[Clang][SVE2.1] Add svpext builtins

[CarolineConcatto]

Summary

As described in: https://github.com/ARM-software/acle/pull/257

[sdesmalen]

clang/include/clang/Basic/arm_sve.td

16

s/follow(eg.:2.x)/follow (e.g. 2.x)/

clang/lib/CodeGen/CGBuiltin.cpp

9856

Please add SVE to the name, i.e. FormSVEMultiVectorResult

That said, if Call is not a multi-vector builtin then it just returns the result value. So perhaps this is better named FormSVEBuiltinResult() ?

9856

Can you add a doxygen comment to this function describing what it does?

9859–9861

Can you implement this with an early exit, e.g.

auto *StructTy = dyn_cast<StructType>(Call->getType());
if (!StructTy)
  return Call;

auto *VTy = dyn_cast<ScalableVectorType>(StructTy->getTypeAtIndex(0U));
if (!VTy)
  return Call;

...

clang/utils/TableGen/SveEmitter.cpp

864

nit: "is never a type in itself" ?

[CarolineConcatto]

Add doxygen comments to FormSVEBuiltinResult function

[CarolineConcatto]

-Rebase

[CarolineConcatto]

-Rebase

clang/include/clang/Basic/arm_sve.td

 //===--- arm_sve.td - ARM SVE compiler interface ------------------------===//
 //
 //  Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 //  See https://llvm.org/LICENSE.txt for license information.
 //  SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 //  This file defines the TableGen definitions from which the ARM SVE header
 //  file will be generated.  See:
 //
 //      https://developer.arm.com/architectures/system-architectures/software-standards/acle
 //
 //===----------------------------------------------------------------------===//
 
 include "arm_sve_sme_incl.td"
 
 ////////////////////////////////////////////////////////////////////////////////
 // Loads
 
 // Load one vector (scalar base)
 def SVLD1   : MInst<"svld1[_{2}]", "dPc", "csilUcUsUiUlhfd", [IsLoad],               MemEltTyDefault, "aarch64_sve_ld1">;
 def SVLD1SB : MInst<"svld1sb_{d}", "dPS", "silUsUiUl",       [IsLoad],               MemEltTyInt8,    "aarch64_sve_ld1">;
 def SVLD1UB : MInst<"svld1ub_{d}", "dPW", "silUsUiUl",       [IsLoad, IsZExtReturn], MemEltTyInt8,    "aarch64_sve_ld1">;

 def SVBEXT_N : SInst<"svbext[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bext_x">;
 def SVBGRP   : SInst<"svbgrp[_{d}]",   "ddd", "UcUsUiUl", MergeNone, "aarch64_sve_bgrp_x">;
 def SVBGRP_N : SInst<"svbgrp[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bgrp_x">;
 }
 
 let TargetGuard = "sve2p1" in {
 def SVFCLAMP   : SInst<"svclamp[_{d}]", "dddd", "hfd", MergeNone, "aarch64_sve_fclamp", [], []>;
 def SVPTRUE_COUNT  : SInst<"svptrue_{d}", "}v", "QcQsQiQl", MergeNone, "aarch64_sve_ptrue_{d}", [IsOverloadNone], []>;
+
+def SVPEXT_SINGLE : SInst<"svpext_lane_{d}", "P}i", "QcQsQiQl", MergeNone, "aarch64_sve_pext", [], [ImmCheck<1, ImmCheck0_3>]>;
+def SVPEXT_X2     : SInst<"svpext_lane_{d}_x2", "2.P}i", "QcQsQiQl", MergeNone, "aarch64_sve_pext_x2", [], [ImmCheck<1, ImmCheck0_1>]>;
 }
 
 let TargetGuard = "sve2p1" in {
 def SVSCLAMP : SInst<"svclamp[_{d}]", "dddd", "csil",     MergeNone, "aarch64_sve_sclamp", [], []>;
 def SVUCLAMP : SInst<"svclamp[_{d}]", "dddd", "UcUsUiUl", MergeNone, "aarch64_sve_uclamp", [], []>;
 def SVCNTP_COUNT : SInst<"svcntp_{d}", "n}i", "QcQsQiQl", MergeNone, "aarch64_sve_cntp_{d}", [IsOverloadNone], [ImmCheck<1, ImmCheck2_4_Mul2>]>;
-
 }

clang/include/clang/Basic/arm_sve_sme_incl.td

 // P: boolean
 // U: unsigned
 // Q: svcount
 
 // Prototype modifiers
 // -------------------
 // prototype: return (arg, arg, ...)
 //
-// 2,3,4: array of default vectors
+// 2,3,4: array of vectors
+// .: indicator for multi-vector modifier that will follow (e.g. 2.x)
 // v: void
 // x: vector of signed integers
 // u: vector of unsigned integers
 // d: default
 // c: const pointer type
 // P: predicate type
 // s: scalar of element type
 // a: scalar of element type (splat to vector type)

clang/lib/CodeGen/CGBuiltin.cpp

Value *CodeGenFunction::EmitSVETupleCreate(const SVETypeFlags &TypeFlags,
   for (unsigned I = 0; I < Ops.size(); I++) {
     Value *Idx = ConstantInt::get(CGM.Int64Ty, I * MinElts);
     Call = Builder.CreateInsertVector(Ty, Call, Ops[I], Idx);
   }
 
   return Call;
 }
 
+Value *CodeGenFunction::FormSVEBuiltinResult(Value *Call) {
+  // Multi-vector results should be broken up into a single (wide) result
+  // vector.
+  auto *StructTy = dyn_cast<StructType>(Call->getType());
+  if (!StructTy)
+    return Call;
+
+  auto *VTy = dyn_cast<ScalableVectorType>(StructTy->getTypeAtIndex(0U));
+  if (!VTy)
+    return Call;
+  unsigned N = StructTy->getNumElements();
+
+  // We may need to emit a cast to a svbool_t
+  bool IsPredTy = VTy->getElementType()->isIntegerTy(1);
+  unsigned MinElts = IsPredTy ? 16 : VTy->getMinNumElements();
+
+  ScalableVectorType *WideVTy =
+      ScalableVectorType::get(VTy->getElementType(), MinElts * N);
+  Value *Ret = llvm::PoisonValue::get(WideVTy);
+  for (unsigned I = 0; I < N; ++I) {
+    Value *SRet = Builder.CreateExtractValue(Call, I);
+    assert(SRet->getType() == VTy && "Unexpected type for result value");
+    Value *Idx = ConstantInt::get(CGM.Int64Ty, I * MinElts);
+
+    if (IsPredTy)
+      SRet = EmitSVEPredicateCast(
+          SRet, ScalableVectorType::get(Builder.getInt1Ty(), 16));
+
+    Ret = Builder.CreateInsertVector(WideVTy, Ret, SRet, Idx);
+  }
+  Call = Ret;
+
+  return Call;
+}
+
 Value *CodeGenFunction::EmitAArch64SVEBuiltinExpr(unsigned BuiltinID,
                                                   const CallExpr *E) {
   // Find out if any arguments are required to be integer constant expressions.
   unsigned ICEArguments = 0;
   ASTContext::GetBuiltinTypeError Error;
   getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
   assert(Error == ASTContext::GE_None && "Should not codegen an error");
 
Function *F = CGM.getIntrinsic(Builtin->LLVMIntrinsic,
                                    getSVEOverloadTypes(TypeFlags, Ty, Ops));
     Value *Call = Builder.CreateCall(F, Ops);
 
     // Predicate results must be converted to svbool_t.
     if (auto PredTy = dyn_cast<llvm::VectorType>(Call->getType()))
       if (PredTy->getScalarType()->isIntegerTy(1))
         Call = EmitSVEPredicateCast(Call, cast<llvm::ScalableVectorType>(Ty));
 
-    return Call;
+    return FormSVEBuiltinResult(Call);
   }
 
   switch (BuiltinID) {
   default:
     return nullptr;
 
   case SVE::BI__builtin_sve_svmov_b_z: {
     // svmov_b_z(pg, op) <=> svand_b_z(pg, op, op)

clang/lib/CodeGen/CodeGenFunction.h

llvm::Value *EmitSVEGatherPrefetch(const SVETypeFlags &TypeFlags,
                                      SmallVectorImpl<llvm::Value *> &Ops,
                                      unsigned IntID);
   llvm::Value *EmitSVEStructLoad(const SVETypeFlags &TypeFlags,
                                  SmallVectorImpl<llvm::Value *> &Ops,
                                  unsigned IntID);
   llvm::Value *EmitSVEStructStore(const SVETypeFlags &TypeFlags,
                                   SmallVectorImpl<llvm::Value *> &Ops,
                                   unsigned IntID);
+  /// FormSVEBuiltinResult - Returns the struct of scalable vectors as a wider
+  /// vector. It extracts the scalable vector from the struct and inserts into
+  /// the wider vector. This avoids the error when allocating space in llvm
+  /// for struct of scalable vectors if a function returns struct.
+  llvm::Value *FormSVEBuiltinResult(llvm::Value *Call);
   llvm::Value *EmitAArch64SVEBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
 
   llvm::Value *EmitSMELd1St1(const SVETypeFlags &TypeFlags,
                              llvm::SmallVectorImpl<llvm::Value *> &Ops,
                              unsigned IntID);
   llvm::Value *EmitSMEReadWrite(const SVETypeFlags &TypeFlags,
                                 llvm::SmallVectorImpl<llvm::Value *> &Ops,
                                 unsigned IntID);

clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_pext.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+// REQUIRES: aarch64-registered-target
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -O1 -Werror -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -O1 -Werror -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+
+#include <arm_sve.h>
+
+// CHECK-LABEL: @test_svpext_lane_c8_0(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.pext.nxv16i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP0]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svpext_lane_c8_0u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.pext.nxv16i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP0]]
+//
+svbool_t test_svpext_lane_c8_0(svcount_t c) {
+  return svpext_lane_c8(c, 0);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c8_3(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.pext.nxv16i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP0]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svpext_lane_c8_3u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.pext.nxv16i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP0]]
+//
+svbool_t test_svpext_lane_c8_3(svcount_t c) {
+  return svpext_lane_c8(c, 3);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c16_0(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.pext.nxv8i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> [[TMP0]])
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z22test_svpext_lane_c16_0u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.pext.nxv8i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> [[TMP0]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+svbool_t test_svpext_lane_c16_0(svcount_t c) {
+  return svpext_lane_c16(c, 0);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c16_3(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.pext.nxv8i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> [[TMP0]])
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z22test_svpext_lane_c16_3u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.pext.nxv8i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv8i1(<vscale x 8 x i1> [[TMP0]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+svbool_t test_svpext_lane_c16_3(svcount_t c) {
+  return svpext_lane_c16(c, 3);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c32_0(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.pext.nxv4i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[TMP0]])
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z22test_svpext_lane_c32_0u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.pext.nxv4i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[TMP0]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+svbool_t test_svpext_lane_c32_0(svcount_t c) {
+  return svpext_lane_c32(c, 0);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c32_3(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.pext.nxv4i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[TMP0]])
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z22test_svpext_lane_c32_3u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.pext.nxv4i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv4i1(<vscale x 4 x i1> [[TMP0]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+svbool_t test_svpext_lane_c32_3(svcount_t c) {
+  return svpext_lane_c32(c, 3);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c64_0(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.pext.nxv2i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[TMP0]])
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z22test_svpext_lane_c64_0u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.pext.nxv2i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[TMP0]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+svbool_t test_svpext_lane_c64_0(svcount_t c) {
+  return svpext_lane_c64(c, 0);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c64_3(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.pext.nxv2i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[TMP0]])
+// CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z22test_svpext_lane_c64_3u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.pext.nxv2i1(target("aarch64.svcount") [[C:%.*]], i32 3)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 16 x i1> @llvm.aarch64.sve.convert.to.svbool.nxv2i1(<vscale x 2 x i1> [[TMP0]])
+// CPP-CHECK-NEXT:    ret <vscale x 16 x i1> [[TMP1]]
+//
+svbool_t test_svpext_lane_c64_3(svcount_t c) {
+  return svpext_lane_c64(c, 3);
+}
+
+// CHECK-LABEL: @test_svpext_lane_c8_x2_0(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i1>, <vscale x 16 x i1> } @llvm.aarch64.sve.pext.x2.nxv16i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i1>, <vscale x 16 x i1> } [[TMP0]], 0
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i1> @llvm.vector.insert.nxv32i1.nxv16i1(<vscale x 32 x i1> poison, <vscale x 16 x i1> [[TMP1]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i1>, <vscale x 16 x i1> } [[TMP0]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i1> @llvm.vector.insert.nxv32i1.nxv16i1(<vscale x 32 x i1> [[TMP2]], <vscale x 16 x i1> [[TMP3]], i64 16)
+// CHECK-NEXT:    ret <vscale x 32 x i1> [[TMP4]]
+//
+// CPP-CHECK-LABEL: @_Z24test_svpext_lane_c8_x2_0u11__SVCount_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call { <vscale x 16 x i1>, <vscale x 16 x i1> } @llvm.aarch64.sve.pext.x2.nxv16i1(target("aarch64.svcount") [[C:%.*]], i32 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = extractvalue { <vscale x 16 x i1>, <vscale x 16 x i1> } [[TMP0]], 0
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 32 x i1> @llvm.vector.insert.nxv32i1.nxv16i1(<vscale x 32 x i1> poison, <vscale x 16 x i1> [[TMP1]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 16 x i1>, <vscale x 16 x i1> } [[TMP0]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 32 x i1> @llvm.vector.insert.nxv32i1.nxv16i1(<vscale x 32 x i1> [[TMP2]], <vscale x 16 x i1> [[TMP3]], i64 16)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x i1> [[TMP4]]
+//
+svboolx2_t test_svpext_lane_c8_x2_0(svcount_t c) {
+  return svpext_lane_c8_x2(c, 0);
+}

clang/test/Sema/aarch64-sve2p1-intrinsics/acle_sve2p1_imm.cpp

 // RUN: %clang_cc1 -triple aarch14-none-linux-gnu -target-feature +sve2p1 -fsyntax-only -verify %s
 
 // REQUIRES: aarch14-registered-target
 
 #include <arm_sve.h>
+void test_svpext_lane_imm_0_3(svcount_t c) {
+  svpext_lane_c8(c, -1);  // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 3]}}
+  svpext_lane_c16(c, -1); // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 3]}}
+  svpext_lane_c32(c, -1); // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 3]}}
+  svpext_lane_c64(c, -1); // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 3]}}
+
+  svpext_lane_c8(c, 4);  // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  svpext_lane_c16(c, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  svpext_lane_c32(c, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+  svpext_lane_c64(c, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
+}
+
+void test_svpext_lane_x2_imm_0_1(svcount_t c) {
+  svpext_lane_c8_x2(c, -1);  // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 1]}}
+  svpext_lane_c16_x2(c, -1); // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 1]}}
+  svpext_lane_c32_x2(c, -1); // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 1]}}
+  svpext_lane_c64_x2(c, -1); // expected-error {{argument value 18446744073709551615 is outside the valid range [0, 1]}}
+
+  svpext_lane_c8_x2(c, 2);  // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  svpext_lane_c16_x2(c, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  svpext_lane_c32_x2(c, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+  svpext_lane_c64_x2(c, 2); // expected-error {{argument value 2 is outside the valid range [0, 1]}}
+}
 
 void test_cntp(svcount_t c) {
   svcntp_c8(c, 1);  // expected-error {{argument value 1 is outside the valid range [2, 4]}}
   svcntp_c11(c, 1); // expected-error {{argument value 1 is outside the valid range [2, 4]}}
   svcntp_c32(c, 1); // expected-error {{argument value 1 is outside the valid range [2, 4]}}
   svcntp_c14(c, 1); // expected-error {{argument value 1 is outside the valid range [2, 4]}}
 
   svcntp_c8(c, 3);  // expected-error {{argument should be a multiple of 2}}
   svcntp_c11(c, 3); // expected-error {{argument should be a multiple of 2}}
   svcntp_c32(c, 3); // expected-error {{argument should be a multiple of 2}}
   svcntp_c14(c, 3); // expected-error {{argument should be a multiple of 2}}
 }
 

clang/utils/TableGen/SveEmitter.cpp

class SVEType {
   bool Float, Signed, Immediate, Void, Constant, Pointer, BFloat;
   bool DefaultType, IsScalable, Predicate, PredicatePattern, PrefetchOp,
       Svcount;
   unsigned Bitwidth, ElementBitwidth, NumVectors;
 
 public:
   SVEType() : SVEType(TypeSpec(), 'v') {}
 
-  SVEType(TypeSpec TS, char CharMod)
+  SVEType(TypeSpec TS, char CharMod, unsigned NumVectors = 1)
       : TS(TS), Float(false), Signed(true), Immediate(false), Void(false),
         Constant(false), Pointer(false), BFloat(false), DefaultType(false),
         IsScalable(true), Predicate(false), PredicatePattern(false),
         PrefetchOp(false), Svcount(false), Bitwidth(128), ElementBitwidth(~0U),
-        NumVectors(1) {
+        NumVectors(NumVectors) {
     if (!TS.empty())
       applyTypespec();
     applyModifier(CharMod);
   }
 
   bool isPointer() const { return Pointer; }
   bool isVoidPointer() const { return Pointer && Void; }
   bool isSigned() const { return Signed; }
public:
   SVEType getBaseType() const { return BaseType; }
 
   StringRef getGuard() const { return Guard; }
   ClassKind getClassKind() const { return Class; }
 
   SVEType getReturnType() const { return Types[0]; }
   ArrayRef<SVEType> getTypes() const { return Types; }
   SVEType getParamType(unsigned I) const { return Types[I + 1]; }
-  unsigned getNumParams() const { return Proto.size() - 1; }
+  unsigned getNumParams() const {
+    return Proto.size() - (2 * std::count(Proto.begin(), Proto.end(), '.')) - 1;
+  }
 
   uint64_t getFlags() const { return Flags; }
   bool isFlagSet(uint64_t Flag) const { return Flags & Flag;}
 
   ArrayRef<ImmCheck> getImmChecks() const { return ImmChecks; }
 
   /// Return the type string for a BUILTIN() macro in Builtins.def.
   std::string getBuiltinTypeStr();
public:
   /// Return true if the intrinsic takes a splat operand.
   bool hasSplat() const {
     // These prototype modifiers are described in arm_sve.td.
     return Proto.find_first_of("ajfrKLR@") != std::string::npos;
   }
 
   /// Return the parameter index of the splat operand.
   unsigned getSplatIdx() const {
-    // These prototype modifiers are described in arm_sve.td.
-    auto Idx = Proto.find_first_of("ajfrKLR@");
-    assert(Idx != std::string::npos && Idx > 0 &&
-           "Prototype has no splat operand");
-    return Idx - 1;
+    unsigned I = 1, Param = 0;
+    for (; I < Proto.size(); ++I, ++Param) {
+      if (Proto[I] == 'a' || Proto[I] == 'j' || Proto[I] == 'f' ||
+          Proto[I] == 'r' || Proto[I] == 'K' || Proto[I] == 'L' ||
+          Proto[I] == 'R' || Proto[I] == '@')
+        break;
+
+      // Multivector modifier can be skipped
+      if (Proto[I] == '.')
+        I += 2;
+    }
+    assert(I != Proto.size() && "Prototype has no splat operand");
+    return Param;
   }
 
   /// Emits the intrinsic declaration to the ostream.
   void emitIntrinsic(raw_ostream &OS, SVEEmitter &Emitter) const;
 
 private:
   std::string getMergeSuffix() const { return MergeSuffix; }
   std::string mangleName(ClassKind LocalCK) const;
default:
       llvm_unreachable("Unhandled type code!");
     }
   }
   assert(ElementBitwidth != ~0U && "Bad element bitwidth!");
 }
 
 void SVEType::applyModifier(char Mod) {
   switch (Mod) {
-  case '2':
-    NumVectors = 2;
-    break;
-  case '3':
-    NumVectors = 3;
-    break;
-  case '4':
-    NumVectors = 4;
-    break;
   case 'v':
     Void = true;
     break;
   case 'd':
     DefaultType = true;
     break;
   case 'c':
     Constant = true;
void SVEType::applyModifier(char Mod) {
   case '}':
     Predicate = false;
     Signed = true;
     Svcount = true;
     NumVectors = 0;
     Float = false;
     BFloat = false;
     break;
+  case '.':
+    llvm_unreachable(". is never a type in itself");
+    break;
   default:
     llvm_unreachable("Unhandled character!");
   }
 }
 
+/// Returns the modifier and number of vectors for the given operand \p Op.
+std::pair<char, unsigned> getProtoModifier(StringRef Proto, unsigned Op) {
+  for (unsigned P = 0; !Proto.empty(); ++P) {
+    unsigned NumVectors = 1;
+    unsigned CharsToSkip = 1;
+    char Mod = Proto[0];
+    if (Mod == '2' || Mod == '3' || Mod == '4') {
+      NumVectors = Mod - '0';
+      Mod = 'd';
+      if (Proto.size() > 1 && Proto[1] == '.') {
+        Mod = Proto[2];
+        CharsToSkip = 3;
+      }
+    }
+
+    if (P == Op)
+      return {Mod, NumVectors};
+
+    Proto = Proto.drop_front(CharsToSkip);
+  }
+  llvm_unreachable("Unexpected Op");
+}
 
 //===----------------------------------------------------------------------===//
 // Intrinsic implementation
 //===----------------------------------------------------------------------===//
 
 Intrinsic::Intrinsic(StringRef Name, StringRef Proto, uint64_t MergeTy,
                      StringRef MergeSuffix, uint64_t MemoryElementTy,
                      StringRef LLVMName, uint64_t Flags,
                      ArrayRef<ImmCheck> Checks, TypeSpec BT, ClassKind Class,
                      SVEEmitter &Emitter, StringRef Guard)
     : Name(Name.str()), LLVMName(LLVMName), Proto(Proto.str()),
       BaseTypeSpec(BT), Class(Class), Guard(Guard.str()),
       MergeSuffix(MergeSuffix.str()), BaseType(BT, 'd'), Flags(Flags),
       ImmChecks(Checks.begin(), Checks.end()) {
   // Types[0] is the return value.
-  for (unsigned I = 0; I < Proto.size(); ++I) {
-    SVEType T(BaseTypeSpec, Proto[I]);
+  for (unsigned I = 0; I < (getNumParams() + 1); ++I) {
+    char Mod;
+    unsigned NumVectors;
+    std::tie(Mod, NumVectors) = getProtoModifier(Proto, I);
+    SVEType T(BaseTypeSpec, Mod, NumVectors);
     Types.push_back(T);
 
     // Add range checks for immediates
     if (I > 0) {
       if (T.isPredicatePattern())
         ImmChecks.emplace_back(
             I - 1, Emitter.getEnumValueForImmCheck("ImmCheck0_31"));
       else if (T.isPrefetchOp())
for (auto TS : TypeSpecs) {
     SmallVector<ImmCheck, 2> ImmChecks;
     for (auto *R : ImmCheckList) {
       int64_t Arg = R->getValueAsInt("Arg");
       int64_t EltSizeArg = R->getValueAsInt("EltSizeArg");
       int64_t Kind = R->getValueAsDef("Kind")->getValueAsInt("Value");
       assert(Arg >= 0 && Kind >= 0 && "Arg and Kind must be nonnegative");
 
       unsigned ElementSizeInBits = 0;
+      char Mod;
+      unsigned NumVectors;
+      std::tie(Mod, NumVectors) = getProtoModifier(Proto, EltSizeArg + 1);
       if (EltSizeArg >= 0)
-        ElementSizeInBits =
-            SVEType(TS, Proto[EltSizeArg + /* offset by return arg */ 1])
-                .getElementSizeInBits();
+        ElementSizeInBits = SVEType(TS, Mod, NumVectors).getElementSizeInBits();
       ImmChecks.push_back(ImmCheck(Arg, Kind, ElementSizeInBits));
     }
 
     Out.push_back(std::make_unique<Intrinsic>(
         Name, Proto, Merge, MergeSuffix, MemEltType, LLVMName, Flags, ImmChecks,
         TS, ClassS, *this, Guard));
 
     // Also generate the short-form (e.g. svadd_m) for the given type-spec.