[AArch64][SVE] Implement unpack intrinsics

[kmclaughlin]

Summary

Implements the following intrinsics:

• int_aarch64_sve_sunpkhi
• int_aarch64_sve_sunpklo
• int_aarch64_sve_uunpkhi
• int_aarch64_sve_uunpklo

This patch also adds AArch64ISD nodes for UNPK instead of implementing
the intrinsics directly, as they are required for a future patch which
implements the sign/zero extension of legal vectors.

This patch includes tests for the Subdivide2Argument type added by D67549

[greened]

lib/Target/AArch64/SVEInstrFormats.td

836

Where is op used? I assume that comes later but it would help to understand where this is going.

[kmclaughlin]

Removed unused SDPatternOperator op from sve_int_perm_unpk class

[kmclaughlin]

lib/Target/AArch64/SVEInstrFormats.td

836

Thanks for pointing this out, op isn't actually used here!

[greened]

LGTM.

include/llvm/IR/IntrinsicsAArch64.td

let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
 
   class AdvSIMD_SVE_CNT_Intrinsic
     : Intrinsic<[LLVMVectorOfBitcastsToInt<0>],
                 [LLVMVectorOfBitcastsToInt<0>,
                  LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
                  llvm_anyvector_ty],
                 [IntrNoMem]>;
 
+  class AdvSIMD_SVE_Unpack_Intrinsic
+   : Intrinsic<[llvm_anyvector_ty],
+               [LLVMSubdivide2VectorType<0>],
+               [IntrNoMem]>;
+
   class AdvSIMD_SVE_PUNPKHI_Intrinsic
     : Intrinsic<[LLVMHalfElementsVectorType<0>],
                 [llvm_anyvector_ty],
                 [IntrNoMem]>;
 
   // This class of intrinsics are not intended to be useful within LLVM IR but
   // are instead here to support some of the more regid parts of the ACLE.
   class Builtin_SVCVT<string name, LLVMType OUT, LLVMType IN>

 
 //
 // Counting bits
 //
 
 def int_aarch64_sve_cnt : AdvSIMD_SVE_CNT_Intrinsic;
 
 //
+// Permutations and selection
+//
+
+def int_aarch64_sve_sunpkhi   : AdvSIMD_SVE_Unpack_Intrinsic;
+def int_aarch64_sve_sunpklo   : AdvSIMD_SVE_Unpack_Intrinsic;
+
+def int_aarch64_sve_uunpkhi   : AdvSIMD_SVE_Unpack_Intrinsic;
+def int_aarch64_sve_uunpklo   : AdvSIMD_SVE_Unpack_Intrinsic;
+
+//
 // Floating-point comparisons
 //
 
 def int_aarch64_sve_fcvtzs_i32f16 : Builtin_SVCVT<"svcvt_s32_f16_m", llvm_nxv4i32_ty, llvm_nxv8f16_ty>;
 
 //
 // Predicate operations
 //
 
 def int_aarch64_sve_punpkhi : AdvSIMD_SVE_PUNPKHI_Intrinsic;
 def int_aarch64_sve_punpklo : AdvSIMD_SVE_PUNPKHI_Intrinsic;
 }

lib/Target/AArch64/AArch64ISelLowering.h

enum NodeType : unsigned {
 
   SMULL,
   UMULL,
 
   // Reciprocal estimates and steps.
   FRECPE, FRECPS,
   FRSQRTE, FRSQRTS,
 
+  SUNPKHI,
+  SUNPKLO,
+  UUNPKHI,
+  UUNPKLO,
+
   // NEON Load/Store with post-increment base updates
   LD2post = ISD::FIRST_TARGET_MEMORY_OPCODE,
   LD3post,
   LD4post,
   ST2post,
   ST3post,
   ST4post,
   LD1x2post,

lib/Target/AArch64/AArch64ISelLowering.cpp

const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case AArch64ISD::FRECPE:            return "AArch64ISD::FRECPE";
   case AArch64ISD::FRECPS:            return "AArch64ISD::FRECPS";
   case AArch64ISD::FRSQRTE:           return "AArch64ISD::FRSQRTE";
   case AArch64ISD::FRSQRTS:           return "AArch64ISD::FRSQRTS";
   case AArch64ISD::STG:               return "AArch64ISD::STG";
   case AArch64ISD::STZG:              return "AArch64ISD::STZG";
   case AArch64ISD::ST2G:              return "AArch64ISD::ST2G";
   case AArch64ISD::STZ2G:             return "AArch64ISD::STZ2G";
+  case AArch64ISD::SUNPKHI:           return "AArch64ISD::SUNPKHI";
+  case AArch64ISD::SUNPKLO:           return "AArch64ISD::SUNPKLO";
+  case AArch64ISD::UUNPKHI:           return "AArch64ISD::UUNPKHI";
+  case AArch64ISD::UUNPKLO:           return "AArch64ISD::UUNPKLO";
   }
   return nullptr;
 }
 
 MachineBasicBlock *
 AArch64TargetLowering::EmitF128CSEL(MachineInstr &MI,
                                     MachineBasicBlock *MBB) const {
   // We materialise the F128CSEL pseudo-instruction as some control flow and a
return DAG.getNode(ISD::UMAX, dl, Op.getValueType(),
                        Op.getOperand(1), Op.getOperand(2));
   case Intrinsic::aarch64_neon_smin:
     return DAG.getNode(ISD::SMIN, dl, Op.getValueType(),
                        Op.getOperand(1), Op.getOperand(2));
   case Intrinsic::aarch64_neon_umin:
     return DAG.getNode(ISD::UMIN, dl, Op.getValueType(),
                        Op.getOperand(1), Op.getOperand(2));
 
+  case Intrinsic::aarch64_sve_sunpkhi:
+    return DAG.getNode(AArch64ISD::SUNPKHI, dl, Op.getValueType(),
+                       Op.getOperand(1));
+  case Intrinsic::aarch64_sve_sunpklo:
+    return DAG.getNode(AArch64ISD::SUNPKLO, dl, Op.getValueType(),
+                       Op.getOperand(1));
+  case Intrinsic::aarch64_sve_uunpkhi:
+    return DAG.getNode(AArch64ISD::UUNPKHI, dl, Op.getValueType(),
+                       Op.getOperand(1));
+  case Intrinsic::aarch64_sve_uunpklo:
+    return DAG.getNode(AArch64ISD::UUNPKLO, dl, Op.getValueType(),
+                       Op.getOperand(1));
+
   case Intrinsic::localaddress: {
     const auto &MF = DAG.getMachineFunction();
     const auto *RegInfo = Subtarget->getRegisterInfo();
     unsigned Reg = RegInfo->getLocalAddressRegister(MF);
     return DAG.getCopyFromReg(DAG.getEntryNode(), dl, Reg,
                               Op.getSimpleValueType());
   }
 

lib/Target/AArch64/AArch64InstrInfo.td

 def AArch64umaxv    : SDNode<"AArch64ISD::UMAXV", SDT_AArch64UnaryVec>;
 
 def SDT_AArch64SETTAG : SDTypeProfile<0, 2, [SDTCisPtrTy<0>, SDTCisPtrTy<1>]>;
 def AArch64stg : SDNode<"AArch64ISD::STG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
 def AArch64stzg : SDNode<"AArch64ISD::STZG", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
 def AArch64st2g : SDNode<"AArch64ISD::ST2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
 def AArch64stz2g : SDNode<"AArch64ISD::STZ2G", SDT_AArch64SETTAG, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
 
+def SDT_AArch64unpk : SDTypeProfile<1, 1, [
+    SDTCisInt<0>, SDTCisInt<1>, SDTCisOpSmallerThanOp<1, 0>
+]>;
+def AArch64sunpkhi : SDNode<"AArch64ISD::SUNPKHI", SDT_AArch64unpk>;
+def AArch64sunpklo : SDNode<"AArch64ISD::SUNPKLO", SDT_AArch64unpk>;
+def AArch64uunpkhi : SDNode<"AArch64ISD::UUNPKHI", SDT_AArch64unpk>;
+def AArch64uunpklo : SDNode<"AArch64ISD::UUNPKLO", SDT_AArch64unpk>;
+
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
 
 // AArch64 Instruction Predicate Definitions.
 // We could compute these on a per-module basis but doing so requires accessing
 // the Function object through the <Target>Subtarget and objections were raised
 // to that (see post-commit review comments for r301750).

lib/Target/AArch64/AArch64SVEInstrInfo.td

let Predicates = [HasSVE] in {
   defm RBIT_ZPmZ : sve_int_perm_rev_rbit<"rbit">;
   defm REVB_ZPmZ : sve_int_perm_rev_revb<"revb">;
   defm REVH_ZPmZ : sve_int_perm_rev_revh<"revh">;
   defm REVW_ZPmZ : sve_int_perm_rev_revw<"revw">;
 
   defm REV_PP : sve_int_perm_reverse_p<"rev">;
   defm REV_ZZ : sve_int_perm_reverse_z<"rev">;
 
-  defm SUNPKLO_ZZ : sve_int_perm_unpk<0b00, "sunpklo">;
-  defm SUNPKHI_ZZ : sve_int_perm_unpk<0b01, "sunpkhi">;
-  defm UUNPKLO_ZZ : sve_int_perm_unpk<0b10, "uunpklo">;
-  defm UUNPKHI_ZZ : sve_int_perm_unpk<0b11, "uunpkhi">;
+  defm SUNPKLO_ZZ : sve_int_perm_unpk<0b00, "sunpklo", AArch64sunpklo>;
+  defm SUNPKHI_ZZ : sve_int_perm_unpk<0b01, "sunpkhi", AArch64sunpkhi>;
+  defm UUNPKLO_ZZ : sve_int_perm_unpk<0b10, "uunpklo", AArch64uunpklo>;
+  defm UUNPKHI_ZZ : sve_int_perm_unpk<0b11, "uunpkhi", AArch64uunpkhi>;
 
   defm PUNPKLO_PP : sve_int_perm_punpk<0b0, "punpklo", int_aarch64_sve_punpklo>;
   defm PUNPKHI_PP : sve_int_perm_punpk<0b1, "punpkhi", int_aarch64_sve_punpkhi>;
 
   defm MOVPRFX_ZPzZ : sve_int_movprfx_pred_zero<0b000, "movprfx">;
   defm MOVPRFX_ZPmZ : sve_int_movprfx_pred_merge<0b001, "movprfx">;
   def MOVPRFX_ZZ : sve_int_bin_cons_misc_0_c<0b00000001, "movprfx", ZPRAny>;
   def FEXPA_ZZ_H : sve_int_bin_cons_misc_0_c<0b01000000, "fexpa", ZPR16>;

lib/Target/AArch64/SVEInstrFormats.td

 multiclass sve_int_perm_reverse_p<string asm> {
   def _B : sve_int_perm_reverse_p<0b00, asm, PPR8>;
   def _H : sve_int_perm_reverse_p<0b01, asm, PPR16>;
   def _S : sve_int_perm_reverse_p<0b10, asm, PPR32>;
   def _D : sve_int_perm_reverse_p<0b11, asm, PPR64>;
 }
 
 class sve_int_perm_unpk<bits<2> sz16_64, bits<2> opc, string asm,
                         ZPRRegOp zprty1, ZPRRegOp zprty2>
 : I<(outs zprty1:$Zd), (ins zprty2:$Zn),
   asm, "\t$Zd, $Zn",
   "", []>, Sched<[]> {
   bits<5> Zd;
   bits<5> Zn;
   let Inst{31-24} = 0b00000101;
   let Inst{23-22} = sz16_64;
   let Inst{21-18} = 0b1100;
   let Inst{17-16} = opc;
   let Inst{15-10} = 0b001110;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zd;
 }
 
-multiclass sve_int_perm_unpk<bits<2> opc, string asm> {
+multiclass sve_int_perm_unpk<bits<2> opc, string asm, SDPatternOperator op> {
   def _H : sve_int_perm_unpk<0b01, opc, asm, ZPR16, ZPR8>;
   def _S : sve_int_perm_unpk<0b10, opc, asm, ZPR32, ZPR16>;
   def _D : sve_int_perm_unpk<0b11, opc, asm, ZPR64, ZPR32>;
+
+  def : SVE_1_Op_Pat<nxv8i16, op, nxv16i8, !cast<Instruction>(NAME # _H)>;
+  def : SVE_1_Op_Pat<nxv4i32, op, nxv8i16, !cast<Instruction>(NAME # _S)>;
+  def : SVE_1_Op_Pat<nxv2i64, op, nxv4i32, !cast<Instruction>(NAME # _D)>;
 }
 
 class sve_int_perm_insrs<bits<2> sz8_64, string asm, ZPRRegOp zprty,
                          RegisterClass srcRegType>
 : I<(outs zprty:$Zdn), (ins zprty:$_Zdn, srcRegType:$Rm),
   asm, "\t$Zdn, $Rm",
   "",
   []>, Sched<[]> {

test/CodeGen/AArch64/sve-intrinsics-perm-select.ll

+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; SUNPKHI
+;
+
+define <vscale x 8 x i16> @sunpkhi_i16(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: sunpkhi_i16
+; CHECK: sunpkhi z0.h, z0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 8 x i16> @llvm.aarch64.sve.sunpkhi.nxv8i16(<vscale x 16 x i8> %a)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @sunpkhi_i32(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sunpkhi_i32
+; CHECK: sunpkhi z0.s, z0.h
+; CHECK-NEXT: ret
+  %res = call <vscale x 4 x i32> @llvm.aarch64.sve.sunpkhi.nxv4i32(<vscale x 8 x i16> %a)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @sunpkhi_i64(<vscale x 4 x i32> %a) {
+; CHECK-LABEL:  sunpkhi_i64
+; CHECK: sunpkhi z0.d, z0.s
+; CHECK-NEXT: ret
+  %res = call <vscale x 2 x i64> @llvm.aarch64.sve.sunpkhi.nxv2i64(<vscale x 4 x i32> %a)
+  ret <vscale x 2 x i64> %res
+}
+
+;
+; SUNPKLO
+;
+
+define <vscale x 8 x i16> @sunpklo_i16(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: sunpklo_i16
+; CHECK: sunpklo z0.h, z0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 8 x i16> @llvm.aarch64.sve.sunpklo.nxv8i16(<vscale x 16 x i8> %a)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @sunpklo_i32(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: sunpklo_i32
+; CHECK: sunpklo z0.s, z0.h
+; CHECK-NEXT: ret
+  %res = call <vscale x 4 x i32> @llvm.aarch64.sve.sunpklo.nxv4i32(<vscale x 8 x i16> %a)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @sunpklo_i64(<vscale x 4 x i32> %a) {
+; CHECK-LABEL:  sunpklo_i64
+; CHECK: sunpklo z0.d, z0.s
+; CHECK-NEXT: ret
+  %res = call <vscale x 2 x i64> @llvm.aarch64.sve.sunpklo.nxv2i64(<vscale x 4 x i32> %a)
+  ret <vscale x 2 x i64> %res
+}
+
+;
+; UUNPKHI
+;
+
+define <vscale x 8 x i16> @uunpkhi_i16(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: uunpkhi_i16
+; CHECK: uunpkhi z0.h, z0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 8 x i16> @llvm.aarch64.sve.uunpkhi.nxv8i16(<vscale x 16 x i8> %a)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @uunpkhi_i32(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: uunpkhi_i32
+; CHECK: uunpkhi z0.s, z0.h
+; CHECK-NEXT: ret
+  %res = call <vscale x 4 x i32> @llvm.aarch64.sve.uunpkhi.nxv4i32(<vscale x 8 x i16> %a)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @uunpkhi_i64(<vscale x 4 x i32> %a) {
+; CHECK-LABEL:  uunpkhi_i64
+; CHECK: uunpkhi z0.d, z0.s
+; CHECK-NEXT: ret
+  %res = call <vscale x 2 x i64> @llvm.aarch64.sve.uunpkhi.nxv2i64(<vscale x 4 x i32> %a)
+  ret <vscale x 2 x i64> %res
+}
+
+;
+; UUNPKLO
+;
+
+define <vscale x 8 x i16> @uunpklo_i16(<vscale x 16 x i8> %a) {
+; CHECK-LABEL: uunpklo_i16
+; CHECK: uunpklo z0.h, z0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 8 x i16> @llvm.aarch64.sve.uunpklo.nxv8i16(<vscale x 16 x i8> %a)
+  ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @uunpklo_i32(<vscale x 8 x i16> %a) {
+; CHECK-LABEL: uunpklo_i32
+; CHECK: uunpklo z0.s, z0.h
+; CHECK-NEXT: ret
+  %res = call <vscale x 4 x i32> @llvm.aarch64.sve.uunpklo.nxv4i32(<vscale x 8 x i16> %a)
+  ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @uunpklo_i64(<vscale x 4 x i32> %a) {
+; CHECK-LABEL:  uunpklo_i64
+; CHECK: uunpklo z0.d, z0.s
+; CHECK-NEXT: ret
+  %res = call <vscale x 2 x i64> @llvm.aarch64.sve.uunpklo.nxv2i64(<vscale x 4 x i32> %a)
+  ret <vscale x 2 x i64> %res
+}
+
+declare <vscale x 8 x i16> @llvm.aarch64.sve.sunpkhi.nxv8i16(<vscale x 16 x i8>)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.sunpkhi.nxv4i32(<vscale x 8 x i16>)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.sunpkhi.nxv2i64(<vscale x 4 x i32>)
+
+declare <vscale x 8 x i16> @llvm.aarch64.sve.sunpklo.nxv8i16(<vscale x 16 x i8>)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.sunpklo.nxv4i32(<vscale x 8 x i16>)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.sunpklo.nxv2i64(<vscale x 4 x i32>)
+
+declare <vscale x 8 x i16> @llvm.aarch64.sve.uunpkhi.nxv8i16(<vscale x 16 x i8>)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.uunpkhi.nxv4i32(<vscale x 8 x i16>)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.uunpkhi.nxv2i64(<vscale x 4 x i32>)
+
+declare <vscale x 8 x i16> @llvm.aarch64.sve.uunpklo.nxv8i16(<vscale x 16 x i8>)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.uunpklo.nxv4i32(<vscale x 8 x i16>)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.uunpklo.nxv2i64(<vscale x 4 x i32>)