[clang] [llvm] [AArch64] Implement intrinsics for SME2 FAMIN/FAMAX (PR #99063)
Momchil Velikov via cfe-commits
cfe-commits at lists.llvm.org
Tue Jul 16 09:56:28 PDT 2024
https://github.com/momchil-velikov created https://github.com/llvm/llvm-project/pull/99063
This patch implements these intrinsics:
``` c
// Variants are also available for:
// [_f32_x2], [_f64_x2],
// [_f16_x4], [_f32_x4], [_f64_x4]
svfloat16x2_t svamax[_f16_x2](svfloat16x2 zd, svfloat16x2_t zm) __arm_streaming;
svfloat16x2_t svamin[_f16_x2](svfloat16x2 zd, svfloat16x2_t zm) __arm_streaming;
```
(cf. https://github.com/ARM-software/acle/pull/324)
>From db16c5c8fb1e8b272d49149016ee3b7ce43c5672 Mon Sep 17 00:00:00 2001
From: Momchil Velikov <momchil.velikov at arm.com>
Date: Tue, 16 Jul 2024 16:49:04 +0100
Subject: [PATCH] [AArch64] Implement intrinsics for SME2 FAMIN/FAMAX
This patch implements these intrinsics:
``` c
// Variants are also available for:
// [_f32_x2], [_f64_x2],
// [_f16_x4], [_f32_x4], [_f64_x4]
svfloat16x2_t svamax[_f16_x2](svfloat16x2 zd, svfloat16x2_t zm) __arm_streaming;
svfloat16x2_t svamin[_f16_x2](svfloat16x2 zd, svfloat16x2_t zm) __arm_streaming;
```
(cf. https://github.com/ARM-software/acle/pull/324)
Co-authored-by: Caroline Concatto <caroline.concatto at arm.com>
---
clang/include/clang/Basic/arm_sve.td | 7 +
.../acle_sme2_faminmax.c | 476 ++++++++++++++++++
llvm/include/llvm/IR/IntrinsicsAArch64.td | 9 +
.../Target/AArch64/AArch64ISelDAGToDAG.cpp | 28 ++
.../AArch64/sme2-intrinsics-faminmax.ll | 241 +++++++++
5 files changed, 761 insertions(+)
create mode 100644 clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c
create mode 100644 llvm/test/CodeGen/AArch64/sme2-intrinsics-faminmax.ll
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 94c093d891156..e415589ee25e5 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -2219,6 +2219,13 @@ let SVETargetGuard = InvalidMode, SMETargetGuard = "sme2" in {
def SVSQDMULH_X4 : SInst<"svqdmulh[_{d}_x4]", "444", "csil", MergeNone, "aarch64_sve_sqdmulh_vgx4", [IsStreaming], []>;
}
+let SVETargetGuard = InvalidMode, SMETargetGuard = "sme2,faminmax" in {
+ def FAMIN_X2 : Inst<"svamin[_{d}_x2]", "222", "hfd", MergeNone, "aarch64_sme_famin_x2", [IsStreaming], []>;
+ def FAMAX_X2 : Inst<"svamax[_{d}_x2]", "222", "hfd", MergeNone, "aarch64_sme_famax_x2", [IsStreaming], []>;
+ def FAMIN_X4 : Inst<"svamin[_{d}_x4]", "444", "hfd", MergeNone, "aarch64_sme_famin_x4", [IsStreaming], []>;
+ def FAMAX_X4 : Inst<"svamax[_{d}_x4]", "444", "hfd", MergeNone, "aarch64_sme_famax_x4", [IsStreaming], []>;
+}
+
let SVETargetGuard = "sve2p1", SMETargetGuard = "sme2" in {
def REINTERPRET_SVBOOL_TO_SVCOUNT : Inst<"svreinterpret[_c]", "}P", "Pc", MergeNone, "", [VerifyRuntimeMode], []>;
def REINTERPRET_SVCOUNT_TO_SVBOOL : Inst<"svreinterpret[_b]", "P}", "Pc", MergeNone, "", [VerifyRuntimeMode], []>;
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c
new file mode 100644
index 0000000000000..5d026f8cde5e0
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_faminmax.c
@@ -0,0 +1,476 @@
+// 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 +sme -target-feature +sme2 -target-feature +faminmax -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 +sme -target-feature +sme2 -target-feature +faminmax -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 +sme -target-feature +sme2 -target-feature +faminmax -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 +sme -target-feature +sme2 -target-feature +faminmax -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 +sme -target-feature +sme2 -target-feature +faminmax -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+#include <arm_sme.h>
+
+#ifdef SVE_OVERLOADED_FORMS
+// A simple used,unused... macro, long enough to represent any SVE builtin.
+#define SVE_ACLE_FUNC(A1,A2_UNUSED) A1
+#else
+#define SVE_ACLE_FUNC(A1,A2) A1##A2
+#endif
+
+
+// Multi, x2
+
+// CHECK-LABEL: @test_svamax_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)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
+// CHECK-NEXT: ret <vscale x 16 x half> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamax_f16_x213svfloat16x2_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT: ret <vscale x 16 x half> [[TMP8]]
+//
+svfloat16x2_t test_svamax_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamax,_f16_x2)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamax_f32_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
+// CHECK-NEXT: ret <vscale x 8 x float> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamax_f32_x213svfloat32x2_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
+// CPP-CHECK-NEXT: ret <vscale x 8 x float> [[TMP8]]
+//
+svfloat32x2_t test_svamax_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamax,_f32_x2)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamax_f64_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
+// CHECK-NEXT: ret <vscale x 4 x double> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamax_f64_x213svfloat64x2_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
+// CPP-CHECK-NEXT: ret <vscale x 4 x double> [[TMP8]]
+//
+svfloat64x2_t test_svamax_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamax,_f64_x2)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamin_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)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
+// CHECK-NEXT: ret <vscale x 16 x half> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamin_f16_x213svfloat16x2_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZM]], i64 8)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT: ret <vscale x 16 x half> [[TMP8]]
+//
+svfloat16x2_t test_svamin_f16_x2(svfloat16x2_t zdn, svfloat16x2_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamin,_f16_x2)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamin_f32_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
+// CHECK-NEXT: ret <vscale x 8 x float> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamin_f32_x213svfloat32x2_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZDN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[ZM]], i64 4)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
+// CPP-CHECK-NEXT: ret <vscale x 8 x float> [[TMP8]]
+//
+svfloat32x2_t test_svamin_f32_x2(svfloat32x2_t zdn, svfloat32x2_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamin,_f32_x2)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamin_f64_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
+// CHECK-NEXT: ret <vscale x 4 x double> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamin_f64_x213svfloat64x2_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZDN]], i64 2)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[ZM]], i64 2)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
+// CPP-CHECK-NEXT: ret <vscale x 4 x double> [[TMP8]]
+//
+svfloat64x2_t test_svamin_f64_x2(svfloat64x2_t zdn, svfloat64x2_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamin,_f64_x2)(zdn, zm);
+}
+
+// Multi, x4
+
+// CHECK-LABEL: @test_svamax_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)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
+// CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
+// CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
+// CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
+// CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
+// CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
+// CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
+// CHECK-NEXT: ret <vscale x 32 x half> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamax_f16_x413svfloat16x4_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CPP-CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
+// CPP-CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
+// CPP-CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
+// CPP-CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
+// CPP-CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
+// CPP-CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
+// CPP-CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
+// CPP-CHECK-NEXT: ret <vscale x 32 x half> [[TMP16]]
+//
+svfloat16x4_t test_svamax_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamax,_f16_x4)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamax_f32_x4(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
+// CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
+// CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
+// CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
+// CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
+// CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
+// CHECK-NEXT: ret <vscale x 16 x float> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamax_f32_x413svfloat32x4_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CPP-CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
+// CPP-CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
+// CPP-CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
+// CPP-CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
+// CPP-CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
+// CPP-CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
+// CPP-CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
+// CPP-CHECK-NEXT: ret <vscale x 16 x float> [[TMP16]]
+//
+svfloat32x4_t test_svamax_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamax,_f32_x4)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamax_f64_x4(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
+// CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
+// CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
+// CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
+// CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
+// CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
+// CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
+// CHECK-NEXT: ret <vscale x 8 x double> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamax_f64_x413svfloat64x4_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CPP-CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
+// CPP-CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
+// CPP-CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
+// CPP-CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
+// CPP-CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
+// CPP-CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
+// CPP-CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
+// CPP-CHECK-NEXT: ret <vscale x 8 x double> [[TMP16]]
+//
+svfloat64x4_t test_svamax_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamax,_f64_x4)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamin_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)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
+// CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
+// CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
+// CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
+// CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
+// CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
+// CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
+// CHECK-NEXT: ret <vscale x 32 x half> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamin_f16_x413svfloat16x4_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 8)
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 16)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZM]], i64 24)
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]])
+// CPP-CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
+// CPP-CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
+// CPP-CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
+// CPP-CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
+// CPP-CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
+// CPP-CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
+// CPP-CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
+// CPP-CHECK-NEXT: ret <vscale x 32 x half> [[TMP16]]
+//
+svfloat16x4_t test_svamin_f16_x4(svfloat16x4_t zdn, svfloat16x4_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamin,_f16_x4)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamin_f32_x4(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
+// CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
+// CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
+// CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
+// CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
+// CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
+// CHECK-NEXT: ret <vscale x 16 x float> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamin_f32_x413svfloat32x4_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZDN]], i64 12)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 4)
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 8)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[ZM]], i64 12)
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]])
+// CPP-CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
+// CPP-CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
+// CPP-CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
+// CPP-CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
+// CPP-CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
+// CPP-CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
+// CPP-CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
+// CPP-CHECK-NEXT: ret <vscale x 16 x float> [[TMP16]]
+//
+svfloat32x4_t test_svamin_f32_x4(svfloat32x4_t zdn, svfloat32x4_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamin,_f32_x4)(zdn, zm);
+}
+
+// CHECK-LABEL: @test_svamin_f64_x4(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
+// CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
+// CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
+// CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
+// CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
+// CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
+// CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
+// CHECK-NEXT: ret <vscale x 8 x double> [[TMP16]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svamin_f64_x413svfloat64x4_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 2)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 4)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZDN]], i64 6)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 2)
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 4)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[ZM]], i64 6)
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]])
+// CPP-CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
+// CPP-CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
+// CPP-CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
+// CPP-CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
+// CPP-CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
+// CPP-CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
+// CPP-CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
+// CPP-CHECK-NEXT: [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
+// CPP-CHECK-NEXT: ret <vscale x 8 x double> [[TMP16]]
+//
+svfloat64x4_t test_svamin_f64_x4(svfloat64x4_t zdn, svfloat64x4_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svamin,_f64_x4)(zdn, zm);
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsAArch64.td b/llvm/include/llvm/IR/IntrinsicsAArch64.td
index 3735bf5222fce..d2922045b585f 100644
--- a/llvm/include/llvm/IR/IntrinsicsAArch64.td
+++ b/llvm/include/llvm/IR/IntrinsicsAArch64.td
@@ -3466,6 +3466,15 @@ let TargetPrefix = "aarch64" in {
def int_aarch64_sve_ # instr # _x4 : SME2_VG4_Multi_Multi_Intrinsic;
}
+ //
+ // Multi-vector floating point absolute min/max number
+ //
+
+ foreach instr = ["famax", "famin"] in {
+ def int_aarch64_sme_ # instr # _x2 : SME2_VG2_Multi_Multi_Intrinsic;
+ def int_aarch64_sme_ # instr # _x4 : SME2_VG4_Multi_Multi_Intrinsic;
+ }
+
//
// Multi-vector vertical dot-products
//
diff --git a/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp b/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
index b9c57d1975b6f..745a132df14fe 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
@@ -5733,6 +5733,34 @@ void AArch64DAGToDAGISel::Select(SDNode *Node) {
AArch64::FMAX_VG4_4Z4Z_S, AArch64::FMAX_VG4_4Z4Z_D}))
SelectDestructiveMultiIntrinsic(Node, 4, true, Op);
return;
+ case Intrinsic::aarch64_sme_famax_x2:
+ if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
+ Node->getValueType(0),
+ {0, AArch64::FAMAX_2Z2Z_H, AArch64::FAMAX_2Z2Z_S,
+ AArch64::FAMAX_2Z2Z_D}))
+ SelectDestructiveMultiIntrinsic(Node, 2, true, Op);
+ return;
+ case Intrinsic::aarch64_sme_famax_x4:
+ if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
+ Node->getValueType(0),
+ {0, AArch64::FAMAX_4Z4Z_H, AArch64::FAMAX_4Z4Z_S,
+ AArch64::FAMAX_4Z4Z_D}))
+ SelectDestructiveMultiIntrinsic(Node, 4, true, Op);
+ return;
+ case Intrinsic::aarch64_sme_famin_x2:
+ if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
+ Node->getValueType(0),
+ {0, AArch64::FAMIN_2Z2Z_H, AArch64::FAMIN_2Z2Z_S,
+ AArch64::FAMIN_2Z2Z_D}))
+ SelectDestructiveMultiIntrinsic(Node, 2, true, Op);
+ return;
+ case Intrinsic::aarch64_sme_famin_x4:
+ if (auto Op = SelectOpcodeFromVT<SelectTypeKind::FP>(
+ Node->getValueType(0),
+ {0, AArch64::FAMIN_4Z4Z_H, AArch64::FAMIN_4Z4Z_S,
+ AArch64::FAMIN_4Z4Z_D}))
+ SelectDestructiveMultiIntrinsic(Node, 4, true, Op);
+ return;
case Intrinsic::aarch64_sve_smin_x2:
if (auto Op = SelectOpcodeFromVT<SelectTypeKind::Int>(
Node->getValueType(0),
diff --git a/llvm/test/CodeGen/AArch64/sme2-intrinsics-faminmax.ll b/llvm/test/CodeGen/AArch64/sme2-intrinsics-faminmax.ll
new file mode 100644
index 0000000000000..ecfbb47ba5571
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sme2-intrinsics-faminmax.ll
@@ -0,0 +1,241 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme -mattr=+sme2 -mattr=+faminmax -force-streaming -verify-machineinstrs < %s | FileCheck %s
+
+; FAMAX (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) {
+; CHECK-LABEL: multi_vec_max_multi_x2_f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z7.d, z4.d
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z6.d, z3.d
+; CHECK-NEXT: mov z4.d, z1.d
+; CHECK-NEXT: famax { z4.h, z5.h }, { z4.h, z5.h }, { z6.h, z7.h }
+; CHECK-NEXT: mov z0.d, z4.d
+; CHECK-NEXT: mov z1.d, z5.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x2.nxv8f16(<vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2)
+ ret { <vscale x 8 x half>, <vscale x 8 x half> } %res
+}
+
+define { <vscale x 4 x float>, <vscale x 4 x float> } @multi_vec_max_multi_x2_f32(<vscale x 4 x float> %unused, <vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2) {
+; CHECK-LABEL: multi_vec_max_multi_x2_f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z7.d, z4.d
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z6.d, z3.d
+; CHECK-NEXT: mov z4.d, z1.d
+; CHECK-NEXT: famax { z4.s, z5.s }, { z4.s, z5.s }, { z6.s, z7.s }
+; CHECK-NEXT: mov z0.d, z4.d
+; CHECK-NEXT: mov z1.d, z5.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x2.nxv4f32(<vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2)
+ ret { <vscale x 4 x float>, <vscale x 4 x float> } %res
+}
+
+define { <vscale x 2 x double>, <vscale x 2 x double> } @multi_vec_max_multi_x2_f64(<vscale x 2 x double> %unused, <vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2) {
+; CHECK-LABEL: multi_vec_max_multi_x2_f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z7.d, z4.d
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z6.d, z3.d
+; CHECK-NEXT: mov z4.d, z1.d
+; CHECK-NEXT: famax { z4.d, z5.d }, { z4.d, z5.d }, { z6.d, z7.d }
+; CHECK-NEXT: mov z0.d, z4.d
+; CHECK-NEXT: mov z1.d, z5.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famax.x2.nxv2f64(<vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2)
+ ret { <vscale x 2 x double>, <vscale x 2 x double> } %res
+}
+
+; FAMAX (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, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2, <vscale x 8 x half> %zm3, <vscale x 8 x half> %zm4) {
+; CHECK-LABEL: multi_vec_max_multi_x4_f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z30.d, z7.d
+; CHECK-NEXT: mov z27.d, z4.d
+; CHECK-NEXT: ptrue p0.h
+; CHECK-NEXT: mov z29.d, z6.d
+; CHECK-NEXT: mov z26.d, z3.d
+; CHECK-NEXT: mov z28.d, z5.d
+; CHECK-NEXT: mov z25.d, z2.d
+; CHECK-NEXT: ld1h { z31.h }, p0/z, [x0]
+; CHECK-NEXT: mov z24.d, z1.d
+; CHECK-NEXT: famax { z24.h - z27.h }, { z24.h - z27.h }, { z28.h - z31.h }
+; CHECK-NEXT: mov z0.d, z24.d
+; CHECK-NEXT: mov z1.d, z25.d
+; CHECK-NEXT: mov z2.d, z26.d
+; CHECK-NEXT: mov z3.d, z27.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famax.x4.nxv8f16(<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)
+ ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } %res
+}
+
+define { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @multi_vec_max_multi_x4_f32(<vscale x 4 x float> %unused, <vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zdn3, <vscale x 4 x float> %zdn4, <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2, <vscale x 4 x float> %zm3, <vscale x 4 x float> %zm4) {
+; CHECK-LABEL: multi_vec_max_multi_x4_f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z30.d, z7.d
+; CHECK-NEXT: mov z27.d, z4.d
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: mov z29.d, z6.d
+; CHECK-NEXT: mov z26.d, z3.d
+; CHECK-NEXT: mov z28.d, z5.d
+; CHECK-NEXT: mov z25.d, z2.d
+; CHECK-NEXT: ld1w { z31.s }, p0/z, [x0]
+; CHECK-NEXT: mov z24.d, z1.d
+; CHECK-NEXT: famax { z24.s - z27.s }, { z24.s - z27.s }, { z28.s - z31.s }
+; CHECK-NEXT: mov z0.d, z24.d
+; CHECK-NEXT: mov z1.d, z25.d
+; CHECK-NEXT: mov z2.d, z26.d
+; CHECK-NEXT: mov z3.d, z27.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famax.x4.nxv4f32(<vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zdn3, <vscale x 4 x float> %zdn4, <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2, <vscale x 4 x float> %zm3, <vscale x 4 x float> %zm4)
+ ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } %res
+}
+
+define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @multi_vec_max_multi_x4_f64(<vscale x 2 x double> %unused, <vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zdn3, <vscale x 2 x double> %zdn4, <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2, <vscale x 2 x double> %zm3, <vscale x 2 x double> %zm4) {
+; CHECK-LABEL: multi_vec_max_multi_x4_f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z30.d, z7.d
+; CHECK-NEXT: mov z27.d, z4.d
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: mov z29.d, z6.d
+; CHECK-NEXT: mov z26.d, z3.d
+; CHECK-NEXT: mov z28.d, z5.d
+; CHECK-NEXT: mov z25.d, z2.d
+; CHECK-NEXT: ld1d { z31.d }, p0/z, [x0]
+; CHECK-NEXT: mov z24.d, z1.d
+; CHECK-NEXT: famax { z24.d - z27.d }, { z24.d - z27.d }, { z28.d - z31.d }
+; CHECK-NEXT: mov z0.d, z24.d
+; CHECK-NEXT: mov z1.d, z25.d
+; CHECK-NEXT: mov z2.d, z26.d
+; CHECK-NEXT: mov z3.d, z27.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }
+ @llvm.aarch64.sme.famax.x4.nxv2f64(<vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zdn3, <vscale x 2 x double> %zdn4,
+ <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2, <vscale x 2 x double> %zm3, <vscale x 2 x double> %zm4)
+ ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
+}
+
+
+; FAMIN (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) {
+; CHECK-LABEL: multi_vec_min_multi_x2_f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z7.d, z4.d
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z6.d, z3.d
+; CHECK-NEXT: mov z4.d, z1.d
+; CHECK-NEXT: famin { z4.h, z5.h }, { z4.h, z5.h }, { z6.h, z7.h }
+; CHECK-NEXT: mov z0.d, z4.d
+; CHECK-NEXT: mov z1.d, z5.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.famin.x2.nxv8f16(<vscale x 8 x half> %zdn1, <vscale x 8 x half> %zdn2, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2)
+ ret { <vscale x 8 x half>, <vscale x 8 x half> } %res
+}
+
+define { <vscale x 4 x float>, <vscale x 4 x float> } @multi_vec_min_multi_x2_f32(<vscale x 4 x float> %unused, <vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2) {
+; CHECK-LABEL: multi_vec_min_multi_x2_f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z7.d, z4.d
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z6.d, z3.d
+; CHECK-NEXT: mov z4.d, z1.d
+; CHECK-NEXT: famin { z4.s, z5.s }, { z4.s, z5.s }, { z6.s, z7.s }
+; CHECK-NEXT: mov z0.d, z4.d
+; CHECK-NEXT: mov z1.d, z5.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.famin.x2.nxv4f32(<vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2)
+ ret { <vscale x 4 x float>, <vscale x 4 x float> } %res
+}
+
+define { <vscale x 2 x double>, <vscale x 2 x double> } @multi_vec_main_multi_x2_f64(<vscale x 2 x double> %unused, <vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2) {
+; CHECK-LABEL: multi_vec_main_multi_x2_f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z7.d, z4.d
+; CHECK-NEXT: mov z5.d, z2.d
+; CHECK-NEXT: mov z6.d, z3.d
+; CHECK-NEXT: mov z4.d, z1.d
+; CHECK-NEXT: famin { z4.d, z5.d }, { z4.d, z5.d }, { z6.d, z7.d }
+; CHECK-NEXT: mov z0.d, z4.d
+; CHECK-NEXT: mov z1.d, z5.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.famin.x2.nxv2f64(<vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2)
+ ret { <vscale x 2 x double>, <vscale x 2 x double> } %res
+}
+
+; FAMIN (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, <vscale x 8 x half> %zm1, <vscale x 8 x half> %zm2, <vscale x 8 x half> %zm3, <vscale x 8 x half> %zm4) {
+; CHECK-LABEL: multi_vec_min_multi_x4_f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z30.d, z7.d
+; CHECK-NEXT: mov z27.d, z4.d
+; CHECK-NEXT: ptrue p0.h
+; CHECK-NEXT: mov z29.d, z6.d
+; CHECK-NEXT: mov z26.d, z3.d
+; CHECK-NEXT: mov z28.d, z5.d
+; CHECK-NEXT: mov z25.d, z2.d
+; CHECK-NEXT: ld1h { z31.h }, p0/z, [x0]
+; CHECK-NEXT: mov z24.d, z1.d
+; CHECK-NEXT: famin { z24.h - z27.h }, { z24.h - z27.h }, { z28.h - z31.h }
+; CHECK-NEXT: mov z0.d, z24.d
+; CHECK-NEXT: mov z1.d, z25.d
+; CHECK-NEXT: mov z2.d, z26.d
+; CHECK-NEXT: mov z3.d, z27.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> }
+ @llvm.aarch64.sme.famin.x4.nxv8f16(<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)
+ ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } %res
+}
+
+define { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @multi_vec_min_multi_x4_f32(<vscale x 4 x float> %unused, <vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zdn3, <vscale x 4 x float> %zdn4, <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2, <vscale x 4 x float> %zm3, <vscale x 4 x float> %zm4) {
+; CHECK-LABEL: multi_vec_min_multi_x4_f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z30.d, z7.d
+; CHECK-NEXT: mov z27.d, z4.d
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: mov z29.d, z6.d
+; CHECK-NEXT: mov z26.d, z3.d
+; CHECK-NEXT: mov z28.d, z5.d
+; CHECK-NEXT: mov z25.d, z2.d
+; CHECK-NEXT: ld1w { z31.s }, p0/z, [x0]
+; CHECK-NEXT: mov z24.d, z1.d
+; CHECK-NEXT: famin { z24.s - z27.s }, { z24.s - z27.s }, { z28.s - z31.s }
+; CHECK-NEXT: mov z0.d, z24.d
+; CHECK-NEXT: mov z1.d, z25.d
+; CHECK-NEXT: mov z2.d, z26.d
+; CHECK-NEXT: mov z3.d, z27.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> }
+ @llvm.aarch64.sme.famin.x4.nxv4f32(<vscale x 4 x float> %zdn1, <vscale x 4 x float> %zdn2, <vscale x 4 x float> %zdn3, <vscale x 4 x float> %zdn4,
+ <vscale x 4 x float> %zm1, <vscale x 4 x float> %zm2, <vscale x 4 x float> %zm3, <vscale x 4 x float> %zm4)
+ ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } %res
+}
+
+define { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @multi_vec_min_multi_x4_f64(<vscale x 2 x double> %unused, <vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zdn3, <vscale x 2 x double> %zdn4, <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2, <vscale x 2 x double> %zm3, <vscale x 2 x double> %zm4) {
+; CHECK-LABEL: multi_vec_min_multi_x4_f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: mov z30.d, z7.d
+; CHECK-NEXT: mov z27.d, z4.d
+; CHECK-NEXT: ptrue p0.d
+; CHECK-NEXT: mov z29.d, z6.d
+; CHECK-NEXT: mov z26.d, z3.d
+; CHECK-NEXT: mov z28.d, z5.d
+; CHECK-NEXT: mov z25.d, z2.d
+; CHECK-NEXT: ld1d { z31.d }, p0/z, [x0]
+; CHECK-NEXT: mov z24.d, z1.d
+; CHECK-NEXT: famin { z24.d - z27.d }, { z24.d - z27.d }, { z28.d - z31.d }
+; CHECK-NEXT: mov z0.d, z24.d
+; CHECK-NEXT: mov z1.d, z25.d
+; CHECK-NEXT: mov z2.d, z26.d
+; CHECK-NEXT: mov z3.d, z27.d
+; CHECK-NEXT: ret
+ %res = call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> }
+ @llvm.aarch64.sme.famin.x4.nxv2f64(<vscale x 2 x double> %zdn1, <vscale x 2 x double> %zdn2, <vscale x 2 x double> %zdn3, <vscale x 2 x double> %zdn4,
+ <vscale x 2 x double> %zm1, <vscale x 2 x double> %zm2, <vscale x 2 x double> %zm3, <vscale x 2 x double> %zm4)
+ ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } %res
+}
More information about the cfe-commits
mailing list