156 lines · c
1// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py2// RUN: %clang_cc1 -triple thumbv8.1m.main-none-none-eabi -target-feature +mve.fp -mfloat-abi hard -O0 -disable-O0-optnone -emit-llvm -o - %s | opt -S -passes='mem2reg,sroa,early-cse<>' | FileCheck %s3// RUN: %clang_cc1 -triple thumbv8.1m.main-none-none-eabi -target-feature +mve.fp -mfloat-abi hard -O0 -disable-O0-optnone -DPOLYMORPHIC -emit-llvm -o - %s | opt -S -passes='mem2reg,sroa,early-cse<>' | FileCheck %s4 5// REQUIRES: aarch64-registered-target || arm-registered-target6 7#include <arm_mve.h>8 9// CHECK-LABEL: @test_vld2q_f16(10// CHECK-NEXT: entry:11// CHECK-NEXT: [[TMP0:%.*]] = call { <8 x half>, <8 x half> } @llvm.arm.mve.vld2q.v8f16.p0(ptr [[ADDR:%.*]])12// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <8 x half>, <8 x half> } [[TMP0]], 013// CHECK-NEXT: [[TMP2:%.*]] = insertvalue [[STRUCT_FLOAT16X8X2_T:%.*]] poison, <8 x half> [[TMP1]], 0, 014// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <8 x half>, <8 x half> } [[TMP0]], 115// CHECK-NEXT: [[TMP4:%.*]] = insertvalue [[STRUCT_FLOAT16X8X2_T]] [[TMP2]], <8 x half> [[TMP3]], 0, 116// CHECK-NEXT: ret [[STRUCT_FLOAT16X8X2_T]] [[TMP4]]17//18float16x8x2_t test_vld2q_f16(const float16_t *addr)19{20#ifdef POLYMORPHIC21 return vld2q(addr);22#else /* POLYMORPHIC */23 return vld2q_f16(addr);24#endif /* POLYMORPHIC */25}26 27// CHECK-LABEL: @test_vld4q_u8(28// CHECK-NEXT: entry:29// CHECK-NEXT: [[TMP0:%.*]] = call { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } @llvm.arm.mve.vld4q.v16i8.p0(ptr [[ADDR:%.*]])30// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP0]], 031// CHECK-NEXT: [[TMP2:%.*]] = insertvalue [[STRUCT_UINT8X16X4_T:%.*]] poison, <16 x i8> [[TMP1]], 0, 032// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP0]], 133// CHECK-NEXT: [[TMP4:%.*]] = insertvalue [[STRUCT_UINT8X16X4_T]] [[TMP2]], <16 x i8> [[TMP3]], 0, 134// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP0]], 235// CHECK-NEXT: [[TMP6:%.*]] = insertvalue [[STRUCT_UINT8X16X4_T]] [[TMP4]], <16 x i8> [[TMP5]], 0, 236// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP0]], 337// CHECK-NEXT: [[TMP8:%.*]] = insertvalue [[STRUCT_UINT8X16X4_T]] [[TMP6]], <16 x i8> [[TMP7]], 0, 338// CHECK-NEXT: ret [[STRUCT_UINT8X16X4_T]] [[TMP8]]39//40uint8x16x4_t test_vld4q_u8(const uint8_t *addr)41{42#ifdef POLYMORPHIC43 return vld4q(addr);44#else /* POLYMORPHIC */45 return vld4q_u8(addr);46#endif /* POLYMORPHIC */47}48 49// CHECK-LABEL: @test_vst2q_u32(50// CHECK-NEXT: entry:51// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_UINT32X4X2_T:%.*]] [[VALUE_COERCE:%.*]], 052// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <4 x i32>] [[TMP0]], 053// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <4 x i32>] [[TMP0]], 154// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_UINT32X4X2_T]] poison, <4 x i32> [[DOTFCA_0_EXTRACT]], 0, 055// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_UINT32X4X2_T]] [[DOTFCA_0_0_INSERT]], <4 x i32> [[DOTFCA_1_EXTRACT]], 0, 156// CHECK-NEXT: call void @llvm.arm.mve.vst2q.p0.v4i32(ptr [[ADDR:%.*]], <4 x i32> [[DOTFCA_0_EXTRACT]], <4 x i32> [[DOTFCA_1_EXTRACT]], i32 0)57// CHECK-NEXT: call void @llvm.arm.mve.vst2q.p0.v4i32(ptr [[ADDR]], <4 x i32> [[DOTFCA_0_EXTRACT]], <4 x i32> [[DOTFCA_1_EXTRACT]], i32 1)58// CHECK-NEXT: ret void59//60void test_vst2q_u32(uint32_t *addr, uint32x4x2_t value)61{62#ifdef POLYMORPHIC63 vst2q(addr, value);64#else /* POLYMORPHIC */65 vst2q_u32(addr, value);66#endif /* POLYMORPHIC */67}68 69// CHECK-LABEL: @test_vst4q_s8(70// CHECK-NEXT: entry:71// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_INT8X16X4_T:%.*]] [[VALUE_COERCE:%.*]], 072// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [4 x <16 x i8>] [[TMP0]], 073// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [4 x <16 x i8>] [[TMP0]], 174// CHECK-NEXT: [[DOTFCA_2_EXTRACT:%.*]] = extractvalue [4 x <16 x i8>] [[TMP0]], 275// CHECK-NEXT: [[DOTFCA_3_EXTRACT:%.*]] = extractvalue [4 x <16 x i8>] [[TMP0]], 376// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_INT8X16X4_T]] poison, <16 x i8> [[DOTFCA_0_EXTRACT]], 0, 077// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_INT8X16X4_T]] [[DOTFCA_0_0_INSERT]], <16 x i8> [[DOTFCA_1_EXTRACT]], 0, 178// CHECK-NEXT: [[DOTFCA_0_2_INSERT:%.*]] = insertvalue [[STRUCT_INT8X16X4_T]] [[DOTFCA_0_1_INSERT]], <16 x i8> [[DOTFCA_2_EXTRACT]], 0, 279// CHECK-NEXT: [[DOTFCA_0_3_INSERT:%.*]] = insertvalue [[STRUCT_INT8X16X4_T]] [[DOTFCA_0_2_INSERT]], <16 x i8> [[DOTFCA_3_EXTRACT]], 0, 380// CHECK-NEXT: call void @llvm.arm.mve.vst4q.p0.v16i8(ptr [[ADDR:%.*]], <16 x i8> [[DOTFCA_0_EXTRACT]], <16 x i8> [[DOTFCA_1_EXTRACT]], <16 x i8> [[DOTFCA_2_EXTRACT]], <16 x i8> [[DOTFCA_3_EXTRACT]], i32 0)81// CHECK-NEXT: call void @llvm.arm.mve.vst4q.p0.v16i8(ptr [[ADDR]], <16 x i8> [[DOTFCA_0_EXTRACT]], <16 x i8> [[DOTFCA_1_EXTRACT]], <16 x i8> [[DOTFCA_2_EXTRACT]], <16 x i8> [[DOTFCA_3_EXTRACT]], i32 1)82// CHECK-NEXT: call void @llvm.arm.mve.vst4q.p0.v16i8(ptr [[ADDR]], <16 x i8> [[DOTFCA_0_EXTRACT]], <16 x i8> [[DOTFCA_1_EXTRACT]], <16 x i8> [[DOTFCA_2_EXTRACT]], <16 x i8> [[DOTFCA_3_EXTRACT]], i32 2)83// CHECK-NEXT: call void @llvm.arm.mve.vst4q.p0.v16i8(ptr [[ADDR]], <16 x i8> [[DOTFCA_0_EXTRACT]], <16 x i8> [[DOTFCA_1_EXTRACT]], <16 x i8> [[DOTFCA_2_EXTRACT]], <16 x i8> [[DOTFCA_3_EXTRACT]], i32 3)84// CHECK-NEXT: ret void85//86void test_vst4q_s8(int8_t *addr, int8x16x4_t value)87{88#ifdef POLYMORPHIC89 vst4q(addr, value);90#else /* POLYMORPHIC */91 vst4q_s8(addr, value);92#endif /* POLYMORPHIC */93}94 95// CHECK-LABEL: @test_vst2q_f16(96// CHECK-NEXT: entry:97// CHECK-NEXT: [[TMP0:%.*]] = extractvalue [[STRUCT_FLOAT16X8X2_T:%.*]] [[VALUE_COERCE:%.*]], 098// CHECK-NEXT: [[DOTFCA_0_EXTRACT:%.*]] = extractvalue [2 x <8 x half>] [[TMP0]], 099// CHECK-NEXT: [[DOTFCA_1_EXTRACT:%.*]] = extractvalue [2 x <8 x half>] [[TMP0]], 1100// CHECK-NEXT: [[DOTFCA_0_0_INSERT:%.*]] = insertvalue [[STRUCT_FLOAT16X8X2_T]] poison, <8 x half> [[DOTFCA_0_EXTRACT]], 0, 0101// CHECK-NEXT: [[DOTFCA_0_1_INSERT:%.*]] = insertvalue [[STRUCT_FLOAT16X8X2_T]] [[DOTFCA_0_0_INSERT]], <8 x half> [[DOTFCA_1_EXTRACT]], 0, 1102// CHECK-NEXT: call void @llvm.arm.mve.vst2q.p0.v8f16(ptr [[ADDR:%.*]], <8 x half> [[DOTFCA_0_EXTRACT]], <8 x half> [[DOTFCA_1_EXTRACT]], i32 0)103// CHECK-NEXT: call void @llvm.arm.mve.vst2q.p0.v8f16(ptr [[ADDR]], <8 x half> [[DOTFCA_0_EXTRACT]], <8 x half> [[DOTFCA_1_EXTRACT]], i32 1)104// CHECK-NEXT: ret void105//106void test_vst2q_f16(float16_t *addr, float16x8x2_t value)107{108#ifdef POLYMORPHIC109 vst2q(addr, value);110#else /* POLYMORPHIC */111 vst2q_f16(addr, value);112#endif /* POLYMORPHIC */113}114 115// CHECK-LABEL: @load_into_variable(116// CHECK-NEXT: entry:117// CHECK-NEXT: [[TMP0:%.*]] = call { <8 x i16>, <8 x i16> } @llvm.arm.mve.vld2q.v8i16.p0(ptr [[ADDR:%.*]])118// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <8 x i16>, <8 x i16> } [[TMP0]], 0119// CHECK-NEXT: [[TMP2:%.*]] = insertvalue [[STRUCT_UINT16X8X2_T:%.*]] poison, <8 x i16> [[TMP1]], 0, 0120// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <8 x i16>, <8 x i16> } [[TMP0]], 1121// CHECK-NEXT: [[TMP4:%.*]] = insertvalue [[STRUCT_UINT16X8X2_T]] [[TMP2]], <8 x i16> [[TMP3]], 0, 1122// CHECK-NEXT: store <8 x i16> [[TMP1]], ptr [[VALUES:%.*]], align 8123// CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds <8 x i16>, ptr [[VALUES]], i32 1124// CHECK-NEXT: store <8 x i16> [[TMP3]], ptr [[ARRAYIDX4]], align 8125// CHECK-NEXT: ret void126//127void load_into_variable(const uint16_t *addr, uint16x8_t *values)128{129 uint16x8x2_t v;130#ifdef POLYMORPHIC131 v = vld2q(addr);132#else /* POLYMORPHIC */133 v = vld2q_u16(addr);134#endif /* POLYMORPHIC */135 values[0] = v.val[0];136 values[1] = v.val[1];137}138 139// CHECK-LABEL: @extract_one_vector(140// CHECK-NEXT: entry:141// CHECK-NEXT: [[TMP0:%.*]] = call { <4 x i32>, <4 x i32> } @llvm.arm.mve.vld2q.v4i32.p0(ptr [[ADDR:%.*]])142// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <4 x i32>, <4 x i32> } [[TMP0]], 0143// CHECK-NEXT: [[TMP2:%.*]] = insertvalue [[STRUCT_INT32X4X2_T:%.*]] poison, <4 x i32> [[TMP1]], 0, 0144// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <4 x i32>, <4 x i32> } [[TMP0]], 1145// CHECK-NEXT: [[TMP4:%.*]] = insertvalue [[STRUCT_INT32X4X2_T]] [[TMP2]], <4 x i32> [[TMP3]], 0, 1146// CHECK-NEXT: ret <4 x i32> [[TMP1]]147//148int32x4_t extract_one_vector(const int32_t *addr)149{150#ifdef POLYMORPHIC151 return vld2q(addr).val[0];152#else /* POLYMORPHIC */153 return vld2q_s32(addr).val[0];154#endif /* POLYMORPHIC */155}156