[clang] [CIR][AArch64] Implement vqshrn_n NEON builtin (PR #195085)

Daniel Alejandro Rodriguez via cfe-commits cfe-commits at lists.llvm.org
Thu Apr 30 10:47:02 PDT 2026


https://github.com/DanAlejandroRodriguez updated https://github.com/llvm/llvm-project/pull/195085

>From e109901c0023af8c9c9968cfc845ef1a206156ae Mon Sep 17 00:00:00 2001
From: IAmCheese1231 <ezhang2 at andrew.cmu.edu>
Date: Thu, 23 Apr 2026 22:01:56 -0400
Subject: [PATCH 1/6] implemented vqshrun builtin, added some basic tests

---
 .../lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp  |  4 ++-
 clang/test/CodeGen/AArch64/neon/vqshrun.c     | 28 +++++++++++++++++++
 2 files changed, 31 insertions(+), 1 deletion(-)
 create mode 100644 clang/test/CodeGen/AArch64/neon/vqshrun.c

diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
index ace8a5737f4bd..52ef0b36a2abe 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -2540,11 +2540,13 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr,
   case NEON::BI__builtin_neon_vmaxnmq_v:
     intrName = "aarch64.neon.fmaxnm";
     return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
+  case NEON::BI__builtin_neon_vqshrun_n_v:
+    intrName = "aarch64.neon.sqshrun";
+    return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
   case NEON::BI__builtin_neon_vmaxnmh_f16:
   case NEON::BI__builtin_neon_vrecpss_f32:
   case NEON::BI__builtin_neon_vrecpsd_f64:
   case NEON::BI__builtin_neon_vrecpsh_f16:
-  case NEON::BI__builtin_neon_vqshrun_n_v:
   case NEON::BI__builtin_neon_vqrshrun_n_v:
   case NEON::BI__builtin_neon_vqshrn_n_v:
   case NEON::BI__builtin_neon_vrshrn_n_v:
diff --git a/clang/test/CodeGen/AArch64/neon/vqshrun.c b/clang/test/CodeGen/AArch64/neon/vqshrun.c
new file mode 100644
index 0000000000000..016377b845183
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/neon/vqshrun.c
@@ -0,0 +1,28 @@
+// RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -emit-llvm -o - %s | FileCheck %s --check-prefix=LLVM
+// RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -fclangir -emit-cir -o - %s | FileCheck %s --check-prefix=CIR
+
+#include <arm_neon.h>
+
+// LLVM-LABEL: @test_vqshrun_n_s16(
+// CIR-LABEL: @test_vqshrun_n_s16(
+uint8x8_t test_vqshrun_n_s16(int16x8_t a) {
+  // CIR: cir.call_llvm_intrinsic "aarch64.neon.sqshrun"
+  // LLVM: call <8 x i8> @llvm.aarch64.neon.sqshrun.v8i8
+  return vqshrun_n_s16(a, 3);
+}
+
+// LLVM-LABEL: @test_vqshrun_n_s32(
+// CIR-LABEL: @test_vqshrun_n_s32(
+uint16x4_t test_vqshrun_n_s32(int32x4_t a) {
+  // CIR: cir.call_llvm_intrinsic "aarch64.neon.sqshrun"
+  // LLVM: call <4 x i16> @llvm.aarch64.neon.sqshrun.v4i16
+  return vqshrun_n_s32(a, 9);
+}
+
+// LLVM-LABEL: @test_vqshrun_n_s64(
+// CIR-LABEL: @test_vqshrun_n_s64(
+uint32x2_t test_vqshrun_n_s64(int64x2_t a) {
+  // CIR: cir.call_llvm_intrinsic "aarch64.neon.sqshrun"
+  // LLVM: call <2 x i32> @llvm.aarch64.neon.sqshrun.v2i32
+  return vqshrun_n_s64(a, 19);
+}
\ No newline at end of file

>From cd1f96bf9a209784fa589095167996a67f75e2a9 Mon Sep 17 00:00:00 2001
From: Jing Wang <jingyan5 at andrew.cmu.edu>
Date: Wed, 29 Apr 2026 12:36:01 -0400
Subject: [PATCH 2/6] [CIR][AArch64] Implement vqshrn_n NEON builtin

---
 .../lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp  |  4 +-
 clang/test/CodeGen/AArch64/neon/vqshrn.c      | 40 +++++++++++++++++++
 2 files changed, 43 insertions(+), 1 deletion(-)
 create mode 100644 clang/test/CodeGen/AArch64/neon/vqshrn.c

diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
index 52ef0b36a2abe..1ffad5dc01d8e 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -2543,12 +2543,14 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr,
   case NEON::BI__builtin_neon_vqshrun_n_v:
     intrName = "aarch64.neon.sqshrun";
     return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
+  case NEON::BI__builtin_neon_vqshrn_n_v:
+    intrName = "aarch64.neon.sqshrn";
+    return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
   case NEON::BI__builtin_neon_vmaxnmh_f16:
   case NEON::BI__builtin_neon_vrecpss_f32:
   case NEON::BI__builtin_neon_vrecpsd_f64:
   case NEON::BI__builtin_neon_vrecpsh_f16:
   case NEON::BI__builtin_neon_vqrshrun_n_v:
-  case NEON::BI__builtin_neon_vqshrn_n_v:
   case NEON::BI__builtin_neon_vrshrn_n_v:
   case NEON::BI__builtin_neon_vqrshrn_n_v:
   case NEON::BI__builtin_neon_vrndah_f16:
diff --git a/clang/test/CodeGen/AArch64/neon/vqshrn.c b/clang/test/CodeGen/AArch64/neon/vqshrn.c
new file mode 100644
index 0000000000000..bcc9f04374cf8
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/neon/vqshrn.c
@@ -0,0 +1,40 @@
+// REQUIRES: aarch64-registered-target || arm-registered-target
+// RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon \
+// RUN:   -disable-O0-optnone -flax-vector-conversions=none \
+// RUN:   -emit-llvm -o - %s | FileCheck %s --check-prefix=LLVM
+// RUN: %if cir-enabled %{ %clang_cc1 -triple arm64-none-linux-gnu \
+// RUN:   -target-feature +neon -disable-O0-optnone \
+// RUN:   -flax-vector-conversions=none -fclangir -emit-cir \
+// RUN:   -o - %s | FileCheck %s --check-prefix=CIR %}
+
+#include <arm_neon.h>
+
+// LLVM-LABEL: @test_vqshrn_n_s16(
+// CIR-LABEL: @vqshrn_n_s16(
+int8x8_t test_vqshrn_n_s16(int16x8_t a) {
+  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.sqshrn"
+  // CIR-SAME: (!cir.vector<8 x !s16i>) -> !cir.vector<8 x !s8i>
+  // LLVM: [[RES:%.*]] = call <8 x i8> @llvm.aarch64.neon.sqshrn.v8i8
+  // LLVM-NEXT: ret <8 x i8> [[RES]]
+  return vqshrn_n_s16(a, 3);
+}
+
+// LLVM-LABEL: @test_vqshrn_n_s32(
+// CIR-LABEL: @vqshrn_n_s32(
+int16x4_t test_vqshrn_n_s32(int32x4_t a) {
+  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.sqshrn"
+  // CIR-SAME: (!cir.vector<4 x !s32i>) -> !cir.vector<4 x !s16i>
+  // LLVM: [[RES:%.*]] = call <4 x i16> @llvm.aarch64.neon.sqshrn.v4i16
+  // LLVM-NEXT: ret <4 x i16> [[RES]]
+  return vqshrn_n_s32(a, 9);
+}
+
+// LLVM-LABEL: @test_vqshrn_n_s64(
+// CIR-LABEL: @vqshrn_n_s64(
+int32x2_t test_vqshrn_n_s64(int64x2_t a) {
+  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.sqshrn"
+  // CIR-SAME: (!cir.vector<2 x !s64i>) -> !cir.vector<2 x !s32i>
+  // LLVM: [[RES:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrn.v2i32
+  // LLVM-NEXT: ret <2 x i32> [[RES]]
+  return vqshrn_n_s64(a, 19);
+}
\ No newline at end of file

>From cb79049db229dcffcae346335ac9d4c77e1121ee Mon Sep 17 00:00:00 2001
From: Shreyas Hari <shreyas at Shreyass-MacBook-Pro-9.local>
Date: Wed, 29 Apr 2026 20:29:03 -0400
Subject: [PATCH 3/6] [CIR][AArch64] Fix vqshrn_n unsigned NEON lowering

---
 .../lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp  |  2 +-
 clang/test/CodeGen/AArch64/neon/vqshrn.c      | 32 ++++++++++++++++++-
 2 files changed, 32 insertions(+), 2 deletions(-)

diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
index 1ffad5dc01d8e..f268ae94a00dc 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -2544,7 +2544,7 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr,
     intrName = "aarch64.neon.sqshrun";
     return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
   case NEON::BI__builtin_neon_vqshrn_n_v:
-    intrName = "aarch64.neon.sqshrn";
+    intrName = usgn ? "aarch64.neon.uqshrn" : "aarch64.neon.sqshrn";
     return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
   case NEON::BI__builtin_neon_vmaxnmh_f16:
   case NEON::BI__builtin_neon_vrecpss_f32:
diff --git a/clang/test/CodeGen/AArch64/neon/vqshrn.c b/clang/test/CodeGen/AArch64/neon/vqshrn.c
index bcc9f04374cf8..9d37a7a213075 100644
--- a/clang/test/CodeGen/AArch64/neon/vqshrn.c
+++ b/clang/test/CodeGen/AArch64/neon/vqshrn.c
@@ -37,4 +37,34 @@ int32x2_t test_vqshrn_n_s64(int64x2_t a) {
   // LLVM: [[RES:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrn.v2i32
   // LLVM-NEXT: ret <2 x i32> [[RES]]
   return vqshrn_n_s64(a, 19);
-}
\ No newline at end of file
+}
+
+// LLVM-LABEL: @test_vqshrn_n_u16(
+// CIR-LABEL: @vqshrn_n_u16(
+uint8x8_t test_vqshrn_n_u16(uint16x8_t a) {
+  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.uqshrn"
+  // CIR-SAME: (!cir.vector<8 x !u16i>) -> !cir.vector<8 x !u8i>
+  // LLVM: [[RES:%.*]] = call <8 x i8> @llvm.aarch64.neon.uqshrn.v8i8
+  // LLVM-NEXT: ret <8 x i8> [[RES]]
+  return vqshrn_n_u16(a, 3);
+}
+
+// LLVM-LABEL: @test_vqshrn_n_u32(
+// CIR-LABEL: @vqshrn_n_u32(
+uint16x4_t test_vqshrn_n_u32(uint32x4_t a) {
+  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.uqshrn"
+  // CIR-SAME: (!cir.vector<4 x !u32i>) -> !cir.vector<4 x !u16i>
+  // LLVM: [[RES:%.*]] = call <4 x i16> @llvm.aarch64.neon.uqshrn.v4i16
+  // LLVM-NEXT: ret <4 x i16> [[RES]]
+  return vqshrn_n_u32(a, 9);
+}
+
+// LLVM-LABEL: @test_vqshrn_n_u64(
+// CIR-LABEL: @vqshrn_n_u64(
+uint32x2_t test_vqshrn_n_u64(uint64x2_t a) {
+  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.uqshrn"
+  // CIR-SAME: (!cir.vector<2 x !u64i>) -> !cir.vector<2 x !u32i>
+  // LLVM: [[RES:%.*]] = call <2 x i32> @llvm.aarch64.neon.uqshrn.v2i32
+  // LLVM-NEXT: ret <2 x i32> [[RES]]
+  return vqshrn_n_u64(a, 19);
+}

>From bc97a2d505fd912ac920c5265d9a6a8a0c8d3981 Mon Sep 17 00:00:00 2001
From: IAmCheese1231 <ezhang2 at andrew.cmu.edu>
Date: Thu, 23 Apr 2026 22:01:56 -0400
Subject: [PATCH 4/6] [CIR][AArch64] Implement vqshrun_n NEON builtin

---
 .../lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp  | 27 ++++++++++++-
 clang/test/CodeGen/AArch64/neon/vqshrun.c     | 40 +++++++++++++++++++
 2 files changed, 66 insertions(+), 1 deletion(-)
 create mode 100644 clang/test/CodeGen/AArch64/neon/vqshrun.c

diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
index 834f66586833b..9717843c223b3 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -2594,7 +2594,32 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr,
   case NEON::BI__builtin_neon_vrecpss_f32:
   case NEON::BI__builtin_neon_vrecpsd_f64:
   case NEON::BI__builtin_neon_vrecpsh_f16:
-  case NEON::BI__builtin_neon_vqshrun_n_v:
+  case NEON::BI__builtin_neon_vqshrun_n_v: {
+    mlir::Type inputTy;
+
+    switch (type.getEltType()) {
+    case NeonTypeFlags::Int8:
+      inputTy = getNeonType(
+          this, NeonTypeFlags(NeonTypeFlags::Int16, false, true), loc);
+      break;
+    case NeonTypeFlags::Int16:
+      inputTy = getNeonType(
+          this, NeonTypeFlags(NeonTypeFlags::Int32, false, true), loc);
+      break;
+    case NeonTypeFlags::Int32:
+      inputTy = getNeonType(
+          this, NeonTypeFlags(NeonTypeFlags::Int64, false, true), loc);
+      break;
+    default:
+      llvm_unreachable("unexpected vqshrun element type");
+    }
+
+    auto shiftTy = ops[1].getType();
+    ops[0] = builder.createBitcast(loc, ops[0], inputTy);
+
+    intrName = "aarch64.neon.sqshrun";
+    return emitNeonCall(cgm, builder, {inputTy, shiftTy}, ops, intrName, ty, loc);
+  }
   case NEON::BI__builtin_neon_vqrshrun_n_v:
   case NEON::BI__builtin_neon_vqshrn_n_v:
   case NEON::BI__builtin_neon_vrshrn_n_v:
diff --git a/clang/test/CodeGen/AArch64/neon/vqshrun.c b/clang/test/CodeGen/AArch64/neon/vqshrun.c
new file mode 100644
index 0000000000000..04cb61df1987e
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/neon/vqshrun.c
@@ -0,0 +1,40 @@
+// REQUIRES: aarch64-registered-target || arm-registered-target
+
+// RUN:                   %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none           -emit-llvm -o - %s | opt -S -passes=mem2reg,sroa,simplifycfg | FileCheck %s --check-prefixes=LLVM
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none -fclangir -emit-llvm -o - %s | opt -S -passes=mem2reg,sroa,simplifycfg | FileCheck %s --check-prefixes=LLVM %}
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none -fclangir -emit-cir  -o - %s |                               FileCheck %s --check-prefixes=CIR %}
+
+#include <arm_neon.h>
+
+// LLVM-LABEL: @test_vqshrun_n_s16(
+// CIR-LABEL: @test_vqshrun_n_s16(
+uint8x8_t test_vqshrun_n_s16(int16x8_t a) {
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<8 x !s16i>, !s32i) -> !cir.vector<8 x !u8i>
+
+// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <8 x i8> @llvm.aarch64.neon.sqshrun.v8i8(<8 x i16> {{.*}}, i32 3)
+// LLVM-NEXT: ret <8 x i8> [[R]]
+  return vqshrun_n_s16(a, 3);
+}
+
+// LLVM-LABEL: @test_vqshrun_n_s32(
+// CIR-LABEL: @test_vqshrun_n_s32(
+uint16x4_t test_vqshrun_n_s32(int32x4_t a) {
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<4 x !s32i>, !s32i) -> !cir.vector<4 x !u16i>
+
+// LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <4 x i16> @llvm.aarch64.neon.sqshrun.v4i16(<4 x i32> {{.*}}, i32 9)
+// LLVM-NEXT: ret <4 x i16> [[R]]
+  return vqshrun_n_s32(a, 9);
+}
+
+// LLVM-LABEL: @test_vqshrun_n_s64(
+// CIR-LABEL: @test_vqshrun_n_s64(
+uint32x2_t test_vqshrun_n_s64(int64x2_t a) {
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<2 x !s64i>, !s32i) -> !cir.vector<2 x !u32i>
+
+// LLVM-SAME: <2 x i64> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrun.v2i32(<2 x i64> {{.*}}, i32 19)
+// LLVM-NEXT: ret <2 x i32> [[R]]
+  return vqshrun_n_s64(a, 19);
+}

>From 0833c1ce9066822a463fd5a113f9635a61197192 Mon Sep 17 00:00:00 2001
From: Daniel Rodriguez <dalejandro111803 at gmail.com>
Date: Thu, 30 Apr 2026 05:28:32 -0400
Subject: [PATCH 5/6] [CIR][AArch64] Fix vqshrn_n type mismatch and strengthen
 tests

---
 .../lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp  | 27 ++++++-
 clang/test/CodeGen/AArch64/neon/vqshrn.c      | 77 ++++++++++---------
 2 files changed, 65 insertions(+), 39 deletions(-)

diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
index f268ae94a00dc..9557e955fdd25 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -2543,9 +2543,32 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr,
   case NEON::BI__builtin_neon_vqshrun_n_v:
     intrName = "aarch64.neon.sqshrun";
     return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
-  case NEON::BI__builtin_neon_vqshrn_n_v:
+  case NEON::BI__builtin_neon_vqshrn_n_v: {
+    mlir::Type inputTy;
+
+    switch (type.getEltType()) {
+    case NeonTypeFlags::Int8:
+      inputTy = getNeonType(
+          this, NeonTypeFlags(NeonTypeFlags::Int16, false, true), loc);
+      break;
+    case NeonTypeFlags::Int16:
+      inputTy = getNeonType(
+          this, NeonTypeFlags(NeonTypeFlags::Int32, false, true), loc);
+      break;
+    case NeonTypeFlags::Int32:
+      inputTy = getNeonType(
+          this, NeonTypeFlags(NeonTypeFlags::Int64, false, true), loc);
+      break;
+    default:
+      llvm_unreachable("unexpected vqshrn element type");
+    }
+
+    auto shiftTy = ops[1].getType();
+    ops[0] = builder.createBitcast(loc, ops[0], inputTy);
+
     intrName = usgn ? "aarch64.neon.uqshrn" : "aarch64.neon.sqshrn";
-    return emitNeonCall(cgm, builder, {ty, ty}, ops, intrName, ty, loc);
+    return emitNeonCall(cgm, builder, {inputTy, shiftTy}, ops, intrName, ty, loc);
+  }
   case NEON::BI__builtin_neon_vmaxnmh_f16:
   case NEON::BI__builtin_neon_vrecpss_f32:
   case NEON::BI__builtin_neon_vrecpsd_f64:
diff --git a/clang/test/CodeGen/AArch64/neon/vqshrn.c b/clang/test/CodeGen/AArch64/neon/vqshrn.c
index 9d37a7a213075..62f8f569b9d03 100644
--- a/clang/test/CodeGen/AArch64/neon/vqshrn.c
+++ b/clang/test/CodeGen/AArch64/neon/vqshrn.c
@@ -1,70 +1,73 @@
 // REQUIRES: aarch64-registered-target || arm-registered-target
-// RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon \
-// RUN:   -disable-O0-optnone -flax-vector-conversions=none \
-// RUN:   -emit-llvm -o - %s | FileCheck %s --check-prefix=LLVM
-// RUN: %if cir-enabled %{ %clang_cc1 -triple arm64-none-linux-gnu \
-// RUN:   -target-feature +neon -disable-O0-optnone \
-// RUN:   -flax-vector-conversions=none -fclangir -emit-cir \
-// RUN:   -o - %s | FileCheck %s --check-prefix=CIR %}
+
+// RUN:                   %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none           -emit-llvm -o - %s | opt -S -passes=mem2reg,sroa,simplifycfg | FileCheck %s --check-prefixes=LLVM
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none -fclangir -emit-llvm -o - %s | opt -S -passes=mem2reg,sroa,simplifycfg | FileCheck %s --check-prefixes=LLVM %}
+// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none -fclangir -emit-cir  -o - %s |                               FileCheck %s --check-prefixes=CIR %}
 
 #include <arm_neon.h>
 
 // LLVM-LABEL: @test_vqshrn_n_s16(
-// CIR-LABEL: @vqshrn_n_s16(
+// CIR-LABEL: @test_vqshrn_n_s16(
 int8x8_t test_vqshrn_n_s16(int16x8_t a) {
-  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.sqshrn"
-  // CIR-SAME: (!cir.vector<8 x !s16i>) -> !cir.vector<8 x !s8i>
-  // LLVM: [[RES:%.*]] = call <8 x i8> @llvm.aarch64.neon.sqshrn.v8i8
-  // LLVM-NEXT: ret <8 x i8> [[RES]]
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrn" {{.*}} : (!cir.vector<8 x !s16i>, !s32i) -> !cir.vector<8 x !s8i>
+
+// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <8 x i8> @llvm.aarch64.neon.sqshrn.v8i8(<8 x i16> {{.*}}, i32 3)
+// LLVM-NEXT: ret <8 x i8> [[R]]
   return vqshrn_n_s16(a, 3);
 }
 
 // LLVM-LABEL: @test_vqshrn_n_s32(
-// CIR-LABEL: @vqshrn_n_s32(
+// CIR-LABEL: @test_vqshrn_n_s32(
 int16x4_t test_vqshrn_n_s32(int32x4_t a) {
-  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.sqshrn"
-  // CIR-SAME: (!cir.vector<4 x !s32i>) -> !cir.vector<4 x !s16i>
-  // LLVM: [[RES:%.*]] = call <4 x i16> @llvm.aarch64.neon.sqshrn.v4i16
-  // LLVM-NEXT: ret <4 x i16> [[RES]]
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrn" {{.*}} : (!cir.vector<4 x !s32i>, !s32i) -> !cir.vector<4 x !s16i>
+
+// LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <4 x i16> @llvm.aarch64.neon.sqshrn.v4i16(<4 x i32> {{.*}}, i32 9)
+// LLVM-NEXT: ret <4 x i16> [[R]]
   return vqshrn_n_s32(a, 9);
 }
 
 // LLVM-LABEL: @test_vqshrn_n_s64(
-// CIR-LABEL: @vqshrn_n_s64(
+// CIR-LABEL: @test_vqshrn_n_s64(
 int32x2_t test_vqshrn_n_s64(int64x2_t a) {
-  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.sqshrn"
-  // CIR-SAME: (!cir.vector<2 x !s64i>) -> !cir.vector<2 x !s32i>
-  // LLVM: [[RES:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrn.v2i32
-  // LLVM-NEXT: ret <2 x i32> [[RES]]
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrn" {{.*}} : (!cir.vector<2 x !s64i>, !s32i) -> !cir.vector<2 x !s32i>
+
+// LLVM-SAME: <2 x i64> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrn.v2i32(<2 x i64> {{.*}}, i32 19)
+// LLVM-NEXT: ret <2 x i32> [[R]]
   return vqshrn_n_s64(a, 19);
 }
 
 // LLVM-LABEL: @test_vqshrn_n_u16(
-// CIR-LABEL: @vqshrn_n_u16(
+// CIR-LABEL: @test_vqshrn_n_u16(
 uint8x8_t test_vqshrn_n_u16(uint16x8_t a) {
-  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.uqshrn"
-  // CIR-SAME: (!cir.vector<8 x !u16i>) -> !cir.vector<8 x !u8i>
-  // LLVM: [[RES:%.*]] = call <8 x i8> @llvm.aarch64.neon.uqshrn.v8i8
-  // LLVM-NEXT: ret <8 x i8> [[RES]]
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.uqshrn" {{.*}} : (!cir.vector<8 x !s16i>, !s32i) -> !cir.vector<8 x !u8i>
+
+// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <8 x i8> @llvm.aarch64.neon.uqshrn.v8i8(<8 x i16> {{.*}}, i32 3)
+// LLVM-NEXT: ret <8 x i8> [[R]]
   return vqshrn_n_u16(a, 3);
 }
 
 // LLVM-LABEL: @test_vqshrn_n_u32(
-// CIR-LABEL: @vqshrn_n_u32(
+// CIR-LABEL: @test_vqshrn_n_u32(
 uint16x4_t test_vqshrn_n_u32(uint32x4_t a) {
-  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.uqshrn"
-  // CIR-SAME: (!cir.vector<4 x !u32i>) -> !cir.vector<4 x !u16i>
-  // LLVM: [[RES:%.*]] = call <4 x i16> @llvm.aarch64.neon.uqshrn.v4i16
-  // LLVM-NEXT: ret <4 x i16> [[RES]]
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.uqshrn" {{.*}} : (!cir.vector<4 x !s32i>, !s32i) -> !cir.vector<4 x !u16i>
+
+// LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <4 x i16> @llvm.aarch64.neon.uqshrn.v4i16(<4 x i32> {{.*}}, i32 9)
+// LLVM-NEXT: ret <4 x i16> [[R]]
   return vqshrn_n_u32(a, 9);
 }
 
 // LLVM-LABEL: @test_vqshrn_n_u64(
-// CIR-LABEL: @vqshrn_n_u64(
+// CIR-LABEL: @test_vqshrn_n_u64(
 uint32x2_t test_vqshrn_n_u64(uint64x2_t a) {
-  // CIR: {{.*}}cir.call_llvm_intrinsic "aarch64.neon.uqshrn"
-  // CIR-SAME: (!cir.vector<2 x !u64i>) -> !cir.vector<2 x !u32i>
-  // LLVM: [[RES:%.*]] = call <2 x i32> @llvm.aarch64.neon.uqshrn.v2i32
-  // LLVM-NEXT: ret <2 x i32> [[RES]]
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.uqshrn" {{.*}} : (!cir.vector<2 x !s64i>, !s32i) -> !cir.vector<2 x !u32i>
+
+// LLVM-SAME: <2 x i64> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <2 x i32> @llvm.aarch64.neon.uqshrn.v2i32(<2 x i64> {{.*}}, i32 19)
+// LLVM-NEXT: ret <2 x i32> [[R]]
   return vqshrn_n_u64(a, 19);
 }

>From 305a48cdded3de01705df1825d9f94eda75f342a Mon Sep 17 00:00:00 2001
From: Daniel Rodriguez <dalejandro111803 at gmail.com>
Date: Thu, 30 Apr 2026 13:46:36 -0400
Subject: [PATCH 6/6] [CIR][AArch64] Address review: combine vqshrn and vqshrun
 tests into one file, fix blank lines

---
 clang/test/CodeGen/AArch64/neon/vqshrn.c  | 36 ++++++++++++++++----
 clang/test/CodeGen/AArch64/neon/vqshrun.c | 40 -----------------------
 2 files changed, 30 insertions(+), 46 deletions(-)
 delete mode 100644 clang/test/CodeGen/AArch64/neon/vqshrun.c

diff --git a/clang/test/CodeGen/AArch64/neon/vqshrn.c b/clang/test/CodeGen/AArch64/neon/vqshrn.c
index 62f8f569b9d03..457e457e61bb5 100644
--- a/clang/test/CodeGen/AArch64/neon/vqshrn.c
+++ b/clang/test/CodeGen/AArch64/neon/vqshrn.c
@@ -10,7 +10,6 @@
 // CIR-LABEL: @test_vqshrn_n_s16(
 int8x8_t test_vqshrn_n_s16(int16x8_t a) {
 // CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrn" {{.*}} : (!cir.vector<8 x !s16i>, !s32i) -> !cir.vector<8 x !s8i>
-
 // LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]]) {{.*}} {
 // LLVM:      [[R:%.*]] = call <8 x i8> @llvm.aarch64.neon.sqshrn.v8i8(<8 x i16> {{.*}}, i32 3)
 // LLVM-NEXT: ret <8 x i8> [[R]]
@@ -21,7 +20,6 @@ int8x8_t test_vqshrn_n_s16(int16x8_t a) {
 // CIR-LABEL: @test_vqshrn_n_s32(
 int16x4_t test_vqshrn_n_s32(int32x4_t a) {
 // CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrn" {{.*}} : (!cir.vector<4 x !s32i>, !s32i) -> !cir.vector<4 x !s16i>
-
 // LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]]) {{.*}} {
 // LLVM:      [[R:%.*]] = call <4 x i16> @llvm.aarch64.neon.sqshrn.v4i16(<4 x i32> {{.*}}, i32 9)
 // LLVM-NEXT: ret <4 x i16> [[R]]
@@ -32,7 +30,6 @@ int16x4_t test_vqshrn_n_s32(int32x4_t a) {
 // CIR-LABEL: @test_vqshrn_n_s64(
 int32x2_t test_vqshrn_n_s64(int64x2_t a) {
 // CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrn" {{.*}} : (!cir.vector<2 x !s64i>, !s32i) -> !cir.vector<2 x !s32i>
-
 // LLVM-SAME: <2 x i64> {{.*}} [[A:%.*]]) {{.*}} {
 // LLVM:      [[R:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrn.v2i32(<2 x i64> {{.*}}, i32 19)
 // LLVM-NEXT: ret <2 x i32> [[R]]
@@ -43,7 +40,6 @@ int32x2_t test_vqshrn_n_s64(int64x2_t a) {
 // CIR-LABEL: @test_vqshrn_n_u16(
 uint8x8_t test_vqshrn_n_u16(uint16x8_t a) {
 // CIR:      cir.call_llvm_intrinsic "aarch64.neon.uqshrn" {{.*}} : (!cir.vector<8 x !s16i>, !s32i) -> !cir.vector<8 x !u8i>
-
 // LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]]) {{.*}} {
 // LLVM:      [[R:%.*]] = call <8 x i8> @llvm.aarch64.neon.uqshrn.v8i8(<8 x i16> {{.*}}, i32 3)
 // LLVM-NEXT: ret <8 x i8> [[R]]
@@ -54,7 +50,6 @@ uint8x8_t test_vqshrn_n_u16(uint16x8_t a) {
 // CIR-LABEL: @test_vqshrn_n_u32(
 uint16x4_t test_vqshrn_n_u32(uint32x4_t a) {
 // CIR:      cir.call_llvm_intrinsic "aarch64.neon.uqshrn" {{.*}} : (!cir.vector<4 x !s32i>, !s32i) -> !cir.vector<4 x !u16i>
-
 // LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]]) {{.*}} {
 // LLVM:      [[R:%.*]] = call <4 x i16> @llvm.aarch64.neon.uqshrn.v4i16(<4 x i32> {{.*}}, i32 9)
 // LLVM-NEXT: ret <4 x i16> [[R]]
@@ -65,9 +60,38 @@ uint16x4_t test_vqshrn_n_u32(uint32x4_t a) {
 // CIR-LABEL: @test_vqshrn_n_u64(
 uint32x2_t test_vqshrn_n_u64(uint64x2_t a) {
 // CIR:      cir.call_llvm_intrinsic "aarch64.neon.uqshrn" {{.*}} : (!cir.vector<2 x !s64i>, !s32i) -> !cir.vector<2 x !u32i>
-
 // LLVM-SAME: <2 x i64> {{.*}} [[A:%.*]]) {{.*}} {
 // LLVM:      [[R:%.*]] = call <2 x i32> @llvm.aarch64.neon.uqshrn.v2i32(<2 x i64> {{.*}}, i32 19)
 // LLVM-NEXT: ret <2 x i32> [[R]]
   return vqshrn_n_u64(a, 19);
 }
+
+// LLVM-LABEL: @test_vqshrun_n_s16(
+// CIR-LABEL: @test_vqshrun_n_s16(
+uint8x8_t test_vqshrun_n_s16(int16x8_t a) {
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<8 x !s16i>, !s32i) -> !cir.vector<8 x !u8i>
+// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <8 x i8> @llvm.aarch64.neon.sqshrun.v8i8(<8 x i16> {{.*}}, i32 3)
+// LLVM-NEXT: ret <8 x i8> [[R]]
+  return vqshrun_n_s16(a, 3);
+}
+
+// LLVM-LABEL: @test_vqshrun_n_s32(
+// CIR-LABEL: @test_vqshrun_n_s32(
+uint16x4_t test_vqshrun_n_s32(int32x4_t a) {
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<4 x !s32i>, !s32i) -> !cir.vector<4 x !u16i>
+// LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <4 x i16> @llvm.aarch64.neon.sqshrun.v4i16(<4 x i32> {{.*}}, i32 9)
+// LLVM-NEXT: ret <4 x i16> [[R]]
+  return vqshrun_n_s32(a, 9);
+}
+
+// LLVM-LABEL: @test_vqshrun_n_s64(
+// CIR-LABEL: @test_vqshrun_n_s64(
+uint32x2_t test_vqshrun_n_s64(int64x2_t a) {
+// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<2 x !s64i>, !s32i) -> !cir.vector<2 x !u32i>
+// LLVM-SAME: <2 x i64> {{.*}} [[A:%.*]]) {{.*}} {
+// LLVM:      [[R:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrun.v2i32(<2 x i64> {{.*}}, i32 19)
+// LLVM-NEXT: ret <2 x i32> [[R]]
+  return vqshrun_n_s64(a, 19);
+}
diff --git a/clang/test/CodeGen/AArch64/neon/vqshrun.c b/clang/test/CodeGen/AArch64/neon/vqshrun.c
deleted file mode 100644
index 04cb61df1987e..0000000000000
--- a/clang/test/CodeGen/AArch64/neon/vqshrun.c
+++ /dev/null
@@ -1,40 +0,0 @@
-// REQUIRES: aarch64-registered-target || arm-registered-target
-
-// RUN:                   %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none           -emit-llvm -o - %s | opt -S -passes=mem2reg,sroa,simplifycfg | FileCheck %s --check-prefixes=LLVM
-// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none -fclangir -emit-llvm -o - %s | opt -S -passes=mem2reg,sroa,simplifycfg | FileCheck %s --check-prefixes=LLVM %}
-// RUN: %if cir-enabled %{%clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon -disable-O0-optnone -flax-vector-conversions=none -fclangir -emit-cir  -o - %s |                               FileCheck %s --check-prefixes=CIR %}
-
-#include <arm_neon.h>
-
-// LLVM-LABEL: @test_vqshrun_n_s16(
-// CIR-LABEL: @test_vqshrun_n_s16(
-uint8x8_t test_vqshrun_n_s16(int16x8_t a) {
-// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<8 x !s16i>, !s32i) -> !cir.vector<8 x !u8i>
-
-// LLVM-SAME: <8 x i16> {{.*}} [[A:%.*]]) {{.*}} {
-// LLVM:      [[R:%.*]] = call <8 x i8> @llvm.aarch64.neon.sqshrun.v8i8(<8 x i16> {{.*}}, i32 3)
-// LLVM-NEXT: ret <8 x i8> [[R]]
-  return vqshrun_n_s16(a, 3);
-}
-
-// LLVM-LABEL: @test_vqshrun_n_s32(
-// CIR-LABEL: @test_vqshrun_n_s32(
-uint16x4_t test_vqshrun_n_s32(int32x4_t a) {
-// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<4 x !s32i>, !s32i) -> !cir.vector<4 x !u16i>
-
-// LLVM-SAME: <4 x i32> {{.*}} [[A:%.*]]) {{.*}} {
-// LLVM:      [[R:%.*]] = call <4 x i16> @llvm.aarch64.neon.sqshrun.v4i16(<4 x i32> {{.*}}, i32 9)
-// LLVM-NEXT: ret <4 x i16> [[R]]
-  return vqshrun_n_s32(a, 9);
-}
-
-// LLVM-LABEL: @test_vqshrun_n_s64(
-// CIR-LABEL: @test_vqshrun_n_s64(
-uint32x2_t test_vqshrun_n_s64(int64x2_t a) {
-// CIR:      cir.call_llvm_intrinsic "aarch64.neon.sqshrun" {{.*}} : (!cir.vector<2 x !s64i>, !s32i) -> !cir.vector<2 x !u32i>
-
-// LLVM-SAME: <2 x i64> {{.*}} [[A:%.*]]) {{.*}} {
-// LLVM:      [[R:%.*]] = call <2 x i32> @llvm.aarch64.neon.sqshrun.v2i32(<2 x i64> {{.*}}, i32 19)
-// LLVM-NEXT: ret <2 x i32> [[R]]
-  return vqshrun_n_s64(a, 19);
-}



More information about the cfe-commits mailing list