[clang] [llvm] [AArch64] Add intrinsics for bflaot16 min/max/minnm/maxnm (PR #90105)

[Hassnaa Hamdi via cfe-commits]

https://github.com/hassnaaHamdi updated https://github.com/llvm/llvm-project/pull/90105

>From 09f69e8550e58d3977ddb22c84517ddd918e4c94 Mon Sep 17 00:00:00 2001
From: Hassnaa Hamdi <hassnaa.hamdi at arm.com>
Date: Thu, 25 Apr 2024 19:02:04 +0000
Subject: [PATCH] [AArch64] Add intrinsics for bfloat16 min/max/minnm/maxnm

Change-Id: I3bbbc2ec2a4155edb494d738b5444bcebc263c08
---
 clang/include/clang/Basic/arm_sve.td          |  10 +-
 .../aarch64-sme2-intrinsics/acle_sme2_max.c   | 150 +++++++++++++++++-
 .../aarch64-sme2-intrinsics/acle_sme2_maxnm.c | 150 +++++++++++++++++-
 .../aarch64-sme2-intrinsics/acle_sme2_min.c   | 150 +++++++++++++++++-
 .../aarch64-sme2-intrinsics/acle_sme2_minnm.c | 150 +++++++++++++++++-
 .../Target/AArch64/AArch64ISelDAGToDAG.cpp    |  74 +++++----
 .../CodeGen/AArch64/sme2-intrinsics-max.ll    | 126 ++++++++++++++-
 .../CodeGen/AArch64/sme2-intrinsics-min.ll    | 127 ++++++++++++++-
 8 files changed, 875 insertions(+), 62 deletions(-)

diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 6cc249837d3f3..2fe7ec7271b2a 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -2094,7 +2094,7 @@ def SVFCLAMP_BF   : SInst<"svclamp[_{d}]", "dddd", "b", MergeNone, "aarch64_sve_
 multiclass MinMaxIntr<string i, string zm, string mul, string t> {
   def SVS # NAME : SInst<"sv" # i # "[" # zm # "_{d}_" # mul # "]", t, "csil",     MergeNone, "aarch64_sve_s" # i # zm # "_" # mul, [IsStreaming], []>;
   def SVU # NAME : SInst<"sv" # i # "[" # zm # "_{d}_" # mul # "]", t, "UcUsUiUl", MergeNone, "aarch64_sve_u" # i # zm # "_" # mul, [IsStreaming], []>;
-  def SVF # NAME : SInst<"sv" # i # "[" # zm # "_{d}_" # mul # "]", t, "hfd",      MergeNone, "aarch64_sve_f" # i # zm # "_" # mul, [IsStreaming], []>;
+  def SVF # NAME : SInst<"sv" # i # "[" # zm # "_{d}_" # mul # "]", t, "bhfd",      MergeNone, "aarch64_sve_f" # i # zm # "_" # mul, [IsStreaming], []>;
 }
 
 let TargetGuard = "sme2" in {
@@ -2112,11 +2112,11 @@ let TargetGuard = "sme2" in {
 }
 
 multiclass SInstMinMaxByVector<string name> {
-  def NAME # _SINGLE_X2 : SInst<"sv" # name # "nm[_single_{d}_x2]", "22d", "hfd", MergeNone, "aarch64_sve_f" # name # "nm_single_x2", [IsStreaming], []>;
-  def NAME # _SINGLE_X4 : SInst<"sv" # name # "nm[_single_{d}_x4]", "44d", "hfd", MergeNone, "aarch64_sve_f" # name # "nm_single_x4", [IsStreaming], []>;
+  def NAME # _SINGLE_X2 : SInst<"sv" # name # "nm[_single_{d}_x2]", "22d", "bhfd", MergeNone, "aarch64_sve_f" # name # "nm_single_x2", [IsStreaming], []>;
+  def NAME # _SINGLE_X4 : SInst<"sv" # name # "nm[_single_{d}_x4]", "44d", "bhfd", MergeNone, "aarch64_sve_f" # name # "nm_single_x4", [IsStreaming], []>;
 
-  def NAME # _X2 : SInst<"sv" # name # "nm[_{d}_x2]", "222", "hfd", MergeNone, "aarch64_sve_f" # name # "nm_x2", [IsStreaming], []>;
-  def NAME # _X4 : SInst<"sv" # name # "nm[_{d}_x4]", "444", "hfd", MergeNone, "aarch64_sve_f" # name # "nm_x4", [IsStreaming], []>;
+  def NAME # _X2 : SInst<"sv" # name # "nm[_{d}_x2]", "222", "bhfd", MergeNone, "aarch64_sve_f" # name # "nm_x2", [IsStreaming], []>;
+  def NAME # _X4 : SInst<"sv" # name # "nm[_{d}_x4]", "444", "bhfd", MergeNone, "aarch64_sve_f" # name # "nm_x4", [IsStreaming], []>;
 }
 
 let TargetGuard = "sme2" in {
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
index a4e2616784efa..1420cadf2d3e9 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
@@ -1,9 +1,9 @@
 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
-// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
 // REQUIRES: aarch64-registered-target
 #include <arm_sme.h>
 
@@ -224,6 +224,32 @@ svuint64x2_t test_svmax_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_str
   return SVE_ACLE_FUNC(svmax,_single_u64_x2)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmax_single_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z25test_svmax_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+svbfloat16x2_t test_svmax_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmax,_single_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmax_single_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -608,6 +634,44 @@ svuint64x4_t test_svmax_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_str
   return SVE_ACLE_FUNC(svmax,_single_u64_x4)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmax_single_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+// CPP-CHECK-LABEL: @_Z25test_svmax_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+svbfloat16x4_t test_svmax_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmax,_single_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmax_single_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
@@ -964,6 +1028,36 @@ svuint64x2_t test_svmax_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streamin
   return SVE_ACLE_FUNC(svmax,_u64_x2)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmax_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svmax_bf16_x214svbfloat16x2_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+svbfloat16x2_t test_svmax_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmax,_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmax_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -1424,6 +1518,52 @@ svuint64x4_t test_svmax_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streamin
   return SVE_ACLE_FUNC(svmax,_u64_x4)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmax_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svmax_bf16_x414svbfloat16x4_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+svbfloat16x4_t test_svmax_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmax,_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmax_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c
index 3e554212cb70b..ce32659c94836 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c
@@ -1,11 +1,11 @@
 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
 // REQUIRES: aarch64-registered-target
 
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
 #include <arm_sme.h>
 
 #ifdef SVE_OVERLOADED_FORMS
@@ -17,6 +17,32 @@
 
 // Single, x2
 
+// CHECK-LABEL: @test_svmaxnm_single_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z27test_svmaxnm_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+svbfloat16x2_t test_svmaxnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmaxnm,_single_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmaxnm_single_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -97,6 +123,44 @@ svfloat64x2_t test_svmaxnm_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __ar
 
 // Single, x4
 
+// CHECK-LABEL: @test_svmaxnm_single_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+// CPP-CHECK-LABEL: @_Z27test_svmaxnm_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+svbfloat16x4_t test_svmaxnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmaxnm,_single_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmaxnm_single_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
@@ -213,6 +277,36 @@ svfloat64x4_t test_svmaxnm_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __ar
 
 // Multi, x2
 
+// CHECK-LABEL: @test_svmaxnm_multi_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z26test_svmaxnm_multi_bf16_x214svbfloat16x2_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+svbfloat16x2_t test_svmaxnm_multi_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmaxnm,_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmaxnm_multi_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -305,6 +399,52 @@ svfloat64x2_t test_svmaxnm_multi_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __a
 
 // Multi, x4
 
+// CHECK-LABEL: @test_svmaxnm_multi_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z26test_svmaxnm_multi_bf16_x414svbfloat16x4_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+svbfloat16x4_t test_svmaxnm_multi_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmaxnm,_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmaxnm_multi_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c
index a438fd395219b..76c2c143a7dc3 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c
@@ -1,9 +1,9 @@
 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
-// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
 // REQUIRES: aarch64-registered-target
 #include <arm_sme.h>
 
@@ -224,6 +224,32 @@ svuint64x2_t test_svmin_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_str
   return SVE_ACLE_FUNC(svmin,_single_u64_x2)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmin_single_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z25test_svmin_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+svbfloat16x2_t test_svmin_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmin,_single_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmin_single_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -608,6 +634,44 @@ svuint64x4_t test_svmin_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_str
   return SVE_ACLE_FUNC(svmin,_single_u64_x4)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmin_single_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+// CPP-CHECK-LABEL: @_Z25test_svmin_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+svbfloat16x4_t test_svmin_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmin,_single_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmin_single_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
@@ -964,6 +1028,36 @@ svuint64x2_t test_svmin_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streamin
   return SVE_ACLE_FUNC(svmin,_u64_x2)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmin_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svmin_bf16_x214svbfloat16x2_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+svbfloat16x2_t test_svmin_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmin,_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmin_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -1424,6 +1518,52 @@ svuint64x4_t test_svmin_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streamin
   return SVE_ACLE_FUNC(svmin,_u64_x4)(zdn, zm);
 }
 
+// CHECK-LABEL: @test_svmin_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svmin_bf16_x414svbfloat16x4_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+svbfloat16x4_t test_svmin_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svmin,_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svmin_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c
index b0cbdc748dc80..4c93c5e86beef 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c
@@ -1,11 +1,11 @@
 // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
 // REQUIRES: aarch64-registered-target
 
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
-// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +b16b16 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
 #include <arm_sme.h>
 
 #ifdef SVE_OVERLOADED_FORMS
@@ -17,6 +17,32 @@
 
 // Single, x2
 
+// CHECK-LABEL: @test_svminnm_single_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z27test_svminnm_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP6]]
+//
+svbfloat16x2_t test_svminnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svminnm,_single_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svminnm_single_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -97,6 +123,44 @@ svfloat64x2_t test_svminnm_single_f64_x2(svfloat64x2_t zdn, svfloat64_t zm) __ar
 
 // Single, x4
 
+// CHECK-LABEL: @test_svminnm_single_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+// CPP-CHECK-LABEL: @_Z27test_svminnm_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP12]]
+//
+svbfloat16x4_t test_svminnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svminnm,_single_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svminnm_single_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
@@ -213,6 +277,36 @@ svfloat64x4_t test_svminnm_single_f64_x4(svfloat64x4_t zdn, svfloat64_t zm) __ar
 
 // Multi, x2
 
+// CHECK-LABEL: @test_svminnm_multi_bf16_x2(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z26test_svminnm_multi_bf16_x214svbfloat16x2_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT:    ret <vscale x 16 x bfloat> [[TMP8]]
+//
+svbfloat16x2_t test_svminnm_multi_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svminnm,_bf16_x2)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svminnm_multi_f16_x2(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -305,6 +399,52 @@ svfloat64x2_t test_svminnm_multi_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __a
 
 // Multi, x4
 
+// CHECK-LABEL: @test_svminnm_multi_bf16_x4(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z26test_svminnm_multi_bf16_x414svbfloat16x4_tS_(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 3
+// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP14]], <vscale x 8 x bfloat> [[TMP15]], i64 24)
+// CPP-CHECK-NEXT:    ret <vscale x 32 x bfloat> [[TMP16]]
+//
+svbfloat16x4_t test_svminnm_multi_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm) __arm_streaming {
+  return SVE_ACLE_FUNC(svminnm,_bf16_x4)(zdn, zm);
+}
+
 // CHECK-LABEL: @test_svminnm_multi_f16_x4(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
diff --git a/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp b/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
index 80272213dd389..25f2e4d7c4de6 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
@@ -1675,6 +1675,7 @@ static unsigned SelectOpcodeFromVT(EVT VT, ArrayRef<unsigned> Opcodes) {
     return 0;
 
   EVT EltVT = VT.getVectorElementType();
+  unsigned Key = VT.getVectorMinNumElements();
   switch (Kind) {
   case SelectTypeKind::AnyType:
     break;
@@ -1688,14 +1689,17 @@ static unsigned SelectOpcodeFromVT(EVT VT, ArrayRef<unsigned> Opcodes) {
       return 0;
     break;
   case SelectTypeKind::FP:
-    if (EltVT != MVT::f16 && EltVT != MVT::f32 && EltVT != MVT::f64)
+    if (EltVT == MVT::bf16)
+      Key = 16;
+    else if (EltVT != MVT::bf16 && EltVT != MVT::f16 && EltVT != MVT::f32 &&
+             EltVT != MVT::f64)
       return 0;
     break;
   }
 
   unsigned Offset;
-  switch (VT.getVectorMinNumElements()) {
-  case 16: // 8-bit
+  switch (Key) {
+  case 16: // 8-bit or bf16
     Offset = 0;
     break;
   case 8: // 16-bit
@@ -5482,8 +5486,8 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmax_single_x2:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAX_VG2_2ZZ_H, AArch64::FMAX_VG2_2ZZ_S,
-               AArch64::FMAX_VG2_2ZZ_D}))
+              {AArch64::BFMAX_VG2_2ZZ_H, AArch64::FMAX_VG2_2ZZ_H,
+               AArch64::FMAX_VG2_2ZZ_S, AArch64::FMAX_VG2_2ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, false, Op);
       return;
     case Intrinsic::aarch64_sve_smax_single_x4:
@@ -5503,8 +5507,8 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmax_single_x4:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAX_VG4_4ZZ_H, AArch64::FMAX_VG4_4ZZ_S,
-               AArch64::FMAX_VG4_4ZZ_D}))
+              {AArch64::BFMAX_VG4_4ZZ_H, AArch64::FMAX_VG4_4ZZ_H,
+               AArch64::FMAX_VG4_4ZZ_S, AArch64::FMAX_VG4_4ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, false, Op);
       return;
     case Intrinsic::aarch64_sve_smin_single_x2:
@@ -5524,8 +5528,8 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmin_single_x2:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMIN_VG2_2ZZ_H, AArch64::FMIN_VG2_2ZZ_S,
-               AArch64::FMIN_VG2_2ZZ_D}))
+              {AArch64::BFMIN_VG2_2ZZ_H, AArch64::FMIN_VG2_2ZZ_H,
+               AArch64::FMIN_VG2_2ZZ_S, AArch64::FMIN_VG2_2ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, false, Op);
       return;
     case Intrinsic::aarch64_sve_smin_single_x4:
@@ -5545,8 +5549,8 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmin_single_x4:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMIN_VG4_4ZZ_H, AArch64::FMIN_VG4_4ZZ_S,
-               AArch64::FMIN_VG4_4ZZ_D}))
+              {AArch64::BFMIN_VG4_4ZZ_H, AArch64::FMIN_VG4_4ZZ_H,
+               AArch64::FMIN_VG4_4ZZ_S, AArch64::FMIN_VG4_4ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, false, Op);
       return;
     case Intrinsic::aarch64_sve_smax_x2:
@@ -5566,8 +5570,8 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmax_x2:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAX_VG2_2Z2Z_H, AArch64::FMAX_VG2_2Z2Z_S,
-               AArch64::FMAX_VG2_2Z2Z_D}))
+              {AArch64::BFMAX_VG2_2Z2Z_H, AArch64::FMAX_VG2_2Z2Z_H,
+               AArch64::FMAX_VG2_2Z2Z_S, AArch64::FMAX_VG2_2Z2Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, true, Op);
       return;
     case Intrinsic::aarch64_sve_smax_x4:
@@ -5587,8 +5591,8 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmax_x4:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAX_VG4_4Z4Z_H, AArch64::FMAX_VG4_4Z4Z_S,
-               AArch64::FMAX_VG4_4Z4Z_D}))
+              {AArch64::BFMAX_VG4_4Z2Z_H, AArch64::FMAX_VG4_4Z4Z_H,
+               AArch64::FMAX_VG4_4Z4Z_S, AArch64::FMAX_VG4_4Z4Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, true, Op);
       return;
     case Intrinsic::aarch64_sve_smin_x2:
@@ -5608,8 +5612,8 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmin_x2:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMIN_VG2_2Z2Z_H, AArch64::FMIN_VG2_2Z2Z_S,
-               AArch64::FMIN_VG2_2Z2Z_D}))
+              {AArch64::BFMIN_VG2_2Z2Z_H, AArch64::FMIN_VG2_2Z2Z_H,
+               AArch64::FMIN_VG2_2Z2Z_S, AArch64::FMIN_VG2_2Z2Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, true, Op);
       return;
     case Intrinsic::aarch64_sve_smin_x4:
@@ -5629,64 +5633,64 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
     case Intrinsic::aarch64_sve_fmin_x4:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMIN_VG4_4Z4Z_H, AArch64::FMIN_VG4_4Z4Z_S,
-               AArch64::FMIN_VG4_4Z4Z_D}))
+              {AArch64::BFMIN_VG4_4Z2Z_H, AArch64::FMIN_VG4_4Z4Z_H,
+               AArch64::FMIN_VG4_4Z4Z_S, AArch64::FMIN_VG4_4Z4Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, true, Op);
       return;
     case Intrinsic::aarch64_sve_fmaxnm_single_x2 :
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAXNM_VG2_2ZZ_H, AArch64::FMAXNM_VG2_2ZZ_S,
-               AArch64::FMAXNM_VG2_2ZZ_D}))
+              {AArch64::BFMAXNM_VG2_2ZZ_H, AArch64::FMAXNM_VG2_2ZZ_H,
+               AArch64::FMAXNM_VG2_2ZZ_S, AArch64::FMAXNM_VG2_2ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, false, Op);
       return;
     case Intrinsic::aarch64_sve_fmaxnm_single_x4 :
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAXNM_VG4_4ZZ_H, AArch64::FMAXNM_VG4_4ZZ_S,
-               AArch64::FMAXNM_VG4_4ZZ_D}))
+              {AArch64::BFMAXNM_VG4_4ZZ_H, AArch64::FMAXNM_VG4_4ZZ_H,
+               AArch64::FMAXNM_VG4_4ZZ_S, AArch64::FMAXNM_VG4_4ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, false, Op);
       return;
     case Intrinsic::aarch64_sve_fminnm_single_x2:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMINNM_VG2_2ZZ_H, AArch64::FMINNM_VG2_2ZZ_S,
-               AArch64::FMINNM_VG2_2ZZ_D}))
+              {AArch64::BFMINNM_VG2_2ZZ_H, AArch64::FMINNM_VG2_2ZZ_H,
+               AArch64::FMINNM_VG2_2ZZ_S, AArch64::FMINNM_VG2_2ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, false, Op);
       return;
     case Intrinsic::aarch64_sve_fminnm_single_x4:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMINNM_VG4_4ZZ_H, AArch64::FMINNM_VG4_4ZZ_S,
-               AArch64::FMINNM_VG4_4ZZ_D}))
+              {AArch64::BFMINNM_VG4_4ZZ_H, AArch64::FMINNM_VG4_4ZZ_H,
+               AArch64::FMINNM_VG4_4ZZ_S, AArch64::FMINNM_VG4_4ZZ_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, false, Op);
       return;
     case Intrinsic::aarch64_sve_fmaxnm_x2:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAXNM_VG2_2Z2Z_H, AArch64::FMAXNM_VG2_2Z2Z_S,
-               AArch64::FMAXNM_VG2_2Z2Z_D}))
+              {AArch64::BFMAXNM_VG2_2Z2Z_H, AArch64::FMAXNM_VG2_2Z2Z_H,
+               AArch64::FMAXNM_VG2_2Z2Z_S, AArch64::FMAXNM_VG2_2Z2Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, true, Op);
       return;
     case Intrinsic::aarch64_sve_fmaxnm_x4:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMAXNM_VG4_4Z4Z_H, AArch64::FMAXNM_VG4_4Z4Z_S,
-               AArch64::FMAXNM_VG4_4Z4Z_D}))
+              {AArch64::BFMAXNM_VG4_4Z2Z_H, AArch64::FMAXNM_VG4_4Z4Z_H,
+               AArch64::FMAXNM_VG4_4Z4Z_S, AArch64::FMAXNM_VG4_4Z4Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, true, Op);
       return;
     case Intrinsic::aarch64_sve_fminnm_x2:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMINNM_VG2_2Z2Z_H, AArch64::FMINNM_VG2_2Z2Z_S,
-               AArch64::FMINNM_VG2_2Z2Z_D}))
+              {AArch64::BFMINNM_VG2_2Z2Z_H, AArch64::FMINNM_VG2_2Z2Z_H,
+               AArch64::FMINNM_VG2_2Z2Z_S, AArch64::FMINNM_VG2_2Z2Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 2, true, Op);
       return;
     case Intrinsic::aarch64_sve_fminnm_x4:
       if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
               Node->getValueType(0),
-              {0, AArch64::FMINNM_VG4_4Z4Z_H, AArch64::FMINNM_VG4_4Z4Z_S,
-               AArch64::FMINNM_VG4_4Z4Z_D}))
+              {AArch64::BFMINNM_VG4_4Z2Z_H, AArch64::FMINNM_VG4_4Z4Z_H,
+               AArch64::FMINNM_VG4_4Z4Z_S, AArch64::FMINNM_VG4_4Z4Z_D}))
         SelectDestructiveMultiIntrinsic(Node, 4, true, Op);
       return;
     case Intrinsic::aarch64_sve_fcvtzs_x2:
diff --git a/llvm/test/CodeGen/AArch64/sme2-intrinsics-max.ll b/llvm/test/CodeGen/AArch64/sme2-intrinsics-max.ll
index 92e8877927ea5..79db677853cb5 100644
--- a/llvm/test/CodeGen/AArch64/sme2-intrinsics-max.ll
+++ b/llvm/test/CodeGen/AArch64/sme2-intrinsics-max.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2 -mattr=+b16b16 -verify-machineinstrs < %s | FileCheck %s
 
 ; SMAX (Single, x2)
 
@@ -109,6 +109,19 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64> } @multi_vec_max_single_x2_u64(<
   ret { <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+; BFMAX (Single, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_max_single_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_max_single_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    bfmax { z0.h, z1.h }, { z0.h, z1.h }, z2.h
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMAX (Single, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_max_single_x2_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm) {
@@ -298,6 +311,21 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2
   ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+; BFMAX (Single, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_max_single_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_max_single_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    bfmax { z0.h - z3.h }, { z0.h - z3.h }, z4.h
+; CHECK-NEXT:    ret
+  %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.fmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMAX (SINGLE, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_max_single_x4_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4, <vscale x 8 x half> %zm) {
@@ -478,6 +506,21 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64> } @multi_vec_max_multi_x2_u64(<v
   ret { <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+; BFMAX (Multi, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_max_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2) {
+; CHECK-LABEL: multi_vec_max_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    bfmax { z0.h, z1.h }, { z0.h, z1.h }, { z2.h, z3.h }
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmax.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMAX (Multi, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_max_multi_x2_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2) {
@@ -729,6 +772,25 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2
   ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+; BFMAX (Multi, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_max_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4) {
+; CHECK-LABEL: multi_vec_max_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z7 killed $z7 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z6 killed $z6 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z5 killed $z5 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z4 killed $z4 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    bfmax { z0.h - z3.h }, { z0.h - z3.h }, { z4.h - z7.h }
+; CHECK-NEXT:    ret
+  %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.fmax.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMAX (Multi, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_max_multi_x4_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4,
@@ -806,6 +868,19 @@ define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <v
   ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMAXNM (Single, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_maxnm_single_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_maxnm_single_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    bfmaxnm { z0.h, z1.h }, { z0.h, z1.h }, z2.h
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMAXNM (Single, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> }  @multi_vec_maxnm_single_x2_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm) {
@@ -847,6 +922,21 @@ define { <vscale x 2 x double>, <vscale x 2 x double> }  @multi_vec_maxnm_single
   ret { <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMAXNM (Single, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_maxnm_single_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_maxnm_single_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    bfmaxnm { z0.h - z3.h }, { z0.h - z3.h }, z4.h
+; CHECK-NEXT:    ret
+  %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.fmaxnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMAXNM (Single, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_maxnm_single_x4_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4, <vscale x 8 x half> %zm) {
@@ -903,6 +993,21 @@ define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <v
   ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMAXNM (Multi, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_maxnm_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2) {
+; CHECK-LABEL: multi_vec_maxnm_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    bfmaxnm { z0.h, z1.h }, { z0.h, z1.h }, { z2.h, z3.h }
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmaxnm.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMAXNM (Multi, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_maxnm_x2_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2) {
@@ -950,6 +1055,25 @@ define { <vscale x 2 x double>, <vscale x 2 x double> } @multi_vec_maxnm_x2_f64(
   ret { <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMAXNM (Multi, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_maxnm_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4) {
+; CHECK-LABEL: multi_vec_maxnm_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z7 killed $z7 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z6 killed $z6 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z5 killed $z5 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z4 killed $z4 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    bfmaxnm { z0.h - z3.h }, { z0.h - z3.h }, { z4.h - z7.h }
+; CHECK-NEXT:    ret
+  %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.fmaxnm.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMAXNM (Multi, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_maxnm_x4_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2, <vscale x 8 x half> %zm3, <vscale x 8 x half> %zm4) {
diff --git a/llvm/test/CodeGen/AArch64/sme2-intrinsics-min.ll b/llvm/test/CodeGen/AArch64/sme2-intrinsics-min.ll
index 363f9ba5d3530..e5c36d42fb135 100644
--- a/llvm/test/CodeGen/AArch64/sme2-intrinsics-min.ll
+++ b/llvm/test/CodeGen/AArch64/sme2-intrinsics-min.ll
@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme2 -mattr=+b16b16 -verify-machineinstrs < %s | FileCheck %s
 
 ; SMIN (Single, x2)
 
@@ -109,6 +109,19 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64> } @multi_vec_min_single_x2_u64(<
   ret { <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+; BFMIN (Single, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_min_single_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_min_single_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    bfmin { z0.h, z1.h }, { z0.h, z1.h }, z2.h
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.single.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMIN (Single, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_min_single_x2_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm) {
@@ -298,6 +311,21 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2
   ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+; BFMIN (Single, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_min_single_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_min_single_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    bfmin { z0.h - z3.h }, { z0.h - z3.h }, z4.h
+; CHECK-NEXT:    ret
+  %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.fmin.single.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMIN (SINGLE, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_min_single_x4_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4, <vscale x 8 x half> %zm) {
@@ -478,6 +506,21 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64> } @multi_vec_min_multi_x2_u64(<v
   ret { <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+; BFMIN (Multi, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_min_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2) {
+; CHECK-LABEL: multi_vec_min_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    bfmin { z0.h, z1.h }, { z0.h, z1.h }, { z2.h, z3.h }
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fmin.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMIN (Multi, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_min_multi_x2_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2) {
@@ -729,6 +772,26 @@ define { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2
   ret { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } %res
 }
 
+
+; BFMIN (Multi, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_min_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4) {
+; CHECK-LABEL: multi_vec_min_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z7 killed $z7 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z6 killed $z6 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z5 killed $z5 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z4 killed $z4 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    bfmin { z0.h - z3.h }, { z0.h - z3.h }, { z4.h - z7.h }
+; CHECK-NEXT:    ret
+  %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.fmin.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMIN (Multi, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_min_multi_x4_f16(<vscale x 8 x half> %unused, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4,
@@ -806,6 +869,19 @@ define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <v
   ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMINNM (Single, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_minnm_single_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_minnm_single_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    bfminnm { z0.h, z1.h }, { z0.h, z1.h }, z2.h
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.single.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMINNM (Single, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> }  @multi_vec_minnm_single_x2_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm) {
@@ -847,6 +923,21 @@ define { <vscale x 2 x double>, <vscale x 2 x double> }  @multi_vec_minnm_single
   ret { <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMINNM (Single, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_minnm_single_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm) {
+; CHECK-LABEL: multi_vec_minnm_single_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    bfminnm { z0.h - z3.h }, { z0.h - z3.h }, z4.h
+; CHECK-NEXT:    ret
+  %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.fminnm.single.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMINNM (Single, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_minnm_single_x4_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4, <vscale x 8 x half> %zm) {
@@ -903,6 +994,21 @@ define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <v
   ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMINNM (Multi, x2)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @multi_vec_minnm_x2_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2) {
+; CHECK-LABEL: multi_vec_minnm_x2_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z2_z3 def $z2_z3
+; CHECK-NEXT:    bfminnm { z0.h, z1.h }, { z0.h, z1.h }, { z2.h, z3.h }
+; CHECK-NEXT:    ret
+  %res = call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.fminnm.x2.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } %res
+}
+
 ; FMINNM (Multi, x2)
 
 define { <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_minnm_x2_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2) {
@@ -950,6 +1056,25 @@ define { <vscale x 2 x double>, <vscale x 2 x double> } @multi_vec_minnm_x2_f64(
   ret { <vscale x 2 x double>, <vscale x 2 x double> } %res
 }
 
+; BFMINNM (Multi, x4)
+
+define { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } @multi_vec_minnm_x4_bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4) {
+; CHECK-LABEL: multi_vec_minnm_x4_bf16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $z3 killed $z3 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z7 killed $z7 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z2 killed $z2 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z6 killed $z6 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z1 killed $z1 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z5 killed $z5 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    // kill: def $z0 killed $z0 killed $z0_z1_z2_z3 def $z0_z1_z2_z3
+; CHECK-NEXT:    // kill: def $z4 killed $z4 killed $z4_z5_z6_z7 def $z4_z5_z6_z7
+; CHECK-NEXT:    bfminnm { z0.h - z3.h }, { z0.h - z3.h }, { z4.h - z7.h }
+; CHECK-NEXT:    ret
+  %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.fminnm.x4.nxv8bf16(<vscale x 8 x bfloat> %zdn1, <vscale x 8 x bfloat> %zdn2, <vscale x 8 x bfloat> %zdn3, <vscale x 8 x bfloat> %zdn4, <vscale x 8 x bfloat> %zm1, <vscale x 8 x bfloat> %zm2, <vscale x 8 x bfloat> %zm3, <vscale x 8 x bfloat> %zm4)
+  ret { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>  } %res
+}
+
 ; FMINNM (Multi, x4)
 
 define { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @multi_vec_minnm_x4_f16(<vscale x 8 x half> %dummy, <vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zdn3, <vscale x 8 x half> %zdn4, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2, <vscale x 8 x half> %zm3, <vscale x 8 x half> %zm4) {