[clang] [llvm] [AArch64] [SVE] Created intrinsics for DUPQ instr. (PR #83260)
via cfe-commits
cfe-commits at lists.llvm.org
Wed Feb 28 05:47:02 PST 2024
https://github.com/Lukacma created https://github.com/llvm/llvm-project/pull/83260
This patch adds clang and llvm support for following intrinsic and maps it to DUPQ instruction:
```
// Variants are also available for:
// _s8, _u16, _s16, _u32, _s32, _u64, _s64
// _bf16, _f16, _f32, _f64
svuint8_t svdup_laneq[_u8](svuint8_t zn, uint64_t imm_idx);
```
>From c5bceafca67d15bad05afa17ac8875421e4259a3 Mon Sep 17 00:00:00 2001
From: Marian Lukac <Marian.Lukac at arm.com>
Date: Wed, 21 Feb 2024 11:11:45 +0000
Subject: [PATCH] [AArch64] [SVE] Created intrinsics for DUPQ instr.
This patch adds 'svdup_laneq[_{d}]' intrinsics and maps them to DUPQ instruction.
---
clang/include/clang/Basic/arm_sve.td | 9 +
.../acle_sve2p1_dupq.c | 213 ++++++++++++++++++
llvm/include/llvm/IR/IntrinsicsAArch64.td | 7 +
.../lib/Target/AArch64/AArch64SVEInstrInfo.td | 2 +-
llvm/lib/Target/AArch64/SVEInstrFormats.td | 12 +-
.../CodeGen/AArch64/sve2p1-intrinsics-dupq.ll | 83 +++++++
6 files changed, 324 insertions(+), 2 deletions(-)
create mode 100644 clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_dupq.c
create mode 100644 llvm/test/CodeGen/AArch64/sve2p1-intrinsics-dupq.ll
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 6da30e08e7521e..6cc249837d3f3d 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -2215,6 +2215,15 @@ let TargetGuard = "sve2p1" in {
def SVTBXQ : SInst<"svtbxq[_{d}]", "dddu", "cUcsUsiUilUlbhfd", MergeNone, "aarch64_sve_tbxq">;
// EXTQ
def EXTQ : SInst<"svextq[_{d}]", "dddk", "cUcsUsiUilUlbhfd", MergeNone, "aarch64_sve_extq", [], [ImmCheck<2, ImmCheck0_15>]>;
+ // DUPQ
+ def SVDUP_LANEQ_B : SInst<"svdup_laneq[_{d}]", "ddi", "cUc", MergeNone, "aarch64_sve_dup_laneq", [IsStreamingCompatible], [ImmCheck<1, ImmCheck0_15>]>;
+ def SVDUP_LANEQ_H : SInst<"svdup_laneq[_{d}]", "ddi", "sUsh", MergeNone, "aarch64_sve_dup_laneq", [IsStreamingCompatible], [ImmCheck<1, ImmCheck0_7>]>;
+ def SVDUP_LANEQ_S : SInst<"svdup_laneq[_{d}]", "ddi", "iUif", MergeNone, "aarch64_sve_dup_laneq", [IsStreamingCompatible], [ImmCheck<1, ImmCheck0_3>]>;
+ def SVDUP_LANEQ_D : SInst<"svdup_laneq[_{d}]", "ddi", "lUld", MergeNone, "aarch64_sve_dup_laneq", [IsStreamingCompatible], [ImmCheck<1, ImmCheck0_1>]>;
+
+ let TargetGuard = "bf16" in {
+ def SVDUP_LANEQ_BF16 : SInst<"svdup_laneq[_{d}]", "ddi", "b", MergeNone, "aarch64_sve_dup_laneq", [IsStreamingCompatible], [ImmCheck<1, ImmCheck0_7>]>;
+ }
// PMOV
// Move to Pred
multiclass PMOV_TO_PRED<string name, string types, string intrinsic, list<FlagType> flags=[], ImmCheckType immCh > {
diff --git a/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_dupq.c b/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_dupq.c
new file mode 100644
index 00000000000000..587a67aa6b7ca2
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_dupq.c
@@ -0,0 +1,213 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
+// REQUIRES: aarch64-registered-target
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2p1 -target-feature +bf16\
+// RUN: -S -Werror -emit-llvm -disable-O0-optnone -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2p1 -target-feature +bf16\
+// RUN: -S -Werror -emit-llvm -disable-O0-optnone -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2p1 -target-feature +bf16\
+// RUN: -S -Werror -emit-llvm -disable-O0-optnone -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 +sve2p1 -target-feature +bf16\
+// RUN: -S -Werror -emit-llvm -disable-O0-optnone -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 +sve2p1 -target-feature +bf16 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+
+
+#include <arm_sve.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
+
+// CHECK-LABEL: define dso_local <vscale x 16 x i8> @test_svdup_laneq_s8
+// CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.dup.laneq.nxv16i8(<vscale x 16 x i8> [[ZN]], i32 0)
+// CHECK-NEXT: ret <vscale x 16 x i8> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i8> @_Z19test_svdup_laneq_s8u10__SVInt8_t
+// CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0:[0-9]+]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.dup.laneq.nxv16i8(<vscale x 16 x i8> [[ZN]], i32 0)
+// CPP-CHECK-NEXT: ret <vscale x 16 x i8> [[TMP0]]
+//
+svint8_t test_svdup_laneq_s8(svint8_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _s8)(zn, 0);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 16 x i8> @test_svdup_laneq_u8
+// CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.dup.laneq.nxv16i8(<vscale x 16 x i8> [[ZN]], i32 15)
+// CHECK-NEXT: ret <vscale x 16 x i8> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 16 x i8> @_Z19test_svdup_laneq_u8u11__SVUint8_t
+// CPP-CHECK-SAME: (<vscale x 16 x i8> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.dup.laneq.nxv16i8(<vscale x 16 x i8> [[ZN]], i32 15)
+// CPP-CHECK-NEXT: ret <vscale x 16 x i8> [[TMP0]]
+//
+svuint8_t test_svdup_laneq_u8(svuint8_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _u8)(zn, 15);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x i16> @test_svdup_laneq_s16
+// CHECK-SAME: (<vscale x 8 x i16> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.dup.laneq.nxv8i16(<vscale x 8 x i16> [[ZN]], i32 1)
+// CHECK-NEXT: ret <vscale x 8 x i16> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i16> @_Z20test_svdup_laneq_s16u11__SVInt16_t
+// CPP-CHECK-SAME: (<vscale x 8 x i16> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.dup.laneq.nxv8i16(<vscale x 8 x i16> [[ZN]], i32 1)
+// CPP-CHECK-NEXT: ret <vscale x 8 x i16> [[TMP0]]
+//
+svint16_t test_svdup_laneq_s16(svint16_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _s16)(zn, 1);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x i16> @test_svdup_laneq_u16
+// CHECK-SAME: (<vscale x 8 x i16> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.dup.laneq.nxv8i16(<vscale x 8 x i16> [[ZN]], i32 7)
+// CHECK-NEXT: ret <vscale x 8 x i16> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 8 x i16> @_Z20test_svdup_laneq_u16u12__SVUint16_t
+// CPP-CHECK-SAME: (<vscale x 8 x i16> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.dup.laneq.nxv8i16(<vscale x 8 x i16> [[ZN]], i32 7)
+// CPP-CHECK-NEXT: ret <vscale x 8 x i16> [[TMP0]]
+//
+svuint16_t test_svdup_laneq_u16(svuint16_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _u16)(zn, 7);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 4 x i32> @test_svdup_laneq_s32
+// CHECK-SAME: (<vscale x 4 x i32> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.dup.laneq.nxv4i32(<vscale x 4 x i32> [[ZN]], i32 2)
+// CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i32> @_Z20test_svdup_laneq_s32u11__SVInt32_t
+// CPP-CHECK-SAME: (<vscale x 4 x i32> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.dup.laneq.nxv4i32(<vscale x 4 x i32> [[ZN]], i32 2)
+// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
+//
+svint32_t test_svdup_laneq_s32(svint32_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _s32)(zn, 2);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 4 x i32> @test_svdup_laneq_u32
+// CHECK-SAME: (<vscale x 4 x i32> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.dup.laneq.nxv4i32(<vscale x 4 x i32> [[ZN]], i32 3)
+// CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 4 x i32> @_Z20test_svdup_laneq_u32u12__SVUint32_t
+// CPP-CHECK-SAME: (<vscale x 4 x i32> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.dup.laneq.nxv4i32(<vscale x 4 x i32> [[ZN]], i32 3)
+// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
+//
+svuint32_t test_svdup_laneq_u32(svuint32_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _u32)(zn, 3);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 2 x i64> @test_svdup_laneq_s64
+// CHECK-SAME: (<vscale x 2 x i64> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.dup.laneq.nxv2i64(<vscale x 2 x i64> [[ZN]], i32 0)
+// CHECK-NEXT: ret <vscale x 2 x i64> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 2 x i64> @_Z20test_svdup_laneq_s64u11__SVInt64_t
+// CPP-CHECK-SAME: (<vscale x 2 x i64> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.dup.laneq.nxv2i64(<vscale x 2 x i64> [[ZN]], i32 0)
+// CPP-CHECK-NEXT: ret <vscale x 2 x i64> [[TMP0]]
+//
+svint64_t test_svdup_laneq_s64(svint64_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _s64)(zn, 0);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 2 x i64> @test_svdup_laneq_u64
+// CHECK-SAME: (<vscale x 2 x i64> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.dup.laneq.nxv2i64(<vscale x 2 x i64> [[ZN]], i32 1)
+// CHECK-NEXT: ret <vscale x 2 x i64> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 2 x i64> @_Z20test_svdup_laneq_u64u12__SVUint64_t
+// CPP-CHECK-SAME: (<vscale x 2 x i64> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.dup.laneq.nxv2i64(<vscale x 2 x i64> [[ZN]], i32 1)
+// CPP-CHECK-NEXT: ret <vscale x 2 x i64> [[TMP0]]
+//
+svuint64_t test_svdup_laneq_u64(svuint64_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _u64)(zn, 1);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x half> @test_svdup_laneq_f16
+// CHECK-SAME: (<vscale x 8 x half> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.dup.laneq.nxv8f16(<vscale x 8 x half> [[ZN]], i32 4)
+// CHECK-NEXT: ret <vscale x 8 x half> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 8 x half> @_Z20test_svdup_laneq_f16u13__SVFloat16_t
+// CPP-CHECK-SAME: (<vscale x 8 x half> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.dup.laneq.nxv8f16(<vscale x 8 x half> [[ZN]], i32 4)
+// CPP-CHECK-NEXT: ret <vscale x 8 x half> [[TMP0]]
+//
+svfloat16_t test_svdup_laneq_f16(svfloat16_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _f16)(zn, 4);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 4 x float> @test_svdup_laneq_f32
+// CHECK-SAME: (<vscale x 4 x float> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.dup.laneq.nxv4f32(<vscale x 4 x float> [[ZN]], i32 1)
+// CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 4 x float> @_Z20test_svdup_laneq_f32u13__SVFloat32_t
+// CPP-CHECK-SAME: (<vscale x 4 x float> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.dup.laneq.nxv4f32(<vscale x 4 x float> [[ZN]], i32 1)
+// CPP-CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
+//
+svfloat32_t test_svdup_laneq_f32(svfloat32_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _f32)(zn, 1);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 2 x double> @test_svdup_laneq_f64
+// CHECK-SAME: (<vscale x 2 x double> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.aarch64.sve.dup.laneq.nxv2f64(<vscale x 2 x double> [[ZN]], i32 1)
+// CHECK-NEXT: ret <vscale x 2 x double> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 2 x double> @_Z20test_svdup_laneq_f64u13__SVFloat64_t
+// CPP-CHECK-SAME: (<vscale x 2 x double> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.aarch64.sve.dup.laneq.nxv2f64(<vscale x 2 x double> [[ZN]], i32 1)
+// CPP-CHECK-NEXT: ret <vscale x 2 x double> [[TMP0]]
+//
+svfloat64_t test_svdup_laneq_f64(svfloat64_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _f64)(zn, 1);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x bfloat> @test_svdup_laneq_bf16
+// CHECK-SAME: (<vscale x 8 x bfloat> [[ZN:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.dup.laneq.nxv8bf16(<vscale x 8 x bfloat> [[ZN]], i32 3)
+// CHECK-NEXT: ret <vscale x 8 x bfloat> [[TMP0]]
+//
+// CPP-CHECK-LABEL: define dso_local <vscale x 8 x bfloat> @_Z21test_svdup_laneq_bf16u14__SVBfloat16_t
+// CPP-CHECK-SAME: (<vscale x 8 x bfloat> [[ZN:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.aarch64.sve.dup.laneq.nxv8bf16(<vscale x 8 x bfloat> [[ZN]], i32 3)
+// CPP-CHECK-NEXT: ret <vscale x 8 x bfloat> [[TMP0]]
+//
+svbfloat16_t test_svdup_laneq_bf16(svbfloat16_t zn) {
+ return SVE_ACLE_FUNC(svdup_laneq, _bf16)(zn, 3);
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsAArch64.td b/llvm/include/llvm/IR/IntrinsicsAArch64.td
index 921e5b95ae03e8..f0657bdbf50110 100644
--- a/llvm/include/llvm/IR/IntrinsicsAArch64.td
+++ b/llvm/include/llvm/IR/IntrinsicsAArch64.td
@@ -1358,6 +1358,12 @@ let TargetPrefix = "aarch64" in { // All intrinsics start with "llvm.aarch64.".
LLVMSubdivide2VectorType<0>,
llvm_i32_ty],
[IntrNoMem, ImmArg<ArgIndex<3>>]>;
+
+ class SVE2_1VectorArgIndexed_Intrinsic
+ : DefaultAttrsIntrinsic<[llvm_anyvector_ty],
+ [LLVMMatchType<0>,
+ llvm_i32_ty],
+ [IntrNoMem, ImmArg<ArgIndex<1>>]>;
class AdvSIMD_SVE_CDOT_LANE_Intrinsic
: DefaultAttrsIntrinsic<[llvm_anyvector_ty],
@@ -1890,6 +1896,7 @@ def int_aarch64_sve_clastb : AdvSIMD_Pred2VectorArg_Intrinsic;
def int_aarch64_sve_clastb_n : AdvSIMD_SVE_ReduceWithInit_Intrinsic;
def int_aarch64_sve_compact : AdvSIMD_Pred1VectorArg_Intrinsic;
def int_aarch64_sve_dupq_lane : AdvSIMD_SVE_DUPQ_Intrinsic;
+def int_aarch64_sve_dup_laneq : SVE2_1VectorArgIndexed_Intrinsic;
def int_aarch64_sve_ext : AdvSIMD_2VectorArgIndexed_Intrinsic;
def int_aarch64_sve_sel : AdvSIMD_Pred2VectorArg_Intrinsic;
def int_aarch64_sve_lasta : AdvSIMD_SVE_Reduce_Intrinsic;
diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 2b0524754b2e9d..8fbe26af09f802 100644
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -4101,7 +4101,7 @@ defm FMINNMQV : sve2p1_fp_reduction_q<0b101, "fminnmqv", int_aarch64_sve_fminnmq
defm FMAXQV : sve2p1_fp_reduction_q<0b110, "fmaxqv", int_aarch64_sve_fmaxqv>;
defm FMINQV : sve2p1_fp_reduction_q<0b111, "fminqv", int_aarch64_sve_fminqv>;
-defm DUPQ_ZZI : sve2p1_dupq<"dupq">;
+defm DUPQ_ZZI : sve2p1_dupq<"dupq", int_aarch64_sve_dup_laneq>;
defm EXTQ_ZZI : sve2p1_extq<"extq", int_aarch64_sve_extq>;
defm PMOV_PZI : sve2p1_vector_to_pred<"pmov", int_aarch64_sve_pmov_to_pred_lane, int_aarch64_sve_pmov_to_pred_lane_zero>;
diff --git a/llvm/lib/Target/AArch64/SVEInstrFormats.td b/llvm/lib/Target/AArch64/SVEInstrFormats.td
index 789ec817d3d8b8..ab43eac9ae4451 100644
--- a/llvm/lib/Target/AArch64/SVEInstrFormats.td
+++ b/llvm/lib/Target/AArch64/SVEInstrFormats.td
@@ -9990,7 +9990,7 @@ class sve2p1_dupq<bits<5> ind_tsz, string mnemonic, ZPRRegOp zprty, Operand ityp
let hasSideEffects = 0;
}
-multiclass sve2p1_dupq<string mnemonic> {
+multiclass sve2p1_dupq<string mnemonic, SDPatternOperator Op> {
def _B : sve2p1_dupq<{?, ?, ?, ?, 1}, mnemonic, ZPR8, VectorIndexB32b> {
bits<4> index;
let Inst{20-17} = index;
@@ -10007,6 +10007,16 @@ multiclass sve2p1_dupq<string mnemonic> {
bits<1> index;
let Inst{20} = index;
}
+
+ def : SVE_2_Op_Imm_Pat<nxv16i8, Op, nxv16i8, i32, VectorIndexB32b_timm, !cast<Instruction>(NAME # _B)>;
+ def : SVE_2_Op_Imm_Pat<nxv8i16, Op, nxv8i16, i32, VectorIndexH32b_timm, !cast<Instruction>(NAME # _H)>;
+ def : SVE_2_Op_Imm_Pat<nxv4i32, Op, nxv4i32, i32, VectorIndexS32b_timm, !cast<Instruction>(NAME # _S)>;
+ def : SVE_2_Op_Imm_Pat<nxv2i64, Op, nxv2i64, i32, VectorIndexD32b_timm, !cast<Instruction>(NAME # _D)>;
+
+ def : SVE_2_Op_Imm_Pat<nxv8f16, Op, nxv8f16, i32, VectorIndexH32b_timm, !cast<Instruction>(NAME # _H)>;
+ def : SVE_2_Op_Imm_Pat<nxv4f32, Op, nxv4f32, i32, VectorIndexS32b_timm, !cast<Instruction>(NAME # _S)>;
+ def : SVE_2_Op_Imm_Pat<nxv2f64, Op, nxv2f64, i32, VectorIndexD32b_timm, !cast<Instruction>(NAME # _D)>;
+ def : SVE_2_Op_Imm_Pat<nxv8bf16, Op, nxv8bf16, i32, VectorIndexH32b_timm, !cast<Instruction>(NAME # _H)>;
}
diff --git a/llvm/test/CodeGen/AArch64/sve2p1-intrinsics-dupq.ll b/llvm/test/CodeGen/AArch64/sve2p1-intrinsics-dupq.ll
new file mode 100644
index 00000000000000..f1a423b02ac2ad
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve2p1-intrinsics-dupq.ll
@@ -0,0 +1,83 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve2p1,+bf16 < %s | FileCheck %s
+
+define <vscale x 16 x i8> @test_dupq_i8 (<vscale x 16 x i8> %zn) {
+; CHECK-LABEL: test_dupq_i8:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.b, z0.b[15]
+; CHECK-NEXT: ret
+ %res = call <vscale x 16 x i8> @llvm.aarch64.sve.dup.laneq.nxv16i8(<vscale x 16 x i8> %zn, i32 15)
+ ret <vscale x 16 x i8> %res
+}
+
+define <vscale x 8 x i16> @test_dupq_i16 (<vscale x 8 x i16> %zn) {
+; CHECK-LABEL: test_dupq_i16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.h, z0.h[7]
+; CHECK-NEXT: ret
+ %res = call <vscale x 8 x i16> @llvm.aarch64.sve.dup.laneq.nxv8i16(<vscale x 8 x i16> %zn, i32 7)
+ ret <vscale x 8 x i16> %res
+}
+
+define <vscale x 4 x i32> @test_dupq__i32 (<vscale x 4 x i32> %zn) {
+; CHECK-LABEL: test_dupq__i32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.s, z0.s[3]
+; CHECK-NEXT: ret
+ %res = call <vscale x 4 x i32> @llvm.aarch64.sve.dup.laneq.nxv4i32(<vscale x 4 x i32> %zn, i32 3)
+ ret <vscale x 4 x i32> %res
+}
+
+define <vscale x 2 x i64> @test_dupq_i64 (<vscale x 2 x i64> %zn) {
+; CHECK-LABEL: test_dupq_i64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.d, z0.d[1]
+; CHECK-NEXT: ret
+ %res = call <vscale x 2 x i64> @llvm.aarch64.sve.dup.laneq.nxv2i64(<vscale x 2 x i64> %zn, i32 1)
+ ret <vscale x 2 x i64> %res
+}
+
+define <vscale x 8 x half> @test_dupq_f16(<vscale x 8 x half> %zn) {
+; CHECK-LABEL: test_dupq_f16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.h, z0.h[4]
+; CHECK-NEXT: ret
+ %res = call <vscale x 8 x half> @llvm.aarch64.sve.dup.laneq.nxv8f16(<vscale x 8 x half> %zn, i32 4)
+ ret <vscale x 8 x half> %res
+}
+
+define <vscale x 4 x float> @test_dupq_f32(<vscale x 4 x float> %zn) {
+; CHECK-LABEL: test_dupq_f32:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.s, z0.s[2]
+; CHECK-NEXT: ret
+ %res = call <vscale x 4 x float> @llvm.aarch64.sve.dup.laneq.nxv4f32(<vscale x 4 x float> %zn, i32 2)
+ ret <vscale x 4 x float> %res
+}
+
+define <vscale x 2 x double> @test_dupq_f64(<vscale x 2 x double> %zn) {
+; CHECK-LABEL: test_dupq_f64:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.d, z0.d[0]
+; CHECK-NEXT: ret
+ %res = call <vscale x 2 x double> @llvm.aarch64.sve.dup.laneq.nxv2f64(<vscale x 2 x double> %zn, i32 0)
+ ret <vscale x 2 x double> %res
+}
+
+define <vscale x 8 x bfloat> @test_dupq_bf16(<vscale x 8 x bfloat> %zn) {
+; CHECK-LABEL: test_dupq_bf16:
+; CHECK: // %bb.0:
+; CHECK-NEXT: dupq z0.h, z0.h[1]
+; CHECK-NEXT: ret
+ %res = call <vscale x 8 x bfloat> @llvm.aarch64.sve.dup.laneq.nxv8bf16(<vscale x 8 x bfloat> %zn, i32 1)
+ ret <vscale x 8 x bfloat> %res
+}
+
+declare <vscale x 16 x i8> @llvm.aarch64.sve.dup.laneq.nxv16i8(<vscale x 16 x i8>, i32)
+declare <vscale x 8 x i16> @llvm.aarch64.sve.dup.laneq.nxv8i16(<vscale x 8 x i16>, i32)
+declare <vscale x 4 x i32> @llvm.aarch64.sve.dup.laneq.nxv4i32(<vscale x 4 x i32>, i32)
+declare <vscale x 2 x i64> @llvm.aarch64.sve.dup.laneq.nxv2i64(<vscale x 2 x i64>, i32)
+declare <vscale x 8 x half> @llvm.aarch64.sve.dup.laneq.nxv8f16(<vscale x 8 x half>, i32)
+declare <vscale x 4 x float> @llvm.aarch64.sve.dup.laneq.nxv4f32(<vscale x 4 x float>, i32)
+declare <vscale x 2 x double> @llvm.aarch64.sve.dup.laneq.nxv2f64(<vscale x 2 x double>, i32)
+declare <vscale x 8 x bfloat> @llvm.aarch64.sve.dup.laneq.nxv8bf16(<vscale x 8 x bfloat>, i32)
More information about the cfe-commits
mailing list