[AArch64][SVE] Add more intrinsics in 'isZeroingInactiveLanes'.

[dewen]

Summary

The REINTERPRET_CAST operation generates redundant and and ptrue instructions. For some instructions, this is redundant, because its inactive lanes are zeroed by construction.

for example. codegen before:

facgt p2.d, p0/z, z4.d, z1.d
ptrue p1.d
and p1.b, p2/z, p2.b, p1.b

After:

facgt p1.d, p0/z, z4.d, z1.d

ref: https://reviews.llvm.org/D129851

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

292

I recently looked at some applications and found that other intrinsics can do similar optimizations.

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

292

@paulwalker-arm @sdesmalen @efriedma @georges @david-arm Can you review this patch for me?

[sdesmalen]

@hassnaa-arm this patch seems to solve the same issue as you're trying to solve in D141397.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

305

Predicate select is slightly different, since there is only a full-vector predicate select (operating on svbool_t), so there are no inactive lanes. It might be worth to point that out in a comment.

Can you please add a test for this case? it seems to be missing.

llvm/test/CodeGen/AArch64/sve2-intrinsics-reinterpret.ll

384

nit: odd indentation

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

305

Okay, thank you. @sdesmalen

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

305

I have updated. Please review. @sdesmalen

[dewen]

llvm/test/CodeGen/AArch64/sve2-intrinsics-reinterpret.ll

384

done

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

305

hi, @sdesmalen, thanks for you review. Could you please review it again?

[sdesmalen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

263–264

This comment only applies to the intrinsic, not to the ISD node.

265–266

We can't make this generalisation to ISD::VSELECT because it doesn't necessarily zero the other lanes, e.g.

define <vscale x 4 x i1> @foo(<vscale x 4 x i1> %pg, <vscale x 4 x i1> %p0, <vscale x 4 x i1> %p1) {
  %1 = select <vscale x 4 x i1> %pg, <vscale x 4 x i1> %p0, <vscale x 4 x i1> %p1
  ret <vscale x 4 x i1> %1
}

Is lowered to:

foo:                                    // @foo
      sel     p0.b, p0, p1.b, p2.b
      ret

If p1.b and p2.b have non-zero values in their inactive lanes, then the result p0.b will also have non-zero values in the inactive lanes.

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

265–266

@sdesmalen, thank you, You're right. I didn't think about it all before. Sorry.

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

265–266

I've deleted this part, thank you again.@sdesmalen

[paulwalker-arm]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

251

Just to add context to my comments below. Here "inactive lanes" means those lanes that are not visible within the DAG. By which I mean the MVTs for SVE's predicate vector types represent a subset of the underlying bits within a predicate register. This function is asking whether the specified operation is known to zero the bits that are not part of this subset. (e.g. nxv2i1 represents every eighth bit of the predicate register 0,8,16... and this function returns true if the operation is guaranteed to zero bits 1-7,9-15...)

301–304

Does trn and uzp zero inactive lanes? My reading of the pseudo code is that all lanes of the result come from one of the inputs. For example when trn1 is used with a .d element type each "lane" is actually a block of the 8 predicate bits that sit within it and they are copied as a block without modification.

Actually I think we've likely got an upstream bug here with how we're lowering the svbool builtins of these instructions to these intrinsics because the builtins fully define the entire predicate register, which when fixed should mean this function will not need to worry about them.

Either way, if you agree I think it's best to remove their entries/tests from this patch.

305–311

I don't believe the logical instructions zero inactive lanes either. You can see this by looking at the tests you've added. The instructions are always "byte" based and so if the input general predicate contains garbage then so will the result. In this instance this is not a problem because from the ACLE's point of view we only support these intrinsics for <vscale x 16 x i1>[1] types and when used with such types they'll be no reinterprets to introduce the unnecessary and and so I think these entries should also be removed.

[1] I know the code generation works for other predicate types, but that's just a fluke based on some stale isel patterns (i.e. they'll never be generated via the clang builtins) that should probably be removed.

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

301–304

I agree with you.@paulwalker-arm

305–311

I don't understand this part, but I think these logic instructions are going to put Inactive elements in the destination predicate register are set to zero. @paulwalker-arm

for example:

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

301–304

hi, @paulwalker-arm, thanks for you review. I've reverted my previous patch. Sorry, I didn't see your comment when I submitted it.

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

305–311

Thank you @paulwalker-arm. I understand my mistake now. I'll update the patch.

[dewen]

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

301–304

I've updated it, can I ask you to review it again? Thank you.@paulwalker-arm

[paulwalker-arm]

Thanks @dewen for the fixes. This looks good to me now. Do you fancy taking a look at fixing the incorrect lowering for the predicate trn, uzp and zip builtins (I suspect we need dedicated intrinsics where the element type is part of the name rather than relying on the operand types) as a separate patch? or would you rather somebody else fix it?

[dewen]

In D141469#4058457, @paulwalker-arm wrote:

Thanks @dewen for the fixes. This looks good to me now. Do you fancy taking a look at fixing the incorrect lowering for the predicate trn, uzp and zip builtins (I suspect we need dedicated intrinsics where the element type is part of the name rather than relying on the operand types) as a separate patch? or would you rather somebody else fix it?

Hi, thank you@paulwalker-arm. I'm interested, but I'm about to start the holiday. Happy New Year.

[paulwalker-arm]

In D141469#4060976, @dewen wrote:

In D141469#4058457, @paulwalker-arm wrote:

Thanks @dewen for the fixes. This looks good to me now. Do you fancy taking a look at fixing the incorrect lowering for the predicate trn, uzp and zip builtins (I suspect we need dedicated intrinsics where the element type is part of the name rather than relying on the operand types) as a separate patch? or would you rather somebody else fix it?

Hi, thank you@paulwalker-arm. I'm interested, but I'm about to start the holiday. Happy New Year.

Ok, I'll bring the fix in house then. Enjoy your holiday.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

static bool isMergePassthruOpcode(unsigned Opc) {
   case AArch64ISD::FSQRT_MERGE_PASSTHRU:
   case AArch64ISD::FRECPX_MERGE_PASSTHRU:
   case AArch64ISD::FABS_MERGE_PASSTHRU:
     return true;
   }
 }
 
 // Returns true if inactive lanes are known to be zeroed by construction.
 static bool isZeroingInactiveLanes(SDValue Op) {
   switch (Op.getOpcode()) {
   default:
     // We guarantee i1 splat_vectors to zero the other lanes by
     // implementing it with ptrue and possibly a punpklo for nxv1i1.
     if (ISD::isConstantSplatVectorAllOnes(Op.getNode()))
       return true;
     return false;
   case AArch64ISD::PTRUE:
   case AArch64ISD::SETCC_MERGE_ZERO:
     return true;
   case ISD::INTRINSIC_WO_CHAIN:
     switch (Op.getConstantOperandVal(0)) {
     default:
       return false;
     case Intrinsic::aarch64_sve_ptrue:
     case Intrinsic::aarch64_sve_pnext:
     case Intrinsic::aarch64_sve_cmpeq:
     case Intrinsic::aarch64_sve_cmpne:
     case Intrinsic::aarch64_sve_cmpge:
     case Intrinsic::aarch64_sve_cmpgt:
     case Intrinsic::aarch64_sve_cmphs:
     case Intrinsic::aarch64_sve_cmphi:
     case Intrinsic::aarch64_sve_cmpeq_wide:
     case Intrinsic::aarch64_sve_cmpne_wide:
     case Intrinsic::aarch64_sve_cmpge_wide:
     case Intrinsic::aarch64_sve_cmpgt_wide:
     case Intrinsic::aarch64_sve_cmplt_wide:
     case Intrinsic::aarch64_sve_cmple_wide:
     case Intrinsic::aarch64_sve_cmphs_wide:
     case Intrinsic::aarch64_sve_cmphi_wide:
     case Intrinsic::aarch64_sve_cmplo_wide:
     case Intrinsic::aarch64_sve_cmpls_wide:
     case Intrinsic::aarch64_sve_fcmpeq:
     case Intrinsic::aarch64_sve_fcmpne:
     case Intrinsic::aarch64_sve_fcmpge:
     case Intrinsic::aarch64_sve_fcmpgt:
     case Intrinsic::aarch64_sve_fcmpuo:
+    case Intrinsic::aarch64_sve_facgt:
+    case Intrinsic::aarch64_sve_facge:
+    case Intrinsic::aarch64_sve_whilege:
+    case Intrinsic::aarch64_sve_whilegt:
+    case Intrinsic::aarch64_sve_whilehi:
+    case Intrinsic::aarch64_sve_whilehs:
+    case Intrinsic::aarch64_sve_whilele:
+    case Intrinsic::aarch64_sve_whilelo:
+    case Intrinsic::aarch64_sve_whilels:
+    case Intrinsic::aarch64_sve_whilelt:
+    case Intrinsic::aarch64_sve_match:
+    case Intrinsic::aarch64_sve_nmatch:
+    case Intrinsic::aarch64_sve_trn1:
+    case Intrinsic::aarch64_sve_trn2:
+    case Intrinsic::aarch64_sve_uzp1:
+    case Intrinsic::aarch64_sve_uzp2:
+    case Intrinsic::aarch64_sve_sel:
+    case Intrinsic::aarch64_sve_orr_z:
+    case Intrinsic::aarch64_sve_orn_z:
+    case Intrinsic::aarch64_sve_and_z:
+    case Intrinsic::aarch64_sve_bic_z:
+    case Intrinsic::aarch64_sve_eor_z:
+    case Intrinsic::aarch64_sve_nor_z:
+    case Intrinsic::aarch64_sve_nand_z:
       return true;
     }
   }
 }
 
 AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
                                              const AArch64Subtarget &STI)
     : TargetLowering(TM), Subtarget(&STI) {

llvm/test/CodeGen/AArch64/sve2-intrinsics-reinterpret.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 < %s | FileCheck %s
+
+define <vscale x 16 x i1> @facgt_fun(<vscale x 2 x i1> %a, <vscale x 2 x double> %b, <vscale x 2 x double> %c) {
+; CHECK-LABEL: facgt_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    facgt p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call <vscale x 2 x i1> @llvm.aarch64.sve.facgt.nxv2f64(<vscale x 2 x i1> %a, <vscale x 2 x double> %b, <vscale x 2 x double> %c)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @facge_fun(<vscale x 2 x i1> %a, <vscale x 2 x double> %b, <vscale x 2 x double> %c) {
+; CHECK-LABEL: facge_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    facge p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call <vscale x 2 x i1> @llvm.aarch64.sve.facge.nxv2f64(<vscale x 2 x i1> %a, <vscale x 2 x double> %b, <vscale x 2 x double> %c)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilege_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilege_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilege p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilege.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilegt_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilegt_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilegt p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilegt.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilehi_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilehi_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilehi p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilehi.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilehs_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilehs_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilehs p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilehs.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilele_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilele_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilele p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilele.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilelo_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilelo_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilelo p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilelo.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilels_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilels_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilels p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilels.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @whilelt_fun(i32 %a, i32 %b) {
+; CHECK-LABEL: whilelt_fun:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    whilelt p0.d, w0, w1
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 2 x i1> @llvm.aarch64.sve.whilelt.nxv2i1.i32(i32 %a, i32 %b)
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %0)
+  ret <vscale x 16 x i1> %1
+}
+
+define <vscale x 16 x i1> @cmpeq_d_fun(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpeq_d_fun:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.cmpeq.nxv2i64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x i64> %a,
+                                                                <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpeq_wide_s_fun(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpeq_wide_s_fun:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmpeq.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpge_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpge_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpge p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.cmpge.nxv2i64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x i64> %a,
+                                                                <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpge_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpge_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpge p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmpge.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpgt_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpgt_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpgt p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.cmpgt.nxv2i64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x i64> %a,
+                                                                <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpgt_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpgt_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpgt p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmpgt.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmphi_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmphi_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmphi p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.cmphi.nxv2i64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x i64> %a,
+                                                                <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmphi_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmphi_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmphi p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmphi.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmphs_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmphs_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmphs p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.cmphs.nxv2i64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x i64> %a,
+                                                                <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmphs_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmphs_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmphs p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmphs.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmple_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmple_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmple p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmple.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmplo_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmplo_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmplo p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmplo.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpls_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpls_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpls p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmpls.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmplt_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmplt_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmplt p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmplt.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpne_d(<vscale x 2 x i1> %pg, <vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpne_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.cmpne.nxv2i64(<vscale x 2 x i1> %pg,
+                                                                <vscale x 2 x i64> %a,
+                                                                <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @cmpne_wide_s(<vscale x 4 x i1> %pg, <vscale x 4 x i32> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: cmpne_wide_s:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmpne p0.s, p0/z, z0.s, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 4 x i1> @llvm.aarch64.sve.cmpne.wide.nxv4i32(<vscale x 4 x i1> %pg,
+                                                                     <vscale x 4 x i32> %a,
+                                                                     <vscale x 2 x i64> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @fcmeq_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmeq_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpeq.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @fcmgt_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmgt_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    fcmgt p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpgt.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @fcmne_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmne_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    fcmne p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpne.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @fcmuo_d(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %b) {
+; CHECK-LABEL: fcmuo_d:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    fcmuo p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.fcmpuo.nxv2f64(<vscale x 2 x i1> %pg,
+                                                                 <vscale x 2 x double> %a,
+                                                                 <vscale x 2 x double> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @match_i16(<vscale x 8 x i1> %pg, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: match_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    match p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 8 x i1> @llvm.aarch64.sve.match.nxv8i16(<vscale x 8 x i1> %pg,
+                                                                <vscale x 8 x i16> %a,
+                                                                <vscale x 8 x i16> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @nmatch_i16(<vscale x 8 x i1> %pg, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: nmatch_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    nmatch p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 8 x i1> @llvm.aarch64.sve.nmatch.nxv8i16(<vscale x 8 x i1> %pg,
+                                                                 <vscale x 8 x i16> %a,
+                                                                 <vscale x 8 x i16> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @trn1_b64(<vscale x 2 x i1> %a, <vscale x 2 x i1> %b) {
+; CHECK-LABEL: trn1_b64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    trn1 p0.d, p0.d, p1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.trn1.nxv2i1(<vscale x 2 x i1> %a,
+                                                              <vscale x 2 x i1> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @trn2_b64(<vscale x 2 x i1> %a, <vscale x 2 x i1> %b) {
+; CHECK-LABEL: trn2_b64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    trn2 p0.d, p0.d, p1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.trn2.nxv2i1(<vscale x 2 x i1> %a,
+                                                              <vscale x 2 x i1> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @uzp1_b64(<vscale x 2 x i1> %a, <vscale x 2 x i1> %b) {
+; CHECK-LABEL: uzp1_b64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uzp1 p0.d, p0.d, p1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.uzp1.nxv2i1(<vscale x 2 x i1> %a,
+                                                              <vscale x 2 x i1> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @uzp2_b64(<vscale x 2 x i1> %a, <vscale x 2 x i1> %b) {
+; CHECK-LABEL: uzp2_b64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uzp2 p0.d, p0.d, p1.d
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.uzp2.nxv2i1(<vscale x 2 x i1> %a,
+                                                              <vscale x 2 x i1> %b)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @orr_2(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd) {
+; CHECK-LABEL: orr_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr p0.b, p0/z, p1.b, p2.b
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.orr.z.nxv2i1(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @orn_2(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd) {
+; CHECK-LABEL: orn_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orn p0.b, p0/z, p1.b, p2.b
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.orn.z.nxv2i1(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out
+}
+
+define <vscale x 16 x i1> @and_2(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd) {
+; CHECK-LABEL: and_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    and p0.b, p0/z, p1.b, p2.b
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.and.z.nxv2i1(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out;
+}
+
+define <vscale x 16 x i1> @bic_2(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd) {
+; CHECK-LABEL: bic_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    bic p0.b, p0/z, p1.b, p2.b
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.bic.z.nxv2i1(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out;
+}
+
+define <vscale x 16 x i1> @eor_2(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd) {
+; CHECK-LABEL: eor_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    eor p0.b, p0/z, p1.b, p2.b
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.eor.z.nxv2i1(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out;
+}
+
+define <vscale x 16 x i1> @nor_2(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd) {
+; CHECK-LABEL: nor_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    nor p0.b, p0/z, p1.b, p2.b
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.nor.z.nxv2i1(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out;
+}
+
+define <vscale x 16 x i1> @nand_2(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd) {
+; CHECK-LABEL: nand_2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    nand p0.b, p0/z, p1.b, p2.b
+; CHECK-NEXT:    ret
+  %1 = call <vscale x 2 x i1> @llvm.aarch64.sve.nand.z.nxv2i1(<vscale x 2 x i1> %Pg, <vscale x 2 x i1> %Pn, <vscale x 2 x i1> %Pd)
+  %out = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> %1)
+  ret <vscale x 16 x i1> %out;
+}
+
+declare <vscale x 2 x i1> @llvm.aarch64.sve.orn.z.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.orr.z.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.nand.z.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.nor.z.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.and.z.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.bic.z.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.eor.z.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.trn2.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.trn1.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.uzp1.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.uzp2.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.facgt.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.facge.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpeq.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpge.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpgt.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpne.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.fcmpuo.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilege.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilegt.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilehi.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilehs.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilele.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilelo.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilels.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.whilelt.nxv2i1.i32(i32, i32)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.cmpeq.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmpeq.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.cmpge.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmpge.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.cmpgt.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmpgt.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.cmphi.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmphi.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.cmphs.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmphs.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmple.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmplo.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmpls.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmplt.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 2 x i1> @llvm.aarch64.sve.cmpne.nxv2i64(<vscale x 2 x i1>, <vscale x 2 x i64>, <vscale x 2 x i64>)
+declare <vscale x 4 x i1> @llvm.aarch64.sve.cmpne.wide.nxv4i32(<vscale x 4 x i1>, <vscale x 4 x i32>, <vscale x 2 x i64>)
+declare <vscale x 8 x i1> @llvm.aarch64.sve.match.nxv8i16(<vscale x 8 x i1>, <vscale x 8 x i16>, <vscale x 8 x i16>)
+declare <vscale x 8 x i1> @llvm.aarch64.sve.nmatch.nxv8i16(<vscale x 8 x i1>, <vscale x 8 x i16>, <vscale x 8 x i16>)
+declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1>)
+declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1>)
+declare <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1>)