[AArch64][SVE] Implement punpk[hi|lo] intrinsics

Summary:
Adds the following two intrinsics:
  - int_aarch64_sve_punpkhi
  - int_aarch64_sve_punpklo

This patch also contains a fix which allows LLVMHalfElementsVectorType
to forward reference overloadable arguments.

Reviewers: sdesmalen, rovka, rengolin

Reviewed By: sdesmalen

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

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D67830

llvm-svn: 373232

include/llvm/IR/IntrinsicsAArch64.td

@@ -768,6 +768,11 @@ let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
                  LLVMMatchType<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>
@@ -792,4 +797,11 @@ def int_aarch64_sve_neg : AdvSIMD_Merged1VectorArg_Intrinsic;
 //
 
 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/IR/Function.cpp

@@ -1211,8 +1211,9 @@ static bool matchIntrinsicType(
     }
     case IITDescriptor::HalfVecArgument:
       // If this is a forward reference, defer the check for later.
-      return D.getArgumentNumber() >= ArgTys.size() ||
+      if (D.getArgumentNumber() >= ArgTys.size())
-             !isa<VectorType>(ArgTys[D.getArgumentNumber()]) ||
+        return IsDeferredCheck || DeferCheck(Ty);
+      return !isa<VectorType>(ArgTys[D.getArgumentNumber()]) ||
              VectorType::getHalfElementsVectorType(
                      cast<VectorType>(ArgTys[D.getArgumentNumber()])) != Ty;
     case IITDescriptor::SameVecWidthArgument: {

lib/Target/AArch64/AArch64SVEInstrInfo.td

@@ -216,8 +216,8 @@ let Predicates = [HasSVE] in {
   defm UUNPKLO_ZZ : sve_int_perm_unpk<0b10, "uunpklo">;
   defm UUNPKHI_ZZ : sve_int_perm_unpk<0b11, "uunpkhi">;
 
-  def  PUNPKLO_PP : sve_int_perm_punpk<0b0, "punpklo">;
+  defm PUNPKLO_PP : sve_int_perm_punpk<0b0, "punpklo", int_aarch64_sve_punpklo>;
-  def  PUNPKHI_PP : sve_int_perm_punpk<0b1, "punpkhi">;
+  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">;

lib/Target/AArch64/SVEInstrFormats.td

@@ -283,6 +283,11 @@ let Predicates = [HasSVE] in {
 // SVE pattern match helpers.
 //===----------------------------------------------------------------------===//
 
+class SVE_1_Op_Pat<ValueType vtd, SDPatternOperator op, ValueType vt1,
+                   Instruction inst>
+: Pat<(vtd (op vt1:$Op1)),
+      (inst $Op1)>;
+
 class SVE_3_Op_Pat<ValueType vtd, SDPatternOperator op, ValueType vt1,
                    ValueType vt2, ValueType vt3, Instruction inst>
 : Pat<(vtd (op vt1:$Op1, vt2:$Op2, vt3:$Op3)),
@@ -4280,6 +4285,14 @@ class sve_int_perm_punpk<bit opc, string asm>
   let Inst{3-0}   = Pd;
 }
 
+multiclass sve_int_perm_punpk<bit opc, string asm, SDPatternOperator op> {
+  def NAME : sve_int_perm_punpk<opc, asm>;
+
+  def : SVE_1_Op_Pat<nxv8i1, op, nxv16i1, !cast<Instruction>(NAME)>;
+  def : SVE_1_Op_Pat<nxv4i1, op, nxv8i1,  !cast<Instruction>(NAME)>;
+  def : SVE_1_Op_Pat<nxv2i1, op, nxv4i1,  !cast<Instruction>(NAME)>;
+}
+
 class sve_int_rdffr_pred<bit s, string asm>
 : I<(outs PPR8:$Pd), (ins PPRAny:$Pg),
   asm, "\t$Pd, $Pg/z",

test/CodeGen/AArch64/sve-intrinsics-pred-operations.ll

@@ -0,0 +1,65 @@
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
+
+;
+; PUNPKHI
+;
+
+define <vscale x 8 x i1> @punpkhi_b16(<vscale x 16 x i1> %a) {
+; CHECK-LABEL: punpkhi_b16
+; CHECK: punpkhi p0.h, p0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 8 x i1> @llvm.aarch64.sve.punpkhi.nxv8i1(<vscale x 16 x i1> %a)
+  ret <vscale x 8 x i1> %res
+}
+
+define <vscale x 4 x i1> @punpkhi_b8(<vscale x 8 x i1> %a) {
+; CHECK-LABEL: punpkhi_b8
+; CHECK: punpkhi p0.h, p0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 4 x i1> @llvm.aarch64.sve.punpkhi.nxv4i1(<vscale x 8 x i1> %a)
+  ret <vscale x 4 x i1> %res
+}
+
+define <vscale x 2 x i1> @punpkhi_b4(<vscale x 4 x i1> %a) {
+; CHECK-LABEL: punpkhi_b4
+; CHECK: punpkhi p0.h, p0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 2 x i1> @llvm.aarch64.sve.punpkhi.nxv2i1(<vscale x 4 x i1> %a)
+  ret <vscale x 2 x i1> %res
+}
+
+;
+; PUNPKLO
+;
+
+define <vscale x 8 x i1> @punpklo_b16(<vscale x 16 x i1> %a) {
+; CHECK-LABEL: punpklo_b16
+; CHECK: punpklo p0.h, p0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 8 x i1> @llvm.aarch64.sve.punpklo.nxv8i1(<vscale x 16 x i1> %a)
+  ret <vscale x 8 x i1> %res
+}
+
+define <vscale x 4 x i1> @punpklo_b8(<vscale x 8 x i1> %a) {
+; CHECK-LABEL: punpklo_b8
+; CHECK: punpklo p0.h, p0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 4 x i1> @llvm.aarch64.sve.punpklo.nxv4i1(<vscale x 8 x i1> %a)
+  ret <vscale x 4 x i1> %res
+}
+
+define <vscale x 2 x i1> @punpklo_b4(<vscale x 4 x i1> %a) {
+; CHECK-LABEL: punpklo_b4
+; CHECK: punpklo p0.h, p0.b
+; CHECK-NEXT: ret
+  %res = call <vscale x 2 x i1> @llvm.aarch64.sve.punpklo.nxv2i1(<vscale x 4 x i1> %a)
+  ret <vscale x 2 x i1> %res
+}
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.punpkhi.nxv8i1(<vscale x 16 x i1>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.punpkhi.nxv4i1(<vscale x 8 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.punpkhi.nxv2i1(<vscale x 4 x i1>)
+
+declare <vscale x 8 x i1> @llvm.aarch64.sve.punpklo.nxv8i1(<vscale x 16 x i1>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.punpklo.nxv4i1(<vscale x 8 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.punpklo.nxv2i1(<vscale x 4 x i1>)