[CIR][CIRGen][Builtin][Neon] Lower neon_vext_v and neon_vextq_v (llvm#951)

ghehg · lanza · commit f781328346ba · 2025-08-10T11:12:18.000-07:00
as title. There are two highlights of the PR 1. The PR introduced a new test file to cover neon intrinsics that move data, which is a big category. This would the 5th neon test file. And we're committed to keep total number of neon test files within 6. This file uses another opt option instcombine, which makes test LLVM code more concise, and our -fclangir generated LLVM code would be identical to OG with this. It looks like OG did some instcombine optimization. 2. `getIntFromMLIRValue` helper function could be substituted by [`mlir::cir::IntAttr getConstOpIntAttr` in CIRGenAtomic.cpp](https://github.com/llvm/clangir/blob/24b24557c98d1c031572a567b658cfb6254f8a89/clang/lib/CIR/CodeGen/CIRGenAtomic.cpp#L337). The function `mlir::cir::IntAttr getConstOpIntAttr` is doing more than `getIntFromMLIRValue`, and there is FIXME in the comment, so not sure if we should just use `mlir::cir::IntAttr getConstOpIntAttr`, either is fine with me.
diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -2195,6 +2195,14 @@ mlir::Value buildNeonCall(unsigned int builtinID, CIRGenFunction &cgf,
   }
 }
 
+/// Get integer from a mlir::Value that is an int constant or a constant op.
+static int64_t getIntValueFromConstOp(mlir::Value val) {
+  auto constOp = mlir::cast<mlir::cir::ConstantOp>(val.getDefiningOp());
+  return (mlir::cast<mlir::cir::IntAttr>(constOp.getValue()))
+      .getValue()
+      .getSExtValue();
+}
+
 mlir::Value CIRGenFunction::buildCommonNeonBuiltinExpr(
     unsigned builtinID, unsigned llvmIntrinsic, unsigned altLLVMIntrinsic,
     const char *nameHint, unsigned modifier, const CallExpr *e,
@@ -2239,6 +2247,18 @@ mlir::Value CIRGenFunction::buildCommonNeonBuiltinExpr(
     // In CIR, integral cast op supports vector of int type truncating.
     return builder.createIntCast(ops[0], ty);
   }
+  case NEON::BI__builtin_neon_vext_v:
+  case NEON::BI__builtin_neon_vextq_v: {
+    int cv = getIntValueFromConstOp(ops[2]);
+    llvm::SmallVector<int64_t, 16> indices;
+    for (unsigned i = 0, e = vTy.getSize(); i != e; ++i)
+      indices.push_back(i + cv);
+
+    ops[0] = builder.createBitcast(ops[0], ty);
+    ops[1] = builder.createBitcast(ops[1], ty);
+    return builder.createVecShuffle(getLoc(e->getExprLoc()), ops[0], ops[1],
+                                    indices);
+  }
   }
 
   // This second switch is for the intrinsics that might have a more generic
diff --git a/clang/test/CIR/CodeGen/AArch64/neon-ext-mov.c b/clang/test/CIR/CodeGen/AArch64/neon-ext-mov.c
@@ -0,0 +1,216 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-android24 -target-feature +neon \
+// RUN:    -fclangir -disable-O0-optnone  -fno-clangir-call-conv-lowering \
+// RUN:  -flax-vector-conversions=none -emit-cir -o %t.cir %s
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+// XFAIL: *
+
+// RUN: %clang_cc1 -triple aarch64-none-linux-android24 -target-feature +neon \
+// RUN:    -fclangir -disable-O0-optnone -fno-clangir-call-conv-lowering \
+// RUN:  -flax-vector-conversions=none -emit-llvm -o - %s \
+// RUN: | opt -S -passes=instcombine,mem2reg,simplifycfg -o %t.ll
+// RUN: FileCheck --check-prefix=LLVM --input-file=%t.ll %s
+
+// REQUIRES: aarch64-registered-target || arm-registered-target
+
+// This test file contains test cases for the intrinsics that move data between
+// registers and vectors, such as mov, get, set, and ext. We dedicate this file 
+// to them becuase they are many. The file neon.c covers some such intrinsics 
+// that are not in this file.  
+
+#include <arm_neon.h>
+
+int8x8_t test_vext_s8(int8x8_t a, int8x8_t b) {
+  return vext_s8(a, b, 2);
+
+  // CIR-LABEL: vext_s8
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s8i x 8>) 
+  // CIR-SAME: [#cir.int<2> : !s32i, #cir.int<3> : !s32i, #cir.int<4> : !s32i, 
+  // CIR-SAME:  #cir.int<5> : !s32i, #cir.int<6> : !s32i, #cir.int<7> : !s32i, 
+  // CIR-SAME:  #cir.int<8> : !s32i, #cir.int<9> : !s32i] : !cir.vector<!s8i x 8>
+
+  // LLVM: {{.*}}test_vext_s8(<8 x i8>{{.*}}[[A:%.*]], <8 x i8>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> [[B]], 
+  // LLVM-SAME: <8 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9>
+  // LLVM: ret <8 x i8> [[RES]]
+}
+
+int8x16_t test_vextq_s8(int8x16_t a, int8x16_t b) {
+  return vextq_s8(a, b, 2);
+
+  // CIR-LABEL: vextq_s8
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s8i x 16>) 
+  // CIR-SAME: [#cir.int<2> : !s32i, #cir.int<3> : !s32i, #cir.int<4> : !s32i, 
+  // CIR-SAME:  #cir.int<5> : !s32i, #cir.int<6> : !s32i, #cir.int<7> : !s32i, 
+  // CIR-SAME:  #cir.int<8> : !s32i, #cir.int<9> : !s32i, #cir.int<10> : !s32i, 
+  // CIR-SAME:  #cir.int<11> : !s32i, #cir.int<12> : !s32i, #cir.int<13> : !s32i,
+  // CIR-SAME:  #cir.int<14> : !s32i, #cir.int<15> : !s32i, #cir.int<16> : !s32i,  
+  // CIR-SAME:  #cir.int<17> : !s32i] : !cir.vector<!s8i x 16>
+
+  // LLVM: {{.*}}test_vextq_s8(<16 x i8>{{.*}}[[A:%.*]], <16 x i8>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <16 x i8> [[A]], <16 x i8> [[B]], 
+  // LLVM-SAME: <16 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9
+  // LLVM-SAME: i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17>
+  // LLVM: ret <16 x i8> [[RES]]
+}
+
+int16x4_t test_vext_s16(int16x4_t a, int16x4_t b) {
+  return vext_s16(a, b, 3);
+
+  // CIR-LABEL: vext_s16
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s16i x 4>) 
+  // CIR-SAME: [#cir.int<3> : !s32i, #cir.int<4> : !s32i, #cir.int<5> : !s32i,
+  // CIR-SAME:  #cir.int<6> : !s32i] : !cir.vector<!s16i x 4>
+
+  // LLVM: {{.*}}test_vext_s16(<4 x i16>{{.*}}[[A:%.*]], <4 x i16>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], 
+  // LLVM-SAME: <4 x i32> <i32 3, i32 4, i32 5, i32 6>
+  // LLVM: ret <4 x i16> [[RES]]
+}
+
+int16x8_t test_vextq_s16(int16x8_t a, int16x8_t b) {
+  return vextq_s16(a, b, 3);
+
+  // CIR-LABEL: vextq_s16
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s16i x 8>) 
+  // CIR-SAME: [#cir.int<3> : !s32i, #cir.int<4> : !s32i, #cir.int<5> : !s32i, 
+  // CIR-SAME:  #cir.int<6> : !s32i, #cir.int<7> : !s32i, #cir.int<8> : !s32i, 
+  // CIR-SAME:  #cir.int<9> : !s32i, #cir.int<10> : !s32i] : !cir.vector<!s16i x 8>
+
+  // LLVM: {{.*}}test_vextq_s16(<8 x i16>{{.*}}[[A:%.*]], <8 x i16>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], 
+  // LLVM-SAME: <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>
+  // LLVM: ret <8 x i16> [[RES]]
+}
+
+
+uint16x4_t test_vext_u16(uint16x4_t a, uint16x4_t b) {
+  return vext_u16(a, b, 3);
+
+  // CIR-LABEL: vext_u16
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!u16i x 4>) 
+  // CIR-SAME: [#cir.int<3> : !s32i, #cir.int<4> : !s32i, #cir.int<5> : !s32i,
+  // CIR-SAME:  #cir.int<6> : !s32i] : !cir.vector<!u16i x 4>
+
+  // LLVM: {{.*}}test_vext_u16(<4 x i16>{{.*}}[[A:%.*]], <4 x i16>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <4 x i16> [[A]], <4 x i16> [[B]], 
+  // LLVM-SAME: <4 x i32> <i32 3, i32 4, i32 5, i32 6>
+  // LLVM: ret <4 x i16> [[RES]]
+}
+
+uint16x8_t test_vextq_u16(uint16x8_t a, uint16x8_t b) {
+  return vextq_u16(a, b, 3);
+
+  // CIR-LABEL: vextq_u16
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!u16i x 8>) 
+  // CIR-SAME: [#cir.int<3> : !s32i, #cir.int<4> : !s32i, #cir.int<5> : !s32i, 
+  // CIR-SAME:  #cir.int<6> : !s32i, #cir.int<7> : !s32i, #cir.int<8> : !s32i, 
+  // CIR-SAME:  #cir.int<9> : !s32i, #cir.int<10> : !s32i] : !cir.vector<!u16i x 8>
+
+  // LLVM: {{.*}}test_vextq_u16(<8 x i16>{{.*}}[[A:%.*]], <8 x i16>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <8 x i16> [[A]], <8 x i16> [[B]], 
+  // LLVM-SAME: <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>
+  // LLVM: ret <8 x i16> [[RES]]
+}
+
+int32x2_t test_vext_s32(int32x2_t a, int32x2_t b) {
+  return vext_s32(a, b, 1);
+
+  // CIR-LABEL: vext_s32
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s32i x 2>) 
+  // CIR-SAME: [#cir.int<1> : !s32i, #cir.int<2> : !s32i] : !cir.vector<!s32i x 2>
+
+  // LLVM: {{.*}}test_vext_s32(<2 x i32>{{.*}}[[A:%.*]], <2 x i32>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> [[B]], 
+  // LLVM-SAME: <2 x i32> <i32 1, i32 2>
+  // LLVM: ret <2 x i32> [[RES]]
+}
+
+int32x4_t test_vextq_s32(int32x4_t a, int32x4_t b) {
+  return vextq_s32(a, b, 1);
+
+  // CIR-LABEL: vextq_s32
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s32i x 4>) 
+  // CIR-SAME: [#cir.int<1> : !s32i, #cir.int<2> : !s32i,
+  // CIR-SAME:  #cir.int<3> : !s32i, #cir.int<4> : !s32i] : !cir.vector<!s32i x 4>
+
+  // LLVM: {{.*}}test_vextq_s32(<4 x i32>{{.*}}[[A:%.*]], <4 x i32>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> [[B]], 
+  // LLVM-SAME: <4 x i32> <i32 1, i32 2, i32 3, i32 4>
+  // LLVM: ret <4 x i32> [[RES]]
+}
+
+int64x1_t test_vext_s64(int64x1_t a, int64x1_t b) {
+  return vext_s64(a, b, 0);
+  
+  // CIR-LABEL: vext_s64
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s64i x 1>) 
+  // CIR-SAME: [#cir.int<0> : !s32i] : !cir.vector<!s64i x 1>
+
+  // LLVM: {{.*}}test_vext_s64(<1 x i64>{{.*}}[[A:%.*]], <1 x i64>{{.*}}[[B:%.*]])
+  // LLVM: ret <1 x i64> [[A]]
+}
+
+int64x2_t test_vextq_s64(int64x2_t a, int64x2_t b) {
+  return vextq_s64(a, b, 1);
+
+  // CIR-LABEL: vextq_s64
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!s64i x 2>) 
+  // CIR-SAME: [#cir.int<1> : !s32i, #cir.int<2> : !s32i] : !cir.vector<!s64i x 2>
+
+  // LLVM: {{.*}}test_vextq_s64(<2 x i64>{{.*}}[[A:%.*]], <2 x i64>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <2 x i64> [[A]], <2 x i64> [[B]], 
+  // LLVM-SAME: <2 x i32> <i32 1, i32 2>
+  // LLVM: ret <2 x i64> [[RES]]
+}
+
+float32x2_t test_vext_f32(float32x2_t a, float32x2_t b) {
+  return vext_f32(a, b, 1);
+
+  // CIR-LABEL: vext_f32
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!cir.float x 2>) 
+  // CIR-SAME: [#cir.int<1> : !s32i, #cir.int<2> : !s32i] : !cir.vector<!cir.float x 2>
+
+  // LLVM: {{.*}}test_vext_f32(<2 x float>{{.*}}[[A:%.*]], <2 x float>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <2 x float> [[A]], <2 x float> [[B]], 
+  // LLVM-SAME: <2 x i32> <i32 1, i32 2>
+  // LLVM: ret <2 x float> [[RES]]
+}
+
+float32x4_t test_vextq_f32(float32x4_t a, float32x4_t b) {
+  return vextq_f32(a, b, 1);
+
+  // CIR-LABEL: vextq_f32
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!cir.float x 4>) 
+  // CIR-SAME: [#cir.int<1> : !s32i, #cir.int<2> : !s32i, #cir.int<3> : !s32i, 
+  // CIR-SAME:  #cir.int<4> : !s32i] : !cir.vector<!cir.float x 4>
+
+  // LLVM: {{.*}}test_vextq_f32(<4 x float>{{.*}}[[A:%.*]], <4 x float>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <4 x float> [[A]], <4 x float> [[B]], 
+  // LLVM-SAME: <4 x i32> <i32 1, i32 2, i32 3, i32 4>
+  // LLVM: ret <4 x float> [[RES]]
+}
+
+
+float64x1_t test_vext_f64(float64x1_t a, float64x1_t b) {
+  return vext_f64(a, b, 0);
+  
+  // CIR-LABEL: vext_f64
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!cir.double x 1>) 
+  // CIR-SAME: [#cir.int<0> : !s32i] : !cir.vector<!cir.double x 1>
+
+  // LLVM: {{.*}}test_vext_f64(<1 x double>{{.*}}[[A:%.*]], <1 x double>{{.*}}[[B:%.*]])
+  // LLVM: ret <1 x double> [[A]]
+}
+
+float64x2_t test_vextq_f64(float64x2_t a, float64x2_t b) {
+  return vextq_f64(a, b, 1);
+
+  // CIR-LABEL: vextq_f64
+  // CIR: {{%.*}}= cir.vec.shuffle({{%.*}}, {{%.*}} : !cir.vector<!cir.double x 2>) 
+  // CIR-SAME: [#cir.int<1> : !s32i, #cir.int<2> : !s32i] : !cir.vector<!cir.double x 2>
+
+  // LLVM: {{.*}}test_vextq_f64(<2 x double>{{.*}}[[A:%.*]], <2 x double>{{.*}}[[B:%.*]])
+  // LLVM: [[RES:%.*]] = shufflevector <2 x double> [[A]], <2 x double> [[B]], 
+  // LLVM-SAME: <2 x i32> <i32 1, i32 2>
+  // LLVM: ret <2 x double> [[RES]]
+}
