[AArch64][SVE] Lower MULHU/MULHS nodes to umulh/smulh instructions

Mark MULHS/MULHU nodes as legal for both scalable and fixed SVE types,
and lower them to the appropriate SVE instructions.

Additionally now that the MULH nodes are legal, integer divides can be
expanded into a more performant code sequence.

Differential Revision: https://reviews.llvm.org/D100487

9 files changed, 1452 insertions, 7 deletions

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 5b94a770556a..22e377ea0b74 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1126,6 +1126,8 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
       setOperationAction(ISD::MGATHER, VT, Custom);
       setOperationAction(ISD::MSCATTER, VT, Custom);
       setOperationAction(ISD::MUL, VT, Custom);
+      setOperationAction(ISD::MULHS, VT, Custom);
+      setOperationAction(ISD::MULHU, VT, Custom);
       setOperationAction(ISD::SPLAT_VECTOR, VT, Custom);
       setOperationAction(ISD::SELECT, VT, Custom);
       setOperationAction(ISD::SETCC, VT, Custom);
@@ -1149,8 +1151,6 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
       setOperationAction(ISD::VECREDUCE_SMAX, VT, Custom);
       setOperationAction(ISD::STEP_VECTOR, VT, Custom);
 
-      setOperationAction(ISD::MULHU, VT, Expand);
-      setOperationAction(ISD::MULHS, VT, Expand);
       setOperationAction(ISD::UMUL_LOHI, VT, Expand);
       setOperationAction(ISD::SMUL_LOHI, VT, Expand);
     }
@@ -1259,6 +1259,10 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
       setOperationAction(ISD::CTTZ, MVT::v1i64, Custom);
       setOperationAction(ISD::MUL, MVT::v1i64, Custom);
       setOperationAction(ISD::MUL, MVT::v2i64, Custom);
+      setOperationAction(ISD::MULHS, MVT::v1i64, Custom);
+      setOperationAction(ISD::MULHS, MVT::v2i64, Custom);
+      setOperationAction(ISD::MULHU, MVT::v1i64, Custom);
+      setOperationAction(ISD::MULHU, MVT::v2i64, Custom);
       setOperationAction(ISD::SDIV, MVT::v8i8, Custom);
       setOperationAction(ISD::SDIV, MVT::v16i8, Custom);
       setOperationAction(ISD::SDIV, MVT::v4i16, Custom);
@@ -1453,6 +1457,8 @@ void AArch64TargetLowering::addTypeForFixedLengthSVE(MVT VT) {
   setOperationAction(ISD::FTRUNC, VT, Custom);
   setOperationAction(ISD::LOAD, VT, Custom);
   setOperationAction(ISD::MUL, VT, Custom);
+  setOperationAction(ISD::MULHS, VT, Custom);
+  setOperationAction(ISD::MULHU, VT, Custom);
   setOperationAction(ISD::OR, VT, Custom);
   setOperationAction(ISD::SDIV, VT, Custom);
   setOperationAction(ISD::SELECT, VT, Custom);
@@ -1799,6 +1805,8 @@ const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
     MAKE_CASE(AArch64ISD::TLSDESC_CALLSEQ)
     MAKE_CASE(AArch64ISD::ADD_PRED)
     MAKE_CASE(AArch64ISD::MUL_PRED)
+    MAKE_CASE(AArch64ISD::MULHS_PRED)
+    MAKE_CASE(AArch64ISD::MULHU_PRED)
     MAKE_CASE(AArch64ISD::SDIV_PRED)
     MAKE_CASE(AArch64ISD::SHL_PRED)
     MAKE_CASE(AArch64ISD::SMAX_PRED)
@@ -4519,6 +4527,12 @@ SDValue AArch64TargetLowering::LowerOperation(SDValue Op,
     return LowerSET_ROUNDING(Op, DAG);
   case ISD::MUL:
     return LowerMUL(Op, DAG);
+  case ISD::MULHS:
+    return LowerToPredicatedOp(Op, DAG, AArch64ISD::MULHS_PRED,
+                               /*OverrideNEON=*/true);
+  case ISD::MULHU:
+    return LowerToPredicatedOp(Op, DAG, AArch64ISD::MULHU_PRED,
+                               /*OverrideNEON=*/true);
   case ISD::INTRINSIC_WO_CHAIN:
     return LowerINTRINSIC_WO_CHAIN(Op, DAG);
   case ISD::STORE:
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
index 63df22326150..a5530c814f7e 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -89,6 +89,8 @@ enum NodeType : unsigned {
   FMUL_PRED,
   FSUB_PRED,
   MUL_PRED,
+  MULHS_PRED,
+  MULHU_PRED,
   SDIV_PRED,
   SHL_PRED,
   SMAX_PRED,
diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 1bd9b1fff207..c2d46441678a 100644
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -192,10 +192,12 @@ def AArch64mul_p  : SDNode<"AArch64ISD::MUL_PRED",  SDT_AArch64Arith>;
 def AArch64sdiv_p : SDNode<"AArch64ISD::SDIV_PRED", SDT_AArch64Arith>;
 def AArch64smax_p : SDNode<"AArch64ISD::SMAX_PRED", SDT_AArch64Arith>;
 def AArch64smin_p : SDNode<"AArch64ISD::SMIN_PRED", SDT_AArch64Arith>;
+def AArch64smulh_p : SDNode<"AArch64ISD::MULHS_PRED", SDT_AArch64Arith>;
 def AArch64sub_p  : SDNode<"AArch64ISD::SUB_PRED",  SDT_AArch64Arith>;
 def AArch64udiv_p : SDNode<"AArch64ISD::UDIV_PRED", SDT_AArch64Arith>;
 def AArch64umax_p : SDNode<"AArch64ISD::UMAX_PRED", SDT_AArch64Arith>;
 def AArch64umin_p : SDNode<"AArch64ISD::UMIN_PRED", SDT_AArch64Arith>;
+def AArch64umulh_p : SDNode<"AArch64ISD::MULHU_PRED", SDT_AArch64Arith>;
 
 def SDT_AArch64IntExtend : SDTypeProfile<1, 4, [
   SDTCisVec<0>, SDTCisVec<1>, SDTCisVec<2>, SDTCisVT<3, OtherVT>, SDTCisVec<4>,
@@ -348,6 +350,8 @@ let Predicates = [HasSVE] in {
   defm UMULH_ZPmZ : sve_int_bin_pred_arit_2<0b011, "umulh", "UMULH_ZPZZ", int_aarch64_sve_umulh, DestructiveBinaryComm>;
 
   defm MUL_ZPZZ   : sve_int_bin_pred_bhsd<AArch64mul_p>;
+  defm SMULH_ZPZZ : sve_int_bin_pred_bhsd<AArch64smulh_p>;
+  defm UMULH_ZPZZ : sve_int_bin_pred_bhsd<AArch64umulh_p>;
 
   defm SDIV_ZPmZ  : sve_int_bin_pred_arit_2_div<0b100, "sdiv",  "SDIV_ZPZZ", int_aarch64_sve_sdiv, DestructiveBinaryCommWithRev, "SDIVR_ZPmZ">;
   defm UDIV_ZPmZ  : sve_int_bin_pred_arit_2_div<0b101, "udiv",  "UDIV_ZPZZ", int_aarch64_sve_udiv, DestructiveBinaryCommWithRev, "UDIVR_ZPmZ">;
@@ -2385,8 +2389,8 @@ let Predicates = [HasSVE2] in {
 
   // SVE2 integer multiply vectors (unpredicated)
   defm MUL_ZZZ    : sve2_int_mul<0b000,  "mul",   null_frag, AArch64mul_p>;
-  defm SMULH_ZZZ  : sve2_int_mul<0b010,  "smulh", null_frag>;
-  defm UMULH_ZZZ  : sve2_int_mul<0b011,  "umulh", null_frag>;
+  defm SMULH_ZZZ  : sve2_int_mul<0b010,  "smulh", null_frag, AArch64smulh_p>;
+  defm UMULH_ZZZ  : sve2_int_mul<0b011,  "umulh", null_frag, AArch64umulh_p>;
   defm PMUL_ZZZ   : sve2_int_mul_single<0b001, "pmul", int_aarch64_sve_pmul>;
 
   // Add patterns for unpredicated version of smulh and umulh.
diff --git a/llvm/test/CodeGen/AArch64/sve-expand-div.ll b/llvm/test/CodeGen/AArch64/sve-expand-div.ll
new file mode 100644
index 000000000000..185054a3ff3e
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-expand-div.ll
@@ -0,0 +1,144 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-linux-gnu < %s | FileCheck %s
+
+; Check that expensive divides are expanded into a more performant sequence
+
+;
+; SDIV
+;
+
+define <vscale x 16 x i8> @sdiv_i8(<vscale x 16 x i8> %a) #0 {
+; CHECK-LABEL: sdiv_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov z1.b, #86 // =0x56
+; CHECK-NEXT:    ptrue p0.b
+; CHECK-NEXT:    smulh z0.b, p0/m, z0.b, z1.b
+; CHECK-NEXT:    lsr z1.b, z0.b, #7
+; CHECK-NEXT:    mov z2.b, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and z1.d, z1.d, z2.d
+; CHECK-NEXT:    add z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %div = sdiv <vscale x 16 x i8> %a, shufflevector (<vscale x 16 x i8> insertelement (<vscale x 16 x i8> undef, i8 3, i32 0), <vscale x 16 x i8> undef, <vscale x 16 x i32> zeroinitializer)
+  ret <vscale x 16 x i8> %div
+}
+
+define <vscale x 8 x i16> @sdiv_i16(<vscale x 8 x i16> %a) #0 {
+; CHECK-LABEL: sdiv_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, #21846
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    mov z1.h, w8
+; CHECK-NEXT:    smulh z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT:    lsr z1.h, z0.h, #15
+; CHECK-NEXT:    mov z2.h, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and z1.d, z1.d, z2.d
+; CHECK-NEXT:    add z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %div = sdiv <vscale x 8 x i16> %a, shufflevector (<vscale x 8 x i16> insertelement (<vscale x 8 x i16> undef, i16 3, i32 0), <vscale x 8 x i16> undef, <vscale x 8 x i32> zeroinitializer)
+  ret <vscale x 8 x i16> %div
+}
+
+define <vscale x 4 x i32> @sdiv_i32(<vscale x 4 x i32> %a) #0 {
+; CHECK-LABEL: sdiv_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, #21846
+; CHECK-NEXT:    movk w8, #21845, lsl #16
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    mov z1.s, w8
+; CHECK-NEXT:    smulh z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT:    lsr z1.s, z0.s, #31
+; CHECK-NEXT:    mov z2.s, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and z1.d, z1.d, z2.d
+; CHECK-NEXT:    add z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %div = sdiv <vscale x 4 x i32> %a, shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> undef, i32 3, i32 0), <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer)
+  ret <vscale x 4 x i32> %div
+}
+
+define <vscale x 2 x i64> @sdiv_i64(<vscale x 2 x i64> %a) #0 {
+; CHECK-LABEL: sdiv_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov x8, #6148914691236517205
+; CHECK-NEXT:    movk x8, #21846
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    mov z1.d, x8
+; CHECK-NEXT:    smulh z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT:    lsr z1.d, z0.d, #63
+; CHECK-NEXT:    mov z2.d, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and z1.d, z1.d, z2.d
+; CHECK-NEXT:    add z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %div = sdiv <vscale x 2 x i64> %a, shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> undef, i64 3, i32 0), <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer)
+  ret <vscale x 2 x i64> %div
+}
+
+;
+; UDIV
+;
+
+define <vscale x 16 x i8> @udiv_i8(<vscale x 16 x i8> %a) #0 {
+; CHECK-LABEL: udiv_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov z1.b, #-85 // =0xffffffffffffffab
+; CHECK-NEXT:    ptrue p0.b
+; CHECK-NEXT:    mov z2.b, #1 // =0x1
+; CHECK-NEXT:    umulh z1.b, p0/m, z1.b, z0.b
+; CHECK-NEXT:    lsr z1.b, z1.b, #1
+; CHECK-NEXT:    cmpeq p0.b, p0/z, z2.b, #3
+; CHECK-NEXT:    sel z0.b, p0, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %div = udiv <vscale x 16 x i8> %a, shufflevector (<vscale x 16 x i8> insertelement (<vscale x 16 x i8> undef, i8 3, i32 0), <vscale x 16 x i8> undef, <vscale x 16 x i32> zeroinitializer)
+  ret <vscale x 16 x i8> %div
+}
+
+define <vscale x 8 x i16> @udiv_i16(<vscale x 8 x i16> %a) #0 {
+; CHECK-LABEL: udiv_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, #-21845
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    mov z2.h, w8
+; CHECK-NEXT:    mov z1.h, #1 // =0x1
+; CHECK-NEXT:    umulh z2.h, p0/m, z2.h, z0.h
+; CHECK-NEXT:    lsr z2.h, z2.h, #1
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z1.h, #3
+; CHECK-NEXT:    sel z0.h, p0, z0.h, z2.h
+; CHECK-NEXT:    ret
+  %div = udiv <vscale x 8 x i16> %a, shufflevector (<vscale x 8 x i16> insertelement (<vscale x 8 x i16> undef, i16 3, i32 0), <vscale x 8 x i16> undef, <vscale x 8 x i32> zeroinitializer)
+  ret <vscale x 8 x i16> %div
+}
+
+define <vscale x 4 x i32> @udiv_i32(<vscale x 4 x i32> %a) #0 {
+; CHECK-LABEL: udiv_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w8, #43691
+; CHECK-NEXT:    movk w8, #43690, lsl #16
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    mov z2.s, w8
+; CHECK-NEXT:    mov z1.s, #3 // =0x3
+; CHECK-NEXT:    umulh z2.s, p0/m, z2.s, z0.s
+; CHECK-NEXT:    lsr z2.s, z2.s, #1
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z1.s, #1
+; CHECK-NEXT:    sel z0.s, p0, z0.s, z2.s
+; CHECK-NEXT:    ret
+  %div = udiv <vscale x 4 x i32> %a, shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> undef, i32 3, i32 0), <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer)
+  ret <vscale x 4 x i32> %div
+}
+
+define <vscale x 2 x i64> @udiv_i64(<vscale x 2 x i64> %a) #0 {
+; CHECK-LABEL: udiv_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov x8, #-6148914691236517206
+; CHECK-NEXT:    movk x8, #43691
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    mov z2.d, x8
+; CHECK-NEXT:    mov z1.d, #3 // =0x3
+; CHECK-NEXT:    umulh z2.d, p0/m, z2.d, z0.d
+; CHECK-NEXT:    lsr z2.d, z2.d, #1
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z1.d, #1
+; CHECK-NEXT:    sel z0.d, p0, z0.d, z2.d
+; CHECK-NEXT:    ret
+  %div = udiv <vscale x 2 x i64> %a, shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> undef, i64 3, i32 0), <vscale x 2 x i64> undef, <vscale x 2 x i32> zeroinitializer)
+  ret <vscale x 2 x i64> %div
+}
+
+attributes #0 = { "target-features"="+sve" }
diff --git a/llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll b/llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll
index b70228260850..1fe10c265f6f 100644
--- a/llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll
+++ b/llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll
@@ -970,7 +970,7 @@ define void @udiv_v32i64(<32 x i64>* %a, <32 x i64>* %b) #0 {
 
 ; This used to crash because isUnaryPredicate and BuildUDIV don't know how
 ; a SPLAT_VECTOR of fixed vector type should be handled.
-define void @udiv_constantsplat_v8i32(<8 x i32>* %a) #0 {
+define void @udiv_constantsplat_v8i32(<8 x i32>* %a) #1 {
 ; CHECK-LABEL: udiv_constantsplat_v8i32:
 ; CHECK: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,4),8)]]
 ; CHECK-NEXT: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
@@ -985,3 +985,4 @@ define void @udiv_constantsplat_v8i32(<8 x i32>* %a) #0 {
 }
 
 attributes #0 = { "target-features"="+sve" }
+attributes #1 = { "target-features"="+sve" minsize }
diff --git a/llvm/test/CodeGen/AArch64/sve-fixed-length-int-mulh.ll b/llvm/test/CodeGen/AArch64/sve-fixed-length-int-mulh.ll
new file mode 100644
index 000000000000..48bae1bb813a
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-fixed-length-int-mulh.ll
@@ -0,0 +1,1006 @@
+; RUN: llc -aarch64-sve-vector-bits-min=128  < %s | FileCheck %s -D#VBYTES=16  -check-prefix=NO_SVE
+; RUN: llc -aarch64-sve-vector-bits-min=256  < %s | FileCheck %s -D#VBYTES=32  -check-prefixes=CHECK,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=384  < %s | FileCheck %s -D#VBYTES=32  -check-prefixes=CHECK,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=512  < %s | FileCheck %s -D#VBYTES=64  -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=640  < %s | FileCheck %s -D#VBYTES=64  -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=768  < %s | FileCheck %s -D#VBYTES=64  -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=896  < %s | FileCheck %s -D#VBYTES=64  -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1024 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1152 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1280 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1408 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1536 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1664 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1792 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=1920 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+; RUN: llc -aarch64-sve-vector-bits-min=2048 < %s | FileCheck %s -D#VBYTES=256 -check-prefixes=CHECK,VBITS_GE_2048,VBITS_GE_1024,VBITS_GE_512,VBITS_GE_256
+
+; VBYTES represents the useful byte size of a vector register from the code
+; generator's point of view. It is clamped to power-of-2 values because
+; only power-of-2 vector lengths are considered legal, regardless of the
+; user specified vector length.
+
+; This test only tests the legal types for a given vector width, as mulh nodes
+; do not get generated for non-legal types.
+
+target triple = "aarch64-unknown-linux-gnu"
+
+; Don't use SVE when its registers are no bigger than NEON.
+; NO_SVE-NOT: ptrue
+
+;
+; SMULH
+;
+
+; Don't use SVE for 64-bit vectors.
+define <8 x i8> @smulh_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
+; CHECK-LABEL: smulh_v8i8:
+; CHECK: smull v0.8h, v0.8b, v1.8b
+; CHECK: ushr v1.8h, v0.8h, #8
+; CHECK: umov w8, v1.h[0]
+; CHECK: fmov s0, w8
+; CHECK: umov w8, v1.h[1]
+; CHECK: mov v0.b[1], w8
+; CHECK: umov w8, v1.h[2]
+; CHECK: mov v0.b[2], w8
+; CHECK: umov w8, v1.h[3]
+; CHECK: mov v0.b[3], w8
+; CHECK: ret
+  %insert = insertelement <8 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
+  %1 = sext <8 x i8> %op1 to <8 x i16>
+  %2 = sext <8 x i8> %op2 to <8 x i16>
+  %mul = mul <8 x i16> %1, %2
+  %shr = lshr <8 x i16> %mul, %splat
+  %res = trunc <8 x i16> %shr to <8 x i8>
+  ret <8 x i8> %res
+}
+
+; Don't use SVE for 128-bit vectors.
+define <16 x i8> @smulh_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
+; CHECK-LABEL: smulh_v16i8:
+; CHECK: smull2 v2.8h, v0.16b, v1.16b
+; CHECK: smull v0.8h, v0.8b, v1.8b
+; CHECK: uzp2 v0.16b, v0.16b, v2.16b
+; CHECK: ret
+  %insert = insertelement <16 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
+  %1 = sext <16 x i8> %op1 to <16 x i16>
+  %2 = sext <16 x i8> %op2 to <16 x i16>
+  %mul = mul <16 x i16> %1, %2
+  %shr = lshr <16 x i16> %mul, %splat
+  %res = trunc <16 x i16> %shr to <16 x i8>
+  ret <16 x i8> %res
+}
+
+define void @smulh_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
+; CHECK-LABEL: smulh_v32i8:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,32)]]
+; VBITS_GE_256-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_256: smulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_256: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <32 x i8>, <32 x i8>* %a
+  %op2 = load <32 x i8>, <32 x i8>* %b
+  %insert = insertelement <32 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
+  %1 = sext <32 x i8> %op1 to <32 x i16>
+  %2 = sext <32 x i8> %op2 to <32 x i16>
+  %mul = mul <32 x i16> %1, %2
+  %shr = lshr <32 x i16> %mul, %splat
+  %res = trunc <32 x i16> %shr to <32 x i8>
+  store <32 x i8> %res, <32 x i8>* %a
+  ret void
+}
+
+define void @smulh_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
+; CHECK-LABEL: smulh_v64i8:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,64)]]
+; VBITS_GE_512-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_512: smulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_512: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <64 x i8>, <64 x i8>* %a
+  %op2 = load <64 x i8>, <64 x i8>* %b
+  %insert = insertelement <64 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <64 x i16> %insert, <64 x i16> undef, <64 x i32> zeroinitializer
+  %1 = sext <64 x i8> %op1 to <64 x i16>
+  %2 = sext <64 x i8> %op2 to <64 x i16>
+  %mul = mul <64 x i16> %1, %2
+  %shr = lshr <64 x i16> %mul, %splat
+  %res = trunc <64 x i16> %shr to <64 x i8>
+  store <64 x i8> %res, <64 x i8>* %a
+  ret void
+}
+
+define void @smulh_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
+; CHECK-LABEL: smulh_v128i8:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,128)]]
+; VBITS_GE_1024-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_1024: smulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_1024: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <128 x i8>, <128 x i8>* %a
+  %op2 = load <128 x i8>, <128 x i8>* %b
+  %insert = insertelement <128 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <128 x i16> %insert, <128 x i16> undef, <128 x i32> zeroinitializer
+  %1 = sext <128 x i8> %op1 to <128 x i16>
+  %2 = sext <128 x i8> %op2 to <128 x i16>
+  %mul = mul <128 x i16> %1, %2
+  %shr = lshr <128 x i16> %mul, %splat
+  %res = trunc <128 x i16> %shr to <128 x i8>
+  store <128 x i8> %res, <128 x i8>* %a
+  ret void
+}
+
+define void @smulh_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
+; CHECK-LABEL: smulh_v256i8:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
+; VBITS_GE_2048-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_2048: smulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_2048: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <256 x i8>, <256 x i8>* %a
+  %op2 = load <256 x i8>, <256 x i8>* %b
+  %insert = insertelement <256 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <256 x i16> %insert, <256 x i16> undef, <256 x i32> zeroinitializer
+  %1 = sext <256 x i8> %op1 to <256 x i16>
+  %2 = sext <256 x i8> %op2 to <256 x i16>
+  %mul = mul <256 x i16> %1, %2
+  %shr = lshr <256 x i16> %mul, %splat
+  %res = trunc <256 x i16> %shr to <256 x i8>
+  store <256 x i8> %res, <256 x i8>* %a
+  ret void
+}
+
+; Don't use SVE for 64-bit vectors.
+define <4 x i16> @smulh_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
+; CHECK-LABEL: smulh_v4i16:
+; CHECK: smull v0.4s, v0.4h, v1.4h
+; CHECK: ushr v0.4s, v0.4s, #16
+; CHECK: mov w8, v0.s[1]
+; CHECK: mov w9, v0.s[2]
+; CHECK: mov w10, v0.s[3]
+; CHECK: mov v0.h[1], w8
+; CHECK: mov v0.h[2], w9
+; CHECK: mov v0.h[3], w10
+; CHECK: ret
+  %insert = insertelement <4 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
+  %1 = sext <4 x i16> %op1 to <4 x i32>
+  %2 = sext <4 x i16> %op2 to <4 x i32>
+  %mul = mul <4 x i32> %1, %2
+  %shr = lshr <4 x i32> %mul, %splat
+  %res = trunc <4 x i32> %shr to <4 x i16>
+  ret <4 x i16> %res
+}
+
+; Don't use SVE for 128-bit vectors.
+define <8 x i16> @smulh_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
+; CHECK-LABEL: smulh_v8i16:
+; CHECK: smull2 v2.4s, v0.8h, v1.8h
+; CHECK: smull v0.4s, v0.4h, v1.4h
+; CHECK: uzp2 v0.8h, v0.8h, v2.8h
+; CHECK: ret
+  %insert = insertelement <8 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
+  %1 = sext <8 x i16> %op1 to <8 x i32>
+  %2 = sext <8 x i16> %op2 to <8 x i32>
+  %mul = mul <8 x i32> %1, %2
+  %shr = lshr <8 x i32> %mul, %splat
+  %res = trunc <8 x i32> %shr to <8 x i16>
+  ret <8 x i16> %res
+}
+
+define void @smulh_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
+; CHECK-LABEL: smulh_v16i16:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,16)]]
+; VBITS_GE_256-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_256: smulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_256: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <16 x i16>, <16 x i16>* %a
+  %op2 = load <16 x i16>, <16 x i16>* %b
+  %insert = insertelement <16 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = sext <16 x i16> %op1 to <16 x i32>
+  %2 = sext <16 x i16> %op2 to <16 x i32>
+  %mul = mul <16 x i32> %1, %2
+  %shr = lshr <16 x i32> %mul, %splat
+  %res = trunc <16 x i32> %shr to <16 x i16>
+  store <16 x i16> %res, <16 x i16>* %a
+  ret void
+}
+
+define void @smulh_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
+; CHECK-LABEL: smulh_v32i16:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,32)]]
+; VBITS_GE_512-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_512: smulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_512: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <32 x i16>, <32 x i16>* %a
+  %op2 = load <32 x i16>, <32 x i16>* %b
+  %insert = insertelement <32 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <32 x i32> %insert, <32 x i32> undef, <32 x i32> zeroinitializer
+  %1 = sext <32 x i16> %op1 to <32 x i32>
+  %2 = sext <32 x i16> %op2 to <32 x i32>
+  %mul = mul <32 x i32> %1, %2
+  %shr = lshr <32 x i32> %mul, %splat
+  %res = trunc <32 x i32> %shr to <32 x i16>
+  store <32 x i16> %res, <32 x i16>* %a
+  ret void
+}
+
+define void @smulh_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
+; CHECK-LABEL: smulh_v64i16:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,64)]]
+; VBITS_GE_1024-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_1024: smulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_1024: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <64 x i16>, <64 x i16>* %a
+  %op2 = load <64 x i16>, <64 x i16>* %b
+  %insert = insertelement <64 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <64 x i32> %insert, <64 x i32> undef, <64 x i32> zeroinitializer
+  %1 = sext <64 x i16> %op1 to <64 x i32>
+  %2 = sext <64 x i16> %op2 to <64 x i32>
+  %mul = mul <64 x i32> %1, %2
+  %shr = lshr <64 x i32> %mul, %splat
+  %res = trunc <64 x i32> %shr to <64 x i16>
+  store <64 x i16> %res, <64 x i16>* %a
+  ret void
+}
+
+define void @smulh_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
+; CHECK-LABEL: smulh_v128i16:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,128)]]
+; VBITS_GE_2048-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_2048: smulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_2048: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <128 x i16>, <128 x i16>* %a
+  %op2 = load <128 x i16>, <128 x i16>* %b
+  %insert = insertelement <128 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <128 x i32> %insert, <128 x i32> undef, <128 x i32> zeroinitializer
+  %1 = sext <128 x i16> %op1 to <128 x i32>
+  %2 = sext <128 x i16> %op2 to <128 x i32>
+  %mul = mul <128 x i32> %1, %2
+  %shr = lshr <128 x i32> %mul, %splat
+  %res = trunc <128 x i32> %shr to <128 x i16>
+  store <128 x i16> %res, <128 x i16>* %a
+  ret void
+}
+
+; Vector i64 multiplications are not legal for NEON so use SVE when available.
+define <2 x i32> @smulh_v2i32(<2 x i32> %op1, <2 x i32> %op2) #0 {
+; CHECK-LABEL: smulh_v2i32:
+; CHECK: sshll v0.2d, v0.2s, #0
+; CHECK: sshll v1.2d, v1.2s, #0
+; CHECK: ptrue p0.d, vl2
+; CHECK: mul z0.d, p0/m, z0.d, z1.d
+; CHECK: shrn v0.2s, v0.2d, #32
+; CHECK: ret
+  %insert = insertelement <2 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
+  %1 = sext <2 x i32> %op1 to <2 x i64>
+  %2 = sext <2 x i32> %op2 to <2 x i64>
+  %mul = mul <2 x i64> %1, %2
+  %shr = lshr <2 x i64> %mul, %splat
+  %res = trunc <2 x i64> %shr to <2 x i32>
+  ret <2 x i32> %res
+}
+
+; Don't use SVE for 128-bit vectors.
+define <4 x i32> @smulh_v4i32(<4 x i32> %op1, <4 x i32> %op2) #0 {
+; CHECK-LABEL: smulh_v4i32:
+; CHECK: smull2 v2.2d, v0.4s, v1.4s
+; CHECK: smull v0.2d, v0.2s, v1.2s
+; CHECK: uzp2 v0.4s, v0.4s, v2.4s
+; CHECK: ret
+  %insert = insertelement <4 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
+  %1 = sext <4 x i32> %op1 to <4 x i64>
+  %2 = sext <4 x i32> %op2 to <4 x i64>
+  %mul = mul <4 x i64> %1, %2
+  %shr = lshr <4 x i64> %mul, %splat
+  %res = trunc <4 x i64> %shr to <4 x i32>
+  ret <4 x i32> %res
+}
+
+define void @smulh_v8i32(<8 x i32>* %a, <8 x i32>* %b) #0 {
+; CHECK-LABEL: smulh_v8i32:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,8)]]
+; VBITS_GE_256-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_256: smulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_256: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <8 x i32>, <8 x i32>* %a
+  %op2 = load <8 x i32>, <8 x i32>* %b
+  %insert = insertelement <8 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
+  %1 = sext <8 x i32> %op1 to <8 x i64>
+  %2 = sext <8 x i32> %op2 to <8 x i64>
+  %mul = mul <8 x i64> %1, %2
+  %shr = lshr <8 x i64> %mul, %splat
+  %res = trunc <8 x i64> %shr to <8 x i32>
+  store <8 x i32> %res, <8 x i32>* %a
+  ret void
+}
+
+define void @smulh_v16i32(<16 x i32>* %a, <16 x i32>* %b) #0 {
+; CHECK-LABEL: smulh_v16i32:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,16)]]
+; VBITS_GE_512-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_512: smulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_512: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <16 x i32>, <16 x i32>* %a
+  %op2 = load <16 x i32>, <16 x i32>* %b
+  %insert = insertelement <16 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <16 x i64> %insert, <16 x i64> undef, <16 x i32> zeroinitializer
+  %1 = sext <16 x i32> %op1 to <16 x i64>
+  %2 = sext <16 x i32> %op2 to <16 x i64>
+  %mul = mul <16 x i64> %1, %2
+  %shr = lshr <16 x i64> %mul, %splat
+  %res = trunc <16 x i64> %shr to <16 x i32>
+  store <16 x i32> %res, <16 x i32>* %a
+  ret void
+}
+
+define void @smulh_v32i32(<32 x i32>* %a, <32 x i32>* %b) #0 {
+; CHECK-LABEL: smulh_v32i32:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,32)]]
+; VBITS_GE_1024-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_1024: smulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_1024: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <32 x i32>, <32 x i32>* %a
+  %op2 = load <32 x i32>, <32 x i32>* %b
+  %insert = insertelement <32 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <32 x i64> %insert, <32 x i64> undef, <32 x i32> zeroinitializer
+  %1 = sext <32 x i32> %op1 to <32 x i64>
+  %2 = sext <32 x i32> %op2 to <32 x i64>
+  %mul = mul <32 x i64> %1, %2
+  %shr = lshr <32 x i64> %mul, %splat
+  %res = trunc <32 x i64> %shr to <32 x i32>
+  store <32 x i32> %res, <32 x i32>* %a
+  ret void
+}
+
+define void @smulh_v64i32(<64 x i32>* %a, <64 x i32>* %b) #0 {
+; CHECK-LABEL: smulh_v64i32:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
+; VBITS_GE_2048-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_2048: smulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_2048: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <64 x i32>, <64 x i32>* %a
+  %op2 = load <64 x i32>, <64 x i32>* %b
+  %insert = insertelement <64 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <64 x i64> %insert, <64 x i64> undef, <64 x i32> zeroinitializer
+  %1 = sext <64 x i32> %op1 to <64 x i64>
+  %2 = sext <64 x i32> %op2 to <64 x i64>
+  %mul = mul <64 x i64> %1, %2
+  %shr = lshr <64 x i64> %mul, %splat
+  %res = trunc <64 x i64> %shr to <64 x i32>
+  store <64 x i32> %res, <64 x i32>* %a
+  ret void
+}
+
+; Vector i64 multiplications are not legal for NEON so use SVE when available.
+define <1 x i64> @smulh_v1i64(<1 x i64> %op1, <1 x i64> %op2) #0 {
+; CHECK-LABEL: smulh_v1i64:
+; CHECK: ptrue p0.d, vl1
+; CHECK: smulh z0.d, p0/m, z0.d, z1.d
+; CHECK: ret
+  %insert = insertelement <1 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <1 x i128> %insert, <1 x i128> undef, <1 x i32> zeroinitializer
+  %1 = sext <1 x i64> %op1 to <1 x i128>
+  %2 = sext <1 x i64> %op2 to <1 x i128>
+  %mul = mul <1 x i128> %1, %2
+  %shr = lshr <1 x i128> %mul, %splat
+  %res = trunc <1 x i128> %shr to <1 x i64>
+  ret <1 x i64> %res
+}
+
+; Vector i64 multiplications are not legal for NEON so use SVE when available.
+define <2 x i64> @smulh_v2i64(<2 x i64> %op1, <2 x i64> %op2) #0 {
+; CHECK-LABEL: smulh_v2i64:
+; CHECK: ptrue p0.d, vl2
+; CHECK: smulh z0.d, p0/m, z0.d, z1.d
+; CHECK: ret
+  %insert = insertelement <2 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <2 x i128> %insert, <2 x i128> undef, <2 x i32> zeroinitializer
+  %1 = sext <2 x i64> %op1 to <2 x i128>
+  %2 = sext <2 x i64> %op2 to <2 x i128>
+  %mul = mul <2 x i128> %1, %2
+  %shr = lshr <2 x i128> %mul, %splat
+  %res = trunc <2 x i128> %shr to <2 x i64>
+  ret <2 x i64> %res
+}
+
+define void @smulh_v4i64(<4 x i64>* %a, <4 x i64>* %b) #0 {
+; CHECK-LABEL: smulh_v4i64:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,4)]]
+; VBITS_GE_256-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_256: smulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_256: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <4 x i64>, <4 x i64>* %a
+  %op2 = load <4 x i64>, <4 x i64>* %b
+  %insert = insertelement <4 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <4 x i128> %insert, <4 x i128> undef, <4 x i32> zeroinitializer
+  %1 = sext <4 x i64> %op1 to <4 x i128>
+  %2 = sext <4 x i64> %op2 to <4 x i128>
+  %mul = mul <4 x i128> %1, %2
+  %shr = lshr <4 x i128> %mul, %splat
+  %res = trunc <4 x i128> %shr to <4 x i64>
+  store <4 x i64> %res, <4 x i64>* %a
+  ret void
+}
+
+define void @smulh_v8i64(<8 x i64>* %a, <8 x i64>* %b) #0 {
+; CHECK-LABEL: smulh_v8i64:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,8)]]
+; VBITS_GE_512-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_512: smulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_512: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <8 x i64>, <8 x i64>* %a
+  %op2 = load <8 x i64>, <8 x i64>* %b
+  %insert = insertelement <8 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <8 x i128> %insert, <8 x i128> undef, <8 x i32> zeroinitializer
+  %1 = sext <8 x i64> %op1 to <8 x i128>
+  %2 = sext <8 x i64> %op2 to <8 x i128>
+  %mul = mul <8 x i128> %1, %2
+  %shr = lshr <8 x i128> %mul, %splat
+  %res = trunc <8 x i128> %shr to <8 x i64>
+  store <8 x i64> %res, <8 x i64>* %a
+  ret void
+}
+
+define void @smulh_v16i64(<16 x i64>* %a, <16 x i64>* %b) #0 {
+; CHECK-LABEL: smulh_v16i64:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,16)]]
+; VBITS_GE_1024-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_1024: smulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_1024: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <16 x i64>, <16 x i64>* %a
+  %op2 = load <16 x i64>, <16 x i64>* %b
+  %insert = insertelement <16 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <16 x i128> %insert, <16 x i128> undef, <16 x i32> zeroinitializer
+  %1 = sext <16 x i64> %op1 to <16 x i128>
+  %2 = sext <16 x i64> %op2 to <16 x i128>
+  %mul = mul <16 x i128> %1, %2
+  %shr = lshr <16 x i128> %mul, %splat
+  %res = trunc <16 x i128> %shr to <16 x i64>
+  store <16 x i64> %res, <16 x i64>* %a
+  ret void
+}
+
+define void @smulh_v32i64(<32 x i64>* %a, <32 x i64>* %b) #0 {
+; CHECK-LABEL: smulh_v32i64:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,32)]]
+; VBITS_GE_2048-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_2048: smulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_2048: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <32 x i64>, <32 x i64>* %a
+  %op2 = load <32 x i64>, <32 x i64>* %b
+  %insert = insertelement <32 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <32 x i128> %insert, <32 x i128> undef, <32 x i32> zeroinitializer
+  %1 = sext <32 x i64> %op1 to <32 x i128>
+  %2 = sext <32 x i64> %op2 to <32 x i128>
+  %mul = mul <32 x i128> %1, %2
+  %shr = lshr <32 x i128> %mul, %splat
+  %res = trunc <32 x i128> %shr to <32 x i64>
+  store <32 x i64> %res, <32 x i64>* %a
+  ret void
+}
+
+;
+; UMULH
+;
+
+; Don't use SVE for 64-bit vectors.
+define <8 x i8> @umulh_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
+; CHECK-LABEL: umulh_v8i8:
+; CHECK: umull v0.8h, v0.8b, v1.8b
+; CHECK: ushr v1.8h, v0.8h, #8
+; CHECK: umov w8, v1.h[0]
+; CHECK: fmov s0, w8
+; CHECK: umov w8, v1.h[1]
+; CHECK: mov v0.b[1], w8
+; CHECK: umov w8, v1.h[2]
+; CHECK: mov v0.b[2], w8
+; CHECK: umov w8, v1.h[3]
+; CHECK: mov v0.b[3], w8
+; CHECK: ret
+  %insert = insertelement <8 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer
+  %1 = zext <8 x i8> %op1 to <8 x i16>
+  %2 = zext <8 x i8> %op2 to <8 x i16>
+  %mul = mul <8 x i16> %1, %2
+  %shr = lshr <8 x i16> %mul, %splat
+  %res = trunc <8 x i16> %shr to <8 x i8>
+  ret <8 x i8> %res
+}
+
+; Don't use SVE for 128-bit vectors.
+define <16 x i8> @umulh_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
+; CHECK-LABEL: umulh_v16i8:
+; CHECK: umull2 v2.8h, v0.16b, v1.16b
+; CHECK: umull v0.8h, v0.8b, v1.8b
+; CHECK: uzp2 v0.16b, v0.16b, v2.16b
+; CHECK: ret
+  %insert = insertelement <16 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer
+  %1 = zext <16 x i8> %op1 to <16 x i16>
+  %2 = zext <16 x i8> %op2 to <16 x i16>
+  %mul = mul <16 x i16> %1, %2
+  %shr = lshr <16 x i16> %mul, %splat
+  %res = trunc <16 x i16> %shr to <16 x i8>
+  ret <16 x i8> %res
+}
+
+define void @umulh_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
+; CHECK-LABEL: umulh_v32i8:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,32)]]
+; VBITS_GE_256-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_256: umulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_256: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <32 x i8>, <32 x i8>* %a
+  %op2 = load <32 x i8>, <32 x i8>* %b
+  %insert = insertelement <32 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer
+  %1 = zext <32 x i8> %op1 to <32 x i16>
+  %2 = zext <32 x i8> %op2 to <32 x i16>
+  %mul = mul <32 x i16> %1, %2
+  %shr = lshr <32 x i16> %mul, %splat
+  %res = trunc <32 x i16> %shr to <32 x i8>
+  store <32 x i8> %res, <32 x i8>* %a
+  ret void
+}
+
+define void @umulh_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
+; CHECK-LABEL: umulh_v64i8:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,64)]]
+; VBITS_GE_512-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_512: umulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_512: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <64 x i8>, <64 x i8>* %a
+  %op2 = load <64 x i8>, <64 x i8>* %b
+  %insert = insertelement <64 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <64 x i16> %insert, <64 x i16> undef, <64 x i32> zeroinitializer
+  %1 = zext <64 x i8> %op1 to <64 x i16>
+  %2 = zext <64 x i8> %op2 to <64 x i16>
+  %mul = mul <64 x i16> %1, %2
+  %shr = lshr <64 x i16> %mul, %splat
+  %res = trunc <64 x i16> %shr to <64 x i8>
+  store <64 x i8> %res, <64 x i8>* %a
+  ret void
+}
+
+define void @umulh_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
+; CHECK-LABEL: umulh_v128i8:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,128)]]
+; VBITS_GE_1024-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_1024: umulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_1024: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <128 x i8>, <128 x i8>* %a
+  %op2 = load <128 x i8>, <128 x i8>* %b
+  %insert = insertelement <128 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <128 x i16> %insert, <128 x i16> undef, <128 x i32> zeroinitializer
+  %1 = zext <128 x i8> %op1 to <128 x i16>
+  %2 = zext <128 x i8> %op2 to <128 x i16>
+  %mul = mul <128 x i16> %1, %2
+  %shr = lshr <128 x i16> %mul, %splat
+  %res = trunc <128 x i16> %shr to <128 x i8>
+  store <128 x i8> %res, <128 x i8>* %a
+  ret void
+}
+
+define void @umulh_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
+; CHECK-LABEL: umulh_v256i8:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
+; VBITS_GE_2048-DAG: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_2048: umulh [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_2048: st1b { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <256 x i8>, <256 x i8>* %a
+  %op2 = load <256 x i8>, <256 x i8>* %b
+  %insert = insertelement <256 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <256 x i16> %insert, <256 x i16> undef, <256 x i32> zeroinitializer
+  %1 = zext <256 x i8> %op1 to <256 x i16>
+  %2 = zext <256 x i8> %op2 to <256 x i16>
+  %mul = mul <256 x i16> %1, %2
+  %shr = lshr <256 x i16> %mul, %splat
+  %res = trunc <256 x i16> %shr to <256 x i8>
+  store <256 x i8> %res, <256 x i8>* %a
+  ret void
+}
+
+; Don't use SVE for 64-bit vectors.
+define <4 x i16> @umulh_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
+; CHECK-LABEL: umulh_v4i16:
+; CHECK: umull v0.4s, v0.4h, v1.4h
+; CHECK: ushr v0.4s, v0.4s, #16
+; CHECK: mov w8, v0.s[1]
+; CHECK: mov w9, v0.s[2]
+; CHECK: mov w10, v0.s[3]
+; CHECK: mov v0.h[1], w8
+; CHECK: mov v0.h[2], w9
+; CHECK: mov v0.h[3], w10
+; CHECK: ret
+  %insert = insertelement <4 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
+  %1 = zext <4 x i16> %op1 to <4 x i32>
+  %2 = zext <4 x i16> %op2 to <4 x i32>
+  %mul = mul <4 x i32> %1, %2
+  %shr = lshr <4 x i32> %mul, %splat
+  %res = trunc <4 x i32> %shr to <4 x i16>
+  ret <4 x i16> %res
+}
+
+; Don't use SVE for 128-bit vectors.
+define <8 x i16> @umulh_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
+; CHECK-LABEL: umulh_v8i16:
+; CHECK: umull2 v2.4s, v0.8h, v1.8h
+; CHECK: umull v0.4s, v0.4h, v1.4h
+; CHECK: uzp2 v0.8h, v0.8h, v2.8h
+; CHECK: ret
+  %insert = insertelement <8 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer
+  %1 = zext <8 x i16> %op1 to <8 x i32>
+  %2 = zext <8 x i16> %op2 to <8 x i32>
+  %mul = mul <8 x i32> %1, %2
+  %shr = lshr <8 x i32> %mul, %splat
+  %res = trunc <8 x i32> %shr to <8 x i16>
+  ret <8 x i16> %res
+}
+
+define void @umulh_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
+; CHECK-LABEL: umulh_v16i16:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,16)]]
+; VBITS_GE_256-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_256: umulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_256: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <16 x i16>, <16 x i16>* %a
+  %op2 = load <16 x i16>, <16 x i16>* %b
+  %insert = insertelement <16 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer
+  %1 = zext <16 x i16> %op1 to <16 x i32>
+  %2 = zext <16 x i16> %op2 to <16 x i32>
+  %mul = mul <16 x i32> %1, %2
+  %shr = lshr <16 x i32> %mul, %splat
+  %res = trunc <16 x i32> %shr to <16 x i16>
+  store <16 x i16> %res, <16 x i16>* %a
+  ret void
+}
+
+define void @umulh_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
+; CHECK-LABEL: umulh_v32i16:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,32)]]
+; VBITS_GE_512-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_512: umulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_512: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <32 x i16>, <32 x i16>* %a
+  %op2 = load <32 x i16>, <32 x i16>* %b
+  %insert = insertelement <32 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <32 x i32> %insert, <32 x i32> undef, <32 x i32> zeroinitializer
+  %1 = zext <32 x i16> %op1 to <32 x i32>
+  %2 = zext <32 x i16> %op2 to <32 x i32>
+  %mul = mul <32 x i32> %1, %2
+  %shr = lshr <32 x i32> %mul, %splat
+  %res = trunc <32 x i32> %shr to <32 x i16>
+  store <32 x i16> %res, <32 x i16>* %a
+  ret void
+}
+
+define void @umulh_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
+; CHECK-LABEL: umulh_v64i16:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,64)]]
+; VBITS_GE_1024-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_1024: umulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_1024: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <64 x i16>, <64 x i16>* %a
+  %op2 = load <64 x i16>, <64 x i16>* %b
+  %insert = insertelement <64 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <64 x i32> %insert, <64 x i32> undef, <64 x i32> zeroinitializer
+  %1 = zext <64 x i16> %op1 to <64 x i32>
+  %2 = zext <64 x i16> %op2 to <64 x i32>
+  %mul = mul <64 x i32> %1, %2
+  %shr = lshr <64 x i32> %mul, %splat
+  %res = trunc <64 x i32> %shr to <64 x i16>
+  store <64 x i16> %res, <64 x i16>* %a
+  ret void
+}
+
+define void @umulh_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
+; CHECK-LABEL: umulh_v128i16:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(VBYTES,128)]]
+; VBITS_GE_2048-DAG: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_2048: umulh [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_2048: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <128 x i16>, <128 x i16>* %a
+  %op2 = load <128 x i16>, <128 x i16>* %b
+  %insert = insertelement <128 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <128 x i32> %insert, <128 x i32> undef, <128 x i32> zeroinitializer
+  %1 = zext <128 x i16> %op1 to <128 x i32>
+  %2 = zext <128 x i16> %op2 to <128 x i32>
+  %mul = mul <128 x i32> %1, %2
+  %shr = lshr <128 x i32> %mul, %splat
+  %res = trunc <128 x i32> %shr to <128 x i16>
+  store <128 x i16> %res, <128 x i16>* %a
+  ret void
+}
+
+; Vector i64 multiplications are not legal for NEON so use SVE when available.
+define <2 x i32> @umulh_v2i32(<2 x i32> %op1, <2 x i32> %op2) #0 {
+; CHECK-LABEL: umulh_v2i32:
+; CHECK: ushll v0.2d, v0.2s, #0
+; CHECK: ushll v1.2d, v1.2s, #0
+; CHECK: ptrue p0.d, vl2
+; CHECK: mul z0.d, p0/m, z0.d, z1.d
+; CHECK: shrn v0.2s, v0.2d, #32
+; CHECK: ret
+  %insert = insertelement <2 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer
+  %1 = zext <2 x i32> %op1 to <2 x i64>
+  %2 = zext <2 x i32> %op2 to <2 x i64>
+  %mul = mul <2 x i64> %1, %2
+  %shr = lshr <2 x i64> %mul, %splat
+  %res = trunc <2 x i64> %shr to <2 x i32>
+  ret <2 x i32> %res
+}
+
+; Don't use SVE for 128-bit vectors.
+define <4 x i32> @umulh_v4i32(<4 x i32> %op1, <4 x i32> %op2) #0 {
+; CHECK-LABEL: umulh_v4i32:
+; CHECK: umull2 v2.2d, v0.4s, v1.4s
+; CHECK: umull v0.2d, v0.2s, v1.2s
+; CHECK: uzp2 v0.4s, v0.4s, v2.4s
+; CHECK: ret
+  %insert = insertelement <4 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer
+  %1 = zext <4 x i32> %op1 to <4 x i64>
+  %2 = zext <4 x i32> %op2 to <4 x i64>
+  %mul = mul <4 x i64> %1, %2
+  %shr = lshr <4 x i64> %mul, %splat
+  %res = trunc <4 x i64> %shr to <4 x i32>
+  ret <4 x i32> %res
+}
+
+define void @umulh_v8i32(<8 x i32>* %a, <8 x i32>* %b) #0 {
+; CHECK-LABEL: umulh_v8i32:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,8)]]
+; VBITS_GE_256-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_256: umulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_256: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <8 x i32>, <8 x i32>* %a
+  %op2 = load <8 x i32>, <8 x i32>* %b
+  %insert = insertelement <8 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer
+  %1 = zext <8 x i32> %op1 to <8 x i64>
+  %2 = zext <8 x i32> %op2 to <8 x i64>
+  %mul = mul <8 x i64> %1, %2
+  %shr = lshr <8 x i64> %mul, %splat
+  %res = trunc <8 x i64> %shr to <8 x i32>
+  store <8 x i32> %res, <8 x i32>* %a
+  ret void
+}
+
+define void @umulh_v16i32(<16 x i32>* %a, <16 x i32>* %b) #0 {
+; CHECK-LABEL: umulh_v16i32:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,16)]]
+; VBITS_GE_512-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_512: umulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_512: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <16 x i32>, <16 x i32>* %a
+  %op2 = load <16 x i32>, <16 x i32>* %b
+  %insert = insertelement <16 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <16 x i64> %insert, <16 x i64> undef, <16 x i32> zeroinitializer
+  %1 = zext <16 x i32> %op1 to <16 x i64>
+  %2 = zext <16 x i32> %op2 to <16 x i64>
+  %mul = mul <16 x i64> %1, %2
+  %shr = lshr <16 x i64> %mul, %splat
+  %res = trunc <16 x i64> %shr to <16 x i32>
+  store <16 x i32> %res, <16 x i32>* %a
+  ret void
+}
+
+define void @umulh_v32i32(<32 x i32>* %a, <32 x i32>* %b) #0 {
+; CHECK-LABEL: umulh_v32i32:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,32)]]
+; VBITS_GE_1024-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_1024: umulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_1024: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <32 x i32>, <32 x i32>* %a
+  %op2 = load <32 x i32>, <32 x i32>* %b
+  %insert = insertelement <32 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <32 x i64> %insert, <32 x i64> undef, <32 x i32> zeroinitializer
+  %1 = zext <32 x i32> %op1 to <32 x i64>
+  %2 = zext <32 x i32> %op2 to <32 x i64>
+  %mul = mul <32 x i64> %1, %2
+  %shr = lshr <32 x i64> %mul, %splat
+  %res = trunc <32 x i64> %shr to <32 x i32>
+  store <32 x i32> %res, <32 x i32>* %a
+  ret void
+}
+
+define void @umulh_v64i32(<64 x i32>* %a, <64 x i32>* %b) #0 {
+; CHECK-LABEL: umulh_v64i32:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
+; VBITS_GE_2048-DAG: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_2048: umulh [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_2048: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <64 x i32>, <64 x i32>* %a
+  %op2 = load <64 x i32>, <64 x i32>* %b
+  %insert = insertelement <64 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <64 x i64> %insert, <64 x i64> undef, <64 x i32> zeroinitializer
+  %1 = zext <64 x i32> %op1 to <64 x i64>
+  %2 = zext <64 x i32> %op2 to <64 x i64>
+  %mul = mul <64 x i64> %1, %2
+  %shr = lshr <64 x i64> %mul, %splat
+  %res = trunc <64 x i64> %shr to <64 x i32>
+  store <64 x i32> %res, <64 x i32>* %a
+  ret void
+}
+
+; Vector i64 multiplications are not legal for NEON so use SVE when available.
+define <1 x i64> @umulh_v1i64(<1 x i64> %op1, <1 x i64> %op2) #0 {
+; CHECK-LABEL: umulh_v1i64:
+; CHECK: ptrue p0.d, vl1
+; CHECK: umulh z0.d, p0/m, z0.d, z1.d
+; CHECK: ret
+  %insert = insertelement <1 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <1 x i128> %insert, <1 x i128> undef, <1 x i32> zeroinitializer
+  %1 = zext <1 x i64> %op1 to <1 x i128>
+  %2 = zext <1 x i64> %op2 to <1 x i128>
+  %mul = mul <1 x i128> %1, %2
+  %shr = lshr <1 x i128> %mul, %splat
+  %res = trunc <1 x i128> %shr to <1 x i64>
+  ret <1 x i64> %res
+}
+
+; Vector i64 multiplications are not legal for NEON so use SVE when available.
+define <2 x i64> @umulh_v2i64(<2 x i64> %op1, <2 x i64> %op2) #0 {
+; CHECK-LABEL: umulh_v2i64:
+; CHECK: ptrue p0.d, vl2
+; CHECK: umulh z0.d, p0/m, z0.d, z1.d
+; CHECK: ret
+  %insert = insertelement <2 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <2 x i128> %insert, <2 x i128> undef, <2 x i32> zeroinitializer
+  %1 = zext <2 x i64> %op1 to <2 x i128>
+  %2 = zext <2 x i64> %op2 to <2 x i128>
+  %mul = mul <2 x i128> %1, %2
+  %shr = lshr <2 x i128> %mul, %splat
+  %res = trunc <2 x i128> %shr to <2 x i64>
+  ret <2 x i64> %res
+}
+
+define void @umulh_v4i64(<4 x i64>* %a, <4 x i64>* %b) #0 {
+; CHECK-LABEL: umulh_v4i64:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,4)]]
+; VBITS_GE_256-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_256-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_256: umulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_256: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_256: ret
+  %op1 = load <4 x i64>, <4 x i64>* %a
+  %op2 = load <4 x i64>, <4 x i64>* %b
+  %insert = insertelement <4 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <4 x i128> %insert, <4 x i128> undef, <4 x i32> zeroinitializer
+  %1 = zext <4 x i64> %op1 to <4 x i128>
+  %2 = zext <4 x i64> %op2 to <4 x i128>
+  %mul = mul <4 x i128> %1, %2
+  %shr = lshr <4 x i128> %mul, %splat
+  %res = trunc <4 x i128> %shr to <4 x i64>
+  store <4 x i64> %res, <4 x i64>* %a
+  ret void
+}
+
+define void @umulh_v8i64(<8 x i64>* %a, <8 x i64>* %b) #0 {
+; CHECK-LABEL: umulh_v8i64:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,8)]]
+; VBITS_GE_512-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_512-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_512: umulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_512: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_512: ret
+  %op1 = load <8 x i64>, <8 x i64>* %a
+  %op2 = load <8 x i64>, <8 x i64>* %b
+  %insert = insertelement <8 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <8 x i128> %insert, <8 x i128> undef, <8 x i32> zeroinitializer
+  %1 = zext <8 x i64> %op1 to <8 x i128>
+  %2 = zext <8 x i64> %op2 to <8 x i128>
+  %mul = mul <8 x i128> %1, %2
+  %shr = lshr <8 x i128> %mul, %splat
+  %res = trunc <8 x i128> %shr to <8 x i64>
+  store <8 x i64> %res, <8 x i64>* %a
+  ret void
+}
+
+define void @umulh_v16i64(<16 x i64>* %a, <16 x i64>* %b) #0 {
+; CHECK-LABEL: umulh_v16i64:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,16)]]
+; VBITS_GE_1024-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_1024-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_1024: umulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_1024: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_1024: ret
+  %op1 = load <16 x i64>, <16 x i64>* %a
+  %op2 = load <16 x i64>, <16 x i64>* %b
+  %insert = insertelement <16 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <16 x i128> %insert, <16 x i128> undef, <16 x i32> zeroinitializer
+  %1 = zext <16 x i64> %op1 to <16 x i128>
+  %2 = zext <16 x i64> %op2 to <16 x i128>
+  %mul = mul <16 x i128> %1, %2
+  %shr = lshr <16 x i128> %mul, %splat
+  %res = trunc <16 x i128> %shr to <16 x i64>
+  store <16 x i64> %res, <16 x i64>* %a
+  ret void
+}
+
+define void @umulh_v32i64(<32 x i64>* %a, <32 x i64>* %b) #0 {
+; CHECK-LABEL: umulh_v32i64:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl[[#min(VBYTES,32)]]
+; VBITS_GE_2048-DAG: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_2048-DAG: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_2048: umulh [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_2048: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_2048: ret
+  %op1 = load <32 x i64>, <32 x i64>* %a
+  %op2 = load <32 x i64>, <32 x i64>* %b
+  %insert = insertelement <32 x i128> undef, i128 64, i128 0
+  %splat = shufflevector <32 x i128> %insert, <32 x i128> undef, <32 x i32> zeroinitializer
+  %1 = zext <32 x i64> %op1 to <32 x i128>
+  %2 = zext <32 x i64> %op2 to <32 x i128>
+  %mul = mul <32 x i128> %1, %2
+  %shr = lshr <32 x i128> %mul, %splat
+  %res = trunc <32 x i128> %shr to <32 x i64>
+  store <32 x i64> %res, <32 x i64>* %a
+  ret void
+}
+attributes #0 = { "target-features"="+sve" }
diff --git a/llvm/test/CodeGen/AArch64/sve-int-arith-imm.ll b/llvm/test/CodeGen/AArch64/sve-int-arith-imm.ll
index fba722e495a7..3b1c332bdd80 100644
--- a/llvm/test/CodeGen/AArch64/sve-int-arith-imm.ll
+++ b/llvm/test/CodeGen/AArch64/sve-int-arith-imm.ll
@@ -776,7 +776,7 @@ define <vscale x 2 x i64> @lsr_i64(<vscale x 2 x i64> %a){
   ret <vscale x 2 x i64> %lshr
 }
 
-define <vscale x 4 x i32> @sdiv_const(<vscale x 4 x i32> %a) {
+define <vscale x 4 x i32> @sdiv_const(<vscale x 4 x i32> %a) #0 {
 ; CHECK-LABEL: sdiv_const:
 ; CHECK:       // %bb.0: // %entry
 ; CHECK-NEXT:    mov z1.s, #3 // =0x3
@@ -788,7 +788,7 @@ entry:
   ret <vscale x 4 x i32> %div
 }
 
-define <vscale x 4 x i32> @udiv_const(<vscale x 4 x i32> %a) {
+define <vscale x 4 x i32> @udiv_const(<vscale x 4 x i32> %a) #0 {
 ; CHECK-LABEL: udiv_const:
 ; CHECK:       // %bb.0: // %entry
 ; CHECK-NEXT:    mov z1.s, #3 // =0x3
@@ -799,3 +799,5 @@ entry:
   %div = udiv <vscale x 4 x i32> %a, shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> undef, i32 3, i32 0), <vscale x 4 x i32> undef, <vscale x 4 x i32> zeroinitializer)
   ret <vscale x 4 x i32> %div
 }
+
+attributes #0 = { minsize }
diff --git a/llvm/test/CodeGen/AArch64/sve-int-mulh-pred.ll b/llvm/test/CodeGen/AArch64/sve-int-mulh-pred.ll
new file mode 100644
index 000000000000..c636f1172711
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve-int-mulh-pred.ll
@@ -0,0 +1,140 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-linux-gnu < %s | FileCheck %s
+
+;
+; SMULH
+;
+
+define <vscale x 16 x i8> @smulh_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) #0 {
+; CHECK-LABEL: smulh_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.b
+; CHECK-NEXT:    smulh z0.b, p0/m, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 16 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <vscale x 16 x i16> %insert, <vscale x 16 x i16> undef, <vscale x 16 x i32> zeroinitializer
+  %1 = sext <vscale x 16 x i8> %a to <vscale x 16 x i16>
+  %2 = sext <vscale x 16 x i8> %b to <vscale x 16 x i16>
+  %mul = mul <vscale x 16 x i16> %1, %2
+  %shr = lshr <vscale x 16 x i16> %mul, %splat
+  %tr = trunc <vscale x 16 x i16> %shr to <vscale x 16 x i8>
+  ret <vscale x 16 x i8> %tr
+}
+
+define <vscale x 8 x i16> @smulh_i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) #0 {
+; CHECK-LABEL: smulh_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    smulh z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 8 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <vscale x 8 x i32> %insert, <vscale x 8 x i32> undef, <vscale x 8 x i32> zeroinitializer
+  %1 = sext <vscale x 8 x i16> %a to <vscale x 8 x i32>
+  %2 = sext <vscale x 8 x i16> %b to <vscale x 8 x i32>
+  %mul = mul <vscale x 8 x i32> %1, %2
+  %shr = lshr <vscale x 8 x i32> %mul, %splat
+  %tr = trunc <vscale x 8 x i32> %shr to <vscale x 8 x i16>
+  ret <vscale x 8 x i16> %tr
+}
+
+define <vscale x 4 x i32> @smulh_i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) #0 {
+; CHECK-LABEL: smulh_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    smulh z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 4 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <vscale x 4 x i64> %insert, <vscale x 4 x i64> undef, <vscale x 4 x i32> zeroinitializer
+  %1 = sext <vscale x 4 x i32> %a to <vscale x 4 x i64>
+  %2 = sext <vscale x 4 x i32> %b to <vscale x 4 x i64>
+  %mul = mul <vscale x 4 x i64> %1, %2
+  %shr = lshr <vscale x 4 x i64> %mul, %splat
+  %tr = trunc <vscale x 4 x i64> %shr to <vscale x 4 x i32>
+  ret <vscale x 4 x i32> %tr
+}
+
+define <vscale x 2 x i64> @smulh_i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) #0 {
+; CHECK-LABEL: smulh_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    smulh z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 2 x i128> undef, i128 64, i64 0
+  %splat = shufflevector <vscale x 2 x i128> %insert, <vscale x 2 x i128> undef, <vscale x 2 x i32> zeroinitializer
+  %1 = sext <vscale x 2 x i64> %a to <vscale x 2 x i128>
+  %2 = sext <vscale x 2 x i64> %b to <vscale x 2 x i128>
+  %mul = mul <vscale x 2 x i128> %1, %2
+  %shr = lshr <vscale x 2 x i128> %mul, %splat
+  %tr = trunc <vscale x 2 x i128> %shr to <vscale x 2 x i64>
+  ret <vscale x 2 x i64> %tr
+}
+
+;
+; UMULH
+;
+
+define <vscale x 16 x i8> @umulh_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) #0 {
+; CHECK-LABEL: umulh_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.b
+; CHECK-NEXT:    umulh z0.b, p0/m, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 16 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <vscale x 16 x i16> %insert, <vscale x 16 x i16> undef, <vscale x 16 x i32> zeroinitializer
+  %1 = zext <vscale x 16 x i8> %a to <vscale x 16 x i16>
+  %2 = zext <vscale x 16 x i8> %b to <vscale x 16 x i16>
+  %mul = mul <vscale x 16 x i16> %1, %2
+  %shr = lshr <vscale x 16 x i16> %mul, %splat
+  %tr = trunc <vscale x 16 x i16> %shr to <vscale x 16 x i8>
+  ret <vscale x 16 x i8> %tr
+}
+
+define <vscale x 8 x i16> @umulh_i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) #0 {
+; CHECK-LABEL: umulh_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.h
+; CHECK-NEXT:    umulh z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 8 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <vscale x 8 x i32> %insert, <vscale x 8 x i32> undef, <vscale x 8 x i32> zeroinitializer
+  %1 = zext <vscale x 8 x i16> %a to <vscale x 8 x i32>
+  %2 = zext <vscale x 8 x i16> %b to <vscale x 8 x i32>
+  %mul = mul <vscale x 8 x i32> %1, %2
+  %shr = lshr <vscale x 8 x i32> %mul, %splat
+  %tr = trunc <vscale x 8 x i32> %shr to <vscale x 8 x i16>
+  ret <vscale x 8 x i16> %tr
+}
+
+define <vscale x 4 x i32> @umulh_i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) #0 {
+; CHECK-LABEL: umulh_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    umulh z0.s, p0/m, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 4 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <vscale x 4 x i64> %insert, <vscale x 4 x i64> undef, <vscale x 4 x i32> zeroinitializer
+  %1 = zext <vscale x 4 x i32> %a to <vscale x 4 x i64>
+  %2 = zext <vscale x 4 x i32> %b to <vscale x 4 x i64>
+  %mul = mul <vscale x 4 x i64> %1, %2
+  %shr = lshr <vscale x 4 x i64> %mul, %splat
+  %tr = trunc <vscale x 4 x i64> %shr to <vscale x 4 x i32>
+  ret <vscale x 4 x i32> %tr
+}
+
+define <vscale x 2 x i64> @umulh_i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) #0 {
+; CHECK-LABEL: umulh_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    umulh z0.d, p0/m, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 2 x i128> undef, i128 64, i64 0
+  %splat = shufflevector <vscale x 2 x i128> %insert, <vscale x 2 x i128> undef, <vscale x 2 x i32> zeroinitializer
+  %1 = zext <vscale x 2 x i64> %a to <vscale x 2 x i128>
+  %2 = zext <vscale x 2 x i64> %b to <vscale x 2 x i128>
+  %mul = mul <vscale x 2 x i128> %1, %2
+  %shr = lshr <vscale x 2 x i128> %mul, %splat
+  %tr = trunc <vscale x 2 x i128> %shr to <vscale x 2 x i64>
+  ret <vscale x 2 x i64> %tr
+}
+
+attributes #0 = { "target-features"="+sve" }
diff --git a/llvm/test/CodeGen/AArch64/sve2-int-mulh.ll b/llvm/test/CodeGen/AArch64/sve2-int-mulh.ll
new file mode 100644
index 000000000000..c929a3faffab
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve2-int-mulh.ll
@@ -0,0 +1,132 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-linux-gnu < %s | FileCheck %s
+
+;
+; SMULH
+;
+
+define <vscale x 16 x i8> @smulh_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) #0 {
+; CHECK-LABEL: smulh_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    smulh z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 16 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <vscale x 16 x i16> %insert, <vscale x 16 x i16> undef, <vscale x 16 x i32> zeroinitializer
+  %1 = sext <vscale x 16 x i8> %a to <vscale x 16 x i16>
+  %2 = sext <vscale x 16 x i8> %b to <vscale x 16 x i16>
+  %mul = mul <vscale x 16 x i16> %1, %2
+  %shr = lshr <vscale x 16 x i16> %mul, %splat
+  %tr = trunc <vscale x 16 x i16> %shr to <vscale x 16 x i8>
+  ret <vscale x 16 x i8> %tr
+}
+
+define <vscale x 8 x i16> @smulh_i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) #0 {
+; CHECK-LABEL: smulh_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    smulh z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 8 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <vscale x 8 x i32> %insert, <vscale x 8 x i32> undef, <vscale x 8 x i32> zeroinitializer
+  %1 = sext <vscale x 8 x i16> %a to <vscale x 8 x i32>
+  %2 = sext <vscale x 8 x i16> %b to <vscale x 8 x i32>
+  %mul = mul <vscale x 8 x i32> %1, %2
+  %shr = lshr <vscale x 8 x i32> %mul, %splat
+  %tr = trunc <vscale x 8 x i32> %shr to <vscale x 8 x i16>
+  ret <vscale x 8 x i16> %tr
+}
+
+define <vscale x 4 x i32> @smulh_i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) #0 {
+; CHECK-LABEL: smulh_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    smulh z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 4 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <vscale x 4 x i64> %insert, <vscale x 4 x i64> undef, <vscale x 4 x i32> zeroinitializer
+  %1 = sext <vscale x 4 x i32> %a to <vscale x 4 x i64>
+  %2 = sext <vscale x 4 x i32> %b to <vscale x 4 x i64>
+  %mul = mul <vscale x 4 x i64> %1, %2
+  %shr = lshr <vscale x 4 x i64> %mul, %splat
+  %tr = trunc <vscale x 4 x i64> %shr to <vscale x 4 x i32>
+  ret <vscale x 4 x i32> %tr
+}
+
+define <vscale x 2 x i64> @smulh_i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) #0 {
+; CHECK-LABEL: smulh_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    smulh z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 2 x i128> undef, i128 64, i64 0
+  %splat = shufflevector <vscale x 2 x i128> %insert, <vscale x 2 x i128> undef, <vscale x 2 x i32> zeroinitializer
+  %1 = sext <vscale x 2 x i64> %a to <vscale x 2 x i128>
+  %2 = sext <vscale x 2 x i64> %b to <vscale x 2 x i128>
+  %mul = mul <vscale x 2 x i128> %1, %2
+  %shr = lshr <vscale x 2 x i128> %mul, %splat
+  %tr = trunc <vscale x 2 x i128> %shr to <vscale x 2 x i64>
+  ret <vscale x 2 x i64> %tr
+}
+
+;
+; UMULH
+;
+
+define <vscale x 16 x i8> @umulh_i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) #0 {
+; CHECK-LABEL: umulh_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    umulh z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 16 x i16> undef, i16 8, i64 0
+  %splat = shufflevector <vscale x 16 x i16> %insert, <vscale x 16 x i16> undef, <vscale x 16 x i32> zeroinitializer
+  %1 = zext <vscale x 16 x i8> %a to <vscale x 16 x i16>
+  %2 = zext <vscale x 16 x i8> %b to <vscale x 16 x i16>
+  %mul = mul <vscale x 16 x i16> %1, %2
+  %shr = lshr <vscale x 16 x i16> %mul, %splat
+  %tr = trunc <vscale x 16 x i16> %shr to <vscale x 16 x i8>
+  ret <vscale x 16 x i8> %tr
+}
+
+define <vscale x 8 x i16> @umulh_i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) #0 {
+; CHECK-LABEL: umulh_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    umulh z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 8 x i32> undef, i32 16, i64 0
+  %splat = shufflevector <vscale x 8 x i32> %insert, <vscale x 8 x i32> undef, <vscale x 8 x i32> zeroinitializer
+  %1 = zext <vscale x 8 x i16> %a to <vscale x 8 x i32>
+  %2 = zext <vscale x 8 x i16> %b to <vscale x 8 x i32>
+  %mul = mul <vscale x 8 x i32> %1, %2
+  %shr = lshr <vscale x 8 x i32> %mul, %splat
+  %tr = trunc <vscale x 8 x i32> %shr to <vscale x 8 x i16>
+  ret <vscale x 8 x i16> %tr
+}
+
+define <vscale x 4 x i32> @umulh_i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) #0 {
+; CHECK-LABEL: umulh_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    umulh z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 4 x i64> undef, i64 32, i64 0
+  %splat = shufflevector <vscale x 4 x i64> %insert, <vscale x 4 x i64> undef, <vscale x 4 x i32> zeroinitializer
+  %1 = zext <vscale x 4 x i32> %a to <vscale x 4 x i64>
+  %2 = zext <vscale x 4 x i32> %b to <vscale x 4 x i64>
+  %mul = mul <vscale x 4 x i64> %1, %2
+  %shr = lshr <vscale x 4 x i64> %mul, %splat
+  %tr = trunc <vscale x 4 x i64> %shr to <vscale x 4 x i32>
+  ret <vscale x 4 x i32> %tr
+}
+
+define <vscale x 2 x i64> @umulh_i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) #0 {
+; CHECK-LABEL: umulh_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    umulh z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+  %insert = insertelement <vscale x 2 x i128> undef, i128 64, i64 0
+  %splat = shufflevector <vscale x 2 x i128> %insert, <vscale x 2 x i128> undef, <vscale x 2 x i32> zeroinitializer
+  %1 = zext <vscale x 2 x i64> %a to <vscale x 2 x i128>
+  %2 = zext <vscale x 2 x i64> %b to <vscale x 2 x i128>
+  %mul = mul <vscale x 2 x i128> %1, %2
+  %shr = lshr <vscale x 2 x i128> %mul, %splat
+  %tr = trunc <vscale x 2 x i128> %shr to <vscale x 2 x i64>
+  ret <vscale x 2 x i64> %tr
+}
+
+attributes #0 = { "target-features"="+sve2" }