212 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc --mtriple=aarch64 -mattr=+fullfp16 < %s | FileCheck %s3; RUN: llc --mtriple=aarch64 < %s | FileCheck %s --check-prefix=CHECKNOFP164 5define half @faddp_2xhalf(<2 x half> %a) {6; CHECK-LABEL: faddp_2xhalf:7; CHECK: // %bb.0: // %entry8; CHECK-NEXT: // kill: def $d0 killed $d0 def $q09; CHECK-NEXT: faddp h0, v0.2h10; CHECK-NEXT: ret11;12; CHECKNOFP16-LABEL: faddp_2xhalf:13; CHECKNOFP16: // %bb.0: // %entry14; CHECKNOFP16-NEXT: // kill: def $d0 killed $d0 def $q015; CHECKNOFP16-NEXT: dup v1.4h, v0.h[1]16; CHECKNOFP16-NEXT: fcvtl v0.4s, v0.4h17; CHECKNOFP16-NEXT: fcvtl v1.4s, v1.4h18; CHECKNOFP16-NEXT: fadd v0.4s, v0.4s, v1.4s19; CHECKNOFP16-NEXT: fcvtn v0.4h, v0.4s20; CHECKNOFP16-NEXT: // kill: def $h0 killed $h0 killed $q021; CHECKNOFP16-NEXT: ret22entry:23 %shift = shufflevector <2 x half> %a, <2 x half> undef, <2 x i32> <i32 1, i32 undef>24 %0 = fadd <2 x half> %a, %shift25 %1 = extractelement <2 x half> %0, i32 026 ret half %127}28 29define half @faddp_2xhalf_commute(<2 x half> %a) {30; CHECK-LABEL: faddp_2xhalf_commute:31; CHECK: // %bb.0: // %entry32; CHECK-NEXT: // kill: def $d0 killed $d0 def $q033; CHECK-NEXT: faddp h0, v0.2h34; CHECK-NEXT: ret35;36; CHECKNOFP16-LABEL: faddp_2xhalf_commute:37; CHECKNOFP16: // %bb.0: // %entry38; CHECKNOFP16-NEXT: // kill: def $d0 killed $d0 def $q039; CHECKNOFP16-NEXT: dup v1.4h, v0.h[1]40; CHECKNOFP16-NEXT: fcvtl v0.4s, v0.4h41; CHECKNOFP16-NEXT: fcvtl v1.4s, v1.4h42; CHECKNOFP16-NEXT: fadd v0.4s, v1.4s, v0.4s43; CHECKNOFP16-NEXT: fcvtn v0.4h, v0.4s44; CHECKNOFP16-NEXT: // kill: def $h0 killed $h0 killed $q045; CHECKNOFP16-NEXT: ret46entry:47 %shift = shufflevector <2 x half> %a, <2 x half> undef, <2 x i32> <i32 1, i32 undef>48 %0 = fadd <2 x half> %shift, %a49 %1 = extractelement <2 x half> %0, i32 050 ret half %151}52 53define half @faddp_4xhalf(<4 x half> %a) {54; CHECK-LABEL: faddp_4xhalf:55; CHECK: // %bb.0: // %entry56; CHECK-NEXT: // kill: def $d0 killed $d0 def $q057; CHECK-NEXT: faddp h0, v0.2h58; CHECK-NEXT: ret59;60; CHECKNOFP16-LABEL: faddp_4xhalf:61; CHECKNOFP16: // %bb.0: // %entry62; CHECKNOFP16-NEXT: // kill: def $d0 killed $d0 def $q063; CHECKNOFP16-NEXT: dup v1.4h, v0.h[1]64; CHECKNOFP16-NEXT: fcvtl v0.4s, v0.4h65; CHECKNOFP16-NEXT: fcvtl v1.4s, v1.4h66; CHECKNOFP16-NEXT: fadd v0.4s, v0.4s, v1.4s67; CHECKNOFP16-NEXT: fcvtn v0.4h, v0.4s68; CHECKNOFP16-NEXT: // kill: def $h0 killed $h0 killed $q069; CHECKNOFP16-NEXT: ret70entry:71 %shift = shufflevector <4 x half> %a, <4 x half> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>72 %0 = fadd <4 x half> %a, %shift73 %1 = extractelement <4 x half> %0, i32 074 ret half %175}76 77define half @faddp_4xhalf_commute(<4 x half> %a) {78; CHECK-LABEL: faddp_4xhalf_commute:79; CHECK: // %bb.0: // %entry80; CHECK-NEXT: // kill: def $d0 killed $d0 def $q081; CHECK-NEXT: faddp h0, v0.2h82; CHECK-NEXT: ret83;84; CHECKNOFP16-LABEL: faddp_4xhalf_commute:85; CHECKNOFP16: // %bb.0: // %entry86; CHECKNOFP16-NEXT: // kill: def $d0 killed $d0 def $q087; CHECKNOFP16-NEXT: dup v1.4h, v0.h[1]88; CHECKNOFP16-NEXT: fcvtl v0.4s, v0.4h89; CHECKNOFP16-NEXT: fcvtl v1.4s, v1.4h90; CHECKNOFP16-NEXT: fadd v0.4s, v1.4s, v0.4s91; CHECKNOFP16-NEXT: fcvtn v0.4h, v0.4s92; CHECKNOFP16-NEXT: // kill: def $h0 killed $h0 killed $q093; CHECKNOFP16-NEXT: ret94entry:95 %shift = shufflevector <4 x half> %a, <4 x half> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>96 %0 = fadd <4 x half> %shift, %a97 %1 = extractelement <4 x half> %0, i32 098 ret half %199}100 101define half @faddp_8xhalf(<8 x half> %a) {102; CHECK-LABEL: faddp_8xhalf:103; CHECK: // %bb.0: // %entry104; CHECK-NEXT: faddp h0, v0.2h105; CHECK-NEXT: ret106;107; CHECKNOFP16-LABEL: faddp_8xhalf:108; CHECKNOFP16: // %bb.0: // %entry109; CHECKNOFP16-NEXT: dup v1.8h, v0.h[1]110; CHECKNOFP16-NEXT: fcvtl v2.4s, v0.4h111; CHECKNOFP16-NEXT: fcvtl2 v0.4s, v0.8h112; CHECKNOFP16-NEXT: fcvtl v3.4s, v1.4h113; CHECKNOFP16-NEXT: fcvtl2 v1.4s, v1.8h114; CHECKNOFP16-NEXT: fadd v2.4s, v2.4s, v3.4s115; CHECKNOFP16-NEXT: fadd v1.4s, v0.4s, v1.4s116; CHECKNOFP16-NEXT: fcvtn v0.4h, v2.4s117; CHECKNOFP16-NEXT: fcvtn2 v0.8h, v1.4s118; CHECKNOFP16-NEXT: // kill: def $h0 killed $h0 killed $q0119; CHECKNOFP16-NEXT: ret120entry:121 %shift = shufflevector <8 x half> %a, <8 x half> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>122 %0 = fadd <8 x half> %a, %shift123 %1 = extractelement <8 x half> %0, i32 0124 ret half %1125}126 127define half @faddp_8xhalf_commute(<8 x half> %a) {128; CHECK-LABEL: faddp_8xhalf_commute:129; CHECK: // %bb.0: // %entry130; CHECK-NEXT: faddp h0, v0.2h131; CHECK-NEXT: ret132;133; CHECKNOFP16-LABEL: faddp_8xhalf_commute:134; CHECKNOFP16: // %bb.0: // %entry135; CHECKNOFP16-NEXT: dup v1.8h, v0.h[1]136; CHECKNOFP16-NEXT: fcvtl v2.4s, v0.4h137; CHECKNOFP16-NEXT: fcvtl2 v0.4s, v0.8h138; CHECKNOFP16-NEXT: fcvtl v3.4s, v1.4h139; CHECKNOFP16-NEXT: fcvtl2 v1.4s, v1.8h140; CHECKNOFP16-NEXT: fadd v2.4s, v3.4s, v2.4s141; CHECKNOFP16-NEXT: fadd v1.4s, v1.4s, v0.4s142; CHECKNOFP16-NEXT: fcvtn v0.4h, v2.4s143; CHECKNOFP16-NEXT: fcvtn2 v0.8h, v1.4s144; CHECKNOFP16-NEXT: // kill: def $h0 killed $h0 killed $q0145; CHECKNOFP16-NEXT: ret146entry:147 %shift = shufflevector <8 x half> %a, <8 x half> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>148 %0 = fadd <8 x half> %shift, %a149 %1 = extractelement <8 x half> %0, i32 0150 ret half %1151}152 153define <8 x half> @addp_v8f16(<8 x half> %a) {154; CHECK-LABEL: addp_v8f16:155; CHECK: // %bb.0: // %entry156; CHECK-NEXT: rev32 v1.8h, v0.8h157; CHECK-NEXT: fadd v0.8h, v1.8h, v0.8h158; CHECK-NEXT: ret159;160; CHECKNOFP16-LABEL: addp_v8f16:161; CHECKNOFP16: // %bb.0: // %entry162; CHECKNOFP16-NEXT: rev32 v1.8h, v0.8h163; CHECKNOFP16-NEXT: fcvtl v2.4s, v0.4h164; CHECKNOFP16-NEXT: fcvtl2 v0.4s, v0.8h165; CHECKNOFP16-NEXT: fcvtl v3.4s, v1.4h166; CHECKNOFP16-NEXT: fcvtl2 v1.4s, v1.8h167; CHECKNOFP16-NEXT: fadd v2.4s, v3.4s, v2.4s168; CHECKNOFP16-NEXT: fadd v1.4s, v1.4s, v0.4s169; CHECKNOFP16-NEXT: fcvtn v0.4h, v2.4s170; CHECKNOFP16-NEXT: fcvtn2 v0.8h, v1.4s171; CHECKNOFP16-NEXT: ret172entry:173 %s = shufflevector <8 x half> %a, <8 x half> poison, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>174 %b = fadd reassoc <8 x half> %s, %a175 ret <8 x half> %b176}177 178define <16 x half> @addp_v16f16(<16 x half> %a) {179; CHECK-LABEL: addp_v16f16:180; CHECK: // %bb.0: // %entry181; CHECK-NEXT: faddp v1.8h, v0.8h, v1.8h182; CHECK-NEXT: zip1 v0.8h, v1.8h, v1.8h183; CHECK-NEXT: zip2 v1.8h, v1.8h, v1.8h184; CHECK-NEXT: ret185;186; CHECKNOFP16-LABEL: addp_v16f16:187; CHECKNOFP16: // %bb.0: // %entry188; CHECKNOFP16-NEXT: rev32 v2.8h, v0.8h189; CHECKNOFP16-NEXT: rev32 v3.8h, v1.8h190; CHECKNOFP16-NEXT: fcvtl v4.4s, v0.4h191; CHECKNOFP16-NEXT: fcvtl v6.4s, v1.4h192; CHECKNOFP16-NEXT: fcvtl2 v0.4s, v0.8h193; CHECKNOFP16-NEXT: fcvtl2 v1.4s, v1.8h194; CHECKNOFP16-NEXT: fcvtl v5.4s, v2.4h195; CHECKNOFP16-NEXT: fcvtl v7.4s, v3.4h196; CHECKNOFP16-NEXT: fcvtl2 v2.4s, v2.8h197; CHECKNOFP16-NEXT: fcvtl2 v3.4s, v3.8h198; CHECKNOFP16-NEXT: fadd v4.4s, v5.4s, v4.4s199; CHECKNOFP16-NEXT: fadd v5.4s, v7.4s, v6.4s200; CHECKNOFP16-NEXT: fadd v2.4s, v2.4s, v0.4s201; CHECKNOFP16-NEXT: fadd v3.4s, v3.4s, v1.4s202; CHECKNOFP16-NEXT: fcvtn v0.4h, v4.4s203; CHECKNOFP16-NEXT: fcvtn v1.4h, v5.4s204; CHECKNOFP16-NEXT: fcvtn2 v0.8h, v2.4s205; CHECKNOFP16-NEXT: fcvtn2 v1.8h, v3.4s206; CHECKNOFP16-NEXT: ret207entry:208 %s = shufflevector <16 x half> %a, <16 x half> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>209 %b = fadd reassoc <16 x half> %s, %a210 ret <16 x half> %b211}212