1 // REQUIRES: aarch64-registered-target 2 // RUN: %clang_cc1 -triple arm64-none-linux-gnu -target-feature +neon \ 3 // RUN: -ffp-contract=fast -emit-llvm -o - %s | opt -S -mem2reg \ 4 // RUN: | FileCheck %s 5 6 // Test new aarch64 intrinsics with poly128 7 // FIXME: Currently, poly128_t equals to uint128, which will be spilt into 8 // two 64-bit GPR(eg X0, X1). Now moving data from X0, X1 to FPR128 will 9 // introduce 2 store and 1 load instructions(store X0, X1 to memory and 10 // then load back to Q0). If target has NEON, this is better replaced by 11 // FMOV or INS. 12 13 #include <arm_neon.h> 14 15 // CHECK-LABEL: define void @test_vstrq_p128(i128* %ptr, i128 %val) #0 { 16 // CHECK: [[TMP0:%.*]] = bitcast i128* %ptr to i8* 17 // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to i128* 18 // CHECK: store i128 %val, i128* [[TMP1]] 19 // CHECK: ret void 20 void test_vstrq_p128(poly128_t * ptr, poly128_t val) { 21 vstrq_p128(ptr, val); 22 23 } 24 25 // CHECK-LABEL: define i128 @test_vldrq_p128(i128* %ptr) #0 { 26 // CHECK: [[TMP0:%.*]] = bitcast i128* %ptr to i8* 27 // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to i128* 28 // CHECK: [[TMP2:%.*]] = load i128, i128* [[TMP1]] 29 // CHECK: ret i128 [[TMP2]] 30 poly128_t test_vldrq_p128(poly128_t * ptr) { 31 return vldrq_p128(ptr); 32 33 } 34 35 // CHECK-LABEL: define void @test_ld_st_p128(i128* %ptr) #0 { 36 // CHECK: [[TMP0:%.*]] = bitcast i128* %ptr to i8* 37 // CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to i128* 38 // CHECK: [[TMP2:%.*]] = load i128, i128* [[TMP1]] 39 // CHECK: [[ADD_PTR:%.*]] = getelementptr inbounds i128, i128* %ptr, i64 1 40 // CHECK: [[TMP3:%.*]] = bitcast i128* [[ADD_PTR]] to i8* 41 // CHECK: [[TMP4:%.*]] = bitcast i8* [[TMP3]] to i128* 42 // CHECK: store i128 [[TMP2]], i128* [[TMP4]] 43 // CHECK: ret void 44 void test_ld_st_p128(poly128_t * ptr) { 45 vstrq_p128(ptr+1, vldrq_p128(ptr)); 46 47 } 48 49 // CHECK-LABEL: define i128 @test_vmull_p64(i64 %a, i64 %b) #0 { 50 // CHECK: [[VMULL_P64_I:%.*]] = call <16 x i8> @llvm.aarch64.neon.pmull64(i64 %a, i64 %b) #2 51 // CHECK: [[VMULL_P641_I:%.*]] = bitcast <16 x i8> [[VMULL_P64_I]] to i128 52 // CHECK: ret i128 [[VMULL_P641_I]] 53 poly128_t test_vmull_p64(poly64_t a, poly64_t b) { 54 return vmull_p64(a, b); 55 } 56 57 // CHECK-LABEL: define i128 @test_vmull_high_p64(<2 x i64> %a, <2 x i64> %b) #0 { 58 // CHECK: [[SHUFFLE_I_I:%.*]] = shufflevector <2 x i64> %a, <2 x i64> %a, <1 x i32> <i32 1> 59 // CHECK: [[TMP0:%.*]] = bitcast <1 x i64> [[SHUFFLE_I_I]] to i64 60 // CHECK: [[SHUFFLE_I7_I:%.*]] = shufflevector <2 x i64> %b, <2 x i64> %b, <1 x i32> <i32 1> 61 // CHECK: [[TMP1:%.*]] = bitcast <1 x i64> [[SHUFFLE_I7_I]] to i64 62 // CHECK: [[VMULL_P64_I_I:%.*]] = call <16 x i8> @llvm.aarch64.neon.pmull64(i64 [[TMP0]], i64 [[TMP1]]) #2 63 // CHECK: [[VMULL_P641_I_I:%.*]] = bitcast <16 x i8> [[VMULL_P64_I_I]] to i128 64 // CHECK: ret i128 [[VMULL_P641_I_I]] 65 poly128_t test_vmull_high_p64(poly64x2_t a, poly64x2_t b) { 66 return vmull_high_p64(a, b); 67 } 68 69 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_s8(<16 x i8> %a) #0 { 70 // CHECK: [[TMP0:%.*]] = bitcast <16 x i8> %a to i128 71 // CHECK: ret i128 [[TMP0]] 72 poly128_t test_vreinterpretq_p128_s8(int8x16_t a) { 73 return vreinterpretq_p128_s8(a); 74 } 75 76 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_s16(<8 x i16> %a) #0 { 77 // CHECK: [[TMP0:%.*]] = bitcast <8 x i16> %a to i128 78 // CHECK: ret i128 [[TMP0]] 79 poly128_t test_vreinterpretq_p128_s16(int16x8_t a) { 80 return vreinterpretq_p128_s16(a); 81 } 82 83 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_s32(<4 x i32> %a) #0 { 84 // CHECK: [[TMP0:%.*]] = bitcast <4 x i32> %a to i128 85 // CHECK: ret i128 [[TMP0]] 86 poly128_t test_vreinterpretq_p128_s32(int32x4_t a) { 87 return vreinterpretq_p128_s32(a); 88 } 89 90 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_s64(<2 x i64> %a) #0 { 91 // CHECK: [[TMP0:%.*]] = bitcast <2 x i64> %a to i128 92 // CHECK: ret i128 [[TMP0]] 93 poly128_t test_vreinterpretq_p128_s64(int64x2_t a) { 94 return vreinterpretq_p128_s64(a); 95 } 96 97 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_u8(<16 x i8> %a) #0 { 98 // CHECK: [[TMP0:%.*]] = bitcast <16 x i8> %a to i128 99 // CHECK: ret i128 [[TMP0]] 100 poly128_t test_vreinterpretq_p128_u8(uint8x16_t a) { 101 return vreinterpretq_p128_u8(a); 102 } 103 104 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_u16(<8 x i16> %a) #0 { 105 // CHECK: [[TMP0:%.*]] = bitcast <8 x i16> %a to i128 106 // CHECK: ret i128 [[TMP0]] 107 poly128_t test_vreinterpretq_p128_u16(uint16x8_t a) { 108 return vreinterpretq_p128_u16(a); 109 } 110 111 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_u32(<4 x i32> %a) #0 { 112 // CHECK: [[TMP0:%.*]] = bitcast <4 x i32> %a to i128 113 // CHECK: ret i128 [[TMP0]] 114 poly128_t test_vreinterpretq_p128_u32(uint32x4_t a) { 115 return vreinterpretq_p128_u32(a); 116 } 117 118 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_u64(<2 x i64> %a) #0 { 119 // CHECK: [[TMP0:%.*]] = bitcast <2 x i64> %a to i128 120 // CHECK: ret i128 [[TMP0]] 121 poly128_t test_vreinterpretq_p128_u64(uint64x2_t a) { 122 return vreinterpretq_p128_u64(a); 123 } 124 125 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_f32(<4 x float> %a) #0 { 126 // CHECK: [[TMP0:%.*]] = bitcast <4 x float> %a to i128 127 // CHECK: ret i128 [[TMP0]] 128 poly128_t test_vreinterpretq_p128_f32(float32x4_t a) { 129 return vreinterpretq_p128_f32(a); 130 } 131 132 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_f64(<2 x double> %a) #0 { 133 // CHECK: [[TMP0:%.*]] = bitcast <2 x double> %a to i128 134 // CHECK: ret i128 [[TMP0]] 135 poly128_t test_vreinterpretq_p128_f64(float64x2_t a) { 136 return vreinterpretq_p128_f64(a); 137 } 138 139 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_p8(<16 x i8> %a) #0 { 140 // CHECK: [[TMP0:%.*]] = bitcast <16 x i8> %a to i128 141 // CHECK: ret i128 [[TMP0]] 142 poly128_t test_vreinterpretq_p128_p8(poly8x16_t a) { 143 return vreinterpretq_p128_p8(a); 144 } 145 146 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_p16(<8 x i16> %a) #0 { 147 // CHECK: [[TMP0:%.*]] = bitcast <8 x i16> %a to i128 148 // CHECK: ret i128 [[TMP0]] 149 poly128_t test_vreinterpretq_p128_p16(poly16x8_t a) { 150 return vreinterpretq_p128_p16(a); 151 } 152 153 // CHECK-LABEL: define i128 @test_vreinterpretq_p128_p64(<2 x i64> %a) #0 { 154 // CHECK: [[TMP0:%.*]] = bitcast <2 x i64> %a to i128 155 // CHECK: ret i128 [[TMP0]] 156 poly128_t test_vreinterpretq_p128_p64(poly64x2_t a) { 157 return vreinterpretq_p128_p64(a); 158 } 159 160 // CHECK-LABEL: define <16 x i8> @test_vreinterpretq_s8_p128(i128 %a) #0 { 161 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <16 x i8> 162 // CHECK: ret <16 x i8> [[TMP0]] 163 int8x16_t test_vreinterpretq_s8_p128(poly128_t a) { 164 return vreinterpretq_s8_p128(a); 165 } 166 167 // CHECK-LABEL: define <8 x i16> @test_vreinterpretq_s16_p128(i128 %a) #0 { 168 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <8 x i16> 169 // CHECK: ret <8 x i16> [[TMP0]] 170 int16x8_t test_vreinterpretq_s16_p128(poly128_t a) { 171 return vreinterpretq_s16_p128(a); 172 } 173 174 // CHECK-LABEL: define <4 x i32> @test_vreinterpretq_s32_p128(i128 %a) #0 { 175 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <4 x i32> 176 // CHECK: ret <4 x i32> [[TMP0]] 177 int32x4_t test_vreinterpretq_s32_p128(poly128_t a) { 178 return vreinterpretq_s32_p128(a); 179 } 180 181 // CHECK-LABEL: define <2 x i64> @test_vreinterpretq_s64_p128(i128 %a) #0 { 182 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <2 x i64> 183 // CHECK: ret <2 x i64> [[TMP0]] 184 int64x2_t test_vreinterpretq_s64_p128(poly128_t a) { 185 return vreinterpretq_s64_p128(a); 186 } 187 188 // CHECK-LABEL: define <16 x i8> @test_vreinterpretq_u8_p128(i128 %a) #0 { 189 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <16 x i8> 190 // CHECK: ret <16 x i8> [[TMP0]] 191 uint8x16_t test_vreinterpretq_u8_p128(poly128_t a) { 192 return vreinterpretq_u8_p128(a); 193 } 194 195 // CHECK-LABEL: define <8 x i16> @test_vreinterpretq_u16_p128(i128 %a) #0 { 196 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <8 x i16> 197 // CHECK: ret <8 x i16> [[TMP0]] 198 uint16x8_t test_vreinterpretq_u16_p128(poly128_t a) { 199 return vreinterpretq_u16_p128(a); 200 } 201 202 // CHECK-LABEL: define <4 x i32> @test_vreinterpretq_u32_p128(i128 %a) #0 { 203 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <4 x i32> 204 // CHECK: ret <4 x i32> [[TMP0]] 205 uint32x4_t test_vreinterpretq_u32_p128(poly128_t a) { 206 return vreinterpretq_u32_p128(a); 207 } 208 209 // CHECK-LABEL: define <2 x i64> @test_vreinterpretq_u64_p128(i128 %a) #0 { 210 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <2 x i64> 211 // CHECK: ret <2 x i64> [[TMP0]] 212 uint64x2_t test_vreinterpretq_u64_p128(poly128_t a) { 213 return vreinterpretq_u64_p128(a); 214 } 215 216 // CHECK-LABEL: define <4 x float> @test_vreinterpretq_f32_p128(i128 %a) #0 { 217 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <4 x float> 218 // CHECK: ret <4 x float> [[TMP0]] 219 float32x4_t test_vreinterpretq_f32_p128(poly128_t a) { 220 return vreinterpretq_f32_p128(a); 221 } 222 223 // CHECK-LABEL: define <2 x double> @test_vreinterpretq_f64_p128(i128 %a) #0 { 224 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <2 x double> 225 // CHECK: ret <2 x double> [[TMP0]] 226 float64x2_t test_vreinterpretq_f64_p128(poly128_t a) { 227 return vreinterpretq_f64_p128(a); 228 } 229 230 // CHECK-LABEL: define <16 x i8> @test_vreinterpretq_p8_p128(i128 %a) #0 { 231 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <16 x i8> 232 // CHECK: ret <16 x i8> [[TMP0]] 233 poly8x16_t test_vreinterpretq_p8_p128(poly128_t a) { 234 return vreinterpretq_p8_p128(a); 235 } 236 237 // CHECK-LABEL: define <8 x i16> @test_vreinterpretq_p16_p128(i128 %a) #0 { 238 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <8 x i16> 239 // CHECK: ret <8 x i16> [[TMP0]] 240 poly16x8_t test_vreinterpretq_p16_p128(poly128_t a) { 241 return vreinterpretq_p16_p128(a); 242 } 243 244 // CHECK-LABEL: define <2 x i64> @test_vreinterpretq_p64_p128(i128 %a) #0 { 245 // CHECK: [[TMP0:%.*]] = bitcast i128 %a to <2 x i64> 246 // CHECK: ret <2 x i64> [[TMP0]] 247 poly64x2_t test_vreinterpretq_p64_p128(poly128_t a) { 248 return vreinterpretq_p64_p128(a); 249 } 250 251 252
