279 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 32; RUN: llc -mtriple=aarch64 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD3; RUN: llc -mtriple=aarch64 -global-isel -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI4 5define double @copysign_f64(double %a, double %b) {6; CHECK-LABEL: copysign_f64:7; CHECK: // %bb.0: // %entry8; CHECK-NEXT: movi v2.2d, #0xffffffffffffffff9; CHECK-NEXT: // kill: def $d0 killed $d0 def $q010; CHECK-NEXT: // kill: def $d1 killed $d1 def $q111; CHECK-NEXT: fneg v2.2d, v2.2d12; CHECK-NEXT: bif v0.16b, v1.16b, v2.16b13; CHECK-NEXT: // kill: def $d0 killed $d0 killed $q014; CHECK-NEXT: ret15entry:16 %c = call double @llvm.copysign.f64(double %a, double %b)17 ret double %c18}19 20define float @copysign_f32(float %a, float %b) {21; CHECK-SD-LABEL: copysign_f32:22; CHECK-SD: // %bb.0: // %entry23; CHECK-SD-NEXT: mvni v2.4s, #128, lsl #2424; CHECK-SD-NEXT: // kill: def $s0 killed $s0 def $q025; CHECK-SD-NEXT: // kill: def $s1 killed $s1 def $q126; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b27; CHECK-SD-NEXT: // kill: def $s0 killed $s0 killed $q028; CHECK-SD-NEXT: ret29;30; CHECK-GI-LABEL: copysign_f32:31; CHECK-GI: // %bb.0: // %entry32; CHECK-GI-NEXT: mvni v2.2s, #128, lsl #2433; CHECK-GI-NEXT: // kill: def $s0 killed $s0 def $d034; CHECK-GI-NEXT: // kill: def $s1 killed $s1 def $d135; CHECK-GI-NEXT: bif v0.8b, v1.8b, v2.8b36; CHECK-GI-NEXT: // kill: def $s0 killed $s0 killed $d037; CHECK-GI-NEXT: ret38entry:39 %c = call float @llvm.copysign.f32(float %a, float %b)40 ret float %c41}42 43define half @copysign_f16(half %a, half %b) {44; CHECK-SD-LABEL: copysign_f16:45; CHECK-SD: // %bb.0: // %entry46; CHECK-SD-NEXT: fcvt s1, h147; CHECK-SD-NEXT: fcvt s0, h048; CHECK-SD-NEXT: mvni v2.4s, #128, lsl #2449; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b50; CHECK-SD-NEXT: fcvt h0, s051; CHECK-SD-NEXT: ret52;53; CHECK-GI-LABEL: copysign_f16:54; CHECK-GI: // %bb.0: // %entry55; CHECK-GI-NEXT: mvni v2.4h, #128, lsl #856; CHECK-GI-NEXT: // kill: def $h0 killed $h0 def $d057; CHECK-GI-NEXT: // kill: def $h1 killed $h1 def $d158; CHECK-GI-NEXT: bif v0.8b, v1.8b, v2.8b59; CHECK-GI-NEXT: // kill: def $h0 killed $h0 killed $d060; CHECK-GI-NEXT: ret61entry:62 %c = call half @llvm.copysign.f16(half %a, half %b)63 ret half %c64}65 66define <2 x double> @copysign_v2f64(<2 x double> %a, <2 x double> %b) {67; CHECK-LABEL: copysign_v2f64:68; CHECK: // %bb.0: // %entry69; CHECK-NEXT: movi v2.2d, #0xffffffffffffffff70; CHECK-NEXT: fneg v2.2d, v2.2d71; CHECK-NEXT: bif v0.16b, v1.16b, v2.16b72; CHECK-NEXT: ret73entry:74 %c = call <2 x double> @llvm.copysign.v2f64(<2 x double> %a, <2 x double> %b)75 ret <2 x double> %c76}77 78define <3 x double> @copysign_v3f64(<3 x double> %a, <3 x double> %b) {79; CHECK-SD-LABEL: copysign_v3f64:80; CHECK-SD: // %bb.0: // %entry81; CHECK-SD-NEXT: movi v6.2d, #0xffffffffffffffff82; CHECK-SD-NEXT: // kill: def $d3 killed $d3 def $q383; CHECK-SD-NEXT: // kill: def $d1 killed $d1 def $q184; CHECK-SD-NEXT: // kill: def $d0 killed $d0 def $q085; CHECK-SD-NEXT: // kill: def $d4 killed $d4 def $q486; CHECK-SD-NEXT: // kill: def $d2 killed $d2 def $q287; CHECK-SD-NEXT: // kill: def $d5 killed $d5 def $q588; CHECK-SD-NEXT: mov v3.d[1], v4.d[0]89; CHECK-SD-NEXT: mov v0.d[1], v1.d[0]90; CHECK-SD-NEXT: fneg v1.2d, v6.2d91; CHECK-SD-NEXT: bif v0.16b, v3.16b, v1.16b92; CHECK-SD-NEXT: bif v2.16b, v5.16b, v1.16b93; CHECK-SD-NEXT: // kill: def $d2 killed $d2 killed $q294; CHECK-SD-NEXT: ext v1.16b, v0.16b, v0.16b, #895; CHECK-SD-NEXT: // kill: def $d0 killed $d0 killed $q096; CHECK-SD-NEXT: // kill: def $d1 killed $d1 killed $q197; CHECK-SD-NEXT: ret98;99; CHECK-GI-LABEL: copysign_v3f64:100; CHECK-GI: // %bb.0: // %entry101; CHECK-GI-NEXT: movi v6.2d, #0xffffffffffffffff102; CHECK-GI-NEXT: // kill: def $d0 killed $d0 def $q0103; CHECK-GI-NEXT: // kill: def $d1 killed $d1 def $q1104; CHECK-GI-NEXT: // kill: def $d3 killed $d3 def $q3105; CHECK-GI-NEXT: // kill: def $d4 killed $d4 def $q4106; CHECK-GI-NEXT: fmov x8, d2107; CHECK-GI-NEXT: fmov x9, d5108; CHECK-GI-NEXT: mov v0.d[1], v1.d[0]109; CHECK-GI-NEXT: mov v3.d[1], v4.d[0]110; CHECK-GI-NEXT: and x8, x8, #0x7fffffffffffffff111; CHECK-GI-NEXT: and x9, x9, #0x8000000000000000112; CHECK-GI-NEXT: fneg v1.2d, v6.2d113; CHECK-GI-NEXT: orr x8, x8, x9114; CHECK-GI-NEXT: fmov d2, x8115; CHECK-GI-NEXT: bif v0.16b, v3.16b, v1.16b116; CHECK-GI-NEXT: mov d1, v0.d[1]117; CHECK-GI-NEXT: // kill: def $d0 killed $d0 killed $q0118; CHECK-GI-NEXT: ret119entry:120 %c = call <3 x double> @llvm.copysign.v3f64(<3 x double> %a, <3 x double> %b)121 ret <3 x double> %c122}123 124define <4 x double> @copysign_v4f64(<4 x double> %a, <4 x double> %b) {125; CHECK-LABEL: copysign_v4f64:126; CHECK: // %bb.0: // %entry127; CHECK-NEXT: movi v4.2d, #0xffffffffffffffff128; CHECK-NEXT: fneg v4.2d, v4.2d129; CHECK-NEXT: bif v0.16b, v2.16b, v4.16b130; CHECK-NEXT: bif v1.16b, v3.16b, v4.16b131; CHECK-NEXT: ret132entry:133 %c = call <4 x double> @llvm.copysign.v4f64(<4 x double> %a, <4 x double> %b)134 ret <4 x double> %c135}136 137define <2 x float> @copysign_v2f32(<2 x float> %a, <2 x float> %b) {138; CHECK-LABEL: copysign_v2f32:139; CHECK: // %bb.0: // %entry140; CHECK-NEXT: mvni v2.2s, #128, lsl #24141; CHECK-NEXT: bif v0.8b, v1.8b, v2.8b142; CHECK-NEXT: ret143entry:144 %c = call <2 x float> @llvm.copysign.v2f32(<2 x float> %a, <2 x float> %b)145 ret <2 x float> %c146}147 148define <3 x float> @copysign_v3f32(<3 x float> %a, <3 x float> %b) {149; CHECK-SD-LABEL: copysign_v3f32:150; CHECK-SD: // %bb.0: // %entry151; CHECK-SD-NEXT: mvni v2.4s, #128, lsl #24152; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b153; CHECK-SD-NEXT: ret154;155; CHECK-GI-LABEL: copysign_v3f32:156; CHECK-GI: // %bb.0: // %entry157; CHECK-GI-NEXT: mov w8, #-2147483648 // =0x80000000158; CHECK-GI-NEXT: mov w9, #2147483647 // =0x7fffffff159; CHECK-GI-NEXT: fmov s2, w9160; CHECK-GI-NEXT: fmov s3, w8161; CHECK-GI-NEXT: mov v2.s[1], w9162; CHECK-GI-NEXT: mov v3.s[1], w8163; CHECK-GI-NEXT: mov v2.s[2], w9164; CHECK-GI-NEXT: mov v3.s[2], w8165; CHECK-GI-NEXT: and v0.16b, v0.16b, v2.16b166; CHECK-GI-NEXT: and v1.16b, v1.16b, v3.16b167; CHECK-GI-NEXT: orr v0.16b, v0.16b, v1.16b168; CHECK-GI-NEXT: ret169entry:170 %c = call <3 x float> @llvm.copysign.v3f32(<3 x float> %a, <3 x float> %b)171 ret <3 x float> %c172}173 174define <4 x float> @copysign_v4f32(<4 x float> %a, <4 x float> %b) {175; CHECK-LABEL: copysign_v4f32:176; CHECK: // %bb.0: // %entry177; CHECK-NEXT: mvni v2.4s, #128, lsl #24178; CHECK-NEXT: bif v0.16b, v1.16b, v2.16b179; CHECK-NEXT: ret180entry:181 %c = call <4 x float> @llvm.copysign.v4f32(<4 x float> %a, <4 x float> %b)182 ret <4 x float> %c183}184 185define <8 x float> @copysign_v8f32(<8 x float> %a, <8 x float> %b) {186; CHECK-LABEL: copysign_v8f32:187; CHECK: // %bb.0: // %entry188; CHECK-NEXT: mvni v4.4s, #128, lsl #24189; CHECK-NEXT: bif v0.16b, v2.16b, v4.16b190; CHECK-NEXT: bif v1.16b, v3.16b, v4.16b191; CHECK-NEXT: ret192entry:193 %c = call <8 x float> @llvm.copysign.v8f32(<8 x float> %a, <8 x float> %b)194 ret <8 x float> %c195}196 197define <7 x half> @copysign_v7f16(<7 x half> %a, <7 x half> %b) {198; CHECK-SD-LABEL: copysign_v7f16:199; CHECK-SD: // %bb.0: // %entry200; CHECK-SD-NEXT: mvni v2.8h, #128, lsl #8201; CHECK-SD-NEXT: bif v0.16b, v1.16b, v2.16b202; CHECK-SD-NEXT: ret203;204; CHECK-GI-LABEL: copysign_v7f16:205; CHECK-GI: // %bb.0: // %entry206; CHECK-GI-NEXT: mov w8, #32768 // =0x8000207; CHECK-GI-NEXT: mov w9, #32767 // =0x7fff208; CHECK-GI-NEXT: fmov s2, w9209; CHECK-GI-NEXT: fmov s3, w8210; CHECK-GI-NEXT: mov v2.h[1], w9211; CHECK-GI-NEXT: mov v3.h[1], w8212; CHECK-GI-NEXT: mov v2.h[2], w9213; CHECK-GI-NEXT: mov v3.h[2], w8214; CHECK-GI-NEXT: mov v2.h[3], w9215; CHECK-GI-NEXT: mov v3.h[3], w8216; CHECK-GI-NEXT: mov v2.h[4], w9217; CHECK-GI-NEXT: mov v3.h[4], w8218; CHECK-GI-NEXT: mov v2.h[5], w9219; CHECK-GI-NEXT: mov v3.h[5], w8220; CHECK-GI-NEXT: mov v2.h[6], w9221; CHECK-GI-NEXT: mov v3.h[6], w8222; CHECK-GI-NEXT: and v0.16b, v0.16b, v2.16b223; CHECK-GI-NEXT: and v1.16b, v1.16b, v3.16b224; CHECK-GI-NEXT: orr v0.16b, v0.16b, v1.16b225; CHECK-GI-NEXT: ret226entry:227 %c = call <7 x half> @llvm.copysign.v7f16(<7 x half> %a, <7 x half> %b)228 ret <7 x half> %c229}230 231define <4 x half> @copysign_v4f16(<4 x half> %a, <4 x half> %b) {232; CHECK-LABEL: copysign_v4f16:233; CHECK: // %bb.0: // %entry234; CHECK-NEXT: mvni v2.4h, #128, lsl #8235; CHECK-NEXT: bif v0.8b, v1.8b, v2.8b236; CHECK-NEXT: ret237entry:238 %c = call <4 x half> @llvm.copysign.v4f16(<4 x half> %a, <4 x half> %b)239 ret <4 x half> %c240}241 242define <8 x half> @copysign_v8f16(<8 x half> %a, <8 x half> %b) {243; CHECK-LABEL: copysign_v8f16:244; CHECK: // %bb.0: // %entry245; CHECK-NEXT: mvni v2.8h, #128, lsl #8246; CHECK-NEXT: bif v0.16b, v1.16b, v2.16b247; CHECK-NEXT: ret248entry:249 %c = call <8 x half> @llvm.copysign.v8f16(<8 x half> %a, <8 x half> %b)250 ret <8 x half> %c251}252 253define <16 x half> @copysign_v16f16(<16 x half> %a, <16 x half> %b) {254; CHECK-LABEL: copysign_v16f16:255; CHECK: // %bb.0: // %entry256; CHECK-NEXT: mvni v4.8h, #128, lsl #8257; CHECK-NEXT: bif v0.16b, v2.16b, v4.16b258; CHECK-NEXT: bif v1.16b, v3.16b, v4.16b259; CHECK-NEXT: ret260entry:261 %c = call <16 x half> @llvm.copysign.v16f16(<16 x half> %a, <16 x half> %b)262 ret <16 x half> %c263}264 265declare <16 x half> @llvm.copysign.v16f16(<16 x half>, <16 x half>)266declare <2 x double> @llvm.copysign.v2f64(<2 x double>, <2 x double>)267declare <2 x float> @llvm.copysign.v2f32(<2 x float>, <2 x float>)268declare <3 x double> @llvm.copysign.v3f64(<3 x double>, <3 x double>)269declare <3 x float> @llvm.copysign.v3f32(<3 x float>, <3 x float>)270declare <4 x double> @llvm.copysign.v4f64(<4 x double>, <4 x double>)271declare <4 x float> @llvm.copysign.v4f32(<4 x float>, <4 x float>)272declare <4 x half> @llvm.copysign.v4f16(<4 x half>, <4 x half>)273declare <7 x half> @llvm.copysign.v7f16(<7 x half>, <7 x half>)274declare <8 x float> @llvm.copysign.v8f32(<8 x float>, <8 x float>)275declare <8 x half> @llvm.copysign.v8f16(<8 x half>, <8 x half>)276declare double @llvm.copysign.f64(double, double)277declare float @llvm.copysign.f32(float, float)278declare half @llvm.copysign.f16(half, half)279