258 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=aarch64-unknown-unknown < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SD3; RUN: llc -mtriple=aarch64-unknown-unknown < %s -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI4 5; CHECK-GI: warning: Instruction selection used fallback path for splat_fdiv_nxv4f326; CHECK-GI-NEXT: warning: Instruction selection used fallback path for splat_three_fdiv_nxv4f327; CHECK-GI-NEXT: warning: Instruction selection used fallback path for splat_fdiv_nxv2f648; CHECK-GI-NEXT: warning: Instruction selection used fallback path for splat_two_fdiv_nxv2f649 10; Following test cases check:11; a / D; b / D; c / D;12; =>13; recip = 1.0 / D; a * recip; b * recip; c * recip;14define void @three_fdiv_float(float %D, float %a, float %b, float %c) {15; CHECK-LABEL: three_fdiv_float:16; CHECK: // %bb.0:17; CHECK-NEXT: fmov s4, #1.0000000018; CHECK-NEXT: fdiv s4, s4, s019; CHECK-NEXT: fmul s0, s1, s420; CHECK-NEXT: fmul s1, s2, s421; CHECK-NEXT: fmul s2, s3, s422; CHECK-NEXT: b foo_3f23 %div = fdiv arcp float %a, %D24 %div1 = fdiv arcp float %b, %D25 %div2 = fdiv arcp float %c, %D26 tail call void @foo_3f(float %div, float %div1, float %div2)27 ret void28}29 30define void @three_fdiv_double(double %D, double %a, double %b, double %c) {31; CHECK-LABEL: three_fdiv_double:32; CHECK: // %bb.0:33; CHECK-NEXT: fmov d4, #1.0000000034; CHECK-NEXT: fdiv d4, d4, d035; CHECK-NEXT: fmul d0, d1, d436; CHECK-NEXT: fmul d1, d2, d437; CHECK-NEXT: fmul d2, d3, d438; CHECK-NEXT: b foo_3d39 %div = fdiv arcp double %a, %D40 %div1 = fdiv arcp double %b, %D41 %div2 = fdiv arcp double %c, %D42 tail call void @foo_3d(double %div, double %div1, double %div2)43 ret void44}45 46define void @three_fdiv_4xfloat(<4 x float> %D, <4 x float> %a, <4 x float> %b, <4 x float> %c) {47; CHECK-LABEL: three_fdiv_4xfloat:48; CHECK: // %bb.0:49; CHECK-NEXT: fmov v4.4s, #1.0000000050; CHECK-NEXT: fdiv v4.4s, v4.4s, v0.4s51; CHECK-NEXT: fmul v0.4s, v1.4s, v4.4s52; CHECK-NEXT: fmul v1.4s, v2.4s, v4.4s53; CHECK-NEXT: fmul v2.4s, v3.4s, v4.4s54; CHECK-NEXT: b foo_3_4xf55 %div = fdiv arcp <4 x float> %a, %D56 %div1 = fdiv arcp <4 x float> %b, %D57 %div2 = fdiv arcp <4 x float> %c, %D58 tail call void @foo_3_4xf(<4 x float> %div, <4 x float> %div1, <4 x float> %div2)59 ret void60}61 62define void @three_fdiv_2xdouble(<2 x double> %D, <2 x double> %a, <2 x double> %b, <2 x double> %c) {63; CHECK-LABEL: three_fdiv_2xdouble:64; CHECK: // %bb.0:65; CHECK-NEXT: fmov v4.2d, #1.0000000066; CHECK-NEXT: fdiv v4.2d, v4.2d, v0.2d67; CHECK-NEXT: fmul v0.2d, v1.2d, v4.2d68; CHECK-NEXT: fmul v1.2d, v2.2d, v4.2d69; CHECK-NEXT: fmul v2.2d, v3.2d, v4.2d70; CHECK-NEXT: b foo_3_2xd71 %div = fdiv arcp <2 x double> %a, %D72 %div1 = fdiv arcp <2 x double> %b, %D73 %div2 = fdiv arcp <2 x double> %c, %D74 tail call void @foo_3_2xd(<2 x double> %div, <2 x double> %div1, <2 x double> %div2)75 ret void76}77 78; Following test cases check we never combine two FDIVs if neither of them79; calculates a reciprocal.80define void @two_fdiv_float(float %D, float %a, float %b) {81; CHECK-LABEL: two_fdiv_float:82; CHECK: // %bb.0:83; CHECK-NEXT: fdiv s3, s1, s084; CHECK-NEXT: fdiv s1, s2, s085; CHECK-NEXT: fmov s0, s386; CHECK-NEXT: b foo_2f87 %div = fdiv arcp float %a, %D88 %div1 = fdiv arcp float %b, %D89 tail call void @foo_2f(float %div, float %div1)90 ret void91}92 93define void @two_fdiv_double(double %D, double %a, double %b) {94; CHECK-LABEL: two_fdiv_double:95; CHECK: // %bb.0:96; CHECK-NEXT: fdiv d3, d1, d097; CHECK-NEXT: fdiv d1, d2, d098; CHECK-NEXT: fmov d0, d399; CHECK-NEXT: b foo_2d100 %div = fdiv arcp double %a, %D101 %div1 = fdiv arcp double %b, %D102 tail call void @foo_2d(double %div, double %div1)103 ret void104}105 106define void @four_fdiv_multi_float(float %D, float %a, float %b, float %c) #0 {107; CHECK-SD-LABEL: four_fdiv_multi_float:108; CHECK-SD: // %bb.0:109; CHECK-SD-NEXT: fmov s4, #1.00000000110; CHECK-SD-NEXT: fdiv s5, s4, s0111; CHECK-SD-NEXT: fmul s4, s1, s5112; CHECK-SD-NEXT: fmul s1, s2, s5113; CHECK-SD-NEXT: fmul s2, s3, s5114; CHECK-SD-NEXT: fmul s3, s0, s5115; CHECK-SD-NEXT: fmov s0, s4116; CHECK-SD-NEXT: b foo_4f117;118; CHECK-GI-LABEL: four_fdiv_multi_float:119; CHECK-GI: // %bb.0:120; CHECK-GI-NEXT: fmov s4, #1.00000000121; CHECK-GI-NEXT: fdiv s5, s4, s0122; CHECK-GI-NEXT: fdiv s4, s0, s0123; CHECK-GI-NEXT: fmul s0, s1, s5124; CHECK-GI-NEXT: fmul s1, s2, s5125; CHECK-GI-NEXT: fmul s2, s3, s5126; CHECK-GI-NEXT: fmov s3, s4127; CHECK-GI-NEXT: b foo_4f128 %div = fdiv arcp float %a, %D129 %div1 = fdiv arcp float %b, %D130 %div2 = fdiv arcp float %c, %D131 %div3 = fdiv arcp float %D, %D132 tail call void @foo_4f(float %div, float %div1, float %div2, float %div3)133 ret void134}135 136define void @splat_three_fdiv_4xfloat(float %D, <4 x float> %a, <4 x float> %b, <4 x float> %c) {137; CHECK-LABEL: splat_three_fdiv_4xfloat:138; CHECK: // %bb.0:139; CHECK-NEXT: // kill: def $s0 killed $s0 def $q0140; CHECK-NEXT: fmov v4.4s, #1.00000000141; CHECK-NEXT: dup v0.4s, v0.s[0]142; CHECK-NEXT: fdiv v4.4s, v4.4s, v0.4s143; CHECK-NEXT: fmul v0.4s, v1.4s, v4.4s144; CHECK-NEXT: fmul v1.4s, v2.4s, v4.4s145; CHECK-NEXT: fmul v2.4s, v3.4s, v4.4s146; CHECK-NEXT: b foo_3_4xf147 %D.ins = insertelement <4 x float> poison, float %D, i64 0148 %splat = shufflevector <4 x float> %D.ins, <4 x float> poison, <4 x i32> zeroinitializer149 %div = fdiv arcp <4 x float> %a, %splat150 %div1 = fdiv arcp <4 x float> %b, %splat151 %div2 = fdiv arcp <4 x float> %c, %splat152 tail call void @foo_3_4xf(<4 x float> %div, <4 x float> %div1, <4 x float> %div2)153 ret void154}155 156define <4 x float> @splat_fdiv_v4f32(float %D, <4 x float> %a) #0 {157; CHECK-SD-LABEL: splat_fdiv_v4f32:158; CHECK-SD: // %bb.0: // %entry159; CHECK-SD-NEXT: // kill: def $s0 killed $s0 def $q0160; CHECK-SD-NEXT: fmov v2.4s, #1.00000000161; CHECK-SD-NEXT: dup v0.4s, v0.s[0]162; CHECK-SD-NEXT: fdiv v0.4s, v2.4s, v0.4s163; CHECK-SD-NEXT: fmul v0.4s, v1.4s, v0.4s164; CHECK-SD-NEXT: ret165;166; CHECK-GI-LABEL: splat_fdiv_v4f32:167; CHECK-GI: // %bb.0: // %entry168; CHECK-GI-NEXT: // kill: def $s0 killed $s0 def $q0169; CHECK-GI-NEXT: dup v0.4s, v0.s[0]170; CHECK-GI-NEXT: fdiv v0.4s, v1.4s, v0.4s171; CHECK-GI-NEXT: ret172entry:173 %D.ins = insertelement <4 x float> poison, float %D, i64 0174 %splat = shufflevector <4 x float> %D.ins, <4 x float> poison, <4 x i32> zeroinitializer175 %div = fdiv arcp <4 x float> %a, %splat176 ret <4 x float> %div177}178 179define <vscale x 4 x float> @splat_fdiv_nxv4f32(float %D, <vscale x 4 x float> %a) #0 {180; CHECK-LABEL: splat_fdiv_nxv4f32:181; CHECK: // %bb.0: // %entry182; CHECK-NEXT: fmov s2, #1.00000000183; CHECK-NEXT: fdiv s0, s2, s0184; CHECK-NEXT: mov z0.s, s0185; CHECK-NEXT: fmul z0.s, z1.s, z0.s186; CHECK-NEXT: ret187entry:188 %D.ins = insertelement <vscale x 4 x float> poison, float %D, i64 0189 %splat = shufflevector <vscale x 4 x float> %D.ins, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer190 %div = fdiv arcp <vscale x 4 x float> %a, %splat191 ret <vscale x 4 x float> %div192}193 194define void @splat_three_fdiv_nxv4f32(float %D, <vscale x 4 x float> %a, <vscale x 4 x float> %b, <vscale x 4 x float> %c) #0 {195; CHECK-LABEL: splat_three_fdiv_nxv4f32:196; CHECK: // %bb.0: // %entry197; CHECK-NEXT: fmov s4, #1.00000000198; CHECK-NEXT: fdiv s0, s4, s0199; CHECK-NEXT: mov z4.s, s0200; CHECK-NEXT: fmul z0.s, z1.s, z4.s201; CHECK-NEXT: fmul z1.s, z2.s, z4.s202; CHECK-NEXT: fmul z2.s, z3.s, z4.s203; CHECK-NEXT: b foo_3_nxv4f32204entry:205 %D.ins = insertelement <vscale x 4 x float> poison, float %D, i64 0206 %splat = shufflevector <vscale x 4 x float> %D.ins, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer207 %div = fdiv arcp <vscale x 4 x float> %a, %splat208 %div1 = fdiv arcp <vscale x 4 x float> %b, %splat209 %div2 = fdiv arcp <vscale x 4 x float> %c, %splat210 tail call void @foo_3_nxv4f32(<vscale x 4 x float> %div, <vscale x 4 x float> %div1, <vscale x 4 x float> %div2)211 ret void212}213 214define <vscale x 2 x double> @splat_fdiv_nxv2f64(double %D, <vscale x 2 x double> %a) #0 {215; CHECK-LABEL: splat_fdiv_nxv2f64:216; CHECK: // %bb.0: // %entry217; CHECK-NEXT: // kill: def $d0 killed $d0 def $z0218; CHECK-NEXT: ptrue p0.d219; CHECK-NEXT: mov z0.d, d0220; CHECK-NEXT: fdivr z0.d, p0/m, z0.d, z1.d221; CHECK-NEXT: ret222entry:223 %D.ins = insertelement <vscale x 2 x double> poison, double %D, i64 0224 %splat = shufflevector <vscale x 2 x double> %D.ins, <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer225 %div = fdiv <vscale x 2 x double> %a, %splat226 ret <vscale x 2 x double> %div227}228 229define void @splat_two_fdiv_nxv2f64(double %D, <vscale x 2 x double> %a, <vscale x 2 x double> %b) #0 {230; CHECK-LABEL: splat_two_fdiv_nxv2f64:231; CHECK: // %bb.0: // %entry232; CHECK-NEXT: fmov d3, #1.00000000233; CHECK-NEXT: fdiv d0, d3, d0234; CHECK-NEXT: mov z3.d, d0235; CHECK-NEXT: fmul z0.d, z1.d, z3.d236; CHECK-NEXT: fmul z1.d, z2.d, z3.d237; CHECK-NEXT: b foo_2_nxv2f64238entry:239 %D.ins = insertelement <vscale x 2 x double> poison, double %D, i64 0240 %splat = shufflevector <vscale x 2 x double> %D.ins, <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer241 %div = fdiv arcp <vscale x 2 x double> %a, %splat242 %div1 = fdiv arcp <vscale x 2 x double> %b, %splat243 tail call void @foo_2_nxv2f64(<vscale x 2 x double> %div, <vscale x 2 x double> %div1)244 ret void245}246 247declare void @foo_3f(float, float, float)248declare void @foo_4f(float, float, float, float)249declare void @foo_3d(double, double, double)250declare void @foo_3_4xf(<4 x float>, <4 x float>, <4 x float>)251declare void @foo_3_2xd(<2 x double>, <2 x double>, <2 x double>)252declare void @foo_2f(float, float)253declare void @foo_2d(double, double)254declare void @foo_3_nxv4f32(<vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>)255declare void @foo_2_nxv2f64(<vscale x 2 x double>, <vscale x 2 x double>)256 257attributes #0 = { "target-features"="+sve" }258