148 lines · plain
1; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - | FileCheck %s2 3%struct.__neon_int8x8x2_t = type { <8 x i8>, <8 x i8> }4%struct.__neon_int16x4x2_t = type { <4 x i16>, <4 x i16> }5%struct.__neon_int32x2x2_t = type { <2 x i32>, <2 x i32> }6%struct.__neon_float32x2x2_t = type { <2 x float>, <2 x float> }7%struct.__neon_int64x1x2_t = type { <1 x i64>, <1 x i64> }8 9%struct.__neon_int8x16x2_t = type { <16 x i8>, <16 x i8> }10%struct.__neon_int16x8x2_t = type { <8 x i16>, <8 x i16> }11%struct.__neon_int32x4x2_t = type { <4 x i32>, <4 x i32> }12%struct.__neon_float32x4x2_t = type { <4 x float>, <4 x float> }13 14define <8 x i8> @vld2i8(ptr %A) nounwind {15;CHECK-LABEL: vld2i8:16;Check the alignment value. Max for this instruction is 128 bits:17;CHECK: vld2.8 {d16, d17}, [{{r[0-9]+|lr}}:64]18 %tmp1 = call %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2.v8i8.p0(ptr %A, i32 8)19 %tmp2 = extractvalue %struct.__neon_int8x8x2_t %tmp1, 020 %tmp3 = extractvalue %struct.__neon_int8x8x2_t %tmp1, 121 %tmp4 = add <8 x i8> %tmp2, %tmp322 ret <8 x i8> %tmp423}24 25define <4 x i16> @vld2i16(ptr %A) nounwind {26;CHECK-LABEL: vld2i16:27;Check the alignment value. Max for this instruction is 128 bits:28;CHECK: vld2.16 {d16, d17}, [{{r[0-9]+|lr}}:128]29 %tmp1 = call %struct.__neon_int16x4x2_t @llvm.arm.neon.vld2.v4i16.p0(ptr %A, i32 32)30 %tmp2 = extractvalue %struct.__neon_int16x4x2_t %tmp1, 031 %tmp3 = extractvalue %struct.__neon_int16x4x2_t %tmp1, 132 %tmp4 = add <4 x i16> %tmp2, %tmp333 ret <4 x i16> %tmp434}35 36define <2 x i32> @vld2i32(ptr %A) nounwind {37;CHECK-LABEL: vld2i32:38;CHECK: vld2.3239 %tmp1 = call %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2.v2i32.p0(ptr %A, i32 1)40 %tmp2 = extractvalue %struct.__neon_int32x2x2_t %tmp1, 041 %tmp3 = extractvalue %struct.__neon_int32x2x2_t %tmp1, 142 %tmp4 = add <2 x i32> %tmp2, %tmp343 ret <2 x i32> %tmp444}45 46define <2 x float> @vld2f(ptr %A) nounwind {47;CHECK-LABEL: vld2f:48;CHECK: vld2.3249 %tmp1 = call %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2.v2f32.p0(ptr %A, i32 1)50 %tmp2 = extractvalue %struct.__neon_float32x2x2_t %tmp1, 051 %tmp3 = extractvalue %struct.__neon_float32x2x2_t %tmp1, 152 %tmp4 = fadd <2 x float> %tmp2, %tmp353 ret <2 x float> %tmp454}55 56;Check for a post-increment updating load. 57define <2 x float> @vld2f_update(ptr %ptr) nounwind {58;CHECK-LABEL: vld2f_update:59;CHECK: vld2.32 {d16, d17}, [{{r[0-9]+|lr}}]!60 %A = load ptr, ptr %ptr61 %tmp1 = call %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2.v2f32.p0(ptr %A, i32 1)62 %tmp2 = extractvalue %struct.__neon_float32x2x2_t %tmp1, 063 %tmp3 = extractvalue %struct.__neon_float32x2x2_t %tmp1, 164 %tmp4 = fadd <2 x float> %tmp2, %tmp365 %tmp5 = getelementptr float, ptr %A, i32 466 store ptr %tmp5, ptr %ptr67 ret <2 x float> %tmp468}69 70define <1 x i64> @vld2i64(ptr %A) nounwind {71;CHECK-LABEL: vld2i64:72;Check the alignment value. Max for this instruction is 128 bits:73;CHECK: vld1.64 {d16, d17}, [{{r[0-9]+|lr}}:128]74 %tmp1 = call %struct.__neon_int64x1x2_t @llvm.arm.neon.vld2.v1i64.p0(ptr %A, i32 32)75 %tmp2 = extractvalue %struct.__neon_int64x1x2_t %tmp1, 076 %tmp3 = extractvalue %struct.__neon_int64x1x2_t %tmp1, 177 %tmp4 = add <1 x i64> %tmp2, %tmp378 ret <1 x i64> %tmp479}80 81define <16 x i8> @vld2Qi8(ptr %A) nounwind {82;CHECK-LABEL: vld2Qi8:83;Check the alignment value. Max for this instruction is 256 bits:84;CHECK: vld2.8 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:64]85 %tmp1 = call %struct.__neon_int8x16x2_t @llvm.arm.neon.vld2.v16i8.p0(ptr %A, i32 8)86 %tmp2 = extractvalue %struct.__neon_int8x16x2_t %tmp1, 087 %tmp3 = extractvalue %struct.__neon_int8x16x2_t %tmp1, 188 %tmp4 = add <16 x i8> %tmp2, %tmp389 ret <16 x i8> %tmp490}91 92;Check for a post-increment updating load with register increment.93define <16 x i8> @vld2Qi8_update(ptr %ptr, i32 %inc) nounwind {94;CHECK-LABEL: vld2Qi8_update:95;CHECK: vld2.8 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:128], r196 %A = load ptr, ptr %ptr97 %tmp1 = call %struct.__neon_int8x16x2_t @llvm.arm.neon.vld2.v16i8.p0(ptr %A, i32 16)98 %tmp2 = extractvalue %struct.__neon_int8x16x2_t %tmp1, 099 %tmp3 = extractvalue %struct.__neon_int8x16x2_t %tmp1, 1100 %tmp4 = add <16 x i8> %tmp2, %tmp3101 %tmp5 = getelementptr i8, ptr %A, i32 %inc102 store ptr %tmp5, ptr %ptr103 ret <16 x i8> %tmp4104}105 106define <8 x i16> @vld2Qi16(ptr %A) nounwind {107;CHECK-LABEL: vld2Qi16:108;Check the alignment value. Max for this instruction is 256 bits:109;CHECK: vld2.16 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:128]110 %tmp1 = call %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2.v8i16.p0(ptr %A, i32 16)111 %tmp2 = extractvalue %struct.__neon_int16x8x2_t %tmp1, 0112 %tmp3 = extractvalue %struct.__neon_int16x8x2_t %tmp1, 1113 %tmp4 = add <8 x i16> %tmp2, %tmp3114 ret <8 x i16> %tmp4115}116 117define <4 x i32> @vld2Qi32(ptr %A) nounwind {118;CHECK-LABEL: vld2Qi32:119;Check the alignment value. Max for this instruction is 256 bits:120;CHECK: vld2.32 {d16, d17, d18, d19}, [{{r[0-9]+|lr}}:256]121 %tmp1 = call %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0(ptr %A, i32 64)122 %tmp2 = extractvalue %struct.__neon_int32x4x2_t %tmp1, 0123 %tmp3 = extractvalue %struct.__neon_int32x4x2_t %tmp1, 1124 %tmp4 = add <4 x i32> %tmp2, %tmp3125 ret <4 x i32> %tmp4126}127 128define <4 x float> @vld2Qf(ptr %A) nounwind {129;CHECK-LABEL: vld2Qf:130;CHECK: vld2.32131 %tmp1 = call %struct.__neon_float32x4x2_t @llvm.arm.neon.vld2.v4f32.p0(ptr %A, i32 1)132 %tmp2 = extractvalue %struct.__neon_float32x4x2_t %tmp1, 0133 %tmp3 = extractvalue %struct.__neon_float32x4x2_t %tmp1, 1134 %tmp4 = fadd <4 x float> %tmp2, %tmp3135 ret <4 x float> %tmp4136}137 138declare %struct.__neon_int8x8x2_t @llvm.arm.neon.vld2.v8i8.p0(ptr, i32) nounwind readonly139declare %struct.__neon_int16x4x2_t @llvm.arm.neon.vld2.v4i16.p0(ptr, i32) nounwind readonly140declare %struct.__neon_int32x2x2_t @llvm.arm.neon.vld2.v2i32.p0(ptr, i32) nounwind readonly141declare %struct.__neon_float32x2x2_t @llvm.arm.neon.vld2.v2f32.p0(ptr, i32) nounwind readonly142declare %struct.__neon_int64x1x2_t @llvm.arm.neon.vld2.v1i64.p0(ptr, i32) nounwind readonly143 144declare %struct.__neon_int8x16x2_t @llvm.arm.neon.vld2.v16i8.p0(ptr, i32) nounwind readonly145declare %struct.__neon_int16x8x2_t @llvm.arm.neon.vld2.v8i16.p0(ptr, i32) nounwind readonly146declare %struct.__neon_int32x4x2_t @llvm.arm.neon.vld2.v4i32.p0(ptr, i32) nounwind readonly147declare %struct.__neon_float32x4x2_t @llvm.arm.neon.vld2.v4f32.p0(ptr, i32) nounwind readonly148