From: Kerry McLaughlin <kerry.mclaughlin@arm.com>
Date: Fri, 18 Oct 2019 09:40:16 +0000 (+0000)
Subject: [AArch64][SVE] Implement unpack intrinsics
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=9d9055cfe2e0e8a48be86cf935b1d8399e043586;p=llvm

[AArch64][SVE] Implement unpack intrinsics

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

Reviewers: sdesmalen, SjoerdMeijer, greened, rengolin, rovka

Reviewed By: greened

Subscribers: tschuett, kristof.beyls, rkruppe, psnobl, cfe-commits, llvm-commits

Differential Revision: https://reviews.llvm.org/D67550

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@375210 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/include/llvm/IR/IntrinsicsAArch64.td b/include/llvm/IR/IntrinsicsAArch64.td
index e0e780aa524..db01700f409 100644
--- a/include/llvm/IR/IntrinsicsAArch64.td
+++ b/include/llvm/IR/IntrinsicsAArch64.td
@@ -775,6 +775,11 @@ let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
                  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],
@@ -826,6 +831,16 @@ def int_aarch64_sve_udot_lane : AdvSIMD_SVE_DOT_Indexed_Intrinsic;
 
 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
 //
diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index ee62b6dfe36..ccd085e649e 100644
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1300,6 +1300,10 @@ const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
   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;
 }
@@ -2838,6 +2842,19 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
     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();
diff --git a/lib/Target/AArch64/AArch64ISelLowering.h b/lib/Target/AArch64/AArch64ISelLowering.h
index 462fcabc994..83109bbdbfc 100644
--- a/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/lib/Target/AArch64/AArch64ISelLowering.h
@@ -191,6 +191,11 @@ enum NodeType : unsigned {
   FRECPE, FRECPS,
   FRSQRTE, FRSQRTS,
 
+  SUNPKHI,
+  SUNPKLO,
+  UUNPKHI,
+  UUNPKLO,
+
   // NEON Load/Store with post-increment base updates
   LD2post = ISD::FIRST_TARGET_MEMORY_OPCODE,
   LD3post,
diff --git a/lib/Target/AArch64/AArch64InstrInfo.td b/lib/Target/AArch64/AArch64InstrInfo.td
index 6154528b6d4..59a1cb25ac6 100644
--- a/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/lib/Target/AArch64/AArch64InstrInfo.td
@@ -421,6 +421,14 @@ def AArch64stzg : SDNode<"AArch64ISD::STZG", SDT_AArch64SETTAG, [SDNPHasChain, S
 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>;
+
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/AArch64/AArch64SVEInstrInfo.td b/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 5477e9f540a..fc75004a787 100644
--- a/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -211,10 +211,10 @@ let Predicates = [HasSVE] in {
   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>;
diff --git a/lib/Target/AArch64/SVEInstrFormats.td b/lib/Target/AArch64/SVEInstrFormats.td
index ec5ebfbbfad..ec83e34a93a 100644
--- a/lib/Target/AArch64/SVEInstrFormats.td
+++ b/lib/Target/AArch64/SVEInstrFormats.td
@@ -848,10 +848,14 @@ class sve_int_perm_unpk<bits<2> sz16_64, bits<2> opc, string asm,
   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,
diff --git a/test/CodeGen/AArch64/sve-intrinsics-perm-select.ll b/test/CodeGen/AArch64/sve-intrinsics-perm-select.ll
new file mode 100644
index 00000000000..c2230b0c550
--- /dev/null
+++ b/test/CodeGen/AArch64/sve-intrinsics-perm-select.ll
@@ -0,0 +1,129 @@
+; 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>)