[SVE] Extend getMemVTFromNode to cover the sret variants of sve.ld2/3/4.

[paulwalker-arm]

Summary

This enables the use of reg+imm addressing modes to match the
non-sret variants of these intrinsics.

[CarolineConcatto]

BTW I pushed this patch to main:
https://reviews.llvm.org/D133023

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

5431

I believe you can remove this test, no?
It should fall into the default switch statement.

[paulwalker-arm]

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

5431

The opcode infers things like the number and type of operands and so this test ensures it's safe to do cast<ConstantSDNode>(Root->getOperand(1))->getZExtValue().

[paulwalker-arm]

ping

[sdesmalen]

This looks sensible to me.

llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp

return getPackedVectorTypeFromPredicateType(
         Ctx, Root->getOperand(1)->getValueType(0), /*NumVec=*/3);
   case AArch64ISD::SVE_LD4_MERGE_ZERO:
     return getPackedVectorTypeFromPredicateType(
         Ctx, Root->getOperand(1)->getValueType(0), /*NumVec=*/4);
   default:
     break;
   }
 
-  if (Opcode != ISD::INTRINSIC_VOID)
+  if (Opcode != ISD::INTRINSIC_VOID && Opcode != ISD::INTRINSIC_W_CHAIN)
     return EVT();
 
-  const unsigned IntNo =
-      cast<ConstantSDNode>(Root->getOperand(1))->getZExtValue();
-  if (IntNo == Intrinsic::aarch64_sme_ldr ||
-      IntNo == Intrinsic::aarch64_sme_str)
-    return MVT::nxv16i8;
-
-  if (IntNo != Intrinsic::aarch64_sve_prf)
+  switch (cast<ConstantSDNode>(Root->getOperand(1))->getZExtValue()) {
+  default:
     return EVT();
-
-  // We are using an SVE prefetch intrinsic. Type must be inferred
-  // from the width of the predicate.
+  case Intrinsic::aarch64_sme_ldr:
+  case Intrinsic::aarch64_sme_str:
+    return MVT::nxv16i8;
+  case Intrinsic::aarch64_sve_prf:
+    // We are using an SVE prefetch intrinsic. Type must be inferred from the
+    // width of the predicate.
     return getPackedVectorTypeFromPredicateType(
         Ctx, Root->getOperand(2)->getValueType(0), /*NumVec=*/1);
+  case Intrinsic::aarch64_sve_ld2_sret:
+    return getPackedVectorTypeFromPredicateType(
+        Ctx, Root->getOperand(2)->getValueType(0), /*NumVec=*/2);
+  case Intrinsic::aarch64_sve_ld3_sret:
+    return getPackedVectorTypeFromPredicateType(
+        Ctx, Root->getOperand(2)->getValueType(0), /*NumVec=*/3);
+  case Intrinsic::aarch64_sve_ld4_sret:
+    return getPackedVectorTypeFromPredicateType(
+        Ctx, Root->getOperand(2)->getValueType(0), /*NumVec=*/4);
+  }
 }
 
 /// SelectAddrModeIndexedSVE - Attempt selection of the addressing mode:
 /// Base + OffImm * sizeof(MemVT) for Min >= OffImm <= Max
 /// where Root is the memory access using N for its address.
 template <int64_t Min, int64_t Max>
 bool AArch64DAGToDAGISel::SelectAddrModeIndexedSVE(SDNode *Root, SDValue N,
                                                    SDValue &Base,

llvm/test/CodeGen/AArch64/sve-intrinsics-ldN-sret-reg+imm-addr-mode.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=aarch64--linux-gnu -mattr=sve < %s | FileCheck %s
 ; RUN: llc -mtriple=aarch64--linux-gnu -mattr=sme < %s | FileCheck %s
 
 ; NOTE: invalid, upper and lower bound immediate values of the regimm
 ; addressing mode are checked only for the byte version of each
 ; instruction (`ld<N>b`), as the code for detecting the immediate is
 ; common to all instructions, and varies only for the number of
 ; elements of the structure store, which is <N> = 2, 3, 4.
 
 ; ld2b
 define { <vscale x 16 x i8>, <vscale x 16 x i8> } @ld2.nxv32i8(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld2.nxv32i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #2
-; CHECK-NEXT:    ld2b { z0.b, z1.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld2b { z0.b, z1.b }, p0/z, [x0, #2, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 2
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8*
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8> } @ld2.nxv32i8_lower_bound(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld2.nxv32i8_lower_bound:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #-16
-; CHECK-NEXT:    ld2b { z0.b, z1.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld2b { z0.b, z1.b }, p0/z, [x0, #-16, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 -16
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8> } @ld2.nxv32i8_upper_bound(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld2.nxv32i8_upper_bound:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #14
-; CHECK-NEXT:    ld2b { z0.b, z1.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld2b { z0.b, z1.b }, p0/z, [x0, #14, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 14
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8> } @ld2.nxv32i8_not_multiple_of_2(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
; CHECK-NEXT:    ret
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 ; ld2h
 define { <vscale x 8 x i16>, <vscale x 8 x i16> } @ld2.nxv16i16(<vscale x 8 x i1> %Pg, <vscale x 8 x i16>* %addr) {
 ; CHECK-LABEL: ld2.nxv16i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #14
-; CHECK-NEXT:    ld2h { z0.h, z1.h }, p0/z, [x8]
+; CHECK-NEXT:    ld2h { z0.h, z1.h }, p0/z, [x0, #14, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* %addr, i64 14
   %base_ptr = bitcast <vscale x 8 x i16>* %base to i16 *
   %res = call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> %Pg, i16 *%base_ptr)
   ret { <vscale x 8 x i16>, <vscale x 8 x i16> } %res
 }
 
 define { <vscale x 8 x half>, <vscale x 8 x half> } @ld2.nxv16f16(<vscale x 8 x i1> %Pg, <vscale x 8 x half>* %addr) {
 ; CHECK-LABEL: ld2.nxv16f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-16
-; CHECK-NEXT:    ld2h { z0.h, z1.h }, p0/z, [x8]
+; CHECK-NEXT:    ld2h { z0.h, z1.h }, p0/z, [x0, #-16, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x half>, <vscale x 8 x half>* %addr, i64 -16
   %base_ptr = bitcast <vscale x 8 x half>* %base to half *
   %res = call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld2.sret.nxv8f16(<vscale x 8 x i1> %Pg, half *%base_ptr)
   ret { <vscale x 8 x half>, <vscale x 8 x half> } %res
 }
 
 define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @ld2.nxv16bf16(<vscale x 8 x i1> %Pg, <vscale x 8 x bfloat>* %addr) #0 {
 ; CHECK-LABEL: ld2.nxv16bf16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #12
-; CHECK-NEXT:    ld2h { z0.h, z1.h }, p0/z, [x8]
+; CHECK-NEXT:    ld2h { z0.h, z1.h }, p0/z, [x0, #12, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* %addr, i64 12
   %base_ptr = bitcast <vscale x 8 x bfloat>* %base to bfloat *
   %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld2.sret.nxv8bf16(<vscale x 8 x i1> %Pg, bfloat *%base_ptr)
   ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
 }
 
 ; ld2w
 define { <vscale x 4 x i32>, <vscale x 4 x i32> } @ld2.nxv8i32(<vscale x 4 x i1> %Pg, <vscale x 4 x i32>* %addr) {
 ; CHECK-LABEL: ld2.nxv8i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #14
-; CHECK-NEXT:    ld2w { z0.s, z1.s }, p0/z, [x8]
+; CHECK-NEXT:    ld2w { z0.s, z1.s }, p0/z, [x0, #14, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* %addr, i64 14
   %base_ptr = bitcast <vscale x 4 x i32>* %base to i32 *
   %res = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> %Pg, i32 *%base_ptr)
   ret { <vscale x 4 x i32>, <vscale x 4 x i32> } %res
 }
 
 define { <vscale x 4 x float>, <vscale x 4 x float> } @ld2.nxv8f32(<vscale x 4 x i1> %Pg, <vscale x 4 x float>* %addr) {
 ; CHECK-LABEL: ld2.nxv8f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-16
-; CHECK-NEXT:    ld2w { z0.s, z1.s }, p0/z, [x8]
+; CHECK-NEXT:    ld2w { z0.s, z1.s }, p0/z, [x0, #-16, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 4 x float>, <vscale x 4 x float>* %addr, i64 -16
   %base_ptr = bitcast <vscale x 4 x float>* %base to float *
   %res = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> %Pg, float *%base_ptr)
   ret { <vscale x 4 x float>, <vscale x 4 x float> } %res
 }
 
 ; ld2d
 define { <vscale x 2 x i64>, <vscale x 2 x i64> } @ld2.nxv4i64(<vscale x 2 x i1> %Pg, <vscale x 2 x i64>* %addr) {
 ; CHECK-LABEL: ld2.nxv4i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #14
-; CHECK-NEXT:    ld2d { z0.d, z1.d }, p0/z, [x8]
+; CHECK-NEXT:    ld2d { z0.d, z1.d }, p0/z, [x0, #14, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* %addr, i64 14
   %base_ptr = bitcast <vscale x 2 x i64>* %base to i64 *
   %res = call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> %Pg, i64 *%base_ptr)
   ret { <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
 define { <vscale x 2 x double>, <vscale x 2 x double> } @ld2.nxv4f64(<vscale x 2 x i1> %Pg, <vscale x 2 x double>* %addr) {
 ; CHECK-LABEL: ld2.nxv4f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-16
-; CHECK-NEXT:    ld2d { z0.d, z1.d }, p0/z, [x8]
+; CHECK-NEXT:    ld2d { z0.d, z1.d }, p0/z, [x0, #-16, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 2 x double>, <vscale x 2 x double>* %addr, i64 -16
   %base_ptr = bitcast <vscale x 2 x double>* %base to double *
   %res = call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> %Pg, double *%base_ptr)
   ret { <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
 ; ld3b
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld3.nxv48i8(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld3.nxv48i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #3
-; CHECK-NEXT:    ld3b { z0.b - z2.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld3b { z0.b - z2.b }, p0/z, [x0, #3, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 3
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld3.nxv48i8_lower_bound(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld3.nxv48i8_lower_bound:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #-24
-; CHECK-NEXT:    ld3b { z0.b - z2.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld3b { z0.b - z2.b }, p0/z, [x0, #-24, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 -24
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld3.nxv48i8_upper_bound(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld3.nxv48i8_upper_bound:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #21
-; CHECK-NEXT:    ld3b { z0.b - z2.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld3b { z0.b - z2.b }, p0/z, [x0, #21, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 21
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld3.nxv48i8_not_multiple_of_3_01(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
; CHECK-NEXT:    ret
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld3.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 ; ld3h
 define { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @ld3.nxv24i16(<vscale x 8 x i1> %Pg, <vscale x 8 x i16> *%addr) {
 ; CHECK-LABEL: ld3.nxv24i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #21
-; CHECK-NEXT:    ld3h { z0.h - z2.h }, p0/z, [x8]
+; CHECK-NEXT:    ld3h { z0.h - z2.h }, p0/z, [x0, #21, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* %addr, i64 21
   %base_ptr = bitcast <vscale x 8 x i16>* %base to i16 *
   %res = call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld3.sret.nxv8i16(<vscale x 8 x i1> %Pg, i16 *%base_ptr)
   ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %res
 }
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @ld3.nxv24f16(<vscale x 8 x i1> %Pg, <vscale x 8 x half> *%addr) {
 ; CHECK-LABEL: ld3.nxv24f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #21
-; CHECK-NEXT:    ld3h { z0.h - z2.h }, p0/z, [x8]
+; CHECK-NEXT:    ld3h { z0.h - z2.h }, p0/z, [x0, #21, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x half>, <vscale x 8 x half>* %addr, i64 21
   %base_ptr = bitcast <vscale x 8 x half>* %base to half *
   %res = call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld3.sret.nxv8f16(<vscale x 8 x i1> %Pg, half *%base_ptr)
   ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } %res
 }
 
 define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @ld3.nxv24bf16(<vscale x 8 x i1> %Pg, <vscale x 8 x bfloat> *%addr) #0 {
 ; CHECK-LABEL: ld3.nxv24bf16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-24
-; CHECK-NEXT:    ld3h { z0.h - z2.h }, p0/z, [x8]
+; CHECK-NEXT:    ld3h { z0.h - z2.h }, p0/z, [x0, #-24, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* %addr, i64 -24
   %base_ptr = bitcast <vscale x 8 x bfloat>* %base to bfloat *
   %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld3.sret.nxv8bf16(<vscale x 8 x i1> %Pg, bfloat *%base_ptr)
   ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
 }
 
 ; ld3w
 define { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @ld3.nxv12i32(<vscale x 4 x i1> %Pg, <vscale x 4 x i32> *%addr) {
 ; CHECK-LABEL: ld3.nxv12i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #21
-; CHECK-NEXT:    ld3w { z0.s - z2.s }, p0/z, [x8]
+; CHECK-NEXT:    ld3w { z0.s - z2.s }, p0/z, [x0, #21, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* %addr, i64 21
   %base_ptr = bitcast <vscale x 4 x i32>* %base to i32 *
   %res = call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld3.sret.nxv4i32(<vscale x 4 x i1> %Pg, i32 *%base_ptr)
   ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } %res
 }
 
 define { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @ld3.nxv12f32(<vscale x 4 x i1> %Pg, <vscale x 4 x float> *%addr) {
 ; CHECK-LABEL: ld3.nxv12f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-24
-; CHECK-NEXT:    ld3w { z0.s - z2.s }, p0/z, [x8]
+; CHECK-NEXT:    ld3w { z0.s - z2.s }, p0/z, [x0, #-24, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 4 x float>, <vscale x 4 x float>* %addr, i64 -24
   %base_ptr = bitcast <vscale x 4 x float>* %base to float *
   %res = call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld3.sret.nxv4f32(<vscale x 4 x i1> %Pg, float *%base_ptr)
   ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } %res
 }
 
 ; ld3d
 define { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @ld3.nxv6i64(<vscale x 2 x i1> %Pg, <vscale x 2 x i64> *%addr) {
 ; CHECK-LABEL: ld3.nxv6i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #21
-; CHECK-NEXT:    ld3d { z0.d - z2.d }, p0/z, [x8]
+; CHECK-NEXT:    ld3d { z0.d - z2.d }, p0/z, [x0, #21, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* %addr, i64 21
   %base_ptr = bitcast <vscale x 2 x i64>* %base to i64 *
   %res = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> %Pg, i64 *%base_ptr)
   ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
 define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @ld3.nxv6f64(<vscale x 2 x i1> %Pg, <vscale x 2 x double> *%addr) {
 ; CHECK-LABEL: ld3.nxv6f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-24
-; CHECK-NEXT:    ld3d { z0.d - z2.d }, p0/z, [x8]
+; CHECK-NEXT:    ld3d { z0.d - z2.d }, p0/z, [x0, #-24, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 2 x double>, <vscale x 2 x double>* %addr, i64 -24
   %base_ptr = bitcast <vscale x 2 x double>* %base to double *
   %res = call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld3.sret.nxv2f64(<vscale x 2 x i1> %Pg, double *%base_ptr)
   ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
 ; ; ld4b
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld4.nxv64i8(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld4.nxv64i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #4
-; CHECK-NEXT:    ld4b { z0.b - z3.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld4b { z0.b - z3.b }, p0/z, [x0, #4, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 4
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld4.nxv64i8_lower_bound(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld4.nxv64i8_lower_bound:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #-32
-; CHECK-NEXT:    ld4b { z0.b - z3.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld4b { z0.b - z3.b }, p0/z, [x0, #-32, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 -32
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld4.nxv64i8_upper_bound(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
 ; CHECK-LABEL: ld4.nxv64i8_upper_bound:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    rdvl x8, #28
-; CHECK-NEXT:    ld4b { z0.b - z3.b }, p0/z, [x0, x8]
+; CHECK-NEXT:    ld4b { z0.b - z3.b }, p0/z, [x0, #28, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 16 x i8>, <vscale x 16 x i8>* %addr, i64 28
   %base_ptr = bitcast <vscale x 16 x i8>* %base to i8 *
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 define { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @ld4.nxv64i8_not_multiple_of_4_01(<vscale x 16 x i1> %Pg, <vscale x 16 x i8> *%addr) {
; xOFFSET = RDVL * 2^-4 * 2^9 = RDVL * 32
   %res = call { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld4.sret.nxv16i8(<vscale x 16 x i1> %Pg, i8 *%base_ptr)
   ret { <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8> } %res
 }
 
 ; ld4h
 define { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @ld4.nxv32i16(<vscale x 8 x i1> %Pg, <vscale x 8 x i16> *%addr) {
 ; CHECK-LABEL: ld4.nxv32i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #8
-; CHECK-NEXT:    ld4h { z0.h - z3.h }, p0/z, [x8]
+; CHECK-NEXT:    ld4h { z0.h - z3.h }, p0/z, [x0, #8, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x i16>, <vscale x 8 x i16>* %addr, i64 8
   %base_ptr = bitcast <vscale x 8 x i16>* %base to i16 *
   %res = call { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld4.sret.nxv8i16(<vscale x 8 x i1> %Pg, i16 *%base_ptr)
   ret { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } %res
 }
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @ld4.nxv32f16(<vscale x 8 x i1> %Pg, <vscale x 8 x half> *%addr) {
 ; CHECK-LABEL: ld4.nxv32f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #28
-; CHECK-NEXT:    ld4h { z0.h - z3.h }, p0/z, [x8]
+; CHECK-NEXT:    ld4h { z0.h - z3.h }, p0/z, [x0, #28, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x half>, <vscale x 8 x half>* %addr, i64 28
   %base_ptr = bitcast <vscale x 8 x half>* %base to half *
   %res = call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sve.ld4.sret.nxv8f16(<vscale x 8 x i1> %Pg, half *%base_ptr)
   ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } %res
 }
 
 define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @ld4.nxv32bf16(<vscale x 8 x i1> %Pg, <vscale x 8 x bfloat> *%addr) #0 {
 ; CHECK-LABEL: ld4.nxv32bf16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-32
-; CHECK-NEXT:    ld4h { z0.h - z3.h }, p0/z, [x8]
+; CHECK-NEXT:    ld4h { z0.h - z3.h }, p0/z, [x0, #-32, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 8 x bfloat>, <vscale x 8 x bfloat>* %addr, i64 -32
   %base_ptr = bitcast <vscale x 8 x bfloat>* %base to bfloat *
   %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.ld4.sret.nxv8bf16(<vscale x 8 x i1> %Pg, bfloat *%base_ptr)
   ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
 }
 
 ; ld4w
 define { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @ld4.nxv16i32(<vscale x 4 x i1> %Pg, <vscale x 4 x i32> *%addr) {
 ; CHECK-LABEL: ld4.nxv16i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #28
-; CHECK-NEXT:    ld4w { z0.s - z3.s }, p0/z, [x8]
+; CHECK-NEXT:    ld4w { z0.s - z3.s }, p0/z, [x0, #28, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 4 x i32>, <vscale x 4 x i32>* %addr, i64 28
   %base_ptr = bitcast <vscale x 4 x i32>* %base to i32 *
   %res = call { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld4.sret.nxv4i32(<vscale x 4 x i1> %Pg, i32 *%base_ptr)
   ret { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } %res
 }
 
 define { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @ld4.nxv16f32(<vscale x 4 x i1> %Pg, <vscale x 4 x float>* %addr) {
 ; CHECK-LABEL: ld4.nxv16f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-32
-; CHECK-NEXT:    ld4w { z0.s - z3.s }, p0/z, [x8]
+; CHECK-NEXT:    ld4w { z0.s - z3.s }, p0/z, [x0, #-32, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 4 x float>, <vscale x 4 x float>* %addr, i64 -32
   %base_ptr = bitcast <vscale x 4 x float>* %base to float *
   %res = call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld4.sret.nxv4f32(<vscale x 4 x i1> %Pg, float *%base_ptr)
   ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } %res
 }
 
 ; ld4d
 define { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @ld4.nxv8i64(<vscale x 2 x i1> %Pg, <vscale x 2 x i64> *%addr) {
 ; CHECK-LABEL: ld4.nxv8i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #28
-; CHECK-NEXT:    ld4d { z0.d - z3.d }, p0/z, [x8]
+; CHECK-NEXT:    ld4d { z0.d - z3.d }, p0/z, [x0, #28, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 2 x i64>, <vscale x 2 x i64>* %addr, i64 28
   %base_ptr = bitcast <vscale x 2 x i64>* %base to i64 *
   %res = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld4.sret.nxv2i64(<vscale x 2 x i1> %Pg, i64 *%base_ptr)
   ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
 define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @ld4.nxv8f64(<vscale x 2 x i1> %Pg, <vscale x 2 x double> *%addr) {
 ; CHECK-LABEL: ld4.nxv8f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    addvl x8, x0, #-32
-; CHECK-NEXT:    ld4d { z0.d - z3.d }, p0/z, [x8]
+; CHECK-NEXT:    ld4d { z0.d - z3.d }, p0/z, [x0, #-32, mul vl]
 ; CHECK-NEXT:    ret
   %base = getelementptr <vscale x 2 x double>, <vscale x 2 x double>* %addr, i64 -32
   %base_ptr = bitcast <vscale x 2 x double>* %base to double *
   %res = call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld4.sret.nxv2f64(<vscale x 2 x i1> %Pg, double * %base_ptr)
   ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
 declare { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1>, i8*)