https://github.com/jthackray updated 
https://github.com/llvm/llvm-project/pull/172025

>From 6009aba4e154ffa8ecda6e9b7b1ab750f5947381 Mon Sep 17 00:00:00 2001
From: Jonathan Thackray <[email protected]>
Date: Mon, 13 Oct 2025 16:00:26 +0100
Subject: [PATCH 1/2] [AArch64][llvm] Add intrinsics for SVE BFSCALE

Add AArch64 intrinsics for BFloat16 floating-point adjust exponent vectors:

```c
  svbfloat16_t svscale[_bf16]_m (svbool_t pg, svbfloat16_t zdn, svint16_t zm);
  svbfloat16_t svscale[_bf16]_x (svbool_t pg, svbfloat16_t zdn, svint16_t zm);
  svbfloat16_t svscale[_bf16]_z (svbool_t pg, svbfloat16_t zdn, svint16_t zm);
  svbfloat16_t svscale[_n_bf16]_m (svbool_t pg, svbfloat16_t zdn, int16_t zm);
  svbfloat16_t svscale[_n_bf16]_x (svbool_t pg, svbfloat16_t zdn, int16_t zm);
  svbfloat16_t svscale[_n_bf16]_z (svbool_t pg, svbfloat16_t zdn, int16_t zm);
```
---
 clang/include/clang/Basic/arm_sve.td          |  10 ++
 .../AArch64/sve-intrinsics/acle_sve_bfscale.c | 140 ++++++++++++++++++
 llvm/lib/Target/AArch64/AArch64InstrInfo.td   |   2 +-
 .../lib/Target/AArch64/AArch64SVEInstrInfo.td |   2 +-
 llvm/lib/Target/AArch64/SVEInstrFormats.td    |  10 +-
 5 files changed, 159 insertions(+), 5 deletions(-)
 create mode 100644 clang/test/CodeGen/AArch64/sve-intrinsics/acle_sve_bfscale.c

diff --git a/clang/include/clang/Basic/arm_sve.td 
b/clang/include/clang/Basic/arm_sve.td
index a4a0ee7ed5f5f..14b1ec29cd596 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -796,6 +796,16 @@ def SVSCALE_N_M : SInst<"svscale[_n_{d}]", "dPdK", "hfd", 
MergeOp1,  "aarch64_sv
 def SVSCALE_N_X : SInst<"svscale[_n_{d}]", "dPdK", "hfd", MergeAny,  
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
 def SVSCALE_N_Z : SInst<"svscale[_n_{d}]", "dPdK", "hfd", MergeZero, 
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
 
+let SVETargetGuard = "sve-bfscale", SMETargetGuard = "sve-bfscale,sme2" in {
+  def SVSCALE_BF_M : SInst<"svscale[_{d}]",   "dPdx", "b", MergeOp1,  
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
+  def SVSCALE_BF_X : SInst<"svscale[_{d}]",   "dPdx", "b", MergeAny,  
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
+  def SVSCALE_BF_Z : SInst<"svscale[_{d}]",   "dPdx", "b", MergeZero, 
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
+
+  def SVSCALE_BF_N_M : SInst<"svscale[_n_{d}]", "dPdK", "b", MergeOp1,  
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
+  def SVSCALE_BF_N_X : SInst<"svscale[_n_{d}]", "dPdK", "b", MergeAny,  
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
+  def SVSCALE_BF_N_Z : SInst<"svscale[_n_{d}]", "dPdK", "b", MergeZero, 
"aarch64_sve_fscale", [VerifyRuntimeMode]>;
+}
+
 defm SVMAD_F  : SInstZPZZZ<"svmad",  "hfd", "aarch64_sve_fmad",  
"aarch64_sve_fmla_u",  [VerifyRuntimeMode, ReverseMergeAnyAccOp]>;
 defm SVMLA_F  : SInstZPZZZ<"svmla",  "hfd", "aarch64_sve_fmla",  
"aarch64_sve_fmla_u", [VerifyRuntimeMode]>;
 defm SVMLS_F  : SInstZPZZZ<"svmls",  "hfd", "aarch64_sve_fmls",  
"aarch64_sve_fmls_u", [VerifyRuntimeMode]>;
diff --git a/clang/test/CodeGen/AArch64/sve-intrinsics/acle_sve_bfscale.c 
b/clang/test/CodeGen/AArch64/sve-intrinsics/acle_sve_bfscale.c
new file mode 100644
index 0000000000000..f3ca61ce34a13
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/sve-intrinsics/acle_sve_bfscale.c
@@ -0,0 +1,140 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+// REQUIRES: aarch64-registered-target
+// RUN: %clang_cc1 -triple aarch64 -target-feature +sve -target-feature 
+sve-bfscale -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S 
-passes=mem2reg,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64 -target-feature +sve -target-feature 
+sve-bfscale -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt 
-S -passes=mem2reg,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve 
-target-feature +sve-bfscale -disable-O0-optnone -Werror -Wall -emit-llvm -o - 
%s | opt -S -passes=mem2reg,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve 
-target-feature +sve-bfscale -disable-O0-optnone -Werror -Wall -emit-llvm -o - 
-x c++ %s | opt -S -passes=mem2reg,tailcallelim | FileCheck %s 
-check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64 -target-feature +sve -target-feature 
+sve-bfscale -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+
+#include <arm_sve.h>
+
+#if defined __ARM_FEATURE_SME
+#define MODE_ATTR __arm_streaming
+#else
+#define MODE_ATTR
+#endif
+
+#ifdef SVE_OVERLOADED_FORMS
+// A simple used,unused... macro, long enough to represent any SVE builtin.
+#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED) A1##A3
+#else
+#define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
+#endif
+
+// CHECK-LABEL: @test_svscale_bf16_z(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], <vscale x 
8 x bfloat> [[OP1:%.*]], <vscale x 8 x bfloat> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+//
+// CPP-CHECK-LABEL: 
@_Z19test_svscale_bf16_zu10__SVBool_tu14__SVBfloat16_tu11__SVInt16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], 
<vscale x 8 x bfloat> [[OP1:%.*]], <vscale x 8 x bfloat> zeroinitializer
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+//
+svbfloat16_t test_svscale_bf16_z(svbool_t pg, svbfloat16_t op1, svint16_t op2) 
MODE_ATTR
+{
+  return SVE_ACLE_FUNC(svscale,_bf16,_z,)(pg, op1, op2);
+}
+
+// CHECK-LABEL: @test_svscale_bf16_m(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+// CPP-CHECK-LABEL: 
@_Z19test_svscale_bf16_mu10__SVBool_tu14__SVBfloat16_tu11__SVInt16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+svbfloat16_t test_svscale_bf16_m(svbool_t pg, svbfloat16_t op1, svint16_t op2) 
MODE_ATTR
+{
+  return SVE_ACLE_FUNC(svscale,_bf16,_m,)(pg, op1, op2);
+}
+
+// CHECK-LABEL: @test_svscale_bf16_x(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+// CPP-CHECK-LABEL: 
@_Z19test_svscale_bf16_xu10__SVBool_tu14__SVBfloat16_tu11__SVInt16_t(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+svbfloat16_t test_svscale_bf16_x(svbool_t pg, svbfloat16_t op1, svint16_t op2) 
MODE_ATTR
+{
+  return SVE_ACLE_FUNC(svscale,_bf16,_x,)(pg, op1, op2);
+}
+
+// CHECK-LABEL: @test_svscale_n_bf16_z(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i16> 
poison, i16 [[OP2:%.*]], i64 0
+// CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i16> 
[[DOTSPLATINSERT]], <vscale x 8 x i16> poison, <vscale x 8 x i32> 
zeroinitializer
+// CHECK-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], <vscale x 
8 x bfloat> [[OP1:%.*]], <vscale x 8 x bfloat> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[TMP1]], <vscale x 8 x i16> [[DOTSPLAT]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svscale_n_bf16_zu10__SVBool_tu14__SVBfloat16_ts(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x 
i16> poison, i16 [[OP2:%.*]], i64 0
+// CPP-CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i16> 
[[DOTSPLATINSERT]], <vscale x 8 x i16> poison, <vscale x 8 x i32> 
zeroinitializer
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], 
<vscale x 8 x bfloat> [[OP1:%.*]], <vscale x 8 x bfloat> zeroinitializer
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[TMP1]], <vscale x 8 x i16> [[DOTSPLAT]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP2]]
+//
+svbfloat16_t test_svscale_n_bf16_z(svbool_t pg, svbfloat16_t op1, int16_t op2) 
MODE_ATTR
+{
+  return SVE_ACLE_FUNC(svscale,_n_bf16,_z,)(pg, op1, op2);
+}
+
+// CHECK-LABEL: @test_svscale_n_bf16_m(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i16> 
poison, i16 [[OP2:%.*]], i64 0
+// CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i16> 
[[DOTSPLATINSERT]], <vscale x 8 x i16> poison, <vscale x 8 x i32> 
zeroinitializer
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[DOTSPLAT]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svscale_n_bf16_mu10__SVBool_tu14__SVBfloat16_ts(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x 
i16> poison, i16 [[OP2:%.*]], i64 0
+// CPP-CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i16> 
[[DOTSPLATINSERT]], <vscale x 8 x i16> poison, <vscale x 8 x i32> 
zeroinitializer
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[DOTSPLAT]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+svbfloat16_t test_svscale_n_bf16_m(svbool_t pg, svbfloat16_t op1, int16_t op2) 
MODE_ATTR
+{
+  return SVE_ACLE_FUNC(svscale,_n_bf16,_m,)(pg, op1, op2);
+}
+
+// CHECK-LABEL: @test_svscale_n_bf16_x(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i16> 
poison, i16 [[OP2:%.*]], i64 0
+// CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i16> 
[[DOTSPLATINSERT]], <vscale x 8 x i16> poison, <vscale x 8 x i32> 
zeroinitializer
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[DOTSPLAT]])
+// CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+// CPP-CHECK-LABEL: @_Z21test_svscale_n_bf16_xu10__SVBool_tu14__SVBfloat16_ts(
+// CPP-CHECK-NEXT:  entry:
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> 
@llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG:%.*]])
+// CPP-CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x 
i16> poison, i16 [[OP2:%.*]], i64 0
+// CPP-CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i16> 
[[DOTSPLATINSERT]], <vscale x 8 x i16> poison, <vscale x 8 x i32> 
zeroinitializer
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> 
@llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x 
bfloat> [[OP1:%.*]], <vscale x 8 x i16> [[DOTSPLAT]])
+// CPP-CHECK-NEXT:    ret <vscale x 8 x bfloat> [[TMP1]]
+//
+svbfloat16_t test_svscale_n_bf16_x(svbool_t pg, svbfloat16_t op1, int16_t op2) 
MODE_ATTR
+{
+  return SVE_ACLE_FUNC(svscale,_n_bf16,_x,)(pg, op1, op2);
+}
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td 
b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 835fd0575b5e2..8de9b92eedf47 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -211,7 +211,7 @@ def HasSME2p2        : Predicate<"Subtarget->isStreaming() 
&& Subtarget->hasSME2
                                  AssemblerPredicateWithAll<(all_of 
FeatureSME2p2), "sme2p2">;
 def HasSVEAES2       : Predicate<"Subtarget->hasSVEAES2()">,
                                  AssemblerPredicateWithAll<(all_of 
FeatureSVEAES2), "sve-aes2">;
-def HasSVEBFSCALE    : Predicate<"Subtarget->isSVEorStreamingSVEAvailable() && 
Subtarget->hasSVEBFSCALE()">,
+def HasSVEBFSCALE    : Predicate<"Subtarget->isSVEorStreamingSVEAvailable() && 
Subtarget->hasSVE_BFSCALE()">,
                                  AssemblerPredicateWithAll<(all_of 
FeatureSVEBFSCALE), "sve-bfscale">;
 def HasSVE_F16F32MM  : Predicate<"Subtarget->isSVEAvailable() && 
Subtarget->hasSVE_F16F32MM()">,
                                  AssemblerPredicateWithAll<(all_of 
FeatureSVE_F16F32MM), "sve-f16f32mm">;
diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td 
b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 33c71bf976672..2766189be6137 100644
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -4515,7 +4515,7 @@ defm BFMAX_ZPZZ   : 
sve_fp_2op_p_zds_zeroing_bfloat<int_aarch64_sve_fmax>;
 } // HasSVEB16B16, HasNonStreamingSVE_or_SME2, UseExperimentalZeroingPseudos
 
 let Predicates = [HasSVEBFSCALE] in {
-  def BFSCALE_ZPZZ : sve_fp_2op_p_zds_bfscale<0b1001, "bfscale", 
DestructiveBinary>;
+  defm BFSCALE_ZPZZ : sve_fp_2op_p_zds_bfscale<0b1001, "bfscale", 
int_aarch64_sve_fscale, DestructiveBinary>;
 } // HasSVEBFSCALE
 
//===----------------------------------------------------------------------===//
 // SME2.1 or SVE2.1 instructions
diff --git a/llvm/lib/Target/AArch64/SVEInstrFormats.td 
b/llvm/lib/Target/AArch64/SVEInstrFormats.td
index 970558c8db52e..b1aae8d930c21 100644
--- a/llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -2332,9 +2332,13 @@ multiclass sve_fp_2op_p_zds_bfloat<bits<4> opc, string 
asm, string Ps,
   def : SVE_3_Op_Pat<nxv8bf16, op, nxv8i1, nxv8bf16, nxv8bf16, 
!cast<Instruction>(NAME)>;
 }
 
-class  sve_fp_2op_p_zds_bfscale<bits<4> opc, string asm,  
DestructiveInstTypeEnum flags>
-: sve_fp_2op_p_zds<0b00, opc, asm, ZPR16>{
-  let DestructiveInstType = flags;
+multiclass sve_fp_2op_p_zds_bfscale<bits<4> opc, string asm, SDPatternOperator 
op,
+                                    DestructiveInstTypeEnum flags> {
+  let DestructiveInstType = flags in {
+  def _H : sve_fp_2op_p_zds<0b00, opc, asm, ZPR16>;
+  }
+
+  def : SVE_3_Op_Pat<nxv8bf16, op, nxv8i1, nxv8bf16, nxv8i16, 
!cast<Instruction>(NAME # _H)>;
 }
 
 multiclass sve_fp_2op_p_zds_zeroing_hsd<SDPatternOperator op> {

>From fabadb6d1c5d370dabf0adabe3b482b427e4744f Mon Sep 17 00:00:00 2001
From: Jonathan Thackray <[email protected]>
Date: Fri, 12 Dec 2025 16:07:18 +0000
Subject: [PATCH 2/2] fixup! [AArch64][llvm] Add intrinsics for SVE BFSCALE

Add lowering testcase
---
 llvm/test/CodeGen/AArch64/sve-intrinsics-fp-arith.ll | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/llvm/test/CodeGen/AArch64/sve-intrinsics-fp-arith.ll 
b/llvm/test/CodeGen/AArch64/sve-intrinsics-fp-arith.ll
index 5491dc274cd1f..e2420cc9d6863 100644
--- a/llvm/test/CodeGen/AArch64/sve-intrinsics-fp-arith.ll
+++ b/llvm/test/CodeGen/AArch64/sve-intrinsics-fp-arith.ll
@@ -1412,6 +1412,17 @@ define <vscale x 2 x double> @fscale_d(<vscale x 2 x i1> 
%pg, <vscale x 2 x doub
   ret <vscale x 2 x double> %out
 }
 
+define <vscale x 8 x half> @bfscale_h(<vscale x 8 x i1> %pg, <vscale x 8 x 
half> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: bfscale_h:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    fscale z0.h, p0/m, z0.h, z1.h
+; CHECK-NEXT:    ret
+  %out = call <vscale x 8 x half> @llvm.aarch64.sve.fscale.nxv8bf16(<vscale x 
8 x i1> %pg,
+                                                                    <vscale x 
8 x half> %a,
+                                                                    <vscale x 
8 x i16> %b)
+  ret <vscale x 8 x half> %out
+}
+
 ;
 ; FSQRT
 ;

_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to