that is true if there is a write
after-read-hazard or false if there is a read-after-write hazard.
This will be used by #100579 and replaces the existing lowering for
whilewr since that isn't needed now we have the intrinsic.
---
llvm/docs/LangRef.rst | 80 ++++
llvm/include/llvm/CodeGen/TargetLowering.h | 5 +
llvm/include/llvm/IR/Intrinsics.td | 5 +
.../SelectionDAG/SelectionDAGBuilder.cpp | 44 ++
.../Target/AArch64/AArch64ISelLowering.cpp | 62 ++-
llvm/lib/Target/AArch64/AArch64ISelLowering.h | 6 +
.../lib/Target/AArch64/AArch64SVEInstrInfo.td | 11 +-
llvm/lib/Target/AArch64/SVEInstrFormats.td | 10 +-
llvm/test/CodeGen/AArch64/alias_mask.ll | 421 ++++++++++++++++++
.../CodeGen/AArch64/alias_mask_scalable.ll | 82 ++++
10 files changed, 715 insertions(+), 11 deletions(-)
create mode 100644 llvm/test/CodeGen/AArch64/alias_mask.ll
create mode 100644 llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 8cc9036d1b67f6..52c2aba61e9027 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -23624,6 +23624,86 @@ Examples:
%active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i64(i64 %elem0, i64 429)
%wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* %3, i32 4, <4 x i1> %active.lane.mask, <4 x i32> poison)
+.. _int_experimental_get_alias_lane_mask:
+
+'``llvm.get.alias.lane.mask.*``' Intrinsics
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+This is an overloaded intrinsic.
+
+::
+
+ declare <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
+ declare <8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
+ declare <16 x i1> @llvm.experimental.get.alias.lane.mask.v16i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
+ declare <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.nxv16i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
+
+
+Overview:
+"""""""""
+
+Create a mask representing lanes that do or not overlap between two pointers across one vector loop iteration.
+
+
+Arguments:
+""""""""""
+
+The first two arguments have the same scalar integer type.
+The final two are immediates and the result is a vector with the i1 element type.
+
+Semantics:
+""""""""""
+
+In the case that ``%writeAfterRead`` is true, the '``llvm.experimental.get.alias.lane.mask.*``' intrinsics are semantically equivalent
+to:
+
+::
+
+ %diff = (%ptrB - %ptrA) / %elementSize
+ %m[i] = (icmp ult i, %diff) || (%diff <= 0)
+
+Otherwise they are semantically equivalent to:
+
+::
+
+ %diff = abs(%ptrB - %ptrA) / %elementSize
+ %m[i] = (icmp ult i, %diff) || (%diff == 0)
+
+where ``%m`` is a vector (mask) of active/inactive lanes with its elements
+indexed by ``i``, and ``%ptrA``, ``%ptrB`` are the two i64 arguments to
+``llvm.experimental.get.alias.lane.mask.*``, ``%elementSize`` is the i32 argument, ``%abs`` is the absolute difference operation, ``%icmp`` is an integer compare and ``ult``
+the unsigned less-than comparison operator. The subtraction between ``%ptrA`` and ``%ptrB`` could be negative. The ``%writeAfterRead`` argument is expected to be true if the ``%ptrB`` is stored to after ``%ptrA`` is read from.
+The above is equivalent to:
+
+::
+
+ %m = @llvm.experimental.get.alias.lane.mask(%ptrA, %ptrB, %elementSize, %writeAfterRead)
+
+This can, for example, be emitted by the loop vectorizer in which case
+``%ptrA`` is a pointer that is read from within the loop, and ``%ptrB`` is a pointer that is stored to within the loop.
+If the difference between these pointers is less than the vector factor, then they overlap (alias) within a loop iteration.
+An example is if ``%ptrA`` is 20 and ``%ptrB`` is 23 with a vector factor of 8, then lanes 3, 4, 5, 6 and 7 of the vector loaded from ``%ptrA``
+share addresses with lanes 0, 1, 2, 3, 4 and 5 from the vector stored to at ``%ptrB``.
+An alias mask of these two pointers should be <1, 1, 1, 0, 0, 0, 0, 0> so that only the non-overlapping lanes are loaded and stored.
+This operation allows many loops to be vectorised when it would otherwise be unsafe to do so.
+
+To account for the fact that only a subset of lanes have been operated on in an iteration,
+the loop's induction variable should be incremented by the popcount of the mask rather than the vector factor.
+
+This mask ``%m`` can e.g. be used in masked load/store instructions.
+
+
+Examples:
+"""""""""
+
+.. code-block:: llvm
+
+ %alias.lane.mask = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64(i64 %ptrA, i64 %ptrB, i32 4, i1 1)
+ %vecA = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* %ptrA, i32 4, <4 x i1> %alias.lane.mask, <4 x i32> poison)
+ [...]
+ call @llvm.masked.store.v4i32.p0v4i32(<4 x i32> %vecA, <4 x i32>* %ptrB, i32 4, <4 x i1> %alias.lane.mask)
.. _int_experimental_vp_splice:
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 38ac90f0c081b3..4e007863497fd1 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -468,6 +468,11 @@ class TargetLoweringBase {
return true;
}
+ /// Return true if the @llvm.experimental.get.alias.lane.mask intrinsic should be expanded using generic code in SelectionDAGBuilder.
+ virtual bool shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT, unsigned EltSize) const {
+ return true;
+ }
+
virtual bool shouldExpandGetVectorLength(EVT CountVT, unsigned VF,
bool IsScalable) const {
return true;
diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
index ee877349a33149..b7a8efbd51b83c 100644
--- a/llvm/include/llvm/IR/Intrinsics.td
+++ b/llvm/include/llvm/IR/Intrinsics.td
@@ -2363,6 +2363,11 @@ let IntrProperties = [IntrNoMem, IntrNoSync, IntrWillReturn, ImmArg<ArgIndex<1>>
llvm_i32_ty]>;
}
+def int_experimental_get_alias_lane_mask:
+ DefaultAttrsIntrinsic<[llvm_anyvector_ty],
+ [llvm_anyint_ty, LLVMMatchType<1>, llvm_anyint_ty, llvm_i1_ty],
+ [IntrNoMem, IntrNoSync, IntrWillReturn, ImmArg<ArgIndex<2>>, ImmArg<ArgIndex<3>>]>;
+
def int_get_active_lane_mask:
DefaultAttrsIntrinsic<[llvm_anyvector_ty],
[llvm_anyint_ty, LLVMMatchType<1>],
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 8a5d7c0b022d90..8e4253b6b7db7b 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -8276,6 +8276,50 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
visitVectorExtractLastActive(I, Intrinsic);
return;
}
+ case Intrinsic::experimental_get_alias_lane_mask: {
+ SDValue SourceValue = getValue(I.getOperand(0));
+ SDValue SinkValue = getValue(I.getOperand(1));
+ SDValue EltSize = getValue(I.getOperand(2));
+ bool IsWriteAfterRead = cast<ConstantSDNode>(getValue(I.getOperand(3)))->getZExtValue() != 0;
+ auto IntrinsicVT = EVT::getEVT(I.getType());
+ auto PtrVT = SourceValue->getValueType(0);
+
+ if (!TLI.shouldExpandGetAliasLaneMask(IntrinsicVT, PtrVT, cast<ConstantSDNode>(EltSize)->getSExtValue())) {
+ visitTargetIntrinsic(I, Intrinsic);
+ return;
+ }
+
+ SDValue Diff = DAG.getNode(ISD::SUB, sdl,
+ PtrVT, SinkValue, SourceValue);
+ if (!IsWriteAfterRead)
+ Diff = DAG.getNode(ISD::ABS, sdl, PtrVT, Diff);
+
+ Diff = DAG.getNode(ISD::SDIV, sdl, PtrVT, Diff, EltSize);
+ SDValue Zero = DAG.getTargetConstant(0, sdl, PtrVT);
+
+ // If the difference is positive then some elements may alias
+ auto CmpVT = TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(),
+ PtrVT);
+ SDValue Cmp = DAG.getSetCC(sdl, CmpVT, Diff, Zero, IsWriteAfterRead ? ISD::SETLE : ISD::SETEQ);
+
+ // Splat the compare result then OR it with a lane mask
+ SDValue Splat = DAG.getSplat(IntrinsicVT, sdl, Cmp);
+
+ SDValue DiffMask;
+ // Don't emit an active lane mask if the target doesn't support it
+ if (TLI.shouldExpandGetActiveLaneMask(IntrinsicVT, PtrVT)) {
+ EVT VecTy = EVT::getVectorVT(*DAG.getContext(), PtrVT,
+ IntrinsicVT.getVectorElementCount());
+ SDValue DiffSplat = DAG.getSplat(VecTy, sdl, Diff);
+ SDValue VectorStep = DAG.getStepVector(sdl, VecTy);
+ DiffMask = DAG.getSetCC(sdl, IntrinsicVT, VectorStep,
+ DiffSplat, ISD::CondCode::SETULT);
+ } else {
+ DiffMask = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, sdl, IntrinsicVT, DAG.getTargetConstant(Intrinsic::get_active_lane_mask, sdl, MVT::i64), Zero, Diff);
+ }
+ SDValue Or = DAG.getNode(ISD::OR, sdl, IntrinsicVT, DiffMask, Splat);
+ setValue(&I, Or);
+ }
}
}
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 8a3a9f75415fbc..83fc20d9b0c226 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -2038,6 +2038,24 @@ bool AArch64TargetLowering::shouldExpandGetActiveLaneMask(EVT ResVT,
return false;
}
+bool AArch64TargetLowering::shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT, unsigned EltSize) const {
+ if (!Subtarget->hasSVE2())
+ return true;
+
+ if (PtrVT != MVT::i64)
+ return true;
+
+ if (VT == MVT::v2i1 || VT == MVT::nxv2i1)
+ return EltSize != 8;
+ if( VT == MVT::v4i1 || VT == MVT::nxv4i1)
+ return EltSize != 4;
+ if (VT == MVT::v8i1 || VT == MVT::nxv8i1)
+ return EltSize != 2;
+ if (VT == MVT::v16i1 || VT == MVT::nxv16i1)
+ return EltSize != 1;
+ return true;
+}
+
bool AArch64TargetLowering::shouldExpandPartialReductionIntrinsic(
const IntrinsicInst *I) const {
if (I->getIntrinsicID() != Intrinsic::experimental_vector_partial_reduce_add)
@@ -2818,6 +2836,8 @@ const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
MAKE_CASE(AArch64ISD::LS64_BUILD)
MAKE_CASE(AArch64ISD::LS64_EXTRACT)
MAKE_CASE(AArch64ISD::TBL)
+ MAKE_CASE(AArch64ISD::WHILEWR)
+ MAKE_CASE(AArch64ISD::WHILERW)
MAKE_CASE(AArch64ISD::FADD_PRED)
MAKE_CASE(AArch64ISD::FADDA_PRED)
MAKE_CASE(AArch64ISD::FADDV_PRED)
@@ -6016,6 +6036,16 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
EVT PtrVT = getPointerTy(DAG.getDataLayout());
return DAG.getNode(AArch64ISD::THREAD_POINTER, dl, PtrVT);
}
+ case Intrinsic::aarch64_sve_whilewr_b:
+ case Intrinsic::aarch64_sve_whilewr_h:
+ case Intrinsic::aarch64_sve_whilewr_s:
+ case Intrinsic::aarch64_sve_whilewr_d:
+ return DAG.getNode(AArch64ISD::WHILEWR, dl, Op.getValueType(), Op.getOperand(1), Op.getOperand(2));
+ case Intrinsic::aarch64_sve_whilerw_b:
+ case Intrinsic::aarch64_sve_whilerw_h:
+ case Intrinsic::aarch64_sve_whilerw_s:
+ case Intrinsic::aarch64_sve_whilerw_d:
+ return DAG.getNode(AArch64ISD::WHILERW, dl, Op.getValueType(), Op.getOperand(1), Op.getOperand(2));
case Intrinsic::aarch64_neon_abs: {
EVT Ty = Op.getValueType();
if (Ty == MVT::i64) {
@@ -6475,16 +6505,39 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
return DAG.getNode(AArch64ISD::USDOT, dl, Op.getValueType(),
Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
}
+ case Intrinsic::experimental_get_alias_lane_mask:
case Intrinsic::get_active_lane_mask: {
+ unsigned IntrinsicID = Intrinsic::aarch64_sve_whilelo;
+ if (IntNo == Intrinsic::experimental_get_alias_lane_mask) {
+ uint64_t EltSize = Op.getOperand(3)->getAsZExtVal();
+ bool IsWriteAfterRead = Op.getOperand(4)->getAsZExtVal() == 1;
+ switch (EltSize) {
+ case 1:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_b : Intrinsic::aarch64_sve_whilerw_b;
+ break;
+ case 2:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_h : Intrinsic::aarch64_sve_whilerw_h;
+ break;
+ case 4:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_s : Intrinsic::aarch64_sve_whilerw_s;
+ break;
+ case 8:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_d : Intrinsic::aarch64_sve_whilerw_d;
+ break;
+ default:
+ llvm_unreachable("Unexpected element size for get.alias.lane.mask");
+ break;
+ }
+ }
SDValue ID =
- DAG.getTargetConstant(Intrinsic::aarch64_sve_whilelo, dl, MVT::i64);
+ DAG.getTargetConstant(IntrinsicID, dl, MVT::i64);
EVT VT = Op.getValueType();
if (VT.isScalableVector())
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT, ID, Op.getOperand(1),
Op.getOperand(2));
- // We can use the SVE whilelo instruction to lower this intrinsic by
+ // We can use the SVE whilelo/whilewr/whilerw instruction to lower this intrinsic by
// creating the appropriate sequence of scalable vector operations and
// then extracting a fixed-width subvector from the scalable vector.
@@ -19872,7 +19925,9 @@ static bool isPredicateCCSettingOp(SDValue N) {
N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilels ||
N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilelt ||
// get_active_lane_mask is lowered to a whilelo instruction.
- N.getConstantOperandVal(0) == Intrinsic::get_active_lane_mask)))
+ N.getConstantOperandVal(0) == Intrinsic::get_active_lane_mask ||
+ // get_alias_lane_mask is lowered to a whilewr/rw instruction.
+ N.getConstantOperandVal(0) == Intrinsic::experimental_get_alias_lane_mask)))
return true;
return false;
@@ -27626,6 +27681,7 @@ void AArch64TargetLowering::ReplaceNodeResults(
return;
}
case Intrinsic::experimental_vector_match:
+ case Intrinsic::experimental_get_alias_lane_mask:
case Intrinsic::get_active_lane_mask: {
if (!VT.isFixedLengthVector() || VT.getVectorElementType() != MVT::i1)
return;
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
index 61579de50db17e..65174498e21237 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -298,6 +298,10 @@ enum NodeType : unsigned {
SMAXV,
UMAXV,
+ // Alias lane masks
+ WHILEWR,
+ WHILERW,
+
SADDV_PRED,
UADDV_PRED,
SMAXV_PRED,
@@ -993,6 +997,8 @@ class AArch64TargetLowering : public TargetLowering {
bool shouldExpandGetActiveLaneMask(EVT VT, EVT OpVT) const override;
+ bool shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT, unsigned EltSize) const override;
+
bool
shouldExpandPartialReductionIntrinsic(const IntrinsicInst *I) const override;
diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 27c88a55919e65..1e133b640366b0 100644
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -140,6 +140,11 @@ def AArch64st1q_scatter : SDNode<"AArch64ISD::SST1Q_PRED", SDT_AArch64_SCATTER_V
// AArch64 SVE/SVE2 - the remaining node definitions
//
+// Alias masks
+def SDT_AArch64Mask : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisInt<1>, SDTCisSameAs<2, 1>, SDTCVecEltisVT<0,i1>]>;
+def AArch64whilewr : SDNode<"AArch64ISD::WHILEWR", SDT_AArch64Mask>;
+def AArch64whilerw : SDNode<"AArch64ISD::WHILERW", SDT_AArch64Mask>;
+
// SVE CNT/INC/RDVL
def sve_rdvl_imm : ComplexPattern<i64, 1, "SelectRDVLImm<-32, 31, 16>">;
def sve_cnth_imm : ComplexPattern<i64, 1, "SelectRDVLImm<1, 16, 8>">;
@@ -3914,9 +3919,9 @@ let Predicates = [HasSVE2_or_SME] in {
defm WHILEHI_PXX : sve_int_while8_rr<0b101, "whilehi", int_aarch64_sve_whilehi, int_aarch64_sve_whilelo>;
// SVE2 pointer conflict compare
- defm WHILEWR_PXX : sve2_int_while_rr<0b0, "whilewr", "int_aarch64_sve_whilewr">;
- defm WHILERW_PXX : sve2_int_while_rr<0b1, "whilerw", "int_aarch64_sve_whilerw">;
-} // End HasSVE2_or_SME
+ defm WHILEWR_PXX : sve2_int_while_rr<0b0, "whilewr", AArch64whilewr>;
+ defm WHILERW_PXX : sve2_int_while_rr<0b1, "whilerw", AArch64whilerw>;
+} // End HasSVE2orSME
let Predicates = [HasSVEAES, HasNonStreamingSVE2_or_SSVE_AES] in {
// SVE2 crypto destructive binary operations
diff --git a/llvm/lib/Target/AArch64/SVEInstrFormats.td b/llvm/lib/Target/AArch64/SVEInstrFormats.td
index 873fbf7dd346bb..2dbf5f4e754eeb 100644
--- a/llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -5835,16 +5835,16 @@ class sve2_int_while_rr<bits<2> sz8_64, bits<1> rw, string asm,
let isWhile = 1;
}
-multiclass sve2_int_while_rr<bits<1> rw, string asm, string op> {
+multiclass sve2_int_while_rr<bits<1> rw, string asm, SDPatternOperator op> {
def _B : sve2_int_while_rr<0b00, rw, asm, PPR8>;
def _H : sve2_int_while_rr<0b01, rw, asm, PPR16>;
def _S : sve2_int_while_rr<0b10, rw, asm, PPR32>;
def _D : sve2_int_while_rr<0b11, rw, asm, PPR64>;
- def : SVE_2_Op_Pat<nxv16i1, !cast<SDPatternOperator>(op # _b), i64, i64, !cast<Instruction>(NAME # _B)>;
- def : SVE_2_Op_Pat<nxv8i1, !cast<SDPatternOperator>(op # _h), i64, i64, !cast<Instruction>(NAME # _H)>;
- def : SVE_2_Op_Pat<nxv4i1, !cast<SDPatternOperator>(op # _s), i64, i64, !cast<Instruction>(NAME # _S)>;
- def : SVE_2_Op_Pat<nxv2i1, !cast<SDPatternOperator>(op # _d), i64, i64, !cast<Instruction>(NAME # _D)>;
+ def : SVE_2_Op_Pat<nxv16i1, op, i64, i64, !cast<Instruction>(NAME # _B)>;
+ def : SVE_2_Op_Pat<nxv8i1, op, i64, i64, !cast<Instruction>(NAME # _H)>;
+ def : SVE_2_Op_Pat<nxv4i1, op, i64, i64, !cast<Instruction>(NAME # _S)>;
+ def : SVE_2_Op_Pat<nxv2i1, op, i64, i64, !cast<Instruction>(NAME # _D)>;
}
//===----------------------------------------------------------------------===//
diff --git a/llvm/test/CodeGen/AArch64/alias_mask.ll b/llvm/test/CodeGen/AArch64/alias_mask.ll
new file mode 100644
index 00000000000000..59ad2bf82db92e
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/alias_mask.ll
@@ -0,0 +1,421 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s | FileCheck %s --check-prefix=CHECK-SVE
+; RUN: llc -mtriple=aarch64 %s | FileCheck %s --check-prefix=CHECK-NOSVE
+
+define <16 x i1> @whilewr_8(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilewr_8:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilewr p0.b, x0, x1
+; CHECK-SVE-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $q0 killed $q0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilewr_8:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI0_0
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI0_1
+; CHECK-NOSVE-NEXT: sub x9, x1, x0
+; CHECK-NOSVE-NEXT: ldr q0, [x8, :lo12:.LCPI0_0]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI0_2
+; CHECK-NOSVE-NEXT: ldr q1, [x10, :lo12:.LCPI0_1]
+; CHECK-NOSVE-NEXT: ldr q3, [x8, :lo12:.LCPI0_2]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI0_4
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI0_3
+; CHECK-NOSVE-NEXT: ldr q5, [x8, :lo12:.LCPI0_4]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI0_5
+; CHECK-NOSVE-NEXT: dup v2.2d, x9
+; CHECK-NOSVE-NEXT: ldr q4, [x10, :lo12:.LCPI0_3]
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI0_6
+; CHECK-NOSVE-NEXT: ldr q6, [x8, :lo12:.LCPI0_5]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI0_7
+; CHECK-NOSVE-NEXT: ldr q7, [x10, :lo12:.LCPI0_6]
+; CHECK-NOSVE-NEXT: cmp x9, #1
+; CHECK-NOSVE-NEXT: ldr q16, [x8, :lo12:.LCPI0_7]
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v2.2d, v0.2d
+; CHECK-NOSVE-NEXT: cmhi v1.2d, v2.2d, v1.2d
+; CHECK-NOSVE-NEXT: cmhi v3.2d, v2.2d, v3.2d
+; CHECK-NOSVE-NEXT: cmhi v4.2d, v2.2d, v4.2d
+; CHECK-NOSVE-NEXT: cmhi v5.2d, v2.2d, v5.2d
+; CHECK-NOSVE-NEXT: cmhi v6.2d, v2.2d, v6.2d
+; CHECK-NOSVE-NEXT: cmhi v7.2d, v2.2d, v7.2d
+; CHECK-NOSVE-NEXT: cmhi v2.2d, v2.2d, v16.2d
+; CHECK-NOSVE-NEXT: uzp1 v0.4s, v1.4s, v0.4s
+; CHECK-NOSVE-NEXT: cset w8, lt
+; CHECK-NOSVE-NEXT: uzp1 v1.4s, v4.4s, v3.4s
+; CHECK-NOSVE-NEXT: uzp1 v3.4s, v6.4s, v5.4s
+; CHECK-NOSVE-NEXT: uzp1 v2.4s, v2.4s, v7.4s
+; CHECK-NOSVE-NEXT: uzp1 v0.8h, v1.8h, v0.8h
+; CHECK-NOSVE-NEXT: uzp1 v1.8h, v2.8h, v3.8h
+; CHECK-NOSVE-NEXT: uzp1 v0.16b, v1.16b, v0.16b
+; CHECK-NOSVE-NEXT: dup v1.16b, w8
+; CHECK-NOSVE-NEXT: orr v0.16b, v0.16b, v1.16b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 1)
+ ret <16 x i1> %0
+}
+
+define <8 x i1> @whilewr_16(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilewr_16:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilewr p0.b, x0, x1
+; CHECK-SVE-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilewr_16:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: sub x8, x1, x0
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI1_0
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI1_1
+; CHECK-NOSVE-NEXT: add x8, x8, x8, lsr #63
+; CHECK-NOSVE-NEXT: adrp x11, .LCPI1_2
+; CHECK-NOSVE-NEXT: ldr q1, [x9, :lo12:.LCPI1_0]
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI1_3
+; CHECK-NOSVE-NEXT: ldr q2, [x10, :lo12:.LCPI1_1]
+; CHECK-NOSVE-NEXT: ldr q3, [x11, :lo12:.LCPI1_2]
+; CHECK-NOSVE-NEXT: asr x8, x8, #1
+; CHECK-NOSVE-NEXT: ldr q4, [x9, :lo12:.LCPI1_3]
+; CHECK-NOSVE-NEXT: dup v0.2d, x8
+; CHECK-NOSVE-NEXT: cmp x8, #1
+; CHECK-NOSVE-NEXT: cset w8, lt
+; CHECK-NOSVE-NEXT: cmhi v1.2d, v0.2d, v1.2d
+; CHECK-NOSVE-NEXT: cmhi v2.2d, v0.2d, v2.2d
+; CHECK-NOSVE-NEXT: cmhi v3.2d, v0.2d, v3.2d
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v0.2d, v4.2d
+; CHECK-NOSVE-NEXT: uzp1 v1.4s, v2.4s, v1.4s
+; CHECK-NOSVE-NEXT: uzp1 v0.4s, v0.4s, v3.4s
+; CHECK-NOSVE-NEXT: uzp1 v0.8h, v0.8h, v1.8h
+; CHECK-NOSVE-NEXT: dup v1.8b, w8
+; CHECK-NOSVE-NEXT: xtn v0.8b, v0.8h
+; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 1)
+ ret <8 x i1> %0
+}
+
+define <4 x i1> @whilewr_32(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilewr_32:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilewr p0.h, x0, x1
+; CHECK-SVE-NEXT: mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilewr_32:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: sub x9, x1, x0
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI2_0
+; CHECK-NOSVE-NEXT: add x10, x9, #3
+; CHECK-NOSVE-NEXT: cmp x9, #0
+; CHECK-NOSVE-NEXT: ldr q1, [x8, :lo12:.LCPI2_0]
+; CHECK-NOSVE-NEXT: csel x9, x10, x9, lt
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI2_1
+; CHECK-NOSVE-NEXT: asr x9, x9, #2
+; CHECK-NOSVE-NEXT: ldr q2, [x10, :lo12:.LCPI2_1]
+; CHECK-NOSVE-NEXT: dup v0.2d, x9
+; CHECK-NOSVE-NEXT: cmp x9, #1
+; CHECK-NOSVE-NEXT: cset w8, lt
+; CHECK-NOSVE-NEXT: cmhi v1.2d, v0.2d, v1.2d
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v0.2d, v2.2d
+; CHECK-NOSVE-NEXT: uzp1 v0.4s, v0.4s, v1.4s
+; CHECK-NOSVE-NEXT: dup v1.4h, w8
+; CHECK-NOSVE-NEXT: xtn v0.4h, v0.4s
+; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 1)
+ ret <4 x i1> %0
+}
+
+define <2 x i1> @whilewr_64(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilewr_64:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilewr p0.s, x0, x1
+; CHECK-SVE-NEXT: mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilewr_64:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: sub x9, x1, x0
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI3_0
+; CHECK-NOSVE-NEXT: add x10, x9, #7
+; CHECK-NOSVE-NEXT: cmp x9, #0
+; CHECK-NOSVE-NEXT: ldr q1, [x8, :lo12:.LCPI3_0]
+; CHECK-NOSVE-NEXT: csel x9, x10, x9, lt
+; CHECK-NOSVE-NEXT: asr x9, x9, #3
+; CHECK-NOSVE-NEXT: dup v0.2d, x9
+; CHECK-NOSVE-NEXT: cmp x9, #1
+; CHECK-NOSVE-NEXT: cset w8, lt
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v0.2d, v1.2d
+; CHECK-NOSVE-NEXT: dup v1.2s, w8
+; CHECK-NOSVE-NEXT: xtn v0.2s, v0.2d
+; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 1)
+ ret <2 x i1> %0
+}
+
+define <16 x i1> @whilerw_8(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilerw_8:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilerw p0.b, x0, x1
+; CHECK-SVE-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $q0 killed $q0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilerw_8:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI4_0
+; CHECK-NOSVE-NEXT: subs x9, x1, x0
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI4_1
+; CHECK-NOSVE-NEXT: ldr q0, [x8, :lo12:.LCPI4_0]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI4_2
+; CHECK-NOSVE-NEXT: cneg x9, x9, mi
+; CHECK-NOSVE-NEXT: ldr q2, [x8, :lo12:.LCPI4_2]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI4_3
+; CHECK-NOSVE-NEXT: ldr q1, [x10, :lo12:.LCPI4_1]
+; CHECK-NOSVE-NEXT: ldr q4, [x8, :lo12:.LCPI4_3]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI4_4
+; CHECK-NOSVE-NEXT: dup v3.2d, x9
+; CHECK-NOSVE-NEXT: ldr q5, [x8, :lo12:.LCPI4_4]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI4_5
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI4_6
+; CHECK-NOSVE-NEXT: ldr q6, [x8, :lo12:.LCPI4_5]
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI4_7
+; CHECK-NOSVE-NEXT: ldr q7, [x10, :lo12:.LCPI4_6]
+; CHECK-NOSVE-NEXT: ldr q16, [x8, :lo12:.LCPI4_7]
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v3.2d, v0.2d
+; CHECK-NOSVE-NEXT: cmhi v1.2d, v3.2d, v1.2d
+; CHECK-NOSVE-NEXT: cmhi v2.2d, v3.2d, v2.2d
+; CHECK-NOSVE-NEXT: cmhi v4.2d, v3.2d, v4.2d
+; CHECK-NOSVE-NEXT: cmhi v5.2d, v3.2d, v5.2d
+; CHECK-NOSVE-NEXT: cmhi v6.2d, v3.2d, v6.2d
+; CHECK-NOSVE-NEXT: cmhi v7.2d, v3.2d, v7.2d
+; CHECK-NOSVE-NEXT: cmhi v3.2d, v3.2d, v16.2d
+; CHECK-NOSVE-NEXT: uzp1 v0.4s, v1.4s, v0.4s
+; CHECK-NOSVE-NEXT: cmp x9, #0
+; CHECK-NOSVE-NEXT: uzp1 v1.4s, v4.4s, v2.4s
+; CHECK-NOSVE-NEXT: cset w8, eq
+; CHECK-NOSVE-NEXT: uzp1 v2.4s, v6.4s, v5.4s
+; CHECK-NOSVE-NEXT: uzp1 v3.4s, v3.4s, v7.4s
+; CHECK-NOSVE-NEXT: uzp1 v0.8h, v1.8h, v0.8h
+; CHECK-NOSVE-NEXT: uzp1 v1.8h, v3.8h, v2.8h
+; CHECK-NOSVE-NEXT: uzp1 v0.16b, v1.16b, v0.16b
+; CHECK-NOSVE-NEXT: dup v1.16b, w8
+; CHECK-NOSVE-NEXT: orr v0.16b, v0.16b, v1.16b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 0)
+ ret <16 x i1> %0
+}
+
+define <8 x i1> @whilerw_16(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilerw_16:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilerw p0.b, x0, x1
+; CHECK-SVE-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilerw_16:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: subs x8, x1, x0
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI5_0
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI5_1
+; CHECK-NOSVE-NEXT: cneg x8, x8, mi
+; CHECK-NOSVE-NEXT: adrp x11, .LCPI5_2
+; CHECK-NOSVE-NEXT: ldr q1, [x9, :lo12:.LCPI5_0]
+; CHECK-NOSVE-NEXT: add x8, x8, x8, lsr #63
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI5_3
+; CHECK-NOSVE-NEXT: ldr q2, [x10, :lo12:.LCPI5_1]
+; CHECK-NOSVE-NEXT: ldr q3, [x11, :lo12:.LCPI5_2]
+; CHECK-NOSVE-NEXT: ldr q4, [x9, :lo12:.LCPI5_3]
+; CHECK-NOSVE-NEXT: asr x8, x8, #1
+; CHECK-NOSVE-NEXT: dup v0.2d, x8
+; CHECK-NOSVE-NEXT: cmp x8, #0
+; CHECK-NOSVE-NEXT: cset w8, eq
+; CHECK-NOSVE-NEXT: cmhi v1.2d, v0.2d, v1.2d
+; CHECK-NOSVE-NEXT: cmhi v2.2d, v0.2d, v2.2d
+; CHECK-NOSVE-NEXT: cmhi v3.2d, v0.2d, v3.2d
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v0.2d, v4.2d
+; CHECK-NOSVE-NEXT: uzp1 v1.4s, v2.4s, v1.4s
+; CHECK-NOSVE-NEXT: uzp1 v0.4s, v0.4s, v3.4s
+; CHECK-NOSVE-NEXT: uzp1 v0.8h, v0.8h, v1.8h
+; CHECK-NOSVE-NEXT: dup v1.8b, w8
+; CHECK-NOSVE-NEXT: xtn v0.8b, v0.8h
+; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 0)
+ ret <8 x i1> %0
+}
+
+define <4 x i1> @whilerw_32(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilerw_32:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilerw p0.h, x0, x1
+; CHECK-SVE-NEXT: mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilerw_32:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: subs x9, x1, x0
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI6_0
+; CHECK-NOSVE-NEXT: cneg x9, x9, mi
+; CHECK-NOSVE-NEXT: ldr q1, [x8, :lo12:.LCPI6_0]
+; CHECK-NOSVE-NEXT: add x10, x9, #3
+; CHECK-NOSVE-NEXT: cmp x9, #0
+; CHECK-NOSVE-NEXT: csel x9, x10, x9, lt
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI6_1
+; CHECK-NOSVE-NEXT: asr x9, x9, #2
+; CHECK-NOSVE-NEXT: ldr q2, [x10, :lo12:.LCPI6_1]
+; CHECK-NOSVE-NEXT: dup v0.2d, x9
+; CHECK-NOSVE-NEXT: cmp x9, #0
+; CHECK-NOSVE-NEXT: cset w8, eq
+; CHECK-NOSVE-NEXT: cmhi v1.2d, v0.2d, v1.2d
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v0.2d, v2.2d
+; CHECK-NOSVE-NEXT: uzp1 v0.4s, v0.4s, v1.4s
+; CHECK-NOSVE-NEXT: dup v1.4h, w8
+; CHECK-NOSVE-NEXT: xtn v0.4h, v0.4s
+; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 0)
+ ret <4 x i1> %0
+}
+
+define <2 x i1> @whilerw_64(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: whilerw_64:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: whilerw p0.s, x0, x1
+; CHECK-SVE-NEXT: mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: // kill: def $d0 killed $d0 killed $z0
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: whilerw_64:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: subs x9, x1, x0
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI7_0
+; CHECK-NOSVE-NEXT: cneg x9, x9, mi
+; CHECK-NOSVE-NEXT: ldr q1, [x8, :lo12:.LCPI7_0]
+; CHECK-NOSVE-NEXT: add x10, x9, #7
+; CHECK-NOSVE-NEXT: cmp x9, #0
+; CHECK-NOSVE-NEXT: csel x9, x10, x9, lt
+; CHECK-NOSVE-NEXT: asr x9, x9, #3
+; CHECK-NOSVE-NEXT: dup v0.2d, x9
+; CHECK-NOSVE-NEXT: cmp x9, #0
+; CHECK-NOSVE-NEXT: cset w8, eq
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v0.2d, v1.2d
+; CHECK-NOSVE-NEXT: dup v1.2s, w8
+; CHECK-NOSVE-NEXT: xtn v0.2s, v0.2d
+; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 0)
+ ret <2 x i1> %0
+}
+
+define <16 x i1> @not_whilewr_wrong_eltsize(i64 %a, i64 %b) {
+; CHECK-SVE-LABEL: not_whilewr_wrong_eltsize:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: add x8, x8, x8, lsr #63
+; CHECK-SVE-NEXT: asr x8, x8, #1
+; CHECK-SVE-NEXT: cmp x8, #1
+; CHECK-SVE-NEXT: cset w9, lt
+; CHECK-SVE-NEXT: whilelo p0.b, #0, x8
+; CHECK-SVE-NEXT: dup v0.16b, w9
+; CHECK-SVE-NEXT: mov z1.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: orr v0.16b, v1.16b, v0.16b
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: not_whilewr_wrong_eltsize:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: sub x8, x1, x0
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_0
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI8_1
+; CHECK-NOSVE-NEXT: add x8, x8, x8, lsr #63
+; CHECK-NOSVE-NEXT: ldr q0, [x9, :lo12:.LCPI8_0]
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_2
+; CHECK-NOSVE-NEXT: ldr q2, [x9, :lo12:.LCPI8_2]
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_4
+; CHECK-NOSVE-NEXT: ldr q1, [x10, :lo12:.LCPI8_1]
+; CHECK-NOSVE-NEXT: asr x8, x8, #1
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI8_3
+; CHECK-NOSVE-NEXT: ldr q5, [x9, :lo12:.LCPI8_4]
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_6
+; CHECK-NOSVE-NEXT: ldr q3, [x10, :lo12:.LCPI8_3]
+; CHECK-NOSVE-NEXT: adrp x10, .LCPI8_5
+; CHECK-NOSVE-NEXT: dup v4.2d, x8
+; CHECK-NOSVE-NEXT: ldr q7, [x9, :lo12:.LCPI8_6]
+; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_7
+; CHECK-NOSVE-NEXT: ldr q6, [x10, :lo12:.LCPI8_5]
+; CHECK-NOSVE-NEXT: ldr q16, [x9, :lo12:.LCPI8_7]
+; CHECK-NOSVE-NEXT: cmp x8, #1
+; CHECK-NOSVE-NEXT: cset w8, lt
+; CHECK-NOSVE-NEXT: cmhi v0.2d, v4.2d, v0.2d
+; CHECK-NOSVE-NEXT: cmhi v1.2d, v4.2d, v1.2d
+; CHECK-NOSVE-NEXT: cmhi v2.2d, v4.2d, v2.2d
+; CHECK-NOSVE-NEXT: cmhi v3.2d, v4.2d, v3.2d
+; CHECK-NOSVE-NEXT: cmhi v5.2d, v4.2d, v5.2d
+; CHECK-NOSVE-NEXT: cmhi v6.2d, v4.2d, v6.2d
+; CHECK-NOSVE-NEXT: cmhi v7.2d, v4.2d, v7.2d
+; CHECK-NOSVE-NEXT: cmhi v4.2d, v4.2d, v16.2d
+; CHECK-NOSVE-NEXT: uzp1 v0.4s, v1.4s, v0.4s
+; CHECK-NOSVE-NEXT: uzp1 v1.4s, v3.4s, v2.4s
+; CHECK-NOSVE-NEXT: uzp1 v2.4s, v6.4s, v5.4s
+; CHECK-NOSVE-NEXT: uzp1 v3.4s, v4.4s, v7.4s
+; CHECK-NOSVE-NEXT: uzp1 v0.8h, v1.8h, v0.8h
+; CHECK-NOSVE-NEXT: uzp1 v1.8h, v3.8h, v2.8h
+; CHECK-NOSVE-NEXT: uzp1 v0.16b, v1.16b, v0.16b
+; CHECK-NOSVE-NEXT: dup v1.16b, w8
+; CHECK-NOSVE-NEXT: orr v0.16b, v0.16b, v1.16b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 1)
+ ret <16 x i1> %0
+}
+
+define <2 x i1> @not_whilerw_ptr32(i32 %a, i32 %b) {
+; CHECK-SVE-LABEL: not_whilerw_ptr32:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: subs w8, w1, w0
+; CHECK-SVE-NEXT: cneg w8, w8, mi
+; CHECK-SVE-NEXT: add w9, w8, #7
+; CHECK-SVE-NEXT: cmp w8, #0
+; CHECK-SVE-NEXT: csel w8, w9, w8, lt
+; CHECK-SVE-NEXT: asr w8, w8, #3
+; CHECK-SVE-NEXT: cmp w8, #0
+; CHECK-SVE-NEXT: cset w9, eq
+; CHECK-SVE-NEXT: whilelo p0.s, #0, w8
+; CHECK-SVE-NEXT: dup v0.2s, w9
+; CHECK-SVE-NEXT: mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-SVE-NEXT: orr v0.8b, v1.8b, v0.8b
+; CHECK-SVE-NEXT: ret
+;
+; CHECK-NOSVE-LABEL: not_whilerw_ptr32:
+; CHECK-NOSVE: // %bb.0: // %entry
+; CHECK-NOSVE-NEXT: subs w9, w1, w0
+; CHECK-NOSVE-NEXT: adrp x8, .LCPI9_0
+; CHECK-NOSVE-NEXT: cneg w9, w9, mi
+; CHECK-NOSVE-NEXT: ldr d1, [x8, :lo12:.LCPI9_0]
+; CHECK-NOSVE-NEXT: add w10, w9, #7
+; CHECK-NOSVE-NEXT: cmp w9, #0
+; CHECK-NOSVE-NEXT: csel w9, w10, w9, lt
+; CHECK-NOSVE-NEXT: asr w9, w9, #3
+; CHECK-NOSVE-NEXT: dup v0.2s, w9
+; CHECK-NOSVE-NEXT: cmp w9, #0
+; CHECK-NOSVE-NEXT: cset w8, eq
+; CHECK-NOSVE-NEXT: dup v2.2s, w8
+; CHECK-NOSVE-NEXT: cmhi v0.2s, v0.2s, v1.2s
+; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v2.8b
+; CHECK-NOSVE-NEXT: ret
+entry:
+ %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i32.i32(i32 %a, i32 %b, i32 8, i1 0)
+ ret <2 x i1> %0
+}
diff --git a/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
new file mode 100644
index 00000000000000..4912bc0f59d40d
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
@@ -0,0 +1,82 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s | FileCheck %s
+
+define <vscale x 16 x i1> @whilewr_8(i64 %a, i64 %b) {
+; CHECK-LABEL: whilewr_8:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilewr p0.b, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 1)
+ ret <vscale x 16 x i1> %0
+}
+
+define <vscale x 8 x i1> @whilewr_16(i64 %a, i64 %b) {
+; CHECK-LABEL: whilewr_16:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilewr p0.h, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 1)
+ ret <vscale x 8 x i1> %0
+}
+
+define <vscale x 4 x i1> @whilewr_32(i64 %a, i64 %b) {
+; CHECK-LABEL: whilewr_32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilewr p0.s, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 1)
+ ret <vscale x 4 x i1> %0
+}
+
+define <vscale x 2 x i1> @whilewr_64(i64 %a, i64 %b) {
+; CHECK-LABEL: whilewr_64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilewr p0.d, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 1)
+ ret <vscale x 2 x i1> %0
+}
+
+define <vscale x 16 x i1> @whilerw_8(i64 %a, i64 %b) {
+; CHECK-LABEL: whilerw_8:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilerw p0.b, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 0)
+ ret <vscale x 16 x i1> %0
+}
+
+define <vscale x 8 x i1> @whilerw_16(i64 %a, i64 %b) {
+; CHECK-LABEL: whilerw_16:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilerw p0.h, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 0)
+ ret <vscale x 8 x i1> %0
+}
+
+define <vscale x 4 x i1> @whilerw_32(i64 %a, i64 %b) {
+; CHECK-LABEL: whilerw_32:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilerw p0.s, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 0)
+ ret <vscale x 4 x i1> %0
+}
+
+define <vscale x 2 x i1> @whilerw_64(i64 %a, i64 %b) {
+; CHECK-LABEL: whilerw_64:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilerw p0.d, x0, x1
+; CHECK-NEXT: ret
+entry:
+ %0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 0)
+ ret <vscale x 2 x i1> %0
+}
>From 9886cfd5b1f36030c181b7d366ed9f3fbe2fbd71 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 21 Nov 2024 13:53:11 +0000
Subject: [PATCH 02/10] Fix intrinsic signature in docs
---
llvm/docs/LangRef.rst | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 52c2aba61e9027..7d2b346e3ac38d 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -23626,7 +23626,7 @@ Examples:
.. _int_experimental_get_alias_lane_mask:
-'``llvm.get.alias.lane.mask.*``' Intrinsics
+'``llvm.experimental.get.alias.lane.mask.*``' Intrinsics
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Syntax:
@@ -23635,10 +23635,10 @@ This is an overloaded intrinsic.
::
- declare <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
- declare <8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
- declare <16 x i1> @llvm.experimental.get.alias.lane.mask.v16i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
- declare <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.nxv16i1.i64(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
+ declare <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64.i64(i64 %ptrA, i64 %ptrB, i64 immarg %elementSize, i1 immarg %writeAfterRead)
+ declare <8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %ptrA, i64 %ptrB, i64 immarg %elementSize, i1 immarg %writeAfterRead)
+ declare <16 x i1> @llvm.experimental.get.alias.lane.mask.v16i1.i64.i32(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
+ declare <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.nxv16i1.i64.i32(i64 %ptrA, i64 %ptrB, i32 immarg %elementSize, i1 immarg %writeAfterRead)
Overview:
>From c7bee377f329e59ea185df5205f2cf9d52c88715 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 21 Nov 2024 14:57:06 +0000
Subject: [PATCH 03/10] Add -o - to test run lines
---
llvm/docs/LangRef.rst | 4 ++--
llvm/test/CodeGen/AArch64/alias_mask.ll | 4 ++--
llvm/test/CodeGen/AArch64/alias_mask_scalable.ll | 2 +-
3 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 7d2b346e3ac38d..62b9c14543c753 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -23673,7 +23673,7 @@ Otherwise they are semantically equivalent to:
where ``%m`` is a vector (mask) of active/inactive lanes with its elements
indexed by ``i``, and ``%ptrA``, ``%ptrB`` are the two i64 arguments to
-``llvm.experimental.get.alias.lane.mask.*``, ``%elementSize`` is the i32 argument, ``%abs`` is the absolute difference operation, ``%icmp`` is an integer compare and ``ult``
+``llvm.experimental.get.alias.lane.mask.*``, ``%elementSize`` is the first immediate argument, ``%abs`` is the absolute difference operation, ``%icmp`` is an integer compare and ``ult``
the unsigned less-than comparison operator. The subtraction between ``%ptrA`` and ``%ptrB`` could be negative. The ``%writeAfterRead`` argument is expected to be true if the ``%ptrB`` is stored to after ``%ptrA`` is read from.
The above is equivalent to:
@@ -23700,7 +23700,7 @@ Examples:
.. code-block:: llvm
- %alias.lane.mask = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64(i64 %ptrA, i64 %ptrB, i32 4, i1 1)
+ %alias.lane.mask = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64.i32(i64 %ptrA, i64 %ptrB, i32 4, i1 1)
%vecA = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* %ptrA, i32 4, <4 x i1> %alias.lane.mask, <4 x i32> poison)
[...]
call @llvm.masked.store.v4i32.p0v4i32(<4 x i32> %vecA, <4 x i32>* %ptrB, i32 4, <4 x i1> %alias.lane.mask)
diff --git a/llvm/test/CodeGen/AArch64/alias_mask.ll b/llvm/test/CodeGen/AArch64/alias_mask.ll
index 59ad2bf82db92e..84a22822f1702b 100644
--- a/llvm/test/CodeGen/AArch64/alias_mask.ll
+++ b/llvm/test/CodeGen/AArch64/alias_mask.ll
@@ -1,6 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s | FileCheck %s --check-prefix=CHECK-SVE
-; RUN: llc -mtriple=aarch64 %s | FileCheck %s --check-prefix=CHECK-NOSVE
+; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s -o - | FileCheck %s --check-prefix=CHECK-SVE
+; RUN: llc -mtriple=aarch64 %s -o - | FileCheck %s --check-prefix=CHECK-NOSVE
define <16 x i1> @whilewr_8(i64 %a, i64 %b) {
; CHECK-SVE-LABEL: whilewr_8:
diff --git a/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
index 4912bc0f59d40d..e4ef5292dee277 100644
--- a/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
+++ b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
@@ -1,5 +1,5 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s | FileCheck %s
+; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s -o - | FileCheck %s
define <vscale x 16 x i1> @whilewr_8(i64 %a, i64 %b) {
; CHECK-LABEL: whilewr_8:
>From 2a39fc427de17d4c2801ab8c0844398c41a8cc2f Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 21 Nov 2024 17:01:47 +0000
Subject: [PATCH 04/10] Format
---
llvm/include/llvm/CodeGen/TargetLowering.h | 6 +-
.../SelectionDAG/SelectionDAGBuilder.cpp | 34 ++++++-----
.../Target/AArch64/AArch64ISelLowering.cpp | 58 +++++++++++--------
llvm/lib/Target/AArch64/AArch64ISelLowering.h | 3 +-
4 files changed, 59 insertions(+), 42 deletions(-)
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 4e007863497fd1..3fe5b3d108ba10 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -468,8 +468,10 @@ class TargetLoweringBase {
return true;
}
- /// Return true if the @llvm.experimental.get.alias.lane.mask intrinsic should be expanded using generic code in SelectionDAGBuilder.
- virtual bool shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT, unsigned EltSize) const {
+ /// Return true if the @llvm.experimental.get.alias.lane.mask intrinsic should
+ /// be expanded using generic code in SelectionDAGBuilder.
+ virtual bool shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT,
+ unsigned EltSize) const {
return true;
}
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 8e4253b6b7db7b..3be7dece56e481 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -8280,17 +8280,19 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
SDValue SourceValue = getValue(I.getOperand(0));
SDValue SinkValue = getValue(I.getOperand(1));
SDValue EltSize = getValue(I.getOperand(2));
- bool IsWriteAfterRead = cast<ConstantSDNode>(getValue(I.getOperand(3)))->getZExtValue() != 0;
+ bool IsWriteAfterRead =
+ cast<ConstantSDNode>(getValue(I.getOperand(3)))->getZExtValue() != 0;
auto IntrinsicVT = EVT::getEVT(I.getType());
auto PtrVT = SourceValue->getValueType(0);
- if (!TLI.shouldExpandGetAliasLaneMask(IntrinsicVT, PtrVT, cast<ConstantSDNode>(EltSize)->getSExtValue())) {
+ if (!TLI.shouldExpandGetAliasLaneMask(
+ IntrinsicVT, PtrVT,
+ cast<ConstantSDNode>(EltSize)->getSExtValue())) {
visitTargetIntrinsic(I, Intrinsic);
return;
}
- SDValue Diff = DAG.getNode(ISD::SUB, sdl,
- PtrVT, SinkValue, SourceValue);
+ SDValue Diff = DAG.getNode(ISD::SUB, sdl, PtrVT, SinkValue, SourceValue);
if (!IsWriteAfterRead)
Diff = DAG.getNode(ISD::ABS, sdl, PtrVT, Diff);
@@ -8298,9 +8300,10 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
SDValue Zero = DAG.getTargetConstant(0, sdl, PtrVT);
// If the difference is positive then some elements may alias
- auto CmpVT = TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(),
- PtrVT);
- SDValue Cmp = DAG.getSetCC(sdl, CmpVT, Diff, Zero, IsWriteAfterRead ? ISD::SETLE : ISD::SETEQ);
+ auto CmpVT =
+ TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), PtrVT);
+ SDValue Cmp = DAG.getSetCC(sdl, CmpVT, Diff, Zero,
+ IsWriteAfterRead ? ISD::SETLE : ISD::SETEQ);
// Splat the compare result then OR it with a lane mask
SDValue Splat = DAG.getSplat(IntrinsicVT, sdl, Cmp);
@@ -8308,14 +8311,17 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
SDValue DiffMask;
// Don't emit an active lane mask if the target doesn't support it
if (TLI.shouldExpandGetActiveLaneMask(IntrinsicVT, PtrVT)) {
- EVT VecTy = EVT::getVectorVT(*DAG.getContext(), PtrVT,
- IntrinsicVT.getVectorElementCount());
- SDValue DiffSplat = DAG.getSplat(VecTy, sdl, Diff);
- SDValue VectorStep = DAG.getStepVector(sdl, VecTy);
- DiffMask = DAG.getSetCC(sdl, IntrinsicVT, VectorStep,
- DiffSplat, ISD::CondCode::SETULT);
+ EVT VecTy = EVT::getVectorVT(*DAG.getContext(), PtrVT,
+ IntrinsicVT.getVectorElementCount());
+ SDValue DiffSplat = DAG.getSplat(VecTy, sdl, Diff);
+ SDValue VectorStep = DAG.getStepVector(sdl, VecTy);
+ DiffMask = DAG.getSetCC(sdl, IntrinsicVT, VectorStep, DiffSplat,
+ ISD::CondCode::SETULT);
} else {
- DiffMask = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, sdl, IntrinsicVT, DAG.getTargetConstant(Intrinsic::get_active_lane_mask, sdl, MVT::i64), Zero, Diff);
+ DiffMask = DAG.getNode(
+ ISD::INTRINSIC_WO_CHAIN, sdl, IntrinsicVT,
+ DAG.getTargetConstant(Intrinsic::get_active_lane_mask, sdl, MVT::i64),
+ Zero, Diff);
}
SDValue Or = DAG.getNode(ISD::OR, sdl, IntrinsicVT, DiffMask, Splat);
setValue(&I, Or);
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 83fc20d9b0c226..47a005e558da49 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -2038,7 +2038,8 @@ bool AArch64TargetLowering::shouldExpandGetActiveLaneMask(EVT ResVT,
return false;
}
-bool AArch64TargetLowering::shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT, unsigned EltSize) const {
+bool AArch64TargetLowering::shouldExpandGetAliasLaneMask(
+ EVT VT, EVT PtrVT, unsigned EltSize) const {
if (!Subtarget->hasSVE2())
return true;
@@ -2047,7 +2048,7 @@ bool AArch64TargetLowering::shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT, unsi
if (VT == MVT::v2i1 || VT == MVT::nxv2i1)
return EltSize != 8;
- if( VT == MVT::v4i1 || VT == MVT::nxv4i1)
+ if (VT == MVT::v4i1 || VT == MVT::nxv4i1)
return EltSize != 4;
if (VT == MVT::v8i1 || VT == MVT::nxv8i1)
return EltSize != 2;
@@ -6040,12 +6041,14 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
case Intrinsic::aarch64_sve_whilewr_h:
case Intrinsic::aarch64_sve_whilewr_s:
case Intrinsic::aarch64_sve_whilewr_d:
- return DAG.getNode(AArch64ISD::WHILEWR, dl, Op.getValueType(), Op.getOperand(1), Op.getOperand(2));
+ return DAG.getNode(AArch64ISD::WHILEWR, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
case Intrinsic::aarch64_sve_whilerw_b:
case Intrinsic::aarch64_sve_whilerw_h:
case Intrinsic::aarch64_sve_whilerw_s:
case Intrinsic::aarch64_sve_whilerw_d:
- return DAG.getNode(AArch64ISD::WHILERW, dl, Op.getValueType(), Op.getOperand(1), Op.getOperand(2));
+ return DAG.getNode(AArch64ISD::WHILERW, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
case Intrinsic::aarch64_neon_abs: {
EVT Ty = Op.getValueType();
if (Ty == MVT::i64) {
@@ -6512,34 +6515,38 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
uint64_t EltSize = Op.getOperand(3)->getAsZExtVal();
bool IsWriteAfterRead = Op.getOperand(4)->getAsZExtVal() == 1;
switch (EltSize) {
- case 1:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_b : Intrinsic::aarch64_sve_whilerw_b;
- break;
- case 2:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_h : Intrinsic::aarch64_sve_whilerw_h;
- break;
- case 4:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_s : Intrinsic::aarch64_sve_whilerw_s;
- break;
- case 8:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_d : Intrinsic::aarch64_sve_whilerw_d;
- break;
- default:
- llvm_unreachable("Unexpected element size for get.alias.lane.mask");
- break;
+ case 1:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_b
+ : Intrinsic::aarch64_sve_whilerw_b;
+ break;
+ case 2:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_h
+ : Intrinsic::aarch64_sve_whilerw_h;
+ break;
+ case 4:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_s
+ : Intrinsic::aarch64_sve_whilerw_s;
+ break;
+ case 8:
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_d
+ : Intrinsic::aarch64_sve_whilerw_d;
+ break;
+ default:
+ llvm_unreachable("Unexpected element size for get.alias.lane.mask");
+ break;
}
}
- SDValue ID =
- DAG.getTargetConstant(IntrinsicID, dl, MVT::i64);
+ SDValue ID = DAG.getTargetConstant(IntrinsicID, dl, MVT::i64);
EVT VT = Op.getValueType();
if (VT.isScalableVector())
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT, ID, Op.getOperand(1),
Op.getOperand(2));
- // We can use the SVE whilelo/whilewr/whilerw instruction to lower this intrinsic by
- // creating the appropriate sequence of scalable vector operations and
- // then extracting a fixed-width subvector from the scalable vector.
+ // We can use the SVE whilelo/whilewr/whilerw instruction to lower this
+ // intrinsic by creating the appropriate sequence of scalable vector
+ // operations and then extracting a fixed-width subvector from the scalable
+ // vector.
EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT);
EVT WhileVT = ContainerVT.changeElementType(MVT::i1);
@@ -19927,7 +19934,8 @@ static bool isPredicateCCSettingOp(SDValue N) {
// get_active_lane_mask is lowered to a whilelo instruction.
N.getConstantOperandVal(0) == Intrinsic::get_active_lane_mask ||
// get_alias_lane_mask is lowered to a whilewr/rw instruction.
- N.getConstantOperandVal(0) == Intrinsic::experimental_get_alias_lane_mask)))
+ N.getConstantOperandVal(0) ==
+ Intrinsic::experimental_get_alias_lane_mask)))
return true;
return false;
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
index 65174498e21237..3df8f2da365f31 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -997,7 +997,8 @@ class AArch64TargetLowering : public TargetLowering {
bool shouldExpandGetActiveLaneMask(EVT VT, EVT OpVT) const override;
- bool shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT, unsigned EltSize) const override;
+ bool shouldExpandGetAliasLaneMask(EVT VT, EVT PtrVT,
+ unsigned EltSize) const override;
bool
shouldExpandPartialReductionIntrinsic(const IntrinsicInst *I) const override;
>From 79c1d4f3573d78c00b39a36cb4c4f5e9167be69c Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Tue, 26 Nov 2024 10:22:07 +0000
Subject: [PATCH 05/10] Extend underline
---
llvm/docs/LangRef.rst | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 62b9c14543c753..ad7dc65c0ee722 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -23627,7 +23627,7 @@ Examples:
.. _int_experimental_get_alias_lane_mask:
'``llvm.experimental.get.alias.lane.mask.*``' Intrinsics
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Syntax:
"""""""
>From 845e9bcb68081249744560210b188d195bf0cedf Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 28 Nov 2024 13:38:46 +0000
Subject: [PATCH 06/10] Add SVE run line
---
.../CodeGen/AArch64/alias_mask_scalable.ll | 179 ++++++++++++++----
1 file changed, 146 insertions(+), 33 deletions(-)
diff --git a/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
index e4ef5292dee277..be5ec8b2a82bf2 100644
--- a/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
+++ b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
@@ -1,81 +1,194 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
-; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s -o - | FileCheck %s
+; RUN: llc -mtriple=aarch64 -mattr=+sve2 %s -o - | FileCheck %s --check-prefix=CHECK-SVE2
+; RUN: llc -mtriple=aarch64 -mattr=+sve %s -o - | FileCheck %s --check-prefix=CHECK-SVE
define <vscale x 16 x i1> @whilewr_8(i64 %a, i64 %b) {
-; CHECK-LABEL: whilewr_8:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilewr p0.b, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilewr_8:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilewr p0.b, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilewr_8:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: cmp x8, #1
+; CHECK-SVE-NEXT: cset w9, lt
+; CHECK-SVE-NEXT: whilelo p0.b, #0, x8
+; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p1.b, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 1)
ret <vscale x 16 x i1> %0
}
define <vscale x 8 x i1> @whilewr_16(i64 %a, i64 %b) {
-; CHECK-LABEL: whilewr_16:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilewr p0.h, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilewr_16:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilewr p0.h, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilewr_16:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: add x8, x8, x8, lsr #63
+; CHECK-SVE-NEXT: asr x8, x8, #1
+; CHECK-SVE-NEXT: cmp x8, #1
+; CHECK-SVE-NEXT: cset w9, lt
+; CHECK-SVE-NEXT: whilelo p0.h, #0, x8
+; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p1.h, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 1)
ret <vscale x 8 x i1> %0
}
define <vscale x 4 x i1> @whilewr_32(i64 %a, i64 %b) {
-; CHECK-LABEL: whilewr_32:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilewr p0.s, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilewr_32:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilewr p0.s, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilewr_32:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: add x9, x8, #3
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: csel x8, x9, x8, lt
+; CHECK-SVE-NEXT: asr x8, x8, #2
+; CHECK-SVE-NEXT: cmp x8, #1
+; CHECK-SVE-NEXT: cset w9, lt
+; CHECK-SVE-NEXT: whilelo p1.s, #0, x8
+; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p0.s, xzr, x9
+; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 1)
ret <vscale x 4 x i1> %0
}
define <vscale x 2 x i1> @whilewr_64(i64 %a, i64 %b) {
-; CHECK-LABEL: whilewr_64:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilewr p0.d, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilewr_64:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilewr p0.d, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilewr_64:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: add x9, x8, #7
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: csel x8, x9, x8, lt
+; CHECK-SVE-NEXT: asr x8, x8, #3
+; CHECK-SVE-NEXT: cmp x8, #1
+; CHECK-SVE-NEXT: cset w9, lt
+; CHECK-SVE-NEXT: whilelo p1.d, #0, x8
+; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p0.d, xzr, x9
+; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 1)
ret <vscale x 2 x i1> %0
}
define <vscale x 16 x i1> @whilerw_8(i64 %a, i64 %b) {
-; CHECK-LABEL: whilerw_8:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilerw p0.b, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilerw_8:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilerw p0.b, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilerw_8:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: cneg x8, x8, mi
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: cset w9, eq
+; CHECK-SVE-NEXT: whilelo p0.b, #0, x8
+; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p1.b, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 0)
ret <vscale x 16 x i1> %0
}
define <vscale x 8 x i1> @whilerw_16(i64 %a, i64 %b) {
-; CHECK-LABEL: whilerw_16:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilerw p0.h, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilerw_16:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilerw p0.h, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilerw_16:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: cneg x8, x8, mi
+; CHECK-SVE-NEXT: add x8, x8, x8, lsr #63
+; CHECK-SVE-NEXT: asr x8, x8, #1
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: cset w9, eq
+; CHECK-SVE-NEXT: whilelo p0.h, #0, x8
+; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p1.h, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 8 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 0)
ret <vscale x 8 x i1> %0
}
define <vscale x 4 x i1> @whilerw_32(i64 %a, i64 %b) {
-; CHECK-LABEL: whilerw_32:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilerw p0.s, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilerw_32:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilerw p0.s, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilerw_32:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: cneg x8, x8, mi
+; CHECK-SVE-NEXT: add x9, x8, #3
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: csel x8, x9, x8, lt
+; CHECK-SVE-NEXT: asr x8, x8, #2
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: cset w9, eq
+; CHECK-SVE-NEXT: whilelo p1.s, #0, x8
+; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p0.s, xzr, x9
+; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 0)
ret <vscale x 4 x i1> %0
}
define <vscale x 2 x i1> @whilerw_64(i64 %a, i64 %b) {
-; CHECK-LABEL: whilerw_64:
-; CHECK: // %bb.0: // %entry
-; CHECK-NEXT: whilerw p0.d, x0, x1
-; CHECK-NEXT: ret
+; CHECK-SVE2-LABEL: whilerw_64:
+; CHECK-SVE2: // %bb.0: // %entry
+; CHECK-SVE2-NEXT: whilerw p0.d, x0, x1
+; CHECK-SVE2-NEXT: ret
+;
+; CHECK-SVE-LABEL: whilerw_64:
+; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: cneg x8, x8, mi
+; CHECK-SVE-NEXT: add x9, x8, #7
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: csel x8, x9, x8, lt
+; CHECK-SVE-NEXT: asr x8, x8, #3
+; CHECK-SVE-NEXT: cmp x8, #0
+; CHECK-SVE-NEXT: cset w9, eq
+; CHECK-SVE-NEXT: whilelo p1.d, #0, x8
+; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
+; CHECK-SVE-NEXT: whilelo p0.d, xzr, x9
+; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: ret
entry:
%0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 0)
ret <vscale x 2 x i1> %0
>From ce43e5a858af5ee7b90dfc0f6c4c405927952b09 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 28 Nov 2024 16:23:15 +0000
Subject: [PATCH 07/10] 80 columns in lang ref and remove some extra usage info
---
llvm/docs/LangRef.rst | 35 +++++++++++++++++++----------------
1 file changed, 19 insertions(+), 16 deletions(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index ad7dc65c0ee722..6c81088368c513 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -23644,7 +23644,8 @@ This is an overloaded intrinsic.
Overview:
"""""""""
-Create a mask representing lanes that do or not overlap between two pointers across one vector loop iteration.
+Create a mask representing lanes that do or not overlap between two pointers
+across one vector loop iteration.
Arguments:
@@ -23656,8 +23657,9 @@ The final two are immediates and the result is a vector with the i1 element type
Semantics:
""""""""""
-In the case that ``%writeAfterRead`` is true, the '``llvm.experimental.get.alias.lane.mask.*``' intrinsics are semantically equivalent
-to:
+In the case that ``%writeAfterRead`` is true, the
+'``llvm.experimental.get.alias.lane.mask.*``' intrinsics are semantically
+equivalent to:
::
@@ -23673,8 +23675,12 @@ Otherwise they are semantically equivalent to:
where ``%m`` is a vector (mask) of active/inactive lanes with its elements
indexed by ``i``, and ``%ptrA``, ``%ptrB`` are the two i64 arguments to
-``llvm.experimental.get.alias.lane.mask.*``, ``%elementSize`` is the first immediate argument, ``%abs`` is the absolute difference operation, ``%icmp`` is an integer compare and ``ult``
-the unsigned less-than comparison operator. The subtraction between ``%ptrA`` and ``%ptrB`` could be negative. The ``%writeAfterRead`` argument is expected to be true if the ``%ptrB`` is stored to after ``%ptrA`` is read from.
+``llvm.experimental.get.alias.lane.mask.*``, ``%elementSize`` is the first
+immediate argument, ``%abs`` is the absolute difference operation, ``%icmp`` is
+an integer compare and ``ult`` the unsigned less-than comparison operator. The
+subtraction between ``%ptrA`` and ``%ptrB`` could be negative. The
+``%writeAfterRead`` argument is expected to be true if the ``%ptrB`` is stored
+to after ``%ptrA`` is read from.
The above is equivalent to:
::
@@ -23682,17 +23688,14 @@ The above is equivalent to:
%m = @llvm.experimental.get.alias.lane.mask(%ptrA, %ptrB, %elementSize, %writeAfterRead)
This can, for example, be emitted by the loop vectorizer in which case
-``%ptrA`` is a pointer that is read from within the loop, and ``%ptrB`` is a pointer that is stored to within the loop.
-If the difference between these pointers is less than the vector factor, then they overlap (alias) within a loop iteration.
-An example is if ``%ptrA`` is 20 and ``%ptrB`` is 23 with a vector factor of 8, then lanes 3, 4, 5, 6 and 7 of the vector loaded from ``%ptrA``
-share addresses with lanes 0, 1, 2, 3, 4 and 5 from the vector stored to at ``%ptrB``.
-An alias mask of these two pointers should be <1, 1, 1, 0, 0, 0, 0, 0> so that only the non-overlapping lanes are loaded and stored.
-This operation allows many loops to be vectorised when it would otherwise be unsafe to do so.
-
-To account for the fact that only a subset of lanes have been operated on in an iteration,
-the loop's induction variable should be incremented by the popcount of the mask rather than the vector factor.
-
-This mask ``%m`` can e.g. be used in masked load/store instructions.
+``%ptrA`` is a pointer that is read from within the loop, and ``%ptrB`` is a
+pointer that is stored to within the loop.
+If the difference between these pointers is less than the vector factor, then
+they overlap (alias) within a loop iteration.
+An example is if ``%ptrA`` is 20 and ``%ptrB`` is 23 with a vector factor of 8,
+then lanes 3, 4, 5, 6 and 7 of the vector loaded from ``%ptrA``
+share addresses with lanes 0, 1, 2, 3, 4 and 5 from the vector stored to at
+``%ptrB``.
Examples:
>From 69fc569bb02cff16a2cfa1c3dfaa8d61be01fa15 Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Fri, 29 Nov 2024 15:04:55 +0000
Subject: [PATCH 08/10] Add note about poison return
---
llvm/docs/LangRef.rst | 3 +++
1 file changed, 3 insertions(+)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 6c81088368c513..6b664dc70b7239 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -23697,6 +23697,9 @@ then lanes 3, 4, 5, 6 and 7 of the vector loaded from ``%ptrA``
share addresses with lanes 0, 1, 2, 3, 4 and 5 from the vector stored to at
``%ptrB``.
+The intrinsic will return poison if ``%ptrA`` and ``%ptrB`` are within
+VF * ``%elementSize`` of each other and ``%ptrA`` + VF * ``%elementSize`` wraps.
+
Examples:
"""""""""
>From ec8f18329dc2e1fd0b86681155cb783609ece0a6 Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Mon, 2 Dec 2024 15:33:31 +0000
Subject: [PATCH 09/10] Reorganise poison note
---
llvm/docs/LangRef.rst | 20 +++++++++-----------
1 file changed, 9 insertions(+), 11 deletions(-)
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
index 6b664dc70b7239..2d17c6591bb405 100644
--- a/llvm/docs/LangRef.rst
+++ b/llvm/docs/LangRef.rst
@@ -23657,7 +23657,9 @@ The final two are immediates and the result is a vector with the i1 element type
Semantics:
""""""""""
-In the case that ``%writeAfterRead`` is true, the
+The intrinsic will return poison if ``%ptrA`` and ``%ptrB`` are within
+VF * ``%elementSize`` of each other and ``%ptrA`` + VF * ``%elementSize`` wraps.
+In other cases when ``%writeAfterRead`` is true, the
'``llvm.experimental.get.alias.lane.mask.*``' intrinsics are semantically
equivalent to:
@@ -23666,7 +23668,9 @@ equivalent to:
%diff = (%ptrB - %ptrA) / %elementSize
%m[i] = (icmp ult i, %diff) || (%diff <= 0)
-Otherwise they are semantically equivalent to:
+When the return value is not poison and ``%writeAfterRead`` is false, the
+'``llvm.experimental.get.alias.lane.mask.*``' intrinsics are semantically
+equivalent to:
::
@@ -23675,12 +23679,9 @@ Otherwise they are semantically equivalent to:
where ``%m`` is a vector (mask) of active/inactive lanes with its elements
indexed by ``i``, and ``%ptrA``, ``%ptrB`` are the two i64 arguments to
-``llvm.experimental.get.alias.lane.mask.*``, ``%elementSize`` is the first
-immediate argument, ``%abs`` is the absolute difference operation, ``%icmp`` is
-an integer compare and ``ult`` the unsigned less-than comparison operator. The
-subtraction between ``%ptrA`` and ``%ptrB`` could be negative. The
-``%writeAfterRead`` argument is expected to be true if the ``%ptrB`` is stored
-to after ``%ptrA`` is read from.
+``llvm.experimental.get.alias.lane.mask.*`` and ``%elementSize`` is the first
+immediate argument. The ``%writeAfterRead`` argument is expected to be true if
+``%ptrB`` is stored to after ``%ptrA`` is read from.
The above is equivalent to:
::
@@ -23697,9 +23698,6 @@ then lanes 3, 4, 5, 6 and 7 of the vector loaded from ``%ptrA``
share addresses with lanes 0, 1, 2, 3, 4 and 5 from the vector stored to at
``%ptrB``.
-The intrinsic will return poison if ``%ptrA`` and ``%ptrB`` are within
-VF * ``%elementSize`` of each other and ``%ptrA`` + VF * ``%elementSize`` wraps.
-
Examples:
"""""""""
>From d09333964e614da1701fcf9aaa7ecaa5ea814678 Mon Sep 17 00:00:00 2001
From: Sam Tebbs <samuel.tebbs at arm.com>
Date: Fri, 10 Jan 2025 11:37:37 +0000
Subject: [PATCH 10/10] Rework lowering location
---
llvm/include/llvm/CodeGen/ISDOpcodes.h | 5 +
llvm/include/llvm/IR/IntrinsicsAArch64.td | 10 +-
.../SelectionDAG/LegalizeIntegerTypes.cpp | 22 ++
llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h | 5 +
.../SelectionDAG/LegalizeVectorOps.cpp | 41 ++++
.../SelectionDAG/SelectionDAGBuilder.cpp | 53 +----
.../SelectionDAG/SelectionDAGDumper.cpp | 3 +
llvm/lib/CodeGen/TargetLoweringBase.cpp | 3 +
.../Target/AArch64/AArch64ISelLowering.cpp | 123 ++++++----
llvm/lib/Target/AArch64/AArch64ISelLowering.h | 1 +
llvm/test/CodeGen/AArch64/alias_mask.ll | 120 ++--------
.../CodeGen/AArch64/alias_mask_scalable.ll | 210 ++++++++++++++----
12 files changed, 358 insertions(+), 238 deletions(-)
diff --git a/llvm/include/llvm/CodeGen/ISDOpcodes.h b/llvm/include/llvm/CodeGen/ISDOpcodes.h
index fd8784a4c10034..b717a329378928 100644
--- a/llvm/include/llvm/CodeGen/ISDOpcodes.h
+++ b/llvm/include/llvm/CodeGen/ISDOpcodes.h
@@ -1484,6 +1484,11 @@ enum NodeType {
// Operands: Mask
VECTOR_FIND_LAST_ACTIVE,
+ // The `llvm.experimental.get.alias.lane.mask.*` intrinsics
+ // Operands: Load pointer, Store pointer, Element size, Write after read
+ // Output: Mask
+ EXPERIMENTAL_ALIAS_LANE_MASK,
+
// llvm.clear_cache intrinsic
// Operands: Input Chain, Start Addres, End Address
// Outputs: Output Chain
diff --git a/llvm/include/llvm/IR/IntrinsicsAArch64.td b/llvm/include/llvm/IR/IntrinsicsAArch64.td
index cc7a81e15f6609..79b8d3bfa96ba9 100644
--- a/llvm/include/llvm/IR/IntrinsicsAArch64.td
+++ b/llvm/include/llvm/IR/IntrinsicsAArch64.td
@@ -2862,14 +2862,8 @@ def int_aarch64_sve_stnt1_pn_x4 : SVE2p1_Store_PN_X4_Intrinsic;
// SVE2 - Contiguous conflict detection
//
-def int_aarch64_sve_whilerw_b : SVE2_CONFLICT_DETECT_Intrinsic;
-def int_aarch64_sve_whilerw_h : SVE2_CONFLICT_DETECT_Intrinsic;
-def int_aarch64_sve_whilerw_s : SVE2_CONFLICT_DETECT_Intrinsic;
-def int_aarch64_sve_whilerw_d : SVE2_CONFLICT_DETECT_Intrinsic;
-def int_aarch64_sve_whilewr_b : SVE2_CONFLICT_DETECT_Intrinsic;
-def int_aarch64_sve_whilewr_h : SVE2_CONFLICT_DETECT_Intrinsic;
-def int_aarch64_sve_whilewr_s : SVE2_CONFLICT_DETECT_Intrinsic;
-def int_aarch64_sve_whilewr_d : SVE2_CONFLICT_DETECT_Intrinsic;
+def int_aarch64_sve_whilerw : SVE2_CONFLICT_DETECT_Intrinsic;
+def int_aarch64_sve_whilewr : SVE2_CONFLICT_DETECT_Intrinsic;
// Scalable Matrix Extension (SME) Intrinsics
let TargetPrefix = "aarch64" in {
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index b0a624680231e9..bb5b535a06dc28 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -55,6 +55,9 @@ void DAGTypeLegalizer::PromoteIntegerResult(SDNode *N, unsigned ResNo) {
N->dump(&DAG); dbgs() << "\n";
#endif
report_fatal_error("Do not know how to promote this operator!");
+ case ISD::EXPERIMENTAL_ALIAS_LANE_MASK:
+ Res = PromoteIntRes_EXPERIMENTAL_ALIAS_LANE_MASK(N);
+ break;
case ISD::MERGE_VALUES:Res = PromoteIntRes_MERGE_VALUES(N, ResNo); break;
case ISD::AssertSext: Res = PromoteIntRes_AssertSext(N); break;
case ISD::AssertZext: Res = PromoteIntRes_AssertZext(N); break;
@@ -359,6 +362,14 @@ SDValue DAGTypeLegalizer::PromoteIntRes_MERGE_VALUES(SDNode *N,
return GetPromotedInteger(Op);
}
+SDValue
+DAGTypeLegalizer::PromoteIntRes_EXPERIMENTAL_ALIAS_LANE_MASK(SDNode *N) {
+ EVT VT = N->getValueType(0);
+ EVT NewVT = TLI.getTypeToTransformTo(*DAG.getContext(), VT);
+ return DAG.getNode(ISD::EXPERIMENTAL_ALIAS_LANE_MASK, SDLoc(N), NewVT,
+ N->ops());
+}
+
SDValue DAGTypeLegalizer::PromoteIntRes_AssertSext(SDNode *N) {
// Sign-extend the new bits, and continue the assertion.
SDValue Op = SExtPromotedInteger(N->getOperand(0));
@@ -2076,6 +2087,9 @@ bool DAGTypeLegalizer::PromoteIntegerOperand(SDNode *N, unsigned OpNo) {
case ISD::VECTOR_FIND_LAST_ACTIVE:
Res = PromoteIntOp_VECTOR_FIND_LAST_ACTIVE(N, OpNo);
break;
+ case ISD::EXPERIMENTAL_ALIAS_LANE_MASK:
+ Res = DAGTypeLegalizer::PromoteIntOp_EXPERIMENTAL_ALIAS_LANE_MASK(N, OpNo);
+ break;
}
// If the result is null, the sub-method took care of registering results etc.
@@ -2824,6 +2838,14 @@ SDValue DAGTypeLegalizer::PromoteIntOp_VECTOR_FIND_LAST_ACTIVE(SDNode *N,
return SDValue(DAG.UpdateNodeOperands(N, NewOps), 0);
}
+SDValue
+DAGTypeLegalizer::PromoteIntOp_EXPERIMENTAL_ALIAS_LANE_MASK(SDNode *N,
+ unsigned OpNo) {
+ SmallVector<SDValue, 4> NewOps(N->ops());
+ NewOps[OpNo] = GetPromotedInteger(N->getOperand(OpNo));
+ return SDValue(DAG.UpdateNodeOperands(N, NewOps), 0);
+}
+
//===----------------------------------------------------------------------===//
// Integer Result Expansion
//===----------------------------------------------------------------------===//
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index f13f70e66cfaa6..e6a8d2ed65bd90 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -379,6 +379,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
SDValue PromoteIntRes_IS_FPCLASS(SDNode *N);
SDValue PromoteIntRes_PATCHPOINT(SDNode *N);
SDValue PromoteIntRes_VECTOR_FIND_LAST_ACTIVE(SDNode *N);
+ SDValue PromoteIntRes_EXPERIMENTAL_ALIAS_LANE_MASK(SDNode *N);
// Integer Operand Promotion.
bool PromoteIntegerOperand(SDNode *N, unsigned OpNo);
@@ -429,7 +430,11 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
SDValue PromoteIntOp_VP_STRIDED(SDNode *N, unsigned OpNo);
SDValue PromoteIntOp_VP_SPLICE(SDNode *N, unsigned OpNo);
SDValue PromoteIntOp_VECTOR_HISTOGRAM(SDNode *N, unsigned OpNo);
+<<<<<<< HEAD
SDValue PromoteIntOp_VECTOR_FIND_LAST_ACTIVE(SDNode *N, unsigned OpNo);
+=======
+ SDValue PromoteIntOp_EXPERIMENTAL_ALIAS_LANE_MASK(SDNode *N, unsigned OpNo);
+>>>>>>> 24d33ca82b86 (Rework lowering location)
void SExtOrZExtPromotedOperands(SDValue &LHS, SDValue &RHS);
void PromoteSetCCOperands(SDValue &LHS,SDValue &RHS, ISD::CondCode Code);
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
index 6ad08bce44b0a4..71b9cc3c7a6716 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -138,6 +138,7 @@ class VectorLegalizer {
SDValue ExpandVP_FNEG(SDNode *Node);
SDValue ExpandVP_FABS(SDNode *Node);
SDValue ExpandVP_FCOPYSIGN(SDNode *Node);
+ SDValue ExpandEXPERIMENTAL_ALIAS_LANE_MASK(SDNode *N);
SDValue ExpandSELECT(SDNode *Node);
std::pair<SDValue, SDValue> ExpandLoad(SDNode *N);
SDValue ExpandStore(SDNode *N);
@@ -467,6 +468,7 @@ SDValue VectorLegalizer::LegalizeOp(SDValue Op) {
case ISD::VECTOR_COMPRESS:
case ISD::SCMP:
case ISD::UCMP:
+ case ISD::EXPERIMENTAL_ALIAS_LANE_MASK:
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
break;
case ISD::SMULFIX:
@@ -1233,6 +1235,9 @@ void VectorLegalizer::Expand(SDNode *Node, SmallVectorImpl<SDValue> &Results) {
case ISD::UCMP:
Results.push_back(TLI.expandCMP(Node, DAG));
return;
+ case ISD::EXPERIMENTAL_ALIAS_LANE_MASK:
+ Results.push_back(ExpandEXPERIMENTAL_ALIAS_LANE_MASK(Node));
+ return;
case ISD::FADD:
case ISD::FMUL:
@@ -1740,6 +1745,42 @@ SDValue VectorLegalizer::ExpandVP_FCOPYSIGN(SDNode *Node) {
return DAG.getNode(ISD::BITCAST, DL, VT, CopiedSign);
}
+SDValue VectorLegalizer::ExpandEXPERIMENTAL_ALIAS_LANE_MASK(SDNode *N) {
+ SDLoc DL(N);
+ SDValue SourceValue = N->getOperand(0);
+ SDValue SinkValue = N->getOperand(1);
+ SDValue EltSize = N->getOperand(2);
+
+ bool IsWriteAfterRead =
+ cast<ConstantSDNode>(N->getOperand(3))->getZExtValue() != 0;
+ auto VT = N->getValueType(0);
+ auto PtrVT = SourceValue->getValueType(0);
+
+ SDValue Diff = DAG.getNode(ISD::SUB, DL, PtrVT, SinkValue, SourceValue);
+ if (!IsWriteAfterRead)
+ Diff = DAG.getNode(ISD::ABS, DL, PtrVT, Diff);
+
+ Diff = DAG.getNode(ISD::SDIV, DL, PtrVT, Diff, EltSize);
+ SDValue Zero = DAG.getTargetConstant(0, DL, PtrVT);
+
+ // If the difference is positive then some elements may alias
+ auto CmpVT = TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(),
+ Diff.getValueType());
+ SDValue Cmp = DAG.getSetCC(DL, CmpVT, Diff, Zero,
+ IsWriteAfterRead ? ISD::SETLE : ISD::SETEQ);
+
+ EVT SplatTY =
+ EVT::getVectorVT(*DAG.getContext(), PtrVT, VT.getVectorElementCount());
+ SDValue DiffSplat = DAG.getSplat(SplatTY, DL, Diff);
+ SDValue VectorStep = DAG.getStepVector(DL, SplatTY);
+ SDValue DiffMask =
+ DAG.getSetCC(DL, VT, VectorStep, DiffSplat, ISD::CondCode::SETULT);
+
+ // Splat the compare result then OR it with a lane mask
+ SDValue Splat = DAG.getSplat(VT, DL, Cmp);
+ return DAG.getNode(ISD::OR, DL, VT, DiffMask, Splat);
+}
+
void VectorLegalizer::ExpandFP_TO_UINT(SDNode *Node,
SmallVectorImpl<SDValue> &Results) {
// Attempt to expand using TargetLowering.
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 3be7dece56e481..1813e31d2ee0cd 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -8277,54 +8277,13 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
return;
}
case Intrinsic::experimental_get_alias_lane_mask: {
- SDValue SourceValue = getValue(I.getOperand(0));
- SDValue SinkValue = getValue(I.getOperand(1));
- SDValue EltSize = getValue(I.getOperand(2));
- bool IsWriteAfterRead =
- cast<ConstantSDNode>(getValue(I.getOperand(3)))->getZExtValue() != 0;
auto IntrinsicVT = EVT::getEVT(I.getType());
- auto PtrVT = SourceValue->getValueType(0);
-
- if (!TLI.shouldExpandGetAliasLaneMask(
- IntrinsicVT, PtrVT,
- cast<ConstantSDNode>(EltSize)->getSExtValue())) {
- visitTargetIntrinsic(I, Intrinsic);
- return;
- }
-
- SDValue Diff = DAG.getNode(ISD::SUB, sdl, PtrVT, SinkValue, SourceValue);
- if (!IsWriteAfterRead)
- Diff = DAG.getNode(ISD::ABS, sdl, PtrVT, Diff);
-
- Diff = DAG.getNode(ISD::SDIV, sdl, PtrVT, Diff, EltSize);
- SDValue Zero = DAG.getTargetConstant(0, sdl, PtrVT);
-
- // If the difference is positive then some elements may alias
- auto CmpVT =
- TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), PtrVT);
- SDValue Cmp = DAG.getSetCC(sdl, CmpVT, Diff, Zero,
- IsWriteAfterRead ? ISD::SETLE : ISD::SETEQ);
-
- // Splat the compare result then OR it with a lane mask
- SDValue Splat = DAG.getSplat(IntrinsicVT, sdl, Cmp);
-
- SDValue DiffMask;
- // Don't emit an active lane mask if the target doesn't support it
- if (TLI.shouldExpandGetActiveLaneMask(IntrinsicVT, PtrVT)) {
- EVT VecTy = EVT::getVectorVT(*DAG.getContext(), PtrVT,
- IntrinsicVT.getVectorElementCount());
- SDValue DiffSplat = DAG.getSplat(VecTy, sdl, Diff);
- SDValue VectorStep = DAG.getStepVector(sdl, VecTy);
- DiffMask = DAG.getSetCC(sdl, IntrinsicVT, VectorStep, DiffSplat,
- ISD::CondCode::SETULT);
- } else {
- DiffMask = DAG.getNode(
- ISD::INTRINSIC_WO_CHAIN, sdl, IntrinsicVT,
- DAG.getTargetConstant(Intrinsic::get_active_lane_mask, sdl, MVT::i64),
- Zero, Diff);
- }
- SDValue Or = DAG.getNode(ISD::OR, sdl, IntrinsicVT, DiffMask, Splat);
- setValue(&I, Or);
+ SmallVector<SDValue, 4> Ops;
+ for (auto &Op : I.operands())
+ Ops.push_back(getValue(Op));
+ SDValue Mask =
+ DAG.getNode(ISD::EXPERIMENTAL_ALIAS_LANE_MASK, sdl, IntrinsicVT, Ops);
+ setValue(&I, Mask);
}
}
}
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
index f63c8dd3df1c83..be59ff7b055dc6 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -570,6 +570,9 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::VECTOR_FIND_LAST_ACTIVE:
return "find_last_active";
+ case ISD::EXPERIMENTAL_ALIAS_LANE_MASK:
+ return "alias_mask";
+
// Vector Predication
#define BEGIN_REGISTER_VP_SDNODE(SDID, LEGALARG, NAME, ...) \
case ISD::SDID: \
diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
index 73af0a9a714074..db245956d65348 100644
--- a/llvm/lib/CodeGen/TargetLoweringBase.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -821,6 +821,9 @@ void TargetLoweringBase::initActions() {
// Masked vector extracts default to expand.
setOperationAction(ISD::VECTOR_FIND_LAST_ACTIVE, VT, Expand);
+ // Aliasing lanes mask default to expand
+ setOperationAction(ISD::EXPERIMENTAL_ALIAS_LANE_MASK, VT, Expand);
+
// FP environment operations default to expand.
setOperationAction(ISD::GET_FPENV, VT, Expand);
setOperationAction(ISD::SET_FPENV, VT, Expand);
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 47a005e558da49..4c6e4e45774213 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1807,6 +1807,13 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
setOperationAction(ISD::INTRINSIC_WO_CHAIN, VT, Custom);
}
+ if (Subtarget->hasSVE2() || (Subtarget->hasSME() && Subtarget->isStreaming())) {
+ for (auto VT : {MVT::v2i32, MVT::v4i16, MVT::v8i8, MVT::v16i8, MVT::nxv2i1,
+ MVT::nxv4i1, MVT::nxv8i1, MVT::nxv16i1}) {
+ setOperationAction(ISD::EXPERIMENTAL_ALIAS_LANE_MASK, VT, Custom);
+ }
+ }
+
// Handle operations that are only available in non-streaming SVE mode.
if (Subtarget->isSVEAvailable()) {
for (auto VT : {MVT::nxv16i8, MVT::nxv8i16, MVT::nxv4i32, MVT::nxv2i64,
@@ -5284,6 +5291,59 @@ SDValue AArch64TargetLowering::LowerFSINCOS(SDValue Op,
static MVT getSVEContainerType(EVT ContentTy);
+SDValue AArch64TargetLowering::LowerALIAS_LANE_MASK(SDValue Op,
+ SelectionDAG &DAG) const {
+ SDLoc DL(Op);
+ unsigned IntrinsicID = 0;
+ uint64_t EltSize = Op.getOperand(2)->getAsZExtVal();
+ bool IsWriteAfterRead = Op.getOperand(3)->getAsZExtVal() == 1;
+ IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr
+ : Intrinsic::aarch64_sve_whilerw;
+ EVT VT = Op.getValueType();
+ MVT SimpleVT = VT.getSimpleVT();
+ // Make sure that the promoted mask size and element size match
+ switch (EltSize) {
+ case 1:
+ assert((SimpleVT == MVT::v16i8 || SimpleVT == MVT::nxv16i1) &&
+ "Unexpected mask or element size");
+ break;
+ case 2:
+ assert((SimpleVT == MVT::v8i8 || SimpleVT == MVT::nxv8i1) &&
+ "Unexpected mask or element size");
+ break;
+ case 4:
+ assert((SimpleVT == MVT::v4i16 || SimpleVT == MVT::nxv4i1) &&
+ "Unexpected mask or element size");
+ break;
+ case 8:
+ assert((SimpleVT == MVT::v2i32 || SimpleVT == MVT::nxv2i1) &&
+ "Unexpected mask or element size");
+ break;
+ default:
+ llvm_unreachable("Unexpected element size for get.alias.lane.mask");
+ break;
+ }
+ SDValue ID = DAG.getTargetConstant(IntrinsicID, DL, MVT::i64);
+
+ if (VT.isScalableVector())
+ return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, VT, ID, Op.getOperand(0),
+ Op.getOperand(1));
+
+ // We can use the SVE whilewr/whilerw instruction to lower this
+ // intrinsic by creating the appropriate sequence of scalable vector
+ // operations and then extracting a fixed-width subvector from the scalable
+ // vector.
+
+ EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT);
+ EVT WhileVT = ContainerVT.changeElementType(MVT::i1);
+
+ SDValue Mask = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, WhileVT, ID,
+ Op.getOperand(0), Op.getOperand(1));
+ SDValue MaskAsInt = DAG.getNode(ISD::SIGN_EXTEND, DL, ContainerVT, Mask);
+ return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, VT, MaskAsInt,
+ DAG.getVectorIdxConstant(0, DL));
+}
+
SDValue AArch64TargetLowering::LowerBITCAST(SDValue Op,
SelectionDAG &DAG) const {
EVT OpVT = Op.getValueType();
@@ -6037,16 +6097,10 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
EVT PtrVT = getPointerTy(DAG.getDataLayout());
return DAG.getNode(AArch64ISD::THREAD_POINTER, dl, PtrVT);
}
- case Intrinsic::aarch64_sve_whilewr_b:
- case Intrinsic::aarch64_sve_whilewr_h:
- case Intrinsic::aarch64_sve_whilewr_s:
- case Intrinsic::aarch64_sve_whilewr_d:
+ case Intrinsic::aarch64_sve_whilewr:
return DAG.getNode(AArch64ISD::WHILEWR, dl, Op.getValueType(),
Op.getOperand(1), Op.getOperand(2));
- case Intrinsic::aarch64_sve_whilerw_b:
- case Intrinsic::aarch64_sve_whilerw_h:
- case Intrinsic::aarch64_sve_whilerw_s:
- case Intrinsic::aarch64_sve_whilerw_d:
+ case Intrinsic::aarch64_sve_whilerw:
return DAG.getNode(AArch64ISD::WHILERW, dl, Op.getValueType(),
Op.getOperand(1), Op.getOperand(2));
case Intrinsic::aarch64_neon_abs: {
@@ -6511,31 +6565,6 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
case Intrinsic::experimental_get_alias_lane_mask:
case Intrinsic::get_active_lane_mask: {
unsigned IntrinsicID = Intrinsic::aarch64_sve_whilelo;
- if (IntNo == Intrinsic::experimental_get_alias_lane_mask) {
- uint64_t EltSize = Op.getOperand(3)->getAsZExtVal();
- bool IsWriteAfterRead = Op.getOperand(4)->getAsZExtVal() == 1;
- switch (EltSize) {
- case 1:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_b
- : Intrinsic::aarch64_sve_whilerw_b;
- break;
- case 2:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_h
- : Intrinsic::aarch64_sve_whilerw_h;
- break;
- case 4:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_s
- : Intrinsic::aarch64_sve_whilerw_s;
- break;
- case 8:
- IntrinsicID = IsWriteAfterRead ? Intrinsic::aarch64_sve_whilewr_d
- : Intrinsic::aarch64_sve_whilerw_d;
- break;
- default:
- llvm_unreachable("Unexpected element size for get.alias.lane.mask");
- break;
- }
- }
SDValue ID = DAG.getTargetConstant(IntrinsicID, dl, MVT::i64);
EVT VT = Op.getValueType();
@@ -6543,7 +6572,7 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT, ID, Op.getOperand(1),
Op.getOperand(2));
- // We can use the SVE whilelo/whilewr/whilerw instruction to lower this
+ // We can use the SVE whilelo instruction to lower this
// intrinsic by creating the appropriate sequence of scalable vector
// operations and then extracting a fixed-width subvector from the scalable
// vector.
@@ -7374,6 +7403,8 @@ SDValue AArch64TargetLowering::LowerOperation(SDValue Op,
default:
llvm_unreachable("unimplemented operand");
return SDValue();
+ case ISD::EXPERIMENTAL_ALIAS_LANE_MASK:
+ return LowerALIAS_LANE_MASK(Op, DAG);
case ISD::BITCAST:
return LowerBITCAST(Op, DAG);
case ISD::GlobalAddress:
@@ -19921,7 +19952,8 @@ static SDValue getPTest(SelectionDAG &DAG, EVT VT, SDValue Pg, SDValue Op,
AArch64CC::CondCode Cond);
static bool isPredicateCCSettingOp(SDValue N) {
- if ((N.getOpcode() == ISD::SETCC) ||
+ if ((N.getOpcode() == ISD::SETCC ||
+ N.getOpcode() == ISD::EXPERIMENTAL_ALIAS_LANE_MASK) ||
(N.getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
(N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilege ||
N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilegt ||
@@ -19932,10 +19964,7 @@ static bool isPredicateCCSettingOp(SDValue N) {
N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilels ||
N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilelt ||
// get_active_lane_mask is lowered to a whilelo instruction.
- N.getConstantOperandVal(0) == Intrinsic::get_active_lane_mask ||
- // get_alias_lane_mask is lowered to a whilewr/rw instruction.
- N.getConstantOperandVal(0) ==
- Intrinsic::experimental_get_alias_lane_mask)))
+ N.getConstantOperandVal(0) == Intrinsic::get_active_lane_mask)))
return true;
return false;
@@ -27633,6 +27662,22 @@ void AArch64TargetLowering::ReplaceNodeResults(
// CONCAT_VECTORS -- but delegate to common code for result type
// legalisation
return;
+ case ISD::EXPERIMENTAL_ALIAS_LANE_MASK: {
+ EVT VT = N->getValueType(0);
+ if (!VT.isFixedLengthVector() || VT.getVectorElementType() != MVT::i1)
+ return;
+
+ // NOTE: Only trivial type promotion is supported.
+ EVT NewVT = getTypeToTransformTo(*DAG.getContext(), VT);
+ if (NewVT.getVectorNumElements() != VT.getVectorNumElements())
+ return;
+
+ SDLoc DL(N);
+ auto V =
+ DAG.getNode(ISD::EXPERIMENTAL_ALIAS_LANE_MASK, DL, NewVT, N->ops());
+ Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, VT, V));
+ return;
+ }
case ISD::INTRINSIC_WO_CHAIN: {
EVT VT = N->getValueType(0);
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
index 3df8f2da365f31..075470a9199114 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -1214,6 +1214,7 @@ class AArch64TargetLowering : public TargetLowering {
SDValue LowerXOR(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const;
+ SDValue LowerALIAS_LANE_MASK(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerBITCAST(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerVSCALE(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const;
diff --git a/llvm/test/CodeGen/AArch64/alias_mask.ll b/llvm/test/CodeGen/AArch64/alias_mask.ll
index 84a22822f1702b..9b344f03da077b 100644
--- a/llvm/test/CodeGen/AArch64/alias_mask.ll
+++ b/llvm/test/CodeGen/AArch64/alias_mask.ll
@@ -48,10 +48,12 @@ define <16 x i1> @whilewr_8(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: uzp1 v1.8h, v2.8h, v3.8h
; CHECK-NOSVE-NEXT: uzp1 v0.16b, v1.16b, v0.16b
; CHECK-NOSVE-NEXT: dup v1.16b, w8
+; CHECK-NOSVE-NEXT: shl v0.16b, v0.16b, #7
+; CHECK-NOSVE-NEXT: cmlt v0.16b, v0.16b, #0
; CHECK-NOSVE-NEXT: orr v0.16b, v0.16b, v1.16b
; CHECK-NOSVE-NEXT: ret
entry:
- %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 1)
+ %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v16i1.i64.i64(i64 %a, i64 %b, i64 1, i1 1)
ret <16 x i1> %0
}
@@ -88,6 +90,8 @@ define <8 x i1> @whilewr_16(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: uzp1 v0.8h, v0.8h, v1.8h
; CHECK-NOSVE-NEXT: dup v1.8b, w8
; CHECK-NOSVE-NEXT: xtn v0.8b, v0.8h
+; CHECK-NOSVE-NEXT: shl v0.8b, v0.8b, #7
+; CHECK-NOSVE-NEXT: cmlt v0.8b, v0.8b, #0
; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
; CHECK-NOSVE-NEXT: ret
entry:
@@ -125,7 +129,7 @@ define <4 x i1> @whilewr_32(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
; CHECK-NOSVE-NEXT: ret
entry:
- %0 = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 1)
+ %0 = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64.i64(i64 %a, i64 %b, i64 4, i1 1)
ret <4 x i1> %0
}
@@ -155,7 +159,7 @@ define <2 x i1> @whilewr_64(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
; CHECK-NOSVE-NEXT: ret
entry:
- %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 1)
+ %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v2i1.i64.i64(i64 %a, i64 %b, i64 8, i1 1)
ret <2 x i1> %0
}
@@ -206,10 +210,12 @@ define <16 x i1> @whilerw_8(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: uzp1 v1.8h, v3.8h, v2.8h
; CHECK-NOSVE-NEXT: uzp1 v0.16b, v1.16b, v0.16b
; CHECK-NOSVE-NEXT: dup v1.16b, w8
+; CHECK-NOSVE-NEXT: shl v0.16b, v0.16b, #7
+; CHECK-NOSVE-NEXT: cmlt v0.16b, v0.16b, #0
; CHECK-NOSVE-NEXT: orr v0.16b, v0.16b, v1.16b
; CHECK-NOSVE-NEXT: ret
entry:
- %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 0)
+ %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v16i1.i64.i64(i64 %a, i64 %b, i64 1, i1 0)
ret <16 x i1> %0
}
@@ -247,6 +253,8 @@ define <8 x i1> @whilerw_16(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: uzp1 v0.8h, v0.8h, v1.8h
; CHECK-NOSVE-NEXT: dup v1.8b, w8
; CHECK-NOSVE-NEXT: xtn v0.8b, v0.8h
+; CHECK-NOSVE-NEXT: shl v0.8b, v0.8b, #7
+; CHECK-NOSVE-NEXT: cmlt v0.8b, v0.8b, #0
; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
; CHECK-NOSVE-NEXT: ret
entry:
@@ -285,7 +293,7 @@ define <4 x i1> @whilerw_32(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
; CHECK-NOSVE-NEXT: ret
entry:
- %0 = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 0)
+ %0 = call <4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64.i64(i64 %a, i64 %b, i64 4, i1 0)
ret <4 x i1> %0
}
@@ -316,106 +324,6 @@ define <2 x i1> @whilerw_64(i64 %a, i64 %b) {
; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v1.8b
; CHECK-NOSVE-NEXT: ret
entry:
- %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 0)
- ret <2 x i1> %0
-}
-
-define <16 x i1> @not_whilewr_wrong_eltsize(i64 %a, i64 %b) {
-; CHECK-SVE-LABEL: not_whilewr_wrong_eltsize:
-; CHECK-SVE: // %bb.0: // %entry
-; CHECK-SVE-NEXT: sub x8, x1, x0
-; CHECK-SVE-NEXT: add x8, x8, x8, lsr #63
-; CHECK-SVE-NEXT: asr x8, x8, #1
-; CHECK-SVE-NEXT: cmp x8, #1
-; CHECK-SVE-NEXT: cset w9, lt
-; CHECK-SVE-NEXT: whilelo p0.b, #0, x8
-; CHECK-SVE-NEXT: dup v0.16b, w9
-; CHECK-SVE-NEXT: mov z1.b, p0/z, #-1 // =0xffffffffffffffff
-; CHECK-SVE-NEXT: orr v0.16b, v1.16b, v0.16b
-; CHECK-SVE-NEXT: ret
-;
-; CHECK-NOSVE-LABEL: not_whilewr_wrong_eltsize:
-; CHECK-NOSVE: // %bb.0: // %entry
-; CHECK-NOSVE-NEXT: sub x8, x1, x0
-; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_0
-; CHECK-NOSVE-NEXT: adrp x10, .LCPI8_1
-; CHECK-NOSVE-NEXT: add x8, x8, x8, lsr #63
-; CHECK-NOSVE-NEXT: ldr q0, [x9, :lo12:.LCPI8_0]
-; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_2
-; CHECK-NOSVE-NEXT: ldr q2, [x9, :lo12:.LCPI8_2]
-; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_4
-; CHECK-NOSVE-NEXT: ldr q1, [x10, :lo12:.LCPI8_1]
-; CHECK-NOSVE-NEXT: asr x8, x8, #1
-; CHECK-NOSVE-NEXT: adrp x10, .LCPI8_3
-; CHECK-NOSVE-NEXT: ldr q5, [x9, :lo12:.LCPI8_4]
-; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_6
-; CHECK-NOSVE-NEXT: ldr q3, [x10, :lo12:.LCPI8_3]
-; CHECK-NOSVE-NEXT: adrp x10, .LCPI8_5
-; CHECK-NOSVE-NEXT: dup v4.2d, x8
-; CHECK-NOSVE-NEXT: ldr q7, [x9, :lo12:.LCPI8_6]
-; CHECK-NOSVE-NEXT: adrp x9, .LCPI8_7
-; CHECK-NOSVE-NEXT: ldr q6, [x10, :lo12:.LCPI8_5]
-; CHECK-NOSVE-NEXT: ldr q16, [x9, :lo12:.LCPI8_7]
-; CHECK-NOSVE-NEXT: cmp x8, #1
-; CHECK-NOSVE-NEXT: cset w8, lt
-; CHECK-NOSVE-NEXT: cmhi v0.2d, v4.2d, v0.2d
-; CHECK-NOSVE-NEXT: cmhi v1.2d, v4.2d, v1.2d
-; CHECK-NOSVE-NEXT: cmhi v2.2d, v4.2d, v2.2d
-; CHECK-NOSVE-NEXT: cmhi v3.2d, v4.2d, v3.2d
-; CHECK-NOSVE-NEXT: cmhi v5.2d, v4.2d, v5.2d
-; CHECK-NOSVE-NEXT: cmhi v6.2d, v4.2d, v6.2d
-; CHECK-NOSVE-NEXT: cmhi v7.2d, v4.2d, v7.2d
-; CHECK-NOSVE-NEXT: cmhi v4.2d, v4.2d, v16.2d
-; CHECK-NOSVE-NEXT: uzp1 v0.4s, v1.4s, v0.4s
-; CHECK-NOSVE-NEXT: uzp1 v1.4s, v3.4s, v2.4s
-; CHECK-NOSVE-NEXT: uzp1 v2.4s, v6.4s, v5.4s
-; CHECK-NOSVE-NEXT: uzp1 v3.4s, v4.4s, v7.4s
-; CHECK-NOSVE-NEXT: uzp1 v0.8h, v1.8h, v0.8h
-; CHECK-NOSVE-NEXT: uzp1 v1.8h, v3.8h, v2.8h
-; CHECK-NOSVE-NEXT: uzp1 v0.16b, v1.16b, v0.16b
-; CHECK-NOSVE-NEXT: dup v1.16b, w8
-; CHECK-NOSVE-NEXT: orr v0.16b, v0.16b, v1.16b
-; CHECK-NOSVE-NEXT: ret
-entry:
- %0 = call <16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 2, i1 1)
- ret <16 x i1> %0
-}
-
-define <2 x i1> @not_whilerw_ptr32(i32 %a, i32 %b) {
-; CHECK-SVE-LABEL: not_whilerw_ptr32:
-; CHECK-SVE: // %bb.0: // %entry
-; CHECK-SVE-NEXT: subs w8, w1, w0
-; CHECK-SVE-NEXT: cneg w8, w8, mi
-; CHECK-SVE-NEXT: add w9, w8, #7
-; CHECK-SVE-NEXT: cmp w8, #0
-; CHECK-SVE-NEXT: csel w8, w9, w8, lt
-; CHECK-SVE-NEXT: asr w8, w8, #3
-; CHECK-SVE-NEXT: cmp w8, #0
-; CHECK-SVE-NEXT: cset w9, eq
-; CHECK-SVE-NEXT: whilelo p0.s, #0, w8
-; CHECK-SVE-NEXT: dup v0.2s, w9
-; CHECK-SVE-NEXT: mov z1.s, p0/z, #-1 // =0xffffffffffffffff
-; CHECK-SVE-NEXT: orr v0.8b, v1.8b, v0.8b
-; CHECK-SVE-NEXT: ret
-;
-; CHECK-NOSVE-LABEL: not_whilerw_ptr32:
-; CHECK-NOSVE: // %bb.0: // %entry
-; CHECK-NOSVE-NEXT: subs w9, w1, w0
-; CHECK-NOSVE-NEXT: adrp x8, .LCPI9_0
-; CHECK-NOSVE-NEXT: cneg w9, w9, mi
-; CHECK-NOSVE-NEXT: ldr d1, [x8, :lo12:.LCPI9_0]
-; CHECK-NOSVE-NEXT: add w10, w9, #7
-; CHECK-NOSVE-NEXT: cmp w9, #0
-; CHECK-NOSVE-NEXT: csel w9, w10, w9, lt
-; CHECK-NOSVE-NEXT: asr w9, w9, #3
-; CHECK-NOSVE-NEXT: dup v0.2s, w9
-; CHECK-NOSVE-NEXT: cmp w9, #0
-; CHECK-NOSVE-NEXT: cset w8, eq
-; CHECK-NOSVE-NEXT: dup v2.2s, w8
-; CHECK-NOSVE-NEXT: cmhi v0.2s, v0.2s, v1.2s
-; CHECK-NOSVE-NEXT: orr v0.8b, v0.8b, v2.8b
-; CHECK-NOSVE-NEXT: ret
-entry:
- %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i32.i32(i32 %a, i32 %b, i32 8, i1 0)
+ %0 = call <2 x i1> @llvm.experimental.get.alias.lane.mask.v2i1.i64.i64(i64 %a, i64 %b, i64 8, i1 0)
ret <2 x i1> %0
}
diff --git a/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
index be5ec8b2a82bf2..a7c9c5e3cdd33b 100644
--- a/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
+++ b/llvm/test/CodeGen/AArch64/alias_mask_scalable.ll
@@ -10,16 +10,57 @@ define <vscale x 16 x i1> @whilewr_8(i64 %a, i64 %b) {
;
; CHECK-SVE-LABEL: whilewr_8:
; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-SVE-NEXT: addvl sp, sp, #-1
+; CHECK-SVE-NEXT: str p7, [sp, #4, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: str p6, [sp, #5, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: str p5, [sp, #6, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: str p4, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x08, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 8 * VG
+; CHECK-SVE-NEXT: .cfi_offset w29, -16
+; CHECK-SVE-NEXT: index z0.d, #0, #1
; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: ptrue p0.d
+; CHECK-SVE-NEXT: mov z2.d, x8
+; CHECK-SVE-NEXT: mov z1.d, z0.d
+; CHECK-SVE-NEXT: mov z3.d, z0.d
+; CHECK-SVE-NEXT: cmphi p1.d, p0/z, z2.d, z0.d
+; CHECK-SVE-NEXT: incd z0.d, all, mul #4
+; CHECK-SVE-NEXT: incd z1.d
+; CHECK-SVE-NEXT: incd z3.d, all, mul #2
+; CHECK-SVE-NEXT: cmphi p5.d, p0/z, z2.d, z0.d
+; CHECK-SVE-NEXT: mov z4.d, z1.d
+; CHECK-SVE-NEXT: cmphi p2.d, p0/z, z2.d, z1.d
+; CHECK-SVE-NEXT: incd z1.d, all, mul #4
+; CHECK-SVE-NEXT: cmphi p3.d, p0/z, z2.d, z3.d
+; CHECK-SVE-NEXT: incd z3.d, all, mul #4
+; CHECK-SVE-NEXT: incd z4.d, all, mul #2
+; CHECK-SVE-NEXT: cmphi p6.d, p0/z, z2.d, z1.d
+; CHECK-SVE-NEXT: cmphi p7.d, p0/z, z2.d, z3.d
+; CHECK-SVE-NEXT: uzp1 p1.s, p1.s, p2.s
+; CHECK-SVE-NEXT: cmphi p4.d, p0/z, z2.d, z4.d
+; CHECK-SVE-NEXT: incd z4.d, all, mul #4
+; CHECK-SVE-NEXT: uzp1 p2.s, p5.s, p6.s
+; CHECK-SVE-NEXT: ldr p6, [sp, #5, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: ldr p5, [sp, #6, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z2.d, z4.d
+; CHECK-SVE-NEXT: uzp1 p3.s, p3.s, p4.s
; CHECK-SVE-NEXT: cmp x8, #1
-; CHECK-SVE-NEXT: cset w9, lt
-; CHECK-SVE-NEXT: whilelo p0.b, #0, x8
-; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: ldr p4, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: cset w8, lt
+; CHECK-SVE-NEXT: uzp1 p1.h, p1.h, p3.h
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: uzp1 p0.s, p7.s, p0.s
+; CHECK-SVE-NEXT: ldr p7, [sp, #4, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: uzp1 p0.h, p2.h, p0.h
+; CHECK-SVE-NEXT: uzp1 p0.b, p1.b, p0.b
; CHECK-SVE-NEXT: whilelo p1.b, xzr, x8
; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
+; CHECK-SVE-NEXT: addvl sp, sp, #1
+; CHECK-SVE-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-SVE-NEXT: ret
entry:
- %0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 1)
+ %0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v16i1.i64.i64(i64 %a, i64 %b, i64 1, i1 1)
ret <vscale x 16 x i1> %0
}
@@ -31,13 +72,28 @@ define <vscale x 8 x i1> @whilewr_16(i64 %a, i64 %b) {
;
; CHECK-SVE-LABEL: whilewr_16:
; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: index z0.d, #0, #1
; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: ptrue p0.d
; CHECK-SVE-NEXT: add x8, x8, x8, lsr #63
; CHECK-SVE-NEXT: asr x8, x8, #1
+; CHECK-SVE-NEXT: mov z1.d, z0.d
+; CHECK-SVE-NEXT: mov z2.d, z0.d
+; CHECK-SVE-NEXT: mov z3.d, x8
+; CHECK-SVE-NEXT: incd z1.d
+; CHECK-SVE-NEXT: incd z2.d, all, mul #2
+; CHECK-SVE-NEXT: cmphi p1.d, p0/z, z3.d, z0.d
+; CHECK-SVE-NEXT: mov z4.d, z1.d
+; CHECK-SVE-NEXT: cmphi p2.d, p0/z, z3.d, z1.d
+; CHECK-SVE-NEXT: cmphi p3.d, p0/z, z3.d, z2.d
+; CHECK-SVE-NEXT: incd z4.d, all, mul #2
+; CHECK-SVE-NEXT: uzp1 p1.s, p1.s, p2.s
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z3.d, z4.d
; CHECK-SVE-NEXT: cmp x8, #1
-; CHECK-SVE-NEXT: cset w9, lt
-; CHECK-SVE-NEXT: whilelo p0.h, #0, x8
-; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: cset w8, lt
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: uzp1 p0.s, p3.s, p0.s
+; CHECK-SVE-NEXT: uzp1 p0.h, p1.h, p0.h
; CHECK-SVE-NEXT: whilelo p1.h, xzr, x8
; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
; CHECK-SVE-NEXT: ret
@@ -54,20 +110,27 @@ define <vscale x 4 x i1> @whilewr_32(i64 %a, i64 %b) {
;
; CHECK-SVE-LABEL: whilewr_32:
; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: index z0.d, #0, #1
; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: ptrue p0.d
; CHECK-SVE-NEXT: add x9, x8, #3
; CHECK-SVE-NEXT: cmp x8, #0
; CHECK-SVE-NEXT: csel x8, x9, x8, lt
; CHECK-SVE-NEXT: asr x8, x8, #2
+; CHECK-SVE-NEXT: mov z1.d, z0.d
+; CHECK-SVE-NEXT: mov z2.d, x8
+; CHECK-SVE-NEXT: incd z1.d
+; CHECK-SVE-NEXT: cmphi p1.d, p0/z, z2.d, z0.d
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z2.d, z1.d
; CHECK-SVE-NEXT: cmp x8, #1
-; CHECK-SVE-NEXT: cset w9, lt
-; CHECK-SVE-NEXT: whilelo p1.s, #0, x8
-; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
-; CHECK-SVE-NEXT: whilelo p0.s, xzr, x9
-; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: cset w8, lt
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: uzp1 p0.s, p1.s, p0.s
+; CHECK-SVE-NEXT: whilelo p1.s, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
; CHECK-SVE-NEXT: ret
entry:
- %0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 1)
+ %0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64.i64(i64 %a, i64 %b, i64 4, i1 1)
ret <vscale x 4 x i1> %0
}
@@ -80,19 +143,22 @@ define <vscale x 2 x i1> @whilewr_64(i64 %a, i64 %b) {
; CHECK-SVE-LABEL: whilewr_64:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: sub x8, x1, x0
+; CHECK-SVE-NEXT: index z0.d, #0, #1
+; CHECK-SVE-NEXT: ptrue p0.d
; CHECK-SVE-NEXT: add x9, x8, #7
; CHECK-SVE-NEXT: cmp x8, #0
; CHECK-SVE-NEXT: csel x8, x9, x8, lt
; CHECK-SVE-NEXT: asr x8, x8, #3
+; CHECK-SVE-NEXT: mov z1.d, x8
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z1.d, z0.d
; CHECK-SVE-NEXT: cmp x8, #1
-; CHECK-SVE-NEXT: cset w9, lt
-; CHECK-SVE-NEXT: whilelo p1.d, #0, x8
-; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
-; CHECK-SVE-NEXT: whilelo p0.d, xzr, x9
-; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: cset w8, lt
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: whilelo p1.d, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
; CHECK-SVE-NEXT: ret
entry:
- %0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 1)
+ %0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v2i1.i64.i64(i64 %a, i64 %b, i64 8, i1 1)
ret <vscale x 2 x i1> %0
}
@@ -104,17 +170,60 @@ define <vscale x 16 x i1> @whilerw_8(i64 %a, i64 %b) {
;
; CHECK-SVE-LABEL: whilerw_8:
; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-SVE-NEXT: addvl sp, sp, #-1
+; CHECK-SVE-NEXT: str p7, [sp, #4, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: str p6, [sp, #5, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: str p5, [sp, #6, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: str p4, [sp, #7, mul vl] // 2-byte Folded Spill
+; CHECK-SVE-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x08, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 8 * VG
+; CHECK-SVE-NEXT: .cfi_offset w29, -16
+; CHECK-SVE-NEXT: index z0.d, #0, #1
; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: ptrue p0.d
; CHECK-SVE-NEXT: cneg x8, x8, mi
+; CHECK-SVE-NEXT: mov z1.d, x8
+; CHECK-SVE-NEXT: mov z2.d, z0.d
+; CHECK-SVE-NEXT: mov z4.d, z0.d
+; CHECK-SVE-NEXT: mov z5.d, z0.d
+; CHECK-SVE-NEXT: cmphi p2.d, p0/z, z1.d, z0.d
+; CHECK-SVE-NEXT: incd z2.d
+; CHECK-SVE-NEXT: incd z4.d, all, mul #2
+; CHECK-SVE-NEXT: incd z5.d, all, mul #4
+; CHECK-SVE-NEXT: mov z3.d, z2.d
+; CHECK-SVE-NEXT: cmphi p1.d, p0/z, z1.d, z2.d
+; CHECK-SVE-NEXT: incd z2.d, all, mul #4
+; CHECK-SVE-NEXT: cmphi p3.d, p0/z, z1.d, z4.d
+; CHECK-SVE-NEXT: incd z4.d, all, mul #4
+; CHECK-SVE-NEXT: cmphi p4.d, p0/z, z1.d, z5.d
+; CHECK-SVE-NEXT: incd z3.d, all, mul #2
+; CHECK-SVE-NEXT: cmphi p5.d, p0/z, z1.d, z2.d
+; CHECK-SVE-NEXT: cmphi p7.d, p0/z, z1.d, z4.d
+; CHECK-SVE-NEXT: uzp1 p1.s, p2.s, p1.s
+; CHECK-SVE-NEXT: mov z0.d, z3.d
+; CHECK-SVE-NEXT: cmphi p6.d, p0/z, z1.d, z3.d
+; CHECK-SVE-NEXT: uzp1 p2.s, p4.s, p5.s
+; CHECK-SVE-NEXT: ldr p5, [sp, #6, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: ldr p4, [sp, #7, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: incd z0.d, all, mul #4
+; CHECK-SVE-NEXT: uzp1 p3.s, p3.s, p6.s
+; CHECK-SVE-NEXT: ldr p6, [sp, #5, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z1.d, z0.d
+; CHECK-SVE-NEXT: uzp1 p1.h, p1.h, p3.h
; CHECK-SVE-NEXT: cmp x8, #0
-; CHECK-SVE-NEXT: cset w9, eq
-; CHECK-SVE-NEXT: whilelo p0.b, #0, x8
-; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: cset w8, eq
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: uzp1 p0.s, p7.s, p0.s
+; CHECK-SVE-NEXT: ldr p7, [sp, #4, mul vl] // 2-byte Folded Reload
+; CHECK-SVE-NEXT: uzp1 p0.h, p2.h, p0.h
+; CHECK-SVE-NEXT: uzp1 p0.b, p1.b, p0.b
; CHECK-SVE-NEXT: whilelo p1.b, xzr, x8
; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
+; CHECK-SVE-NEXT: addvl sp, sp, #1
+; CHECK-SVE-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-SVE-NEXT: ret
entry:
- %0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 1, i1 0)
+ %0 = call <vscale x 16 x i1> @llvm.experimental.get.alias.lane.mask.v16i1.i64.i64(i64 %a, i64 %b, i64 1, i1 0)
ret <vscale x 16 x i1> %0
}
@@ -126,14 +235,29 @@ define <vscale x 8 x i1> @whilerw_16(i64 %a, i64 %b) {
;
; CHECK-SVE-LABEL: whilerw_16:
; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: index z0.d, #0, #1
; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: ptrue p0.d
; CHECK-SVE-NEXT: cneg x8, x8, mi
; CHECK-SVE-NEXT: add x8, x8, x8, lsr #63
+; CHECK-SVE-NEXT: mov z1.d, z0.d
+; CHECK-SVE-NEXT: mov z2.d, z0.d
; CHECK-SVE-NEXT: asr x8, x8, #1
+; CHECK-SVE-NEXT: mov z3.d, x8
+; CHECK-SVE-NEXT: incd z1.d
+; CHECK-SVE-NEXT: incd z2.d, all, mul #2
+; CHECK-SVE-NEXT: cmphi p1.d, p0/z, z3.d, z0.d
+; CHECK-SVE-NEXT: mov z4.d, z1.d
+; CHECK-SVE-NEXT: cmphi p2.d, p0/z, z3.d, z1.d
+; CHECK-SVE-NEXT: cmphi p3.d, p0/z, z3.d, z2.d
+; CHECK-SVE-NEXT: incd z4.d, all, mul #2
+; CHECK-SVE-NEXT: uzp1 p1.s, p1.s, p2.s
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z3.d, z4.d
; CHECK-SVE-NEXT: cmp x8, #0
-; CHECK-SVE-NEXT: cset w9, eq
-; CHECK-SVE-NEXT: whilelo p0.h, #0, x8
-; CHECK-SVE-NEXT: sbfx x8, x9, #0, #1
+; CHECK-SVE-NEXT: cset w8, eq
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: uzp1 p0.s, p3.s, p0.s
+; CHECK-SVE-NEXT: uzp1 p0.h, p1.h, p0.h
; CHECK-SVE-NEXT: whilelo p1.h, xzr, x8
; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
; CHECK-SVE-NEXT: ret
@@ -150,21 +274,28 @@ define <vscale x 4 x i1> @whilerw_32(i64 %a, i64 %b) {
;
; CHECK-SVE-LABEL: whilerw_32:
; CHECK-SVE: // %bb.0: // %entry
+; CHECK-SVE-NEXT: index z0.d, #0, #1
; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: ptrue p0.d
; CHECK-SVE-NEXT: cneg x8, x8, mi
; CHECK-SVE-NEXT: add x9, x8, #3
; CHECK-SVE-NEXT: cmp x8, #0
; CHECK-SVE-NEXT: csel x8, x9, x8, lt
+; CHECK-SVE-NEXT: mov z1.d, z0.d
; CHECK-SVE-NEXT: asr x8, x8, #2
+; CHECK-SVE-NEXT: mov z2.d, x8
+; CHECK-SVE-NEXT: incd z1.d
+; CHECK-SVE-NEXT: cmphi p1.d, p0/z, z2.d, z0.d
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z2.d, z1.d
; CHECK-SVE-NEXT: cmp x8, #0
-; CHECK-SVE-NEXT: cset w9, eq
-; CHECK-SVE-NEXT: whilelo p1.s, #0, x8
-; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
-; CHECK-SVE-NEXT: whilelo p0.s, xzr, x9
-; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: cset w8, eq
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: uzp1 p0.s, p1.s, p0.s
+; CHECK-SVE-NEXT: whilelo p1.s, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
; CHECK-SVE-NEXT: ret
entry:
- %0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 4, i1 0)
+ %0 = call <vscale x 4 x i1> @llvm.experimental.get.alias.lane.mask.v4i1.i64.i64(i64 %a, i64 %b, i64 4, i1 0)
ret <vscale x 4 x i1> %0
}
@@ -177,19 +308,22 @@ define <vscale x 2 x i1> @whilerw_64(i64 %a, i64 %b) {
; CHECK-SVE-LABEL: whilerw_64:
; CHECK-SVE: // %bb.0: // %entry
; CHECK-SVE-NEXT: subs x8, x1, x0
+; CHECK-SVE-NEXT: index z0.d, #0, #1
+; CHECK-SVE-NEXT: ptrue p0.d
; CHECK-SVE-NEXT: cneg x8, x8, mi
; CHECK-SVE-NEXT: add x9, x8, #7
; CHECK-SVE-NEXT: cmp x8, #0
; CHECK-SVE-NEXT: csel x8, x9, x8, lt
; CHECK-SVE-NEXT: asr x8, x8, #3
+; CHECK-SVE-NEXT: mov z1.d, x8
+; CHECK-SVE-NEXT: cmphi p0.d, p0/z, z1.d, z0.d
; CHECK-SVE-NEXT: cmp x8, #0
-; CHECK-SVE-NEXT: cset w9, eq
-; CHECK-SVE-NEXT: whilelo p1.d, #0, x8
-; CHECK-SVE-NEXT: sbfx x9, x9, #0, #1
-; CHECK-SVE-NEXT: whilelo p0.d, xzr, x9
-; CHECK-SVE-NEXT: mov p0.b, p1/m, p1.b
+; CHECK-SVE-NEXT: cset w8, eq
+; CHECK-SVE-NEXT: sbfx x8, x8, #0, #1
+; CHECK-SVE-NEXT: whilelo p1.d, xzr, x8
+; CHECK-SVE-NEXT: sel p0.b, p0, p0.b, p1.b
; CHECK-SVE-NEXT: ret
entry:
- %0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v8i1.i64.i64(i64 %a, i64 %b, i64 8, i1 0)
+ %0 = call <vscale x 2 x i1> @llvm.experimental.get.alias.lane.mask.v2i1.i64.i64(i64 %a, i64 %b, i64 8, i1 0)
ret <vscale x 2 x i1> %0
}
More information about the llvm-commits
mailing list