[SVE] Change some bfloat lane intrinsics to use i32 immediates

[david-arm]

Summary

Almost all of the other SVE LLVM IR intrinsics take i32 values
for lane indices or other immediates. We should bring the bfloat
intrinsics in line with that. It will also make it easier to
add support for the SVE2.1 float intrinsics in future, since
they reuse the same underlying instruction classes.

I've maintained backwards compatibility with the old i64 variants
and used the autoupgrade mechanism.

[david-arm]

• Changed patch to use autoupgrade mechanism.

[sdesmalen]

llvm/include/llvm/IR/IntrinsicsAArch64.td

2522–2524

do you also want to remove the old intrinsics?

[david-arm]

• Actually done the upgrade stuff properly this time. :)

[david-arm]

llvm/include/llvm/IR/IntrinsicsAArch64.td

2522–2524

Good spot! Turns out that not only had I done this wrong, but I'd also missed out upgrades for bfmlalb/t too. :)

[paulwalker-arm]

llvm/include/llvm/IR/IntrinsicsAArch64.td

2522–2524

Having _i32 in the name is confusing because it'll come out as .i32 when printed in IR, which looks like a type suffix but in this case it's actually part of the name.

With that said, do you have to change the name? That seems unfortunate given this is a bug fix.

If it's absolutely necessary then I suggestion using _v2 to signify this is the second version of this intrinsic.

[david-arm]

llvm/include/llvm/IR/IntrinsicsAArch64.td

2522–2524

Sadly @paulwalker-arm the auto-upgrade mechanism forbids you from upgrading to an intrinsic of the same name. I don't really know why we have that restriction though, since in theory I could decide to upgrade only when the index is a i64 type. I'll use _v2 then!

[david-arm]

I did try to avoid having to create an intrinsic with a different name, but I was thwarted at every turn! It is with great regret that I report these two problems:

1. When you write an IR test file containing a declaration of the same intrinsic with a different type and try to use the upgrade mechanism the compiler crashes in llvm::Intrinsic::getDeclaration. This is presumably because we've declared an intrinsic in the module with a different signature to that in IntrinsicsAArch64.td and it seems deeply unhappy about it.
2. Secondly, in llvm::UpgradeIntrinsicFunction we would also hit the assert assert(F != NewFn && "Intrinsic function upgraded to the same function");

Let us bemoan the evil that befalls us @paulwalker-arm!

[paulwalker-arm]

llvm/include/llvm/IR/IntrinsicsAArch64.td

1517

I suppose you don't actually need to change the name of this class.

llvm/lib/IR/AutoUpgrade.cpp

3977

This may as well be 4 given we know the exact number of arguments.

clang/include/clang/Basic/arm_sve.td

 let ArchGuard = "defined(__ARM_FEATURE_SVE_BF16)" in {
   def SVBFDOT        : SInst<"svbfdot[_{0}]",        "MMdd",  "b", MergeNone, "aarch64_sve_bfdot",        [IsOverloadNone]>;
   def SVBFMLALB      : SInst<"svbfmlalb[_{0}]",      "MMdd",  "b", MergeNone, "aarch64_sve_bfmlalb",      [IsOverloadNone]>;
   def SVBFMLALT      : SInst<"svbfmlalt[_{0}]",      "MMdd",  "b", MergeNone, "aarch64_sve_bfmlalt",      [IsOverloadNone]>;
   def SVBFMMLA       : SInst<"svbfmmla[_{0}]",       "MMdd",  "b", MergeNone, "aarch64_sve_bfmmla",       [IsOverloadNone]>;
   def SVBFDOT_N      : SInst<"svbfdot[_n_{0}]",      "MMda",  "b", MergeNone, "aarch64_sve_bfdot",        [IsOverloadNone]>;
   def SVBFMLAL_N     : SInst<"svbfmlalb[_n_{0}]",    "MMda",  "b", MergeNone, "aarch64_sve_bfmlalb",      [IsOverloadNone]>;
   def SVBFMLALT_N    : SInst<"svbfmlalt[_n_{0}]",    "MMda",  "b", MergeNone, "aarch64_sve_bfmlalt",      [IsOverloadNone]>;
-  def SVBFDOT_LANE   : SInst<"svbfdot_lane[_{0}]",   "MMddn", "b", MergeNone, "aarch64_sve_bfdot_lane",   [IsOverloadNone], [ImmCheck<3, ImmCheck0_3>]>;
-  def SVBFMLALB_LANE : SInst<"svbfmlalb_lane[_{0}]", "MMddn", "b", MergeNone, "aarch64_sve_bfmlalb_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
-  def SVBFMLALT_LANE : SInst<"svbfmlalt_lane[_{0}]", "MMddn", "b", MergeNone, "aarch64_sve_bfmlalt_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
+  def SVBFDOT_LANE   : SInst<"svbfdot_lane[_{0}]",   "MMddi", "b", MergeNone, "aarch64_sve_bfdot_lane_v2",   [IsOverloadNone], [ImmCheck<3, ImmCheck0_3>]>;
+  def SVBFMLALB_LANE : SInst<"svbfmlalb_lane[_{0}]", "MMddi", "b", MergeNone, "aarch64_sve_bfmlalb_lane_v2", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
+  def SVBFMLALT_LANE : SInst<"svbfmlalt_lane[_{0}]", "MMddi", "b", MergeNone, "aarch64_sve_bfmlalt_lane_v2", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Stores
 
 // Store one vector (scalar base)
 def SVST1    : MInst<"svst1[_{d}]",  "vPpd", "csilUcUsUiUlhfd", [IsStore], MemEltTyDefault, "aarch64_sve_st1">;
 def SVST1B_S : MInst<"svst1b[_{d}]", "vPAd", "sil",             [IsStore], MemEltTyInt8,    "aarch64_sve_st1">;

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_bfdot.c

 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_bfdot_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfdot, _f32, , )(x, y, z);
 }
 
 // CHECK-LABEL: @test_bfdot_lane_0_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 0)
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z21test_bfdot_lane_0_f32u13__SVFloat32_tu14__SVBFloat16_tu14__SVBFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 0)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_bfdot_lane_0_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfdot_lane, _f32, , )(x, y, z, 0);
 }
 
 // CHECK-LABEL: @test_bfdot_lane_3_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 3)
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 3)
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z21test_bfdot_lane_3_f32u13__SVFloat32_tu14__SVBFloat16_tu14__SVBFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 3)
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 3)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_bfdot_lane_3_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfdot_lane, _f32, , )(x, y, z, 3);
 }
 
 // CHECK-LABEL: @test_bfdot_n_f32(
 // CHECK-NEXT:  entry:

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_bfmlalb.c

 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_svbfmlalb_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfmlalb, _f32, , )(x, y, z);
 }
 
 // CHECK-LABEL: @test_bfmlalb_lane_0_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 0)
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z23test_bfmlalb_lane_0_f32u13__SVFloat32_tu14__SVBFloat16_tu14__SVBFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 0)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_bfmlalb_lane_0_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfmlalb_lane, _f32, , )(x, y, z, 0);
 }
 
 // CHECK-LABEL: @test_bfmlalb_lane_7_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 7)
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 7)
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z23test_bfmlalb_lane_7_f32u13__SVFloat32_tu14__SVBFloat16_tu14__SVBFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 7)
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_bfmlalb_lane_7_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfmlalb_lane, _f32, , )(x, y, z, 7);
 }
 
 // CHECK-LABEL: @test_bfmlalb_n_f32(
 // CHECK-NEXT:  entry:

clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_bfmlalt.c

 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_svbfmlalt_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfmlalt, _f32, , )(x, y, z);
 }
 
 // CHECK-LABEL: @test_bfmlalt_lane_0_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 0)
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z23test_bfmlalt_lane_0_f32u13__SVFloat32_tu14__SVBFloat16_tu14__SVBFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 0)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_bfmlalt_lane_0_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfmlalt_lane, _f32, , )(x, y, z, 0);
 }
 
 // CHECK-LABEL: @test_bfmlalt_lane_7_f32(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 7)
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 7)
 // CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 // CPP-CHECK-LABEL: @_Z23test_bfmlalt_lane_7_f32u13__SVFloat32_tu14__SVBFloat16_tu14__SVBFloat16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i64 7)
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> [[X:%.*]], <vscale x 8 x bfloat> [[Y:%.*]], <vscale x 8 x bfloat> [[Z:%.*]], i32 7)
 // CPP-CHECK-NEXT:    ret <vscale x 4 x float> [[TMP0]]
 //
 svfloat32_t test_bfmlalt_lane_7_f32(svfloat32_t x, svbfloat16_t y, svbfloat16_t z) {
   return SVE_ACLE_FUNC(svbfmlalt_lane, _f32, , )(x, y, z, 7);
 }
 
 // CHECK-LABEL: @test_bfmlalt_n_f32(
 // CHECK-NEXT:  entry:

llvm/include/llvm/IR/IntrinsicsAArch64.td

: DefaultAttrsIntrinsic<[llvm_anyvector_ty],
                 [LLVMMatchType<0>, LLVMSubdivide4VectorType<0>, LLVMSubdivide4VectorType<0>],
                 [IntrNoMem]>;
 
 class SVE_4Vec_BF16
     : DefaultAttrsIntrinsic<[llvm_nxv4f32_ty],
                 [llvm_nxv4f32_ty, llvm_nxv8bf16_ty, llvm_nxv8bf16_ty],
                 [IntrNoMem]>;
 
 class SVE_4Vec_BF16_Indexed
     : DefaultAttrsIntrinsic<[llvm_nxv4f32_ty],
-                [llvm_nxv4f32_ty, llvm_nxv8bf16_ty, llvm_nxv8bf16_ty, llvm_i64_ty],
+                [llvm_nxv4f32_ty, llvm_nxv8bf16_ty, llvm_nxv8bf16_ty, llvm_i32_ty],
                 [IntrNoMem, ImmArg<ArgIndex<3>>]>;
 
 //
 // Loads
 //
 
 def int_aarch64_sve_ld1   : AdvSIMD_1Vec_PredLoad_Intrinsic;
 

 //
 
 def int_aarch64_sve_bfdot   : SVE_4Vec_BF16;
 def int_aarch64_sve_bfmlalb : SVE_4Vec_BF16;
 def int_aarch64_sve_bfmlalt : SVE_4Vec_BF16;
 
 def int_aarch64_sve_bfmmla  : SVE_4Vec_BF16;
 
-def int_aarch64_sve_bfdot_lane   : SVE_4Vec_BF16_Indexed;
-def int_aarch64_sve_bfmlalb_lane : SVE_4Vec_BF16_Indexed;
-def int_aarch64_sve_bfmlalt_lane : SVE_4Vec_BF16_Indexed;
+def int_aarch64_sve_bfdot_lane_v2   : SVE_4Vec_BF16_Indexed;
+def int_aarch64_sve_bfmlalb_lane_v2 : SVE_4Vec_BF16_Indexed;
+def int_aarch64_sve_bfmlalt_lane_v2 : SVE_4Vec_BF16_Indexed;
 }
 
 //
 // SVE2 - Contiguous conflict detection
 //
 
 def int_aarch64_sve_whilerw_b : SVE2_CONFLICT_DETECT_Intrinsic;
 def int_aarch64_sve_whilerw_h : SVE2_CONFLICT_DETECT_Intrinsic;

llvm/lib/IR/AutoUpgrade.cpp

if (Name.startswith("aarch64.neon.frintn")) {
                                         F->arg_begin()->getType());
       return true;
     }
     if (Name.startswith("aarch64.neon.rbit")) {
       NewFn = Intrinsic::getDeclaration(F->getParent(), Intrinsic::bitreverse,
                                         F->arg_begin()->getType());
       return true;
     }
+    if (Name == "aarch64.sve.bfdot.lane") {
+      NewFn = Intrinsic::getDeclaration(F->getParent(),
+                                        Intrinsic::aarch64_sve_bfdot_lane_v2);
+      return true;
+    }
+    if (Name == "aarch64.sve.bfmlalb.lane") {
+      NewFn = Intrinsic::getDeclaration(F->getParent(),
+                                        Intrinsic::aarch64_sve_bfmlalb_lane_v2);
+      return true;
+    }
+    if (Name == "aarch64.sve.bfmlalt.lane") {
+      NewFn = Intrinsic::getDeclaration(F->getParent(),
+                                        Intrinsic::aarch64_sve_bfmlalt_lane_v2);
+      return true;
+    }
     static const Regex LdRegex("^aarch64\\.sve\\.ld[234](.nxv[a-z0-9]+|$)");
     if (LdRegex.match(Name)) {
       Type *ScalarTy =
           dyn_cast<VectorType>(F->getReturnType())->getElementType();
       ElementCount EC =
           dyn_cast<VectorType>(F->arg_begin()->getType())->getElementCount();
       Type *Ty = VectorType::get(ScalarTy, EC);
       Intrinsic::ID ID =

   case Intrinsic::arm_neon_vst4:
   case Intrinsic::arm_neon_vst2lane:
   case Intrinsic::arm_neon_vst3lane:
   case Intrinsic::arm_neon_vst4lane: {
     SmallVector<Value *, 4> Args(CI->args());
     NewCall = Builder.CreateCall(NewFn, Args);
     break;
   }
+  case Intrinsic::aarch64_sve_bfmlalb_lane_v2:
+  case Intrinsic::aarch64_sve_bfmlalt_lane_v2:
+  case Intrinsic::aarch64_sve_bfdot_lane_v2: {
+    LLVMContext &Ctx = F->getParent()->getContext();
+    SmallVector<Value *, 4> Args(CI->args());
+    Args[3] = ConstantInt::get(Type::getInt32Ty(Ctx),
+                               cast<ConstantInt>(Args[3])->getZExtValue());
+    NewCall = Builder.CreateCall(NewFn, Args);
+    break;
+  }
   case Intrinsic::aarch64_sve_ld3_sret:
   case Intrinsic::aarch64_sve_ld4_sret:
   case Intrinsic::aarch64_sve_ld2_sret: {
     StringRef Name = F->getName();
     Name = Name.substr(5);
     unsigned N = StringSwitch<unsigned>(Name)
                      .StartsWith("aarch64.sve.ld2", 2)
                      .StartsWith("aarch64.sve.ld3", 3)

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

let Predicates = [HasSVEorSME] in {
   defm FRINTA_ZPmZ : sve_fp_2op_p_zd_HSD<0b00100, "frinta", AArch64frinta_mt>;
   defm FRINTX_ZPmZ : sve_fp_2op_p_zd_HSD<0b00110, "frintx", AArch64frintx_mt>;
   defm FRINTI_ZPmZ : sve_fp_2op_p_zd_HSD<0b00111, "frinti", AArch64frinti_mt>;
   defm FRECPX_ZPmZ : sve_fp_2op_p_zd_HSD<0b01100, "frecpx", AArch64frecpx_mt>;
   defm FSQRT_ZPmZ  : sve_fp_2op_p_zd_HSD<0b01101, "fsqrt",  AArch64fsqrt_mt>;
 } // End HasSVEorSME
 
 let Predicates = [HasBF16, HasSVEorSME] in {
-  defm BFDOT_ZZZ    : sve_bfloat_dot<"bfdot", int_aarch64_sve_bfdot>;
-  defm BFDOT_ZZI    : sve_bfloat_dot_indexed<"bfdot", int_aarch64_sve_bfdot_lane>;
+  defm BFDOT_ZZZ    : sve_float_dot<0b1, "bfdot", nxv8bf16, int_aarch64_sve_bfdot>;
+  defm BFDOT_ZZI    : sve_float_dot_indexed<0b1, "bfdot", nxv8bf16, int_aarch64_sve_bfdot_lane_v2>;
 } // End HasBF16, HasSVEorSME
 
 let Predicates = [HasBF16, HasSVE] in {
   defm BFMMLA_ZZZ   : sve_bfloat_matmul<"bfmmla", int_aarch64_sve_bfmmla>;
 } // End HasBF16, HasSVE
 
 let Predicates = [HasBF16, HasSVEorSME] in {
   defm BFMLALB_ZZZ : sve_bfloat_matmul_longvecl<0b0, 0b0, "bfmlalb", int_aarch64_sve_bfmlalb>;
   defm BFMLALT_ZZZ : sve_bfloat_matmul_longvecl<0b1, 0b0, "bfmlalt", int_aarch64_sve_bfmlalt>;
-  defm BFMLALB_ZZZI : sve_bfloat_matmul_longvecl_idx<0b0, 0b0, "bfmlalb", int_aarch64_sve_bfmlalb_lane>;
-  defm BFMLALT_ZZZI : sve_bfloat_matmul_longvecl_idx<0b1, 0b0, "bfmlalt", int_aarch64_sve_bfmlalt_lane>;
+  defm BFMLALB_ZZZI : sve_bfloat_matmul_longvecl_idx<0b0, 0b0, "bfmlalb", int_aarch64_sve_bfmlalb_lane_v2>;
+  defm BFMLALT_ZZZI : sve_bfloat_matmul_longvecl_idx<0b1, 0b0, "bfmlalt", int_aarch64_sve_bfmlalt_lane_v2>;
   defm BFCVT_ZPmZ   : sve_bfloat_convert<0b1, "bfcvt",   int_aarch64_sve_fcvt_bf16f32>;
   defm BFCVTNT_ZPmZ : sve_bfloat_convert<0b0, "bfcvtnt", int_aarch64_sve_fcvtnt_bf16f32>;
 } // End HasBF16, HasSVEorSME
 
 let Predicates = [HasSVEorSME] in {
   // InstAliases
   def : InstAlias<"mov $Zd, $Zn",
                   (ORR_ZZZ ZPR64:$Zd, ZPR64:$Zn, ZPR64:$Zn), 1>;

llvm/lib/Target/AArch64/SVEInstrFormats.td

: I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR16:$Zm),
   let Inst{15-10} = 0b100000;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zda;
 
   let Constraints = "$Zda = $_Zda";
   let DestructiveInstType = DestructiveOther;
 }
 
-multiclass sve_bfloat_dot<string asm, SDPatternOperator op> {
-  def NAME : sve_float_dot<0b1, asm>;
-  def : SVE_3_Op_Pat<nxv4f32, op, nxv4f32, nxv8bf16, nxv8bf16 ,!cast<Instruction>(NAME)>;
+multiclass sve_float_dot<bit bf, string asm, ValueType InVT, SDPatternOperator op> {
+  def NAME : sve_float_dot<bf, asm>;
+  def : SVE_3_Op_Pat<nxv4f32, op, nxv4f32, InVT, InVT, !cast<Instruction>(NAME)>;
 }
 
 class sve_float_dot_indexed<bit bf, string asm>
-: I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR3b16:$Zm, VectorIndexS:$iop),
+: I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR3b16:$Zm, VectorIndexS32b:$iop),
     asm, "\t$Zda, $Zn, $Zm$iop", "", []>, Sched<[]> {
   bits<5> Zda;
   bits<5> Zn;
   bits<3> Zm;
   bits<2> iop;
   let Inst{31-23} = 0b011001000;
   let Inst{22}    = bf;
   let Inst{21}    = 0b1;
   let Inst{20-19} = iop;
   let Inst{18-16} = Zm;
   let Inst{15-10} = 0b010000;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zda;
 
   let Constraints = "$Zda = $_Zda";
   let DestructiveInstType = DestructiveOther;
 }
 
-multiclass sve_bfloat_dot_indexed<string asm, SDPatternOperator op> {
-  def NAME : sve_float_dot_indexed<0b1, asm>;
-  def : SVE_4_Op_Imm_Pat<nxv4f32, op, nxv4f32, nxv8bf16, nxv8bf16, i64, VectorIndexS_timm, !cast<Instruction>(NAME)>;
+multiclass sve_float_dot_indexed<bit bf, string asm, ValueType InVT, SDPatternOperator op> {
+  def NAME : sve_float_dot_indexed<bf, asm>;
+  def : SVE_4_Op_Imm_Pat<nxv4f32, op, nxv4f32, InVT, InVT, i32, VectorIndexS32b_timm, !cast<Instruction>(NAME)>;
 }
 
 class sve_bfloat_matmul<string asm>
 : I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR16:$Zm),
   asm, "\t$Zda, $Zn, $Zm", "", []>, Sched<[]> {
   bits<5> Zm;
   bits<5> Zda;
   bits<5> Zn;

 }
 
 multiclass sve_bfloat_matmul_longvecl<bit BT, bit sub, string asm, SDPatternOperator op> {
   def NAME : sve_bfloat_matmul_longvecl<BT, sub, asm>;
   def : SVE_3_Op_Pat<nxv4f32, op, nxv4f32, nxv8bf16, nxv8bf16 ,!cast<Instruction>(NAME)>;
 }
 
 class sve_bfloat_matmul_longvecl_idx<bit BT, bit sub, string asm>
-: I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR3b16:$Zm, VectorIndexH:$iop),
+: I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR3b16:$Zm, VectorIndexH32b:$iop),
     asm, "\t$Zda, $Zn, $Zm$iop", "", []>, Sched<[]> {
   bits<5> Zda;
   bits<5> Zn;
   bits<3> Zm;
   bits<3> iop;
   let Inst{31-21} = 0b01100100111;
   let Inst{20-19} = iop{2-1};
   let Inst{18-16} = Zm;
   let Inst{15-14} = 0b01;
   let Inst{13}    = sub;
   let Inst{12}    = 0b0;
   let Inst{11}    = iop{0};
   let Inst{10}    = BT;
   let Inst{9-5}   = Zn;
   let Inst{4-0}   = Zda;
 
   let Constraints = "$Zda = $_Zda";
   let DestructiveInstType = DestructiveOther;
   let ElementSize = ElementSizeH;
 }
 
 multiclass sve_bfloat_matmul_longvecl_idx<bit BT, bit sub, string asm, SDPatternOperator op> {
   def NAME : sve_bfloat_matmul_longvecl_idx<BT, sub, asm>;
-  def : SVE_4_Op_Imm_Pat<nxv4f32, op, nxv4f32, nxv8bf16, nxv8bf16, i64, VectorIndexH_timm, !cast<Instruction>(NAME)>;
+  def : SVE_4_Op_Imm_Pat<nxv4f32, op, nxv4f32, nxv8bf16, nxv8bf16, i32, VectorIndexH32b_timm, !cast<Instruction>(NAME)>;
 }
 
 class sve_bfloat_convert<bit N, string asm>
 : I<(outs ZPR16:$Zd), (ins ZPR16:$_Zd, PPR3bAny:$Pg, ZPR32:$Zn),
   asm, "\t$Zd, $Pg/m, $Zn", "", []>, Sched<[]> {
   bits<5> Zd;
   bits<3> Pg;
   bits<5> Zn;
class sve2p1_two_way_dot_vv<string mnemonic, bit u>
 
   let Constraints = "$Zda = $_Zda";
   let DestructiveInstType = DestructiveOther;
 }
 
 
 // SVE two-way dot product (indexed)
 class sve2p1_two_way_dot_vvi<string mnemonic, bit u>
-    : I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR3b16:$Zm, VectorIndexS:$i2),
+    : I<(outs ZPR32:$Zda), (ins ZPR32:$_Zda, ZPR16:$Zn, ZPR3b16:$Zm, VectorIndexS32b:$i2),
         mnemonic, "\t$Zda, $Zn, $Zm$i2",
         "", []>, Sched<[]> {
   bits<5> Zda;
   bits<5> Zn;
   bits<3> Zm;
   bits<2> i2;
   let Inst{31-21} = 0b01000100100;
   let Inst{20-19} = i2;

llvm/test/Bitcode/upgrade-aarch64-sve-intrinsics.ll

 ; CHECK-LABEL: @get_tuple2_nxv8i32_elt1
 ; CHECK:  %ext = call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> %tuple, i64 4)
 ; CHECK-NEXT:  ret <vscale x 4 x i32> %ext
 
   %ext = call <vscale x 4 x i32> @llvm.aarch64.sve.tuple.get.nxv8i32(<vscale x 8 x i32> %tuple, i32 1)
   ret <vscale x 4 x i32> %ext
 }
 
+; bfdot
+define <vscale x 4 x float> @bfdot_lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
+; CHECK-LABEL: @bfdot_lane
+; CHECK:       %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 0)
+; CHECK-NEXT:  ret <vscale x 4 x float> %out
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 0)
+  ret <vscale x 4 x float> %out
+}
+
+; bfmlalb
+define <vscale x 4 x float> @bfmlalb_lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
+; CHECK-LABEL: @bfmlalb_lane
+; CHECK:       %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 0)
+; CHECK-NEXT:  ret <vscale x 4 x float> %out
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 0)
+  ret <vscale x 4 x float> %out
+}
+
+; bfmlalt
+define <vscale x 4 x float> @bfmlalt_lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
+; CHECK-LABEL: @bfmlalt_lane
+; CHECK:       %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 0)
+; CHECK-NEXT:  ret <vscale x 4 x float> %out
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 0)
+  ret <vscale x 4 x float> %out
+}
+
 declare  <vscale x 32 x i8> @llvm.aarch64.sve.tuple.create2.nxv32i8.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>)
 declare  <vscale x 32 x i8> @llvm.aarch64.sve.tuple.create2.nxv32i8(<vscale x 16 x i8>, <vscale x 16 x i8>)
 declare  <vscale x 32 x i8> @llvm.aarch64.sve.tuple.create2(<vscale x 16 x i8>, <vscale x 16 x i8>)
 declare  <vscale x 24 x i16> @llvm.aarch64.sve.tuple.create3.nxv24i16.nxv8i16(<vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>)
 declare <vscale x 64 x i8> @llvm.aarch64.sve.tuple.create4.nxv64i8.nxv16i8(<vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>)
 declare <vscale x 16 x i8> @llvm.aarch64.sve.tuple.create1.nxv16i8.nxv16i8(<vscale x 16 x i8>)
 declare <vscale x 8 x i32> @llvm.aarch64.sve.tuple.set.nxv8i32.nxv4i32(<vscale x 8 x i32>, i32, <vscale x 4 x i32>)
 declare <vscale x 4 x i32> @llvm.aarch64.sve.tuple.get.nxv8i32(<vscale x 8 x i32>, i32)
+declare <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i64)
+declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i64)
+declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i64)

llvm/test/CodeGen/AArch64/sve-intrinsics-bfloat.ll

; CHECK-NEXT:    ret
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfdot_lane_0_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfdot_lane_0_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfdot z0.s, z1.h, z2.h[0]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 0)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 0)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfdot_lane_1_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfdot_lane_1_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfdot z0.s, z1.h, z2.h[1]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 1)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 1)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfdot_lane_2_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfdot_lane_2_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfdot z0.s, z1.h, z2.h[2]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 2)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 2)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfdot_lane_3_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfdot_lane_3_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfdot z0.s, z1.h, z2.h[3]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 3)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 3)
   ret <vscale x 4 x float> %out
 }
 
 ;
 ; BFMLALB
 ;
 
 define <vscale x 4 x float> @bfmlalb_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h
 ; CHECK-NEXT:    ret
   %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_0_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_0_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[0]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 0)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 0)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_1_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_1_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[1]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 1)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 1)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_2_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_2_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[2]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 2)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 2)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_3_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_3_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[3]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 3)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 3)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_4_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_4_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[4]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 4)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 4)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_5_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_5_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[5]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 5)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 5)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_6_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_6_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[6]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 6)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 6)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalb_lane_7_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalb_lane_7_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalb z0.s, z1.h, z2.h[7]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 7)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 7)
   ret <vscale x 4 x float> %out
 }
 
 ;
 ; BFMLALT
 ;
 
 define <vscale x 4 x float> @bfmlalt_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h
 ; CHECK-NEXT:    ret
   %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_0_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_0_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[0]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 0)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 0)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_1_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_1_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[1]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 1)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 1)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_2_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_2_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[2]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 2)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 2)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_3_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_3_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[3]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 3)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 3)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_4_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_4_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[4]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 4)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 4)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_5_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_5_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[5]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 5)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 5)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_6_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_6_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[6]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 6)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 6)
   ret <vscale x 4 x float> %out
 }
 
 define <vscale x 4 x float> @bfmlalt_lane_7_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {
 ; CHECK-LABEL: bfmlalt_lane_7_f32:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfmlalt z0.s, z1.h, z2.h[7]
 ; CHECK-NEXT:    ret
-  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i64 7)
+  %out = call <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c, i32 7)
   ret <vscale x 4 x float> %out
 }
 
 ;
 ; BFMMLA
 ;
 
 define <vscale x 4 x float> @bfmmla_f32(<vscale x 4 x float> %a, <vscale x 8 x bfloat> %b, <vscale x 8 x bfloat> %c) nounwind {

 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bfcvtnt z0.h, p0/m, z1.s
 ; CHECK-NEXT:    ret
   %out = call <vscale x 8 x bfloat> @llvm.aarch64.sve.fcvtnt.bf16f32(<vscale x 8 x bfloat> %a, <vscale x 8 x i1> %pg, <vscale x 4 x float> %b)
   ret <vscale x 8 x bfloat> %out
 }
 
 declare <vscale x 4 x float> @llvm.aarch64.sve.bfdot(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>)
-declare <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i64)
+declare <vscale x 4 x float> @llvm.aarch64.sve.bfdot.lane.v2(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i32)
 declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>)
-declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i64)
+declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalb.lane.v2(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i32)
 declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>)
-declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i64)
+declare <vscale x 4 x float> @llvm.aarch64.sve.bfmlalt.lane.v2(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, i32)
 declare <vscale x 4 x float> @llvm.aarch64.sve.bfmmla(<vscale x 4 x float>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>)
 declare <vscale x 8 x bfloat> @llvm.aarch64.sve.fcvt.bf16f32(<vscale x 8 x bfloat>, <vscale x 8 x i1>, <vscale x 4 x float>)
 declare <vscale x 8 x bfloat> @llvm.aarch64.sve.fcvtnt.bf16f32(<vscale x 8 x bfloat>, <vscale x 8 x i1>, <vscale x 4 x float>)