[CIR][CIRGen][Builtin][Neon] Lower vld1_dup and vld1q_dup (llvm#936)

keryell · Oct 19, 2024 · 90a9be6 · 90a9be6
1 parent 666dbcf
commit 90a9be6
Show file tree

Hide file tree

Showing 2 changed files with 213 additions and 1 deletion.
diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp
@@ -3451,7 +3451,11 @@ CIRGenFunction::buildAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
   }
   case NEON::BI__builtin_neon_vld1_dup_v:
   case NEON::BI__builtin_neon_vld1q_dup_v: {
-    llvm_unreachable("NYI");
+    cir::Address ptrAddr = PtrOp0.withElementType(vTy.getEltType());
+    mlir::Value val = builder.createLoad(getLoc(E->getExprLoc()), ptrAddr);
+    mlir::cir::VecSplatOp vecSplat = builder.create<mlir::cir::VecSplatOp>(
+        getLoc(E->getExprLoc()), vTy, val);
+    return vecSplat;
   }
   case NEON::BI__builtin_neon_vst1_lane_v:
   case NEON::BI__builtin_neon_vst1q_lane_v: {

diff --git a/clang/test/CIR/CodeGen/AArch64/neon.c b/clang/test/CIR/CodeGen/AArch64/neon.c
@@ -17577,3 +17577,211 @@ int32x2_t test_vmovn_s64(int64x2_t a) {
   // LLVM: [[VMOVN_I:%.*]] = trunc <2 x i64> [[A]] to <2 x i32>
   // LLVM: ret <2 x i32> [[VMOVN_I]]
 }
+
+uint8x8_t test_vld1_dup_u8(uint8_t const * ptr) {
+  return vld1_dup_u8(ptr);
+}
+
+// CIR-LABEL: vld1_dup_u8
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!u8i>, !u8i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !u8i, !cir.vector<!u8i x 8>
+
+// LLVM: {{.*}}test_vld1_dup_u8(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i8, ptr [[PTR]], align 1
+// LLVM: [[VEC:%.*]] = insertelement <8 x i8> poison, i8 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <8 x i8> [[VEC]], <8 x i8> poison, <8 x i32> zeroinitializer
+
+int8x8_t test_vld1_dup_s8(int8_t const * ptr) {
+  return vld1_dup_s8(ptr);
+}
+
+// CIR-LABEL: test_vld1_dup_s8
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s8i>, !s8i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s8i, !cir.vector<!s8i x 8>
+
+// LLVM: {{.*}}test_vld1_dup_s8(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i8, ptr [[PTR]], align 1
+// LLVM: [[VEC:%.*]] = insertelement <8 x i8> poison, i8 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <8 x i8> [[VEC]], <8 x i8> poison, <8 x i32> zeroinitializer
+
+uint16x4_t test_vld1_dup_u16(uint16_t const * ptr) {
+  return vld1_dup_u16(ptr);
+}
+
+// CIR-LABEL: test_vld1_dup_u16
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!u16i>, !u16i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !u16i, !cir.vector<!u16i x 4>
+
+// LLVM: {{.*}}test_vld1_dup_u16(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i16, ptr [[PTR]], align 2
+// LLVM: [[VEC:%.*]] = insertelement <4 x i16> poison, i16 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <4 x i16> [[VEC]], <4 x i16> poison, <4 x i32> zeroinitializer
+
+int16x4_t test_vld1_dup_s16(int16_t const * ptr) {
+  return vld1_dup_s16(ptr);
+}
+
+// CIR-LABEL: test_vld1_dup_s16
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s16i>, !s16i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s16i, !cir.vector<!s16i x 4>
+
+// LLVM: {{.*}}test_vld1_dup_s16(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i16, ptr [[PTR]], align 2
+// LLVM: [[VEC:%.*]] = insertelement <4 x i16> poison, i16 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <4 x i16> [[VEC]], <4 x i16> poison, <4 x i32> zeroinitializer
+
+int32x2_t test_vld1_dup_s32(int32_t const * ptr) {
+  return vld1_dup_s32(ptr);
+}
+
+// CIR-LABEL: test_vld1_dup_s32
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s32i>, !s32i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s32i, !cir.vector<!s32i x 2>
+
+// LLVM: {{.*}}test_vld1_dup_s32(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i32, ptr [[PTR]], align 4
+// LLVM: [[VEC:%.*]] = insertelement <2 x i32> poison, i32 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <2 x i32> [[VEC]], <2 x i32> poison, <2 x i32> zeroinitializer
+
+int64x1_t test_vld1_dup_s64(int64_t const * ptr) {
+  return vld1_dup_s64(ptr);
+}
+
+// CIR-LABEL: test_vld1_dup_s64
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s64i>, !s64i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s64i, !cir.vector<!s64i x 1>
+
+// LLVM: {{.*}}test_vld1_dup_s64(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i64, ptr [[PTR]], align 8
+// LLVM: [[VEC:%.*]] = insertelement <1 x i64> poison, i64 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <1 x i64> [[VEC]], <1 x i64> poison, <1 x i32> zeroinitializer
+
+float32x2_t test_vld1_dup_f32(float32_t const * ptr) {
+  return vld1_dup_f32(ptr);
+}
+
+// CIR-LABEL: test_vld1_dup_f32
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!cir.float>, !cir.float
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !cir.float, !cir.vector<!cir.float x 2>
+
+// LLVM: {{.*}}test_vld1_dup_f32(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load float, ptr [[PTR]], align 4
+// LLVM: [[VEC:%.*]] = insertelement <2 x float> poison, float [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <2 x float> [[VEC]], <2 x float> poison, <2 x i32> zeroinitializer
+
+float64x1_t test_vld1_dup_f64(float64_t const * ptr) {
+  return vld1_dup_f64(ptr);
+}
+
+// CIR-LABEL: test_vld1_dup_f64
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!cir.double>, !cir.double
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !cir.double, !cir.vector<!cir.double x 1>
+
+// LLVM: {{.*}}test_vld1_dup_f64(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load double, ptr [[PTR]], align 8
+// LLVM: [[VEC:%.*]] = insertelement <1 x double> poison, double [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <1 x double> [[VEC]], <1 x double> poison, <1 x i32> zeroinitializer
+
+uint8x16_t test_vld1q_dup_u8(uint8_t const * ptr) {
+  return vld1q_dup_u8(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_u8
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!u8i>, !u8i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !u8i, !cir.vector<!u8i x 16>
+
+// LLVM: {{.*}}test_vld1q_dup_u8(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i8, ptr [[PTR]], align 1
+// LLVM: [[VEC:%.*]] = insertelement <16 x i8> poison, i8 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <16 x i8> [[VEC]], <16 x i8> poison, <16 x i32> zeroinitializer
+
+int8x16_t test_vld1q_dup_s8(int8_t const * ptr) {
+  return vld1q_dup_s8(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_s8
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s8i>, !s8i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s8i, !cir.vector<!s8i x 16>
+
+// LLVM: {{.*}}test_vld1q_dup_s8(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i8, ptr [[PTR]], align 1
+// LLVM: [[VEC:%.*]] = insertelement <16 x i8> poison, i8 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <16 x i8> [[VEC]], <16 x i8> poison, <16 x i32> zeroinitializer
+
+uint16x8_t test_vld1q_dup_u16(uint16_t const * ptr) {
+  return vld1q_dup_u16(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_u16
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!u16i>, !u16i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !u16i, !cir.vector<!u16i x 8>
+
+// LLVM: {{.*}}test_vld1q_dup_u16(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i16, ptr [[PTR]], align 2
+// LLVM: [[VEC:%.*]] = insertelement <8 x i16> poison, i16 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <8 x i16> [[VEC]], <8 x i16> poison, <8 x i32> zeroinitializer
+
+int16x8_t test_vld1q_dup_s16(int16_t const * ptr) {
+  return vld1q_dup_s16(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_s16
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s16i>, !s16i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s16i, !cir.vector<!s16i x 8>
+
+// LLVM: {{.*}}test_vld1q_dup_s16(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i16, ptr [[PTR]], align 2
+// LLVM: [[VEC:%.*]] = insertelement <8 x i16> poison, i16 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <8 x i16> [[VEC]], <8 x i16> poison, <8 x i32> zeroinitializer
+
+int32x4_t test_vld1q_dup_s32(int32_t const * ptr) {
+  return vld1q_dup_s32(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_s32
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s32i>, !s32i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s32i, !cir.vector<!s32i x 4>
+
+// LLVM: {{.*}}test_vld1q_dup_s32(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i32, ptr [[PTR]], align 4
+// LLVM: [[VEC:%.*]] = insertelement <4 x i32> poison, i32 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <4 x i32> [[VEC]], <4 x i32> poison, <4 x i32> zeroinitializer
+
+int64x2_t test_vld1q_dup_s64(int64_t const * ptr) {
+  return vld1q_dup_s64(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_s64
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!s64i>, !s64i
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !s64i, !cir.vector<!s64i x 2>
+
+// LLVM: {{.*}}test_vld1q_dup_s64(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load i64, ptr [[PTR]], align 8
+// LLVM: [[VEC:%.*]] = insertelement <2 x i64> poison, i64 [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <2 x i64> [[VEC]], <2 x i64> poison, <2 x i32> zeroinitializer
+
+float32x4_t test_vld1q_dup_f32(float32_t const * ptr) {
+  return vld1q_dup_f32(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_f32
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!cir.float>, !cir.float
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !cir.float, !cir.vector<!cir.float x 4>
+
+// LLVM: {{.*}}test_vld1q_dup_f32(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load float, ptr [[PTR]], align 4
+// LLVM: [[VEC:%.*]] = insertelement <4 x float> poison, float [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <4 x float> [[VEC]], <4 x float> poison, <4 x i32> zeroinitializer
+
+float64x2_t test_vld1q_dup_f64(float64_t const * ptr) {
+  return vld1q_dup_f64(ptr);
+}
+
+// CIR-LABEL: test_vld1q_dup_f64
+// CIR: [[VAL:%.*]] = cir.load {{%.*}} : !cir.ptr<!cir.double>, !cir.double
+// CIR: {{%.*}} = cir.vec.splat [[VAL]] : !cir.double, !cir.vector<!cir.double x 2>
+
+// LLVM: {{.*}}test_vld1q_dup_f64(ptr{{.*}}[[PTR:%.*]])
+// LLVM: [[VAL:%.*]] = load double, ptr [[PTR]], align 8
+// LLVM: [[VEC:%.*]] = insertelement <2 x double> poison, double [[VAL]], i64 0
+// LLVM: {{%.*}} = shufflevector <2 x double> [[VEC]], <2 x double> poison, <2 x i32> zeroinitializer