[SVE] Lower fixed length UDIV to scalable

[cameron.mcinally]

Summary

Pretty much just a copy of the SDIV patches (D86114 and D85982) with string replacement.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

if (useSVEForFixedLengthVectors()) {
       setOperationAction(ISD::SDIV, MVT::v2i32, Custom);
       setOperationAction(ISD::SDIV, MVT::v4i32, Custom);
       setOperationAction(ISD::SDIV, MVT::v1i64, Custom);
       setOperationAction(ISD::SDIV, MVT::v2i64, Custom);
       setOperationAction(ISD::SMAX, MVT::v1i64, Custom);
       setOperationAction(ISD::SMAX, MVT::v2i64, Custom);
       setOperationAction(ISD::SMIN, MVT::v1i64, Custom);
       setOperationAction(ISD::SMIN, MVT::v2i64, Custom);
+      setOperationAction(ISD::UDIV, MVT::v8i8, Custom);
+      setOperationAction(ISD::UDIV, MVT::v16i8, Custom);
+      setOperationAction(ISD::UDIV, MVT::v4i16, Custom);
+      setOperationAction(ISD::UDIV, MVT::v8i16, Custom);
+      setOperationAction(ISD::UDIV, MVT::v2i32, Custom);
+      setOperationAction(ISD::UDIV, MVT::v4i32, Custom);
+      setOperationAction(ISD::UDIV, MVT::v1i64, Custom);
+      setOperationAction(ISD::UDIV, MVT::v2i64, Custom);
       setOperationAction(ISD::UMAX, MVT::v1i64, Custom);
       setOperationAction(ISD::UMAX, MVT::v2i64, Custom);
       setOperationAction(ISD::UMIN, MVT::v1i64, Custom);
       setOperationAction(ISD::UMIN, MVT::v2i64, Custom);
     }
   }
 
   PredictableSelectIsExpensive = Subtarget->predictableSelectIsExpensive();
void AArch64TargetLowering::addTypeForFixedLengthSVE(MVT VT) {
   setOperationAction(ISD::SMAX, VT, Custom);
   setOperationAction(ISD::SMIN, VT, Custom);
   setOperationAction(ISD::SPLAT_VECTOR, VT, Custom);
   setOperationAction(ISD::SRA, VT, Custom);
   setOperationAction(ISD::SRL, VT, Custom);
   setOperationAction(ISD::STORE, VT, Custom);
   setOperationAction(ISD::SUB, VT, Custom);
   setOperationAction(ISD::TRUNCATE, VT, Custom);
+  setOperationAction(ISD::UDIV, VT, Custom);
   setOperationAction(ISD::UMAX, VT, Custom);
   setOperationAction(ISD::UMIN, VT, Custom);
   setOperationAction(ISD::XOR, VT, Custom);
   setOperationAction(ISD::ZERO_EXTEND, VT, Custom);
 }
 
 void AArch64TargetLowering::addDRTypeForNEON(MVT VT) {
   addRegisterClass(VT, &AArch64::FPR64RegClass);

llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll

 ; VBITS_GE_2048-NEXT: ret
   %op1 = load <32 x i64>, <32 x i64>* %a
   %op2 = load <32 x i64>, <32 x i64>* %b
   %res = sdiv <32 x i64> %op1, %op2
   store <32 x i64> %res, <32 x i64>* %a
   ret void
 }
 
+;
+; UDIV
+;
+
+; Vector vXi8 udiv are not legal for NEON so use SVE when available.
+define <8 x i8> @udiv_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
+; CHECK-LABEL: udiv_v8i8:
+; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
+; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
+; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,2)]]
+; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
+; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
+; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; CHECK-NEXT: udivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; CHECK-NEXT: udivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, z0.h
+; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
+; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
+; CHECK-NEXT: udiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; CHECK-NEXT: udiv [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; CHECK-NEXT: uzp1 z0.b, [[RES_LO]].b, [[RES_HI]].b
+; CHECK:    ret
+  %res = udiv <8 x i8> %op1, %op2
+  ret <8 x i8> %res
+}
+
+define <16 x i8> @udiv_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
+; CHECK-LABEL: udiv_v16i8:
+; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
+; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
+; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,4)]]
+; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
+; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
+; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; CHECK-NEXT: udivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; CHECK-NEXT: udivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, z0.h
+; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
+; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
+; CHECK-NEXT: udiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; CHECK-NEXT: udiv [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; CHECK-NEXT: uzp1 z0.b, [[RES_LO]].b, [[RES_HI]].b
+; CHECK: ret
+  %res = udiv <16 x i8> %op1, %op2
+  ret <16 x i8> %res
+}
+
+define void @udiv_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
+; CHECK-LABEL: udiv_v32i8:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,32)]]
+; VBITS_GE_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_256: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,8)]]
+; VBITS_GE_256-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_256-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_256-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_256-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
+; VBITS_GE_256-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
+; VBITS_GE_256-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_GE_256-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_GE_256-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_GE_256-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_256-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_GE_256-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_256-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_256-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_256-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_GE_256-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_256-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; VBITS_GE_256-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; VBITS_GE_256-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
+; VBITS_GE_256-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_256-NEXT: ret
+  %op1 = load <32 x i8>, <32 x i8>* %a
+  %op2 = load <32 x i8>, <32 x i8>* %b
+  %res = udiv <32 x i8> %op1, %op2
+  store <32 x i8> %res, <32 x i8>* %a
+  ret void
+}
+
+define void @udiv_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
+; CHECK-LABEL: udiv_v64i8:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,64)]]
+; VBITS_GE_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,16)]]
+; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
+; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_GE_512-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_GE_512-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_512-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_GE_512-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_512-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_GE_512-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_512-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; VBITS_GE_512-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; VBITS_GE_512-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
+; VBITS_GE_512-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %op1 = load <64 x i8>, <64 x i8>* %a
+  %op2 = load <64 x i8>, <64 x i8>* %b
+  %res = udiv <64 x i8> %op1, %op2
+  store <64 x i8> %res, <64 x i8>* %a
+  ret void
+}
+
+define void @udiv_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
+; CHECK-LABEL: udiv_v128i8:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,128)]]
+; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,32)]]
+; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
+; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_GE_1024-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_GE_1024-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_1024-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_GE_1024-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_1024-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_GE_1024-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_1024-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; VBITS_GE_1024-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; VBITS_GE_1024-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
+; VBITS_GE_1024-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <128 x i8>, <128 x i8>* %a
+  %op2 = load <128 x i8>, <128 x i8>* %b
+  %res = udiv <128 x i8> %op1, %op2
+  store <128 x i8> %res, <128 x i8>* %a
+  ret void
+}
+
+define void @udiv_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
+; CHECK-LABEL: udiv_v256i8:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
+; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
+; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
+; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_GE_2048-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_GE_2048-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_2048-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_GE_2048-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_2048-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_GE_2048-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_2048-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; VBITS_GE_2048-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; VBITS_GE_2048-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
+; VBITS_GE_2048-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <256 x i8>, <256 x i8>* %a
+  %op2 = load <256 x i8>, <256 x i8>* %b
+  %res = udiv <256 x i8> %op1, %op2
+  store <256 x i8> %res, <256 x i8>* %a
+  ret void
+}
+
+; Vector vXi16 udiv are not legal for NEON so use SVE when available.
+define <4 x i16> @udiv_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
+; CHECK-LABEL: udiv_v4i16:
+; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
+; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
+; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),2)]]
+; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, z1.h
+; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, z0.h
+; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; CHECK-NEXT: ret
+  %res = udiv <4 x i16> %op1, %op2
+  ret <4 x i16> %res
+}
+
+define <8 x i16> @udiv_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
+; CHECK-LABEL: udiv_v8i16:
+; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
+; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
+; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),4)]]
+; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, z1.h
+; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, z0.h
+; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; CHECK-NEXT: ret
+  %res = udiv <8 x i16> %op1, %op2
+  ret <8 x i16> %res
+}
+
+define void @udiv_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
+; CHECK-LABEL: udiv_v16i16:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),16)]]
+; VBITS_GE_256-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_256-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_256-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),8)]]
+; VBITS_GE_256-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_256-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_256-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_256-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_GE_256-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_256-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; VBITS_GE_256-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_256-NEXT: ret
+  %op1 = load <16 x i16>, <16 x i16>* %a
+  %op2 = load <16 x i16>, <16 x i16>* %b
+  %res = udiv <16 x i16> %op1, %op2
+  store <16 x i16> %res, <16 x i16>* %a
+  ret void
+}
+
+define void @udiv_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
+; CHECK-LABEL: udiv_v32i16:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),32)]]
+; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),16)]]
+; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_512-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_GE_512-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; VBITS_GE_512-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %op1 = load <32 x i16>, <32 x i16>* %a
+  %op2 = load <32 x i16>, <32 x i16>* %b
+  %res = udiv <32 x i16> %op1, %op2
+  store <32 x i16> %res, <32 x i16>* %a
+  ret void
+}
+
+define void @udiv_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
+; CHECK-LABEL: udiv_v64i16:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),64)]]
+; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),32)]]
+; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_1024-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_GE_1024-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; VBITS_GE_1024-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <64 x i16>, <64 x i16>* %a
+  %op2 = load <64 x i16>, <64 x i16>* %b
+  %res = udiv <64 x i16> %op1, %op2
+  store <64 x i16> %res, <64 x i16>* %a
+  ret void
+}
+
+define void @udiv_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
+; CHECK-LABEL: udiv_v128i16:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
+; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
+; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_2048-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_GE_2048-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; VBITS_GE_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <128 x i16>, <128 x i16>* %a
+  %op2 = load <128 x i16>, <128 x i16>* %b
+  %res = udiv <128 x i16> %op1, %op2
+  store <128 x i16> %res, <128 x i16>* %a
+  ret void
+}
+
+; Vector v2i32 udiv are not legal for NEON so use SVE when available.
+define <2 x i32> @udiv_v2i32(<2 x i32> %op1, <2 x i32> %op2) #0 {
+; CHECK-LABEL: udiv_v2i32:
+; CHECK: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,4),2)]]
+; CHECK: udiv z0.s, [[PG]]/m, z0.s, z1.s
+; CHECK: ret
+  %res = udiv <2 x i32> %op1, %op2
+  ret <2 x i32> %res
+}
+
+; Vector v4i32 udiv are not legal for NEON so use SVE when available.
+define <4 x i32> @udiv_v4i32(<4 x i32> %op1, <4 x i32> %op2) #0 {
+; CHECK-LABEL: udiv_v4i32:
+; CHECK: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,4),4)]]
+; CHECK: udiv z0.s, [[PG]]/m, z0.s, z1.s
+; CHECK: ret
+  %res = udiv <4 x i32> %op1, %op2
+  ret <4 x i32> %res
+}
+
+define void @udiv_v8i32(<8 x i32>* %a, <8 x i32>* %b) #0 {
+; CHECK-LABEL: udiv_v8i32:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,4),8)]]
+; VBITS_GE_256-NEXT: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_256-NEXT: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_256-NEXT: udiv [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_256-NEXT: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_256-NEXT: ret
+  %op1 = load <8 x i32>, <8 x i32>* %a
+  %op2 = load <8 x i32>, <8 x i32>* %b
+  %res = udiv <8 x i32> %op1, %op2
+  store <8 x i32> %res, <8 x i32>* %a
+  ret void
+}
+
+define void @udiv_v16i32(<16 x i32>* %a, <16 x i32>* %b) #0 {
+; CHECK-LABEL: udiv_v16i32:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,4),16)]]
+; VBITS_GE_512-NEXT: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: udiv [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_512-NEXT: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %op1 = load <16 x i32>, <16 x i32>* %a
+  %op2 = load <16 x i32>, <16 x i32>* %b
+  %res = udiv <16 x i32> %op1, %op2
+  store <16 x i32> %res, <16 x i32>* %a
+  ret void
+}
+
+define void @udiv_v32i32(<32 x i32>* %a, <32 x i32>* %b) #0 {
+; CHECK-LABEL: udiv_v32i32:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,4),32)]]
+; VBITS_GE_1024-NEXT: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: udiv [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_1024-NEXT: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <32 x i32>, <32 x i32>* %a
+  %op2 = load <32 x i32>, <32 x i32>* %b
+  %res = udiv <32 x i32> %op1, %op2
+  store <32 x i32> %res, <32 x i32>* %a
+  ret void
+}
+
+define void @udiv_v64i32(<64 x i32>* %a, <64 x i32>* %b) #0 {
+; CHECK-LABEL: udiv_v64i32:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,4),64)]]
+; VBITS_GE_2048-NEXT: ld1w { [[OP1:z[0-9]+]].s }, [[PG]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1w { [[OP2:z[0-9]+]].s }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: udiv [[RES:z[0-9]+]].s, [[PG]]/m, [[OP1]].s, [[OP2]].s
+; VBITS_GE_2048-NEXT: st1w { [[RES]].s }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <64 x i32>, <64 x i32>* %a
+  %op2 = load <64 x i32>, <64 x i32>* %b
+  %res = udiv <64 x i32> %op1, %op2
+  store <64 x i32> %res, <64 x i32>* %a
+  ret void
+}
+
+; Vector i64 udiv are not legal for NEON so use SVE when available.
+define <1 x i64> @udiv_v1i64(<1 x i64> %op1, <1 x i64> %op2) #0 {
+; CHECK-LABEL: udiv_v1i64:
+; CHECK: ptrue [[PG:p[0-9]+]].d, vl1
+; CHECK: udiv z0.d, [[PG]]/m, z0.d, z1.d
+; CHECK: ret
+  %res = udiv <1 x i64> %op1, %op2
+  ret <1 x i64> %res
+}
+
+; Vector i64 udiv are not legal for NEON so use SVE when available.
+define <2 x i64> @udiv_v2i64(<2 x i64> %op1, <2 x i64> %op2) #0 {
+; CHECK-LABEL: udiv_v2i64:
+; CHECK: ptrue [[PG:p[0-9]+]].d, vl2
+; CHECK: udiv z0.d, [[PG]]/m, z0.d, z1.d
+; CHECK: ret
+  %res = udiv <2 x i64> %op1, %op2
+  ret <2 x i64> %res
+}
+
+define void @udiv_v4i64(<4 x i64>* %a, <4 x i64>* %b) #0 {
+; CHECK-LABEL: udiv_v4i64:
+; VBITS_GE_256: ptrue [[PG:p[0-9]+]].d, vl[[#min(div(VBYTES,8),4)]]
+; VBITS_GE_256-NEXT: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_256-NEXT: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_256-NEXT: udiv [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_256-NEXT: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_256-NEXT: ret
+  %op1 = load <4 x i64>, <4 x i64>* %a
+  %op2 = load <4 x i64>, <4 x i64>* %b
+  %res = udiv <4 x i64> %op1, %op2
+  store <4 x i64> %res, <4 x i64>* %a
+  ret void
+}
+
+define void @udiv_v8i64(<8 x i64>* %a, <8 x i64>* %b) #0 {
+; CHECK-LABEL: udiv_v8i64:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].d, vl[[#min(div(VBYTES,8),8)]]
+; VBITS_GE_512-NEXT: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_512-NEXT: udiv [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_512-NEXT: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_512-NEXT: ret
+  %op1 = load <8 x i64>, <8 x i64>* %a
+  %op2 = load <8 x i64>, <8 x i64>* %b
+  %res = udiv <8 x i64> %op1, %op2
+  store <8 x i64> %res, <8 x i64>* %a
+  ret void
+}
+
+define void @udiv_v16i64(<16 x i64>* %a, <16 x i64>* %b) #0 {
+; CHECK-LABEL: udiv_v16i64:
+; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].d, vl[[#min(div(VBYTES,8),16)]]
+; VBITS_GE_1024-NEXT: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_1024-NEXT: udiv [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_1024-NEXT: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_1024-NEXT: ret
+  %op1 = load <16 x i64>, <16 x i64>* %a
+  %op2 = load <16 x i64>, <16 x i64>* %b
+  %res = udiv <16 x i64> %op1, %op2
+  store <16 x i64> %res, <16 x i64>* %a
+  ret void
+}
+
+define void @udiv_v32i64(<32 x i64>* %a, <32 x i64>* %b) #0 {
+; CHECK-LABEL: udiv_v32i64:
+; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].d, vl[[#min(div(VBYTES,8),32)]]
+; VBITS_GE_2048-NEXT: ld1d { [[OP1:z[0-9]+]].d }, [[PG]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1d { [[OP2:z[0-9]+]].d }, [[PG]]/z, [x1]
+; VBITS_GE_2048-NEXT: udiv [[RES:z[0-9]+]].d, [[PG]]/m, [[OP1]].d, [[OP2]].d
+; VBITS_GE_2048-NEXT: st1d { [[RES]].d }, [[PG]], [x0]
+; VBITS_GE_2048-NEXT: ret
+  %op1 = load <32 x i64>, <32 x i64>* %a
+  %op2 = load <32 x i64>, <32 x i64>* %b
+  %res = udiv <32 x i64> %op1, %op2
+  store <32 x i64> %res, <32 x i64>* %a
+  ret void
+}
+
 attributes #0 = { "target-features"="+sve" }