172 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s3 4; FDIV5 6define <vscale x 8 x half> @fdiv_8f16(<vscale x 8 x half> %a, <vscale x 8 x half> %b) {7; CHECK-LABEL: fdiv_8f16:8; CHECK: // %bb.0:9; CHECK-NEXT: ptrue p0.h10; CHECK-NEXT: fdiv z0.h, p0/m, z0.h, z1.h11; CHECK-NEXT: ret12 %fdiv = fdiv fast <vscale x 8 x half> %a, %b13 ret <vscale x 8 x half> %fdiv14}15 16define <vscale x 8 x half> @fdiv_recip_8f16(<vscale x 8 x half> %a, <vscale x 8 x half> %b) #0 {17; CHECK-LABEL: fdiv_recip_8f16:18; CHECK: // %bb.0:19; CHECK-NEXT: frecpe z2.h, z1.h20; CHECK-NEXT: frecps z1.h, z1.h, z2.h21; CHECK-NEXT: fmul z1.h, z2.h, z1.h22; CHECK-NEXT: fmul z0.h, z1.h, z0.h23; CHECK-NEXT: ret24 %fdiv = fdiv fast <vscale x 8 x half> %a, %b25 ret <vscale x 8 x half> %fdiv26}27 28define <vscale x 4 x float> @fdiv_4f32(<vscale x 4 x float> %a, <vscale x 4 x float> %b) {29; CHECK-LABEL: fdiv_4f32:30; CHECK: // %bb.0:31; CHECK-NEXT: ptrue p0.s32; CHECK-NEXT: fdiv z0.s, p0/m, z0.s, z1.s33; CHECK-NEXT: ret34 %fdiv = fdiv fast <vscale x 4 x float> %a, %b35 ret <vscale x 4 x float> %fdiv36}37 38define <vscale x 4 x float> @fdiv_recip_4f32(<vscale x 4 x float> %a, <vscale x 4 x float> %b) #0 {39; CHECK-LABEL: fdiv_recip_4f32:40; CHECK: // %bb.0:41; CHECK-NEXT: frecpe z2.s, z1.s42; CHECK-NEXT: frecps z3.s, z1.s, z2.s43; CHECK-NEXT: fmul z2.s, z2.s, z3.s44; CHECK-NEXT: frecps z1.s, z1.s, z2.s45; CHECK-NEXT: fmul z1.s, z2.s, z1.s46; CHECK-NEXT: fmul z0.s, z1.s, z0.s47; CHECK-NEXT: ret48 %fdiv = fdiv fast <vscale x 4 x float> %a, %b49 ret <vscale x 4 x float> %fdiv50}51 52define <vscale x 2 x double> @fdiv_2f64(<vscale x 2 x double> %a, <vscale x 2 x double> %b) {53; CHECK-LABEL: fdiv_2f64:54; CHECK: // %bb.0:55; CHECK-NEXT: ptrue p0.d56; CHECK-NEXT: fdiv z0.d, p0/m, z0.d, z1.d57; CHECK-NEXT: ret58 %fdiv = fdiv fast <vscale x 2 x double> %a, %b59 ret <vscale x 2 x double> %fdiv60}61 62define <vscale x 2 x double> @fdiv_recip_2f64(<vscale x 2 x double> %a, <vscale x 2 x double> %b) #0 {63; CHECK-LABEL: fdiv_recip_2f64:64; CHECK: // %bb.0:65; CHECK-NEXT: frecpe z2.d, z1.d66; CHECK-NEXT: frecps z3.d, z1.d, z2.d67; CHECK-NEXT: fmul z2.d, z2.d, z3.d68; CHECK-NEXT: frecps z3.d, z1.d, z2.d69; CHECK-NEXT: fmul z2.d, z2.d, z3.d70; CHECK-NEXT: frecps z1.d, z1.d, z2.d71; CHECK-NEXT: fmul z1.d, z2.d, z1.d72; CHECK-NEXT: fmul z0.d, z1.d, z0.d73; CHECK-NEXT: ret74 %fdiv = fdiv fast <vscale x 2 x double> %a, %b75 ret <vscale x 2 x double> %fdiv76}77 78; FSQRT79 80define <vscale x 8 x half> @fsqrt_8f16(<vscale x 8 x half> %a) {81; CHECK-LABEL: fsqrt_8f16:82; CHECK: // %bb.0:83; CHECK-NEXT: ptrue p0.h84; CHECK-NEXT: fsqrt z0.h, p0/m, z0.h85; CHECK-NEXT: ret86 %fsqrt = call fast <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half> %a)87 ret <vscale x 8 x half> %fsqrt88}89 90define <vscale x 8 x half> @fsqrt_recip_8f16(<vscale x 8 x half> %a) #0 {91; CHECK-LABEL: fsqrt_recip_8f16:92; CHECK: // %bb.0:93; CHECK-NEXT: frsqrte z1.h, z0.h94; CHECK-NEXT: ptrue p0.h95; CHECK-NEXT: fcmne p0.h, p0/z, z0.h, #0.096; CHECK-NEXT: fmul z2.h, z1.h, z1.h97; CHECK-NEXT: frsqrts z2.h, z0.h, z2.h98; CHECK-NEXT: fmul z1.h, z1.h, z2.h99; CHECK-NEXT: fmul z0.h, p0/m, z0.h, z1.h100; CHECK-NEXT: ret101 %fsqrt = call fast <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half> %a)102 ret <vscale x 8 x half> %fsqrt103}104 105define <vscale x 4 x float> @fsqrt_4f32(<vscale x 4 x float> %a) {106; CHECK-LABEL: fsqrt_4f32:107; CHECK: // %bb.0:108; CHECK-NEXT: ptrue p0.s109; CHECK-NEXT: fsqrt z0.s, p0/m, z0.s110; CHECK-NEXT: ret111 %fsqrt = call fast <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float> %a)112 ret <vscale x 4 x float> %fsqrt113}114 115define <vscale x 4 x float> @fsqrt_recip_4f32(<vscale x 4 x float> %a) #0 {116; CHECK-LABEL: fsqrt_recip_4f32:117; CHECK: // %bb.0:118; CHECK-NEXT: frsqrte z1.s, z0.s119; CHECK-NEXT: ptrue p0.s120; CHECK-NEXT: fcmne p0.s, p0/z, z0.s, #0.0121; CHECK-NEXT: fmul z2.s, z1.s, z1.s122; CHECK-NEXT: frsqrts z2.s, z0.s, z2.s123; CHECK-NEXT: fmul z1.s, z1.s, z2.s124; CHECK-NEXT: fmul z2.s, z1.s, z1.s125; CHECK-NEXT: frsqrts z2.s, z0.s, z2.s126; CHECK-NEXT: fmul z1.s, z1.s, z2.s127; CHECK-NEXT: fmul z0.s, p0/m, z0.s, z1.s128; CHECK-NEXT: ret129 %fsqrt = call fast <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float> %a)130 ret <vscale x 4 x float> %fsqrt131}132 133define <vscale x 2 x double> @fsqrt_2f64(<vscale x 2 x double> %a) {134; CHECK-LABEL: fsqrt_2f64:135; CHECK: // %bb.0:136; CHECK-NEXT: ptrue p0.d137; CHECK-NEXT: fsqrt z0.d, p0/m, z0.d138; CHECK-NEXT: ret139 %fsqrt = call fast <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double> %a)140 ret <vscale x 2 x double> %fsqrt141}142 143define <vscale x 2 x double> @fsqrt_recip_2f64(<vscale x 2 x double> %a) #0 {144; CHECK-LABEL: fsqrt_recip_2f64:145; CHECK: // %bb.0:146; CHECK-NEXT: frsqrte z1.d, z0.d147; CHECK-NEXT: ptrue p0.d148; CHECK-NEXT: fcmne p0.d, p0/z, z0.d, #0.0149; CHECK-NEXT: fmul z2.d, z1.d, z1.d150; CHECK-NEXT: frsqrts z2.d, z0.d, z2.d151; CHECK-NEXT: fmul z1.d, z1.d, z2.d152; CHECK-NEXT: fmul z2.d, z1.d, z1.d153; CHECK-NEXT: frsqrts z2.d, z0.d, z2.d154; CHECK-NEXT: fmul z1.d, z1.d, z2.d155; CHECK-NEXT: fmul z2.d, z1.d, z1.d156; CHECK-NEXT: frsqrts z2.d, z0.d, z2.d157; CHECK-NEXT: fmul z1.d, z1.d, z2.d158; CHECK-NEXT: fmul z0.d, p0/m, z0.d, z1.d159; CHECK-NEXT: ret160 %fsqrt = call fast <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double> %a)161 ret <vscale x 2 x double> %fsqrt162}163 164declare <vscale x 2 x half> @llvm.sqrt.nxv2f16(<vscale x 2 x half>)165declare <vscale x 4 x half> @llvm.sqrt.nxv4f16(<vscale x 4 x half>)166declare <vscale x 8 x half> @llvm.sqrt.nxv8f16(<vscale x 8 x half>)167declare <vscale x 2 x float> @llvm.sqrt.nxv2f32(<vscale x 2 x float>)168declare <vscale x 4 x float> @llvm.sqrt.nxv4f32(<vscale x 4 x float>)169declare <vscale x 2 x double> @llvm.sqrt.nxv2f64(<vscale x 2 x double>)170 171attributes #0 = { "reciprocal-estimates"="all" }172