brintos / llvm-project-archived public Read only

1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -verify-machineinstrs %s -o - | FileCheck %s3 4define arm_aapcs_vfpcc void @test_fadd(ptr noalias nocapture readonly %A, float %B, ptr noalias nocapture %C, i32 %n) {5; CHECK-LABEL: test_fadd:6; CHECK:       @ %bb.0: @ %entry7; CHECK-NEXT:    cmp r2, #18; CHECK-NEXT:    it lt9; CHECK-NEXT:    bxlt lr10; CHECK-NEXT:  .LBB0_1: @ %vector.ph11; CHECK-NEXT:    vmov r3, s012; CHECK-NEXT:  .LBB0_2: @ %vector.body13; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=114; CHECK-NEXT:    vldrw.u32 q0, [r0], #1615; CHECK-NEXT:    subs r2, #416; CHECK-NEXT:    vadd.f32 q0, q0, r317; CHECK-NEXT:    vstrb.8 q0, [r1], #1618; CHECK-NEXT:    bne .LBB0_219; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup20; CHECK-NEXT:    bx lr21entry:22  %i = and i32 %n, 723  %cmp = icmp eq i32 %i, 024  tail call void @llvm.assume(i1 %cmp)25  %cmp18 = icmp sgt i32 %n, 026  br i1 %cmp18, label %vector.ph, label %for.cond.cleanup27 28vector.ph:                                        ; preds = %entry29  %broadcast.splatinsert10 = insertelement <4 x float> undef, float %B, i32 030  %broadcast.splat11 = shufflevector <4 x float> %broadcast.splatinsert10, <4 x float> undef, <4 x i32> zeroinitializer31  br label %vector.body32 33vector.body:                                      ; preds = %vector.body, %vector.ph34  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]35  %i1 = getelementptr inbounds float, ptr %A, i32 %index36  %wide.load = load <4 x float>, ptr %i1, align 437  %i3 = fadd fast <4 x float> %wide.load, %broadcast.splat1138  %i4 = getelementptr inbounds float, ptr %C, i32 %index39  store <4 x float> %i3, ptr %i4, align 440  %index.next = add i32 %index, 441  %i6 = icmp eq i32 %index.next, %n42  br i1 %i6, label %for.cond.cleanup, label %vector.body43 44for.cond.cleanup:                                 ; preds = %vector.body, %entry45  ret void46}47 48define arm_aapcs_vfpcc void @test_fadd_r(ptr noalias nocapture readonly %A, float %B, ptr noalias nocapture %C, i32 %n) {49; CHECK-LABEL: test_fadd_r:50; CHECK:       @ %bb.0: @ %entry51; CHECK-NEXT:    cmp r2, #152; CHECK-NEXT:    it lt53; CHECK-NEXT:    bxlt lr54; CHECK-NEXT:  .LBB1_1: @ %vector.ph55; CHECK-NEXT:    vmov r3, s056; CHECK-NEXT:  .LBB1_2: @ %vector.body57; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=158; CHECK-NEXT:    vldrw.u32 q0, [r0], #1659; CHECK-NEXT:    subs r2, #460; CHECK-NEXT:    vadd.f32 q0, q0, r361; CHECK-NEXT:    vstrb.8 q0, [r1], #1662; CHECK-NEXT:    bne .LBB1_263; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup64; CHECK-NEXT:    bx lr65entry:66  %i = and i32 %n, 767  %cmp = icmp eq i32 %i, 068  tail call void @llvm.assume(i1 %cmp)69  %cmp18 = icmp sgt i32 %n, 070  br i1 %cmp18, label %vector.ph, label %for.cond.cleanup71 72vector.ph:                                        ; preds = %entry73  %broadcast.splatinsert10 = insertelement <4 x float> undef, float %B, i32 074  %broadcast.splat11 = shufflevector <4 x float> %broadcast.splatinsert10, <4 x float> undef, <4 x i32> zeroinitializer75  br label %vector.body76 77vector.body:                                      ; preds = %vector.body, %vector.ph78  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]79  %i1 = getelementptr inbounds float, ptr %A, i32 %index80  %wide.load = load <4 x float>, ptr %i1, align 481  %i3 = fadd fast <4 x float> %broadcast.splat11, %wide.load82  %i4 = getelementptr inbounds float, ptr %C, i32 %index83  store <4 x float> %i3, ptr %i4, align 484  %index.next = add i32 %index, 485  %i6 = icmp eq i32 %index.next, %n86  br i1 %i6, label %for.cond.cleanup, label %vector.body87 88for.cond.cleanup:                                 ; preds = %vector.body, %entry89  ret void90}91 92define arm_aapcs_vfpcc void @test_fmul(ptr noalias nocapture readonly %A, float %B, ptr noalias nocapture %C, i32 %n) {93; CHECK-LABEL: test_fmul:94; CHECK:       @ %bb.0: @ %entry95; CHECK-NEXT:    cmp r2, #196; CHECK-NEXT:    it lt97; CHECK-NEXT:    bxlt lr98; CHECK-NEXT:  .LBB2_1: @ %vector.ph99; CHECK-NEXT:    vmov r3, s0100; CHECK-NEXT:  .LBB2_2: @ %vector.body101; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1102; CHECK-NEXT:    vldrw.u32 q0, [r0], #16103; CHECK-NEXT:    subs r2, #4104; CHECK-NEXT:    vmul.f32 q0, q0, r3105; CHECK-NEXT:    vstrb.8 q0, [r1], #16106; CHECK-NEXT:    bne .LBB2_2107; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup108; CHECK-NEXT:    bx lr109entry:110  %i = and i32 %n, 7111  %cmp = icmp eq i32 %i, 0112  tail call void @llvm.assume(i1 %cmp)113  %cmp18 = icmp sgt i32 %n, 0114  br i1 %cmp18, label %vector.ph, label %for.cond.cleanup115 116vector.ph:                                        ; preds = %entry117  %broadcast.splatinsert10 = insertelement <4 x float> undef, float %B, i32 0118  %broadcast.splat11 = shufflevector <4 x float> %broadcast.splatinsert10, <4 x float> undef, <4 x i32> zeroinitializer119  br label %vector.body120 121vector.body:                                      ; preds = %vector.body, %vector.ph122  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]123  %i1 = getelementptr inbounds float, ptr %A, i32 %index124  %wide.load = load <4 x float>, ptr %i1, align 4125  %i3 = fmul fast <4 x float> %wide.load, %broadcast.splat11126  %i4 = getelementptr inbounds float, ptr %C, i32 %index127  store <4 x float> %i3, ptr %i4, align 4128  %index.next = add i32 %index, 4129  %i6 = icmp eq i32 %index.next, %n130  br i1 %i6, label %for.cond.cleanup, label %vector.body131 132for.cond.cleanup:                                 ; preds = %vector.body, %entry133  ret void134}135 136define arm_aapcs_vfpcc void @test_fmul_r(ptr noalias nocapture readonly %A, float %B, ptr noalias nocapture %C, i32 %n) {137; CHECK-LABEL: test_fmul_r:138; CHECK:       @ %bb.0: @ %entry139; CHECK-NEXT:    cmp r2, #1140; CHECK-NEXT:    it lt141; CHECK-NEXT:    bxlt lr142; CHECK-NEXT:  .LBB3_1: @ %vector.ph143; CHECK-NEXT:    vmov r3, s0144; CHECK-NEXT:  .LBB3_2: @ %vector.body145; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1146; CHECK-NEXT:    vldrw.u32 q0, [r0], #16147; CHECK-NEXT:    subs r2, #4148; CHECK-NEXT:    vmul.f32 q0, q0, r3149; CHECK-NEXT:    vstrb.8 q0, [r1], #16150; CHECK-NEXT:    bne .LBB3_2151; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup152; CHECK-NEXT:    bx lr153entry:154  %i = and i32 %n, 7155  %cmp = icmp eq i32 %i, 0156  tail call void @llvm.assume(i1 %cmp)157  %cmp18 = icmp sgt i32 %n, 0158  br i1 %cmp18, label %vector.ph, label %for.cond.cleanup159 160vector.ph:                                        ; preds = %entry161  %broadcast.splatinsert10 = insertelement <4 x float> undef, float %B, i32 0162  %broadcast.splat11 = shufflevector <4 x float> %broadcast.splatinsert10, <4 x float> undef, <4 x i32> zeroinitializer163  br label %vector.body164 165vector.body:                                      ; preds = %vector.body, %vector.ph166  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]167  %i1 = getelementptr inbounds float, ptr %A, i32 %index168  %wide.load = load <4 x float>, ptr %i1, align 4169  %i3 = fmul fast <4 x float> %broadcast.splat11, %wide.load170  %i4 = getelementptr inbounds float, ptr %C, i32 %index171  store <4 x float> %i3, ptr %i4, align 4172  %index.next = add i32 %index, 4173  %i6 = icmp eq i32 %index.next, %n174  br i1 %i6, label %for.cond.cleanup, label %vector.body175 176for.cond.cleanup:                                 ; preds = %vector.body, %entry177  ret void178}179 180define arm_aapcs_vfpcc void @test_fsub(ptr noalias nocapture readonly %A, float %B, ptr noalias nocapture %C, i32 %n) {181; CHECK-LABEL: test_fsub:182; CHECK:       @ %bb.0: @ %entry183; CHECK-NEXT:    cmp r2, #1184; CHECK-NEXT:    it lt185; CHECK-NEXT:    bxlt lr186; CHECK-NEXT:  .LBB4_1: @ %vector.ph187; CHECK-NEXT:    vmov r3, s0188; CHECK-NEXT:  .LBB4_2: @ %vector.body189; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1190; CHECK-NEXT:    vldrw.u32 q0, [r0], #16191; CHECK-NEXT:    subs r2, #4192; CHECK-NEXT:    vsub.f32 q0, q0, r3193; CHECK-NEXT:    vstrb.8 q0, [r1], #16194; CHECK-NEXT:    bne .LBB4_2195; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup196; CHECK-NEXT:    bx lr197entry:198  %i = and i32 %n, 7199  %cmp = icmp eq i32 %i, 0200  tail call void @llvm.assume(i1 %cmp)201  %cmp18 = icmp sgt i32 %n, 0202  br i1 %cmp18, label %vector.ph, label %for.cond.cleanup203 204vector.ph:                                        ; preds = %entry205  %broadcast.splatinsert10 = insertelement <4 x float> undef, float %B, i32 0206  %broadcast.splat11 = shufflevector <4 x float> %broadcast.splatinsert10, <4 x float> undef, <4 x i32> zeroinitializer207  br label %vector.body208 209vector.body:                                      ; preds = %vector.body, %vector.ph210  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]211  %i1 = getelementptr inbounds float, ptr %A, i32 %index212  %wide.load = load <4 x float>, ptr %i1, align 4213  %i3 = fsub fast <4 x float> %wide.load, %broadcast.splat11214  %i4 = getelementptr inbounds float, ptr %C, i32 %index215  store <4 x float> %i3, ptr %i4, align 4216  %index.next = add i32 %index, 4217  %i6 = icmp eq i32 %index.next, %n218  br i1 %i6, label %for.cond.cleanup, label %vector.body219 220for.cond.cleanup:                                 ; preds = %vector.body, %entry221  ret void222}223 224define arm_aapcs_vfpcc void @test_fsub_r(ptr noalias nocapture readonly %A, float %B, ptr noalias nocapture %C, i32 %n) {225; CHECK-LABEL: test_fsub_r:226; CHECK:       @ %bb.0: @ %entry227; CHECK-NEXT:    cmp r2, #1228; CHECK-NEXT:    it lt229; CHECK-NEXT:    bxlt lr230; CHECK-NEXT:  .LBB5_1: @ %vector.ph231; CHECK-NEXT:    vmov r3, s0232; CHECK-NEXT:    vdup.32 q0, r3233; CHECK-NEXT:  .LBB5_2: @ %vector.body234; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1235; CHECK-NEXT:    vldrw.u32 q1, [r0], #16236; CHECK-NEXT:    subs r2, #4237; CHECK-NEXT:    vsub.f32 q1, q0, q1238; CHECK-NEXT:    vstrb.8 q1, [r1], #16239; CHECK-NEXT:    bne .LBB5_2240; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup241; CHECK-NEXT:    bx lr242entry:243  %i = and i32 %n, 7244  %cmp = icmp eq i32 %i, 0245  tail call void @llvm.assume(i1 %cmp)246  %cmp18 = icmp sgt i32 %n, 0247  br i1 %cmp18, label %vector.ph, label %for.cond.cleanup248 249vector.ph:                                        ; preds = %entry250  %broadcast.splatinsert10 = insertelement <4 x float> undef, float %B, i32 0251  %broadcast.splat11 = shufflevector <4 x float> %broadcast.splatinsert10, <4 x float> undef, <4 x i32> zeroinitializer252  br label %vector.body253 254vector.body:                                      ; preds = %vector.body, %vector.ph255  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]256  %i1 = getelementptr inbounds float, ptr %A, i32 %index257  %wide.load = load <4 x float>, ptr %i1, align 4258  %i3 = fsub fast <4 x float> %broadcast.splat11, %wide.load259  %i4 = getelementptr inbounds float, ptr %C, i32 %index260  store <4 x float> %i3, ptr %i4, align 4261  %index.next = add i32 %index, 4262  %i6 = icmp eq i32 %index.next, %n263  br i1 %i6, label %for.cond.cleanup, label %vector.body264 265for.cond.cleanup:                                 ; preds = %vector.body, %entry266  ret void267}268 269 270define arm_aapcs_vfpcc void @test_fmas(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {271; CHECK-LABEL: test_fmas:272; CHECK:       @ %bb.0: @ %entry273; CHECK-NEXT:    cmp r3, #1274; CHECK-NEXT:    it lt275; CHECK-NEXT:    bxlt lr276; CHECK-NEXT:  .LBB6_1: @ %vector.ph277; CHECK-NEXT:    vmov r12, s0278; CHECK-NEXT:  .LBB6_2: @ %vector.body279; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1280; CHECK-NEXT:    vldrw.u32 q0, [r0], #16281; CHECK-NEXT:    vldrw.u32 q1, [r1], #16282; CHECK-NEXT:    subs r3, #4283; CHECK-NEXT:    vfmas.f32 q1, q0, r12284; CHECK-NEXT:    vstrb.8 q1, [r2], #16285; CHECK-NEXT:    bne .LBB6_2286; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup287; CHECK-NEXT:    bx lr288entry:289  %i = and i32 %n, 7290  %cmp = icmp eq i32 %i, 0291  tail call void @llvm.assume(i1 %cmp)292  %cmp110 = icmp sgt i32 %n, 0293  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup294 295vector.ph:                                        ; preds = %entry296  %broadcast.splatinsert13 = insertelement <4 x float> undef, float %C, i32 0297  %broadcast.splat14 = shufflevector <4 x float> %broadcast.splatinsert13, <4 x float> undef, <4 x i32> zeroinitializer298  br label %vector.body299 300vector.body:                                      ; preds = %vector.body, %vector.ph301  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]302  %i1 = getelementptr inbounds float, ptr %A, i32 %index303  %wide.load = load <4 x float>, ptr %i1, align 4304  %i3 = getelementptr inbounds float, ptr %B, i32 %index305  %wide.load12 = load <4 x float>, ptr %i3, align 4306  %i5 = fmul fast <4 x float> %wide.load12, %wide.load307  %i6 = fadd fast <4 x float> %i5, %broadcast.splat14308  %i7 = getelementptr inbounds float, ptr %D, i32 %index309  store <4 x float> %i6, ptr %i7, align 4310  %index.next = add i32 %index, 4311  %i9 = icmp eq i32 %index.next, %n312  br i1 %i9, label %for.cond.cleanup, label %vector.body313 314for.cond.cleanup:                                 ; preds = %vector.body, %entry315  ret void316}317 318define arm_aapcs_vfpcc void @test_fmas_r(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {319; CHECK-LABEL: test_fmas_r:320; CHECK:       @ %bb.0: @ %entry321; CHECK-NEXT:    cmp r3, #1322; CHECK-NEXT:    it lt323; CHECK-NEXT:    bxlt lr324; CHECK-NEXT:  .LBB7_1: @ %vector.ph325; CHECK-NEXT:    vmov r12, s0326; CHECK-NEXT:  .LBB7_2: @ %vector.body327; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1328; CHECK-NEXT:    vldrw.u32 q0, [r0], #16329; CHECK-NEXT:    vldrw.u32 q1, [r1], #16330; CHECK-NEXT:    subs r3, #4331; CHECK-NEXT:    vfmas.f32 q1, q0, r12332; CHECK-NEXT:    vstrb.8 q1, [r2], #16333; CHECK-NEXT:    bne .LBB7_2334; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup335; CHECK-NEXT:    bx lr336entry:337  %i = and i32 %n, 7338  %cmp = icmp eq i32 %i, 0339  tail call void @llvm.assume(i1 %cmp)340  %cmp110 = icmp sgt i32 %n, 0341  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup342 343vector.ph:                                        ; preds = %entry344  %broadcast.splatinsert13 = insertelement <4 x float> undef, float %C, i32 0345  %broadcast.splat14 = shufflevector <4 x float> %broadcast.splatinsert13, <4 x float> undef, <4 x i32> zeroinitializer346  br label %vector.body347 348vector.body:                                      ; preds = %vector.body, %vector.ph349  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]350  %i1 = getelementptr inbounds float, ptr %A, i32 %index351  %wide.load = load <4 x float>, ptr %i1, align 4352  %i3 = getelementptr inbounds float, ptr %B, i32 %index353  %wide.load12 = load <4 x float>, ptr %i3, align 4354  %i5 = fmul fast <4 x float> %wide.load12, %wide.load355  %i6 = fadd fast <4 x float> %broadcast.splat14, %i5356  %i7 = getelementptr inbounds float, ptr %D, i32 %index357  store <4 x float> %i6, ptr %i7, align 4358  %index.next = add i32 %index, 4359  %i9 = icmp eq i32 %index.next, %n360  br i1 %i9, label %for.cond.cleanup, label %vector.body361 362for.cond.cleanup:                                 ; preds = %vector.body, %entry363  ret void364}365 366define arm_aapcs_vfpcc void @test_fma(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {367; CHECK-LABEL: test_fma:368; CHECK:       @ %bb.0: @ %entry369; CHECK-NEXT:    cmp r3, #1370; CHECK-NEXT:    it lt371; CHECK-NEXT:    bxlt lr372; CHECK-NEXT:  .LBB8_1: @ %vector.ph373; CHECK-NEXT:    vmov r12, s0374; CHECK-NEXT:  .LBB8_2: @ %vector.body375; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1376; CHECK-NEXT:    vldrw.u32 q0, [r0], #16377; CHECK-NEXT:    vldrw.u32 q1, [r1], #16378; CHECK-NEXT:    subs r3, #4379; CHECK-NEXT:    vfma.f32 q1, q0, r12380; CHECK-NEXT:    vstrb.8 q1, [r2], #16381; CHECK-NEXT:    bne .LBB8_2382; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup383; CHECK-NEXT:    bx lr384entry:385  %i = and i32 %n, 7386  %cmp = icmp eq i32 %i, 0387  tail call void @llvm.assume(i1 %cmp)388  %cmp110 = icmp sgt i32 %n, 0389  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup390 391vector.ph:                                        ; preds = %entry392  %broadcast.splatinsert12 = insertelement <4 x float> undef, float %C, i32 0393  %broadcast.splat13 = shufflevector <4 x float> %broadcast.splatinsert12, <4 x float> undef, <4 x i32> zeroinitializer394  br label %vector.body395 396vector.body:                                      ; preds = %vector.body, %vector.ph397  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]398  %i1 = getelementptr inbounds float, ptr %A, i32 %index399  %wide.load = load <4 x float>, ptr %i1, align 4400  %i3 = fmul fast <4 x float> %wide.load, %broadcast.splat13401  %i4 = getelementptr inbounds float, ptr %B, i32 %index402  %wide.load14 = load <4 x float>, ptr %i4, align 4403  %i6 = fadd fast <4 x float> %i3, %wide.load14404  %i7 = getelementptr inbounds float, ptr %D, i32 %index405  store <4 x float> %i6, ptr %i7, align 4406  %index.next = add i32 %index, 4407  %i9 = icmp eq i32 %index.next, %n408  br i1 %i9, label %for.cond.cleanup, label %vector.body409 410for.cond.cleanup:                                 ; preds = %vector.body, %entry411  ret void412}413 414define arm_aapcs_vfpcc void @test_fma_r(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {415; CHECK-LABEL: test_fma_r:416; CHECK:       @ %bb.0: @ %entry417; CHECK-NEXT:    cmp r3, #1418; CHECK-NEXT:    it lt419; CHECK-NEXT:    bxlt lr420; CHECK-NEXT:  .LBB9_1: @ %vector.ph421; CHECK-NEXT:    vmov r12, s0422; CHECK-NEXT:  .LBB9_2: @ %vector.body423; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1424; CHECK-NEXT:    vldrw.u32 q0, [r0], #16425; CHECK-NEXT:    vldrw.u32 q1, [r1], #16426; CHECK-NEXT:    subs r3, #4427; CHECK-NEXT:    vfma.f32 q1, q0, r12428; CHECK-NEXT:    vstrb.8 q1, [r2], #16429; CHECK-NEXT:    bne .LBB9_2430; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup431; CHECK-NEXT:    bx lr432entry:433  %i = and i32 %n, 7434  %cmp = icmp eq i32 %i, 0435  tail call void @llvm.assume(i1 %cmp)436  %cmp110 = icmp sgt i32 %n, 0437  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup438 439vector.ph:                                        ; preds = %entry440  %broadcast.splatinsert12 = insertelement <4 x float> undef, float %C, i32 0441  %broadcast.splat13 = shufflevector <4 x float> %broadcast.splatinsert12, <4 x float> undef, <4 x i32> zeroinitializer442  br label %vector.body443 444vector.body:                                      ; preds = %vector.body, %vector.ph445  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]446  %i1 = getelementptr inbounds float, ptr %A, i32 %index447  %wide.load = load <4 x float>, ptr %i1, align 4448  %i3 = fmul fast <4 x float> %broadcast.splat13, %wide.load449  %i4 = getelementptr inbounds float, ptr %B, i32 %index450  %wide.load14 = load <4 x float>, ptr %i4, align 4451  %i6 = fadd fast <4 x float> %i3, %wide.load14452  %i7 = getelementptr inbounds float, ptr %D, i32 %index453  store <4 x float> %i6, ptr %i7, align 4454  %index.next = add i32 %index, 4455  %i9 = icmp eq i32 %index.next, %n456  br i1 %i9, label %for.cond.cleanup, label %vector.body457 458for.cond.cleanup:                                 ; preds = %vector.body, %entry459  ret void460}461 462 463define arm_aapcs_vfpcc void @test_fmss(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {464; CHECK-LABEL: test_fmss:465; CHECK:       @ %bb.0: @ %entry466; CHECK-NEXT:    cmp r3, #1467; CHECK-NEXT:    it lt468; CHECK-NEXT:    bxlt lr469; CHECK-NEXT:  .LBB10_1: @ %vector.ph470; CHECK-NEXT:    vmov r12, s0471; CHECK-NEXT:    vdup.32 q0, r12472; CHECK-NEXT:    vneg.f32 q0, q0473; CHECK-NEXT:  .LBB10_2: @ %vector.body474; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1475; CHECK-NEXT:    vldrw.u32 q1, [r0], #16476; CHECK-NEXT:    vldrw.u32 q2, [r1], #16477; CHECK-NEXT:    vmov q3, q0478; CHECK-NEXT:    subs r3, #4479; CHECK-NEXT:    vfma.f32 q3, q2, q1480; CHECK-NEXT:    vstrb.8 q3, [r2], #16481; CHECK-NEXT:    bne .LBB10_2482; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup483; CHECK-NEXT:    bx lr484entry:485  %i = and i32 %n, 7486  %cmp = icmp eq i32 %i, 0487  tail call void @llvm.assume(i1 %cmp)488  %cmp110 = icmp sgt i32 %n, 0489  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup490 491vector.ph:                                        ; preds = %entry492  %broadcast.splatinsert13 = insertelement <4 x float> undef, float %C, i32 0493  %broadcast.splat14 = shufflevector <4 x float> %broadcast.splatinsert13, <4 x float> undef, <4 x i32> zeroinitializer494  br label %vector.body495 496vector.body:                                      ; preds = %vector.body, %vector.ph497  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]498  %i1 = getelementptr inbounds float, ptr %A, i32 %index499  %wide.load = load <4 x float>, ptr %i1, align 4500  %i3 = getelementptr inbounds float, ptr %B, i32 %index501  %wide.load12 = load <4 x float>, ptr %i3, align 4502  %i5 = fmul fast <4 x float> %wide.load12, %wide.load503  %i6 = fsub fast <4 x float> %i5, %broadcast.splat14504  %i7 = getelementptr inbounds float, ptr %D, i32 %index505  store <4 x float> %i6, ptr %i7, align 4506  %index.next = add i32 %index, 4507  %i9 = icmp eq i32 %index.next, %n508  br i1 %i9, label %for.cond.cleanup, label %vector.body509 510for.cond.cleanup:                                 ; preds = %vector.body, %entry511  ret void512}513 514define arm_aapcs_vfpcc void @test_fmss_r(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {515; CHECK-LABEL: test_fmss_r:516; CHECK:       @ %bb.0: @ %entry517; CHECK-NEXT:    cmp r3, #1518; CHECK-NEXT:    it lt519; CHECK-NEXT:    bxlt lr520; CHECK-NEXT:  .LBB11_1: @ %vector.ph521; CHECK-NEXT:    vmov r12, s0522; CHECK-NEXT:    vdup.32 q0, r12523; CHECK-NEXT:  .LBB11_2: @ %vector.body524; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1525; CHECK-NEXT:    vldrw.u32 q1, [r0], #16526; CHECK-NEXT:    vldrw.u32 q2, [r1], #16527; CHECK-NEXT:    vmov q3, q0528; CHECK-NEXT:    subs r3, #4529; CHECK-NEXT:    vfms.f32 q3, q2, q1530; CHECK-NEXT:    vstrb.8 q3, [r2], #16531; CHECK-NEXT:    bne .LBB11_2532; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup533; CHECK-NEXT:    bx lr534entry:535  %i = and i32 %n, 7536  %cmp = icmp eq i32 %i, 0537  tail call void @llvm.assume(i1 %cmp)538  %cmp110 = icmp sgt i32 %n, 0539  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup540 541vector.ph:                                        ; preds = %entry542  %broadcast.splatinsert13 = insertelement <4 x float> undef, float %C, i32 0543  %broadcast.splat14 = shufflevector <4 x float> %broadcast.splatinsert13, <4 x float> undef, <4 x i32> zeroinitializer544  br label %vector.body545 546vector.body:                                      ; preds = %vector.body, %vector.ph547  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]548  %i1 = getelementptr inbounds float, ptr %A, i32 %index549  %wide.load = load <4 x float>, ptr %i1, align 4550  %i3 = getelementptr inbounds float, ptr %B, i32 %index551  %wide.load12 = load <4 x float>, ptr %i3, align 4552  %i5 = fmul fast <4 x float> %wide.load12, %wide.load553  %i6 = fsub fast <4 x float> %broadcast.splat14, %i5554  %i7 = getelementptr inbounds float, ptr %D, i32 %index555  store <4 x float> %i6, ptr %i7, align 4556  %index.next = add i32 %index, 4557  %i9 = icmp eq i32 %index.next, %n558  br i1 %i9, label %for.cond.cleanup, label %vector.body559 560for.cond.cleanup:                                 ; preds = %vector.body, %entry561  ret void562}563 564define arm_aapcs_vfpcc void @test_fms(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {565; CHECK-LABEL: test_fms:566; CHECK:       @ %bb.0: @ %entry567; CHECK-NEXT:    cmp r3, #1568; CHECK-NEXT:    it lt569; CHECK-NEXT:    bxlt lr570; CHECK-NEXT:  .LBB12_1: @ %vector.ph571; CHECK-NEXT:    vmov r12, s0572; CHECK-NEXT:  .LBB12_2: @ %vector.body573; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1574; CHECK-NEXT:    vldrw.u32 q0, [r1], #16575; CHECK-NEXT:    vldrw.u32 q1, [r0], #16576; CHECK-NEXT:    subs r3, #4577; CHECK-NEXT:    vneg.f32 q0, q0578; CHECK-NEXT:    vfma.f32 q0, q1, r12579; CHECK-NEXT:    vstrb.8 q0, [r2], #16580; CHECK-NEXT:    bne .LBB12_2581; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup582; CHECK-NEXT:    bx lr583entry:584  %i = and i32 %n, 7585  %cmp = icmp eq i32 %i, 0586  tail call void @llvm.assume(i1 %cmp)587  %cmp110 = icmp sgt i32 %n, 0588  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup589 590vector.ph:                                        ; preds = %entry591  %broadcast.splatinsert12 = insertelement <4 x float> undef, float %C, i32 0592  %broadcast.splat13 = shufflevector <4 x float> %broadcast.splatinsert12, <4 x float> undef, <4 x i32> zeroinitializer593  br label %vector.body594 595vector.body:                                      ; preds = %vector.body, %vector.ph596  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]597  %i1 = getelementptr inbounds float, ptr %A, i32 %index598  %wide.load = load <4 x float>, ptr %i1, align 4599  %i3 = fmul fast <4 x float> %wide.load, %broadcast.splat13600  %i4 = getelementptr inbounds float, ptr %B, i32 %index601  %wide.load14 = load <4 x float>, ptr %i4, align 4602  %i6 = fsub fast <4 x float> %i3, %wide.load14603  %i7 = getelementptr inbounds float, ptr %D, i32 %index604  store <4 x float> %i6, ptr %i7, align 4605  %index.next = add i32 %index, 4606  %i9 = icmp eq i32 %index.next, %n607  br i1 %i9, label %for.cond.cleanup, label %vector.body608 609for.cond.cleanup:                                 ; preds = %vector.body, %entry610  ret void611}612 613define arm_aapcs_vfpcc void @test_fms_r(ptr noalias nocapture readonly %A, ptr noalias nocapture readonly %B, float %C, ptr noalias nocapture %D, i32 %n) {614; CHECK-LABEL: test_fms_r:615; CHECK:       @ %bb.0: @ %entry616; CHECK-NEXT:    cmp r3, #1617; CHECK-NEXT:    it lt618; CHECK-NEXT:    bxlt lr619; CHECK-NEXT:  .LBB13_1: @ %vector.ph620; CHECK-NEXT:    vmov r12, s0621; CHECK-NEXT:  .LBB13_2: @ %vector.body622; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1623; CHECK-NEXT:    vldrw.u32 q0, [r1], #16624; CHECK-NEXT:    vldrw.u32 q1, [r0], #16625; CHECK-NEXT:    subs r3, #4626; CHECK-NEXT:    vneg.f32 q0, q0627; CHECK-NEXT:    vfma.f32 q0, q1, r12628; CHECK-NEXT:    vstrb.8 q0, [r2], #16629; CHECK-NEXT:    bne .LBB13_2630; CHECK-NEXT:  @ %bb.3: @ %for.cond.cleanup631; CHECK-NEXT:    bx lr632entry:633  %i = and i32 %n, 7634  %cmp = icmp eq i32 %i, 0635  tail call void @llvm.assume(i1 %cmp)636  %cmp110 = icmp sgt i32 %n, 0637  br i1 %cmp110, label %vector.ph, label %for.cond.cleanup638 639vector.ph:                                        ; preds = %entry640  %broadcast.splatinsert12 = insertelement <4 x float> undef, float %C, i32 0641  %broadcast.splat13 = shufflevector <4 x float> %broadcast.splatinsert12, <4 x float> undef, <4 x i32> zeroinitializer642  br label %vector.body643 644vector.body:                                      ; preds = %vector.body, %vector.ph645  %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]646  %i1 = getelementptr inbounds float, ptr %A, i32 %index647  %wide.load = load <4 x float>, ptr %i1, align 4648  %i3 = fmul fast <4 x float> %broadcast.splat13, %wide.load649  %i4 = getelementptr inbounds float, ptr %B, i32 %index650  %wide.load14 = load <4 x float>, ptr %i4, align 4651  %i6 = fsub fast <4 x float> %i3, %wide.load14652  %i7 = getelementptr inbounds float, ptr %D, i32 %index653  store <4 x float> %i6, ptr %i7, align 4654  %index.next = add i32 %index, 4655  %i9 = icmp eq i32 %index.next, %n656  br i1 %i9, label %for.cond.cleanup, label %vector.body657 658for.cond.cleanup:                                 ; preds = %vector.body, %entry659  ret void660}661 662 663define dso_local void @test_nested(ptr noalias nocapture %pInT1, ptr noalias nocapture readonly %pOutT1, ptr noalias nocapture readonly %pPRT_in, ptr noalias nocapture readnone %pPRT_pDst, i32 %numRows, i32 %numCols, i32 %l) local_unnamed_addr {664; CHECK-LABEL: test_nested:665; CHECK:       @ %bb.0: @ %for.body.us.preheader666; CHECK-NEXT:    .save {r4, r5, r6, lr}667; CHECK-NEXT:    push {r4, r5, r6, lr}668; CHECK-NEXT:    ldrd lr, r12, [sp, #16]669; CHECK-NEXT:    lsl.w r3, r12, #2670; CHECK-NEXT:  .LBB14_1: @ %for.body.us671; CHECK-NEXT:    @ =>This Loop Header: Depth=1672; CHECK-NEXT:    @ Child Loop BB14_2 Depth 2673; CHECK-NEXT:    ldr r4, [r1]674; CHECK-NEXT:    mov r5, r2675; CHECK-NEXT:    mov r6, r12676; CHECK-NEXT:    vdup.32 q0, r4677; CHECK-NEXT:    mov r4, r0678; CHECK-NEXT:  .LBB14_2: @ %vector.body679; CHECK-NEXT:    @ Parent Loop BB14_1 Depth=1680; CHECK-NEXT:    @ => This Inner Loop Header: Depth=2681; CHECK-NEXT:    vldrw.u32 q1, [r5], #16682; CHECK-NEXT:    vldrw.u32 q2, [r4]683; CHECK-NEXT:    subs r6, #4684; CHECK-NEXT:    vfms.f32 q2, q1, q0685; CHECK-NEXT:    vstrb.8 q2, [r4], #16686; CHECK-NEXT:    bne .LBB14_2687; CHECK-NEXT:  @ %bb.3: @ %for.cond6.for.end_crit_edge.us688; CHECK-NEXT:    @ in Loop: Header=BB14_1 Depth=1689; CHECK-NEXT:    add r0, r3690; CHECK-NEXT:    add r2, r3691; CHECK-NEXT:    adds r1, #4692; CHECK-NEXT:    le lr, .LBB14_1693; CHECK-NEXT:  @ %bb.4: @ %for.end14694; CHECK-NEXT:    pop {r4, r5, r6, pc}695for.body.us.preheader:696  %cmp = icmp sgt i32 %numRows, 0697  tail call void @llvm.assume(i1 %cmp)698  %cmp1 = icmp sgt i32 %numCols, 0699  tail call void @llvm.assume(i1 %cmp1)700  %rem = and i32 %numCols, 7701  %cmp2 = icmp eq i32 %rem, 0702  tail call void @llvm.assume(i1 %cmp2)703  %cmp3 = icmp slt i32 %l, %numCols704  tail call void @llvm.assume(i1 %cmp3)705  br label %for.body.us706 707for.body.us:                                      ; preds = %for.cond6.for.end_crit_edge.us, %for.body.us.preheader708  %pInT1.addr.038.us = phi ptr [ %scevgep40, %for.cond6.for.end_crit_edge.us ], [ %pInT1, %for.body.us.preheader ]709  %i.037.us = phi i32 [ %inc13.us, %for.cond6.for.end_crit_edge.us ], [ 0, %for.body.us.preheader ]710  %pOutT1.addr.036.us = phi ptr [ %incdec.ptr.us, %for.cond6.for.end_crit_edge.us ], [ %pOutT1, %for.body.us.preheader ]711  %pPRT_in.addr.035.us = phi ptr [ %scevgep, %for.cond6.for.end_crit_edge.us ], [ %pPRT_in, %for.body.us.preheader ]712  %scevgep = getelementptr float, ptr %pPRT_in.addr.035.us, i32 %numCols713  %i = load float, ptr %pOutT1.addr.036.us, align 4714  %broadcast.splatinsert47 = insertelement <4 x float> undef, float %i, i32 0715  %broadcast.splat48 = shufflevector <4 x float> %broadcast.splatinsert47, <4 x float> undef, <4 x i32> zeroinitializer716  br label %vector.body717 718vector.body:                                      ; preds = %vector.body, %for.body.us719  %index = phi i32 [ 0, %for.body.us ], [ %index.next, %vector.body ]720  %next.gep = getelementptr float, ptr %pInT1.addr.038.us, i32 %index721  %next.gep45 = getelementptr float, ptr %pPRT_in.addr.035.us, i32 %index722  %wide.load = load <4 x float>, ptr %next.gep, align 4723  %wide.load46 = load <4 x float>, ptr %next.gep45, align 4724  %i3 = fmul fast <4 x float> %wide.load46, %broadcast.splat48725  %i4 = fsub fast <4 x float> %wide.load, %i3726  store <4 x float> %i4, ptr %next.gep, align 4727  %index.next = add i32 %index, 4728  %i5 = icmp eq i32 %index.next, %numCols729  br i1 %i5, label %for.cond6.for.end_crit_edge.us, label %vector.body730 731for.cond6.for.end_crit_edge.us:                   ; preds = %vector.body732  %incdec.ptr.us = getelementptr inbounds float, ptr %pOutT1.addr.036.us, i32 1733  %scevgep40 = getelementptr float, ptr %pInT1.addr.038.us, i32 %numCols734  %inc13.us = add nuw nsw i32 %i.037.us, 1735  %exitcond41 = icmp eq i32 %inc13.us, %numRows736  br i1 %exitcond41, label %for.end14, label %for.body.us737 738for.end14:                                        ; preds = %for.cond6.for.end_crit_edge.us739  ret void740}741 742%struct.arm_fir_instance_f32 = type { i16, ptr, ptr }743define void @arm_fir_f32_1_4_mve(ptr nocapture readonly %S, ptr nocapture readonly %pSrc, ptr %pDst, i32 %blockSize) {744; CHECK-LABEL: arm_fir_f32_1_4_mve:745; CHECK:       @ %bb.0: @ %entry746; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}747; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}748; CHECK-NEXT:    .pad #8749; CHECK-NEXT:    sub sp, #8750; CHECK-NEXT:    ldrh.w r10, [r0]751; CHECK-NEXT:    mov r11, r1752; CHECK-NEXT:    ldr.w r12, [r0, #4]753; CHECK-NEXT:    sub.w r1, r10, #1754; CHECK-NEXT:    cmp r1, #3755; CHECK-NEXT:    bhi .LBB15_6756; CHECK-NEXT:  @ %bb.1: @ %if.then757; CHECK-NEXT:    ldr r4, [r0, #8]758; CHECK-NEXT:    ldrd r7, r6, [r4]759; CHECK-NEXT:    ldrd r5, r8, [r4, #8]760; CHECK-NEXT:    add.w r4, r12, r1, lsl #2761; CHECK-NEXT:    lsrs r1, r3, #2762; CHECK-NEXT:    wls lr, r1, .LBB15_5763; CHECK-NEXT:  @ %bb.2: @ %while.body.lr.ph764; CHECK-NEXT:    bic r1, r3, #3765; CHECK-NEXT:    str r1, [sp] @ 4-byte Spill766; CHECK-NEXT:    add.w r9, r12, #4767; CHECK-NEXT:    add.w r1, r2, r1, lsl #2768; CHECK-NEXT:    str r1, [sp, #4] @ 4-byte Spill769; CHECK-NEXT:    mov r1, r11770; CHECK-NEXT:  .LBB15_3: @ %while.body771; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1772; CHECK-NEXT:    vldrw.u32 q0, [r1], #16773; CHECK-NEXT:    vstrb.8 q0, [r4], #16774; CHECK-NEXT:    vldrw.u32 q0, [r9, #-4]775; CHECK-NEXT:    vldrw.u32 q1, [r9], #16776; CHECK-NEXT:    vmul.f32 q0, q0, r7777; CHECK-NEXT:    vldrw.u32 q2, [r9, #-8]778; CHECK-NEXT:    vfma.f32 q0, q1, r6779; CHECK-NEXT:    vldrw.u32 q1, [r9, #-12]780; CHECK-NEXT:    vfma.f32 q0, q1, r5781; CHECK-NEXT:    vfma.f32 q0, q2, r8782; CHECK-NEXT:    vstrb.8 q0, [r2], #16783; CHECK-NEXT:    le lr, .LBB15_3784; CHECK-NEXT:  @ %bb.4: @ %while.end.loopexit785; CHECK-NEXT:    ldr r1, [sp] @ 4-byte Reload786; CHECK-NEXT:    ldr r2, [sp, #4] @ 4-byte Reload787; CHECK-NEXT:    add.w r12, r12, r1, lsl #2788; CHECK-NEXT:    add.w r11, r11, r1, lsl #2789; CHECK-NEXT:  .LBB15_5: @ %while.end790; CHECK-NEXT:    and r1, r3, #3791; CHECK-NEXT:    vldrw.u32 q0, [r11]792; CHECK-NEXT:    vctp.32 r1793; CHECK-NEXT:    vpst794; CHECK-NEXT:    vstrwt.32 q0, [r4]795; CHECK-NEXT:    vldrw.u32 q0, [r12]796; CHECK-NEXT:    vldrw.u32 q1, [r12, #4]797; CHECK-NEXT:    vmul.f32 q0, q0, r7798; CHECK-NEXT:    vfma.f32 q0, q1, r6799; CHECK-NEXT:    vldrw.u32 q1, [r12, #8]800; CHECK-NEXT:    vfma.f32 q0, q1, r5801; CHECK-NEXT:    vldrw.u32 q1, [r12, #12]802; CHECK-NEXT:    vfma.f32 q0, q1, r8803; CHECK-NEXT:    vpst804; CHECK-NEXT:    vstrwt.32 q0, [r2]805; CHECK-NEXT:    ldr.w r12, [r0, #4]806; CHECK-NEXT:  .LBB15_6: @ %if.end807; CHECK-NEXT:    add.w r0, r12, r3, lsl #2808; CHECK-NEXT:    lsr.w r1, r10, #2809; CHECK-NEXT:    wls lr, r1, .LBB15_10810; CHECK-NEXT:  @ %bb.7: @ %while.body51.preheader811; CHECK-NEXT:    bic r2, r10, #3812; CHECK-NEXT:    adds r1, r2, r3813; CHECK-NEXT:    mov r3, r12814; CHECK-NEXT:    add.w r1, r12, r1, lsl #2815; CHECK-NEXT:  .LBB15_8: @ %while.body51816; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1817; CHECK-NEXT:    vldrw.u32 q0, [r0], #16818; CHECK-NEXT:    vstrb.8 q0, [r3], #16819; CHECK-NEXT:    le lr, .LBB15_8820; CHECK-NEXT:  @ %bb.9: @ %while.end55.loopexit821; CHECK-NEXT:    add.w r12, r12, r2, lsl #2822; CHECK-NEXT:    mov r0, r1823; CHECK-NEXT:  .LBB15_10: @ %while.end55824; CHECK-NEXT:    ands r1, r10, #3825; CHECK-NEXT:    beq .LBB15_12826; CHECK-NEXT:  @ %bb.11: @ %if.then59827; CHECK-NEXT:    vldrw.u32 q0, [r0]828; CHECK-NEXT:    vctp.32 r1829; CHECK-NEXT:    vpst830; CHECK-NEXT:    vstrwt.32 q0, [r12]831; CHECK-NEXT:  .LBB15_12: @ %if.end61832; CHECK-NEXT:    add sp, #8833; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}834entry:835  %pState1 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 1836  %i = load ptr, ptr %pState1, align 4837  %pCoeffs2 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 2838  %i1 = load ptr, ptr %pCoeffs2, align 4839  %numTaps3 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 0840  %i2 = load i16, ptr %numTaps3, align 4841  %conv = zext i16 %i2 to i32842  %sub = add nsw i32 %conv, -1843  %cmp = icmp ult i32 %sub, 4844  br i1 %cmp, label %if.then, label %if.end845 846if.then:                                          ; preds = %entry847  %arrayidx = getelementptr inbounds float, ptr %i, i32 %sub848  %incdec.ptr = getelementptr inbounds float, ptr %i1, i32 1849  %i3 = load float, ptr %i1, align 4850  %incdec.ptr6 = getelementptr inbounds float, ptr %i1, i32 2851  %i4 = load float, ptr %incdec.ptr, align 4852  %incdec.ptr7 = getelementptr inbounds float, ptr %i1, i32 3853  %i5 = load float, ptr %incdec.ptr6, align 4854  %i6 = load float, ptr %incdec.ptr7, align 4855  %shr = lshr i32 %blockSize, 2856  %cmp9146 = icmp eq i32 %shr, 0857  %.pre161 = insertelement <4 x float> undef, float %i3, i32 0858  %.pre162 = shufflevector <4 x float> %.pre161, <4 x float> undef, <4 x i32> zeroinitializer859  %.pre163 = insertelement <4 x float> undef, float %i4, i32 0860  %.pre164 = shufflevector <4 x float> %.pre163, <4 x float> undef, <4 x i32> zeroinitializer861  %.pre165 = insertelement <4 x float> undef, float %i5, i32 0862  %.pre166 = shufflevector <4 x float> %.pre165, <4 x float> undef, <4 x i32> zeroinitializer863  %.pre167 = insertelement <4 x float> undef, float %i6, i32 0864  %.pre168 = shufflevector <4 x float> %.pre167, <4 x float> undef, <4 x i32> zeroinitializer865  br i1 %cmp9146, label %while.end, label %while.body.lr.ph866 867while.body.lr.ph:                                 ; preds = %if.then868  %i7 = and i32 %blockSize, -4869  %scevgep158 = getelementptr float, ptr %pDst, i32 %i7870  br label %while.body871 872while.body:                                       ; preds = %while.body, %while.body.lr.ph873  %pStateCur.0151 = phi ptr [ %arrayidx, %while.body.lr.ph ], [ %add.ptr, %while.body ]874  %pSamples.0150 = phi ptr [ %i, %while.body.lr.ph ], [ %add.ptr24, %while.body ]875  %pOutput.0149 = phi ptr [ %pDst, %while.body.lr.ph ], [ %add.ptr23, %while.body ]876  %pTempSrc.0148 = phi ptr [ %pSrc, %while.body.lr.ph ], [ %add.ptr11, %while.body ]877  %blkCnt.0147 = phi i32 [ %shr, %while.body.lr.ph ], [ %dec, %while.body ]878  %i9 = load <4 x float>, ptr %pTempSrc.0148, align 4879  store <4 x float> %i9, ptr %pStateCur.0151, align 4880  %add.ptr = getelementptr inbounds float, ptr %pStateCur.0151, i32 4881  %add.ptr11 = getelementptr inbounds float, ptr %pTempSrc.0148, i32 4882  %i12 = load <4 x float>, ptr %pSamples.0150, align 4883  %i13 = fmul fast <4 x float> %i12, %.pre162884  %arrayidx12 = getelementptr inbounds float, ptr %pSamples.0150, i32 1885  %i15 = load <4 x float>, ptr %arrayidx12, align 4886  %mul = fmul fast <4 x float> %i15, %.pre164887  %add = fadd fast <4 x float> %mul, %i13888  %arrayidx13 = getelementptr inbounds float, ptr %pSamples.0150, i32 2889  %i17 = load <4 x float>, ptr %arrayidx13, align 4890  %mul16 = fmul fast <4 x float> %i17, %.pre166891  %add17 = fadd fast <4 x float> %add, %mul16892  %arrayidx18 = getelementptr inbounds float, ptr %pSamples.0150, i32 3893  %i19 = load <4 x float>, ptr %arrayidx18, align 4894  %mul21 = fmul fast <4 x float> %i19, %.pre168895  %add22 = fadd fast <4 x float> %add17, %mul21896  store <4 x float> %add22, ptr %pOutput.0149, align 4897  %add.ptr23 = getelementptr inbounds float, ptr %pOutput.0149, i32 4898  %add.ptr24 = getelementptr inbounds float, ptr %pSamples.0150, i32 4899  %dec = add nsw i32 %blkCnt.0147, -1900  %cmp9 = icmp eq i32 %dec, 0901  br i1 %cmp9, label %while.end.loopexit, label %while.body902 903while.end.loopexit:                               ; preds = %while.body904  %scevgep157 = getelementptr float, ptr %pSrc, i32 %i7905  %scevgep159 = getelementptr float, ptr %i, i32 %i7906  br label %while.end907 908while.end:                                        ; preds = %while.end.loopexit, %if.then909  %pTempSrc.0.lcssa = phi ptr [ %scevgep157, %while.end.loopexit ], [ %pSrc, %if.then ]910  %pOutput.0.lcssa = phi ptr [ %scevgep158, %while.end.loopexit ], [ %pDst, %if.then ]911  %pSamples.0.lcssa = phi ptr [ %scevgep159, %while.end.loopexit ], [ %i, %if.then ]912  %pStateCur.0.lcssa = phi ptr [ %add.ptr, %while.end.loopexit ], [ %arrayidx, %if.then ]913  %and = and i32 %blockSize, 3914  %i21 = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %and)915  %i23 = load <4 x float>, ptr %pTempSrc.0.lcssa, align 4916  tail call void @llvm.masked.store.v4f32.p0(<4 x float> %i23, ptr %pStateCur.0.lcssa, i32 4, <4 x i1> %i21)917  %i26 = load <4 x float>, ptr %pSamples.0.lcssa, align 4918  %i27 = fmul fast <4 x float> %i26, %.pre162919  %arrayidx29 = getelementptr inbounds float, ptr %pSamples.0.lcssa, i32 1920  %i29 = load <4 x float>, ptr %arrayidx29, align 4921  %mul32 = fmul fast <4 x float> %i29, %.pre164922  %add33 = fadd fast <4 x float> %mul32, %i27923  %arrayidx34 = getelementptr inbounds float, ptr %pSamples.0.lcssa, i32 2924  %i31 = load <4 x float>, ptr %arrayidx34, align 4925  %mul37 = fmul fast <4 x float> %i31, %.pre166926  %add38 = fadd fast <4 x float> %add33, %mul37927  %arrayidx39 = getelementptr inbounds float, ptr %pSamples.0.lcssa, i32 3928  %i33 = load <4 x float>, ptr %arrayidx39, align 4929  %mul42 = fmul fast <4 x float> %i33, %.pre168930  %add43 = fadd fast <4 x float> %add38, %mul42931  tail call void @llvm.masked.store.v4f32.p0(<4 x float> %add43, ptr %pOutput.0.lcssa, i32 4, <4 x i1> %i21)932  %.pre = load ptr, ptr %pState1, align 4933  br label %if.end934 935if.end:                                           ; preds = %while.end, %entry936  %i35 = phi ptr [ %.pre, %while.end ], [ %i, %entry ]937  %arrayidx45 = getelementptr inbounds float, ptr %i35, i32 %blockSize938  %shr47 = lshr i32 %conv, 2939  %cmp49141 = icmp eq i32 %shr47, 0940  br i1 %cmp49141, label %while.end55, label %while.body51.preheader941 942while.body51.preheader:                           ; preds = %if.end943  %i36 = and i32 %conv, 65532944  %i37 = add i32 %i36, %blockSize945  %scevgep = getelementptr float, ptr %i35, i32 %i37946  br label %while.body51947 948while.body51:                                     ; preds = %while.body51, %while.body51.preheader949  %pTempSrc.1144 = phi ptr [ %add.ptr52, %while.body51 ], [ %arrayidx45, %while.body51.preheader ]950  %pTempDest.0143 = phi ptr [ %add.ptr53, %while.body51 ], [ %i35, %while.body51.preheader ]951  %blkCnt.1142 = phi i32 [ %dec54, %while.body51 ], [ %shr47, %while.body51.preheader ]952  %i39 = load <4 x float>, ptr %pTempSrc.1144, align 4953  store <4 x float> %i39, ptr %pTempDest.0143, align 4954  %add.ptr52 = getelementptr inbounds float, ptr %pTempSrc.1144, i32 4955  %add.ptr53 = getelementptr inbounds float, ptr %pTempDest.0143, i32 4956  %dec54 = add nsw i32 %blkCnt.1142, -1957  %cmp49 = icmp eq i32 %dec54, 0958  br i1 %cmp49, label %while.end55.loopexit, label %while.body51959 960while.end55.loopexit:                             ; preds = %while.body51961  %scevgep156 = getelementptr float, ptr %i35, i32 %i36962  br label %while.end55963 964while.end55:                                      ; preds = %while.end55.loopexit, %if.end965  %pTempDest.0.lcssa = phi ptr [ %i35, %if.end ], [ %scevgep156, %while.end55.loopexit ]966  %pTempSrc.1.lcssa = phi ptr [ %arrayidx45, %if.end ], [ %scevgep, %while.end55.loopexit ]967  %and56 = and i32 %conv, 3968  %cmp57 = icmp eq i32 %and56, 0969  br i1 %cmp57, label %if.end61, label %if.then59970 971if.then59:                                        ; preds = %while.end55972  %i41 = tail call <4 x i1> @llvm.arm.mve.vctp32(i32 %and56)973  %i43 = load <4 x float>, ptr %pTempSrc.1.lcssa, align 4974  tail call void @llvm.masked.store.v4f32.p0(<4 x float> %i43, ptr %pTempDest.0.lcssa, i32 4, <4 x i1> %i41)975  br label %if.end61976 977if.end61:                                         ; preds = %if.then59, %while.end55978  ret void979}980 981 982define void @fir(ptr nocapture readonly %S, ptr nocapture readonly %pSrc, ptr nocapture %pDst, i32 %blockSize) {983; CHECK-LABEL: fir:984; CHECK:       @ %bb.0: @ %entry985; CHECK-NEXT:    cmp r3, #8986; CHECK-NEXT:    blo.w .LBB16_13987; CHECK-NEXT:  @ %bb.1: @ %if.then988; CHECK-NEXT:    lsrs.w r12, r3, #2989; CHECK-NEXT:    it eq990; CHECK-NEXT:    bxeq lr991; CHECK-NEXT:  .LBB16_2: @ %while.body.lr.ph992; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}993; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}994; CHECK-NEXT:    .pad #4995; CHECK-NEXT:    sub sp, #4996; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13}997; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13}998; CHECK-NEXT:    .pad #24999; CHECK-NEXT:    sub sp, #241000; CHECK-NEXT:    ldrh r6, [r0]1001; CHECK-NEXT:    movs r4, #11002; CHECK-NEXT:    ldrd r7, r10, [r0, #4]1003; CHECK-NEXT:    sub.w r0, r6, #81004; CHECK-NEXT:    add.w r3, r0, r0, lsr #291005; CHECK-NEXT:    and r0, r0, #71006; CHECK-NEXT:    asrs r5, r3, #31007; CHECK-NEXT:    cmp r5, #11008; CHECK-NEXT:    it gt1009; CHECK-NEXT:    asrgt r4, r3, #31010; CHECK-NEXT:    add.w r3, r7, r6, lsl #21011; CHECK-NEXT:    sub.w r9, r3, #41012; CHECK-NEXT:    rsbs r3, r6, #01013; CHECK-NEXT:    str r4, [sp] @ 4-byte Spill1014; CHECK-NEXT:    str r6, [sp, #8] @ 4-byte Spill1015; CHECK-NEXT:    str r3, [sp, #4] @ 4-byte Spill1016; CHECK-NEXT:    str r0, [sp, #12] @ 4-byte Spill1017; CHECK-NEXT:    b .LBB16_61018; CHECK-NEXT:  .LBB16_3: @ %while.end.loopexit1019; CHECK-NEXT:    @ in Loop: Header=BB16_6 Depth=11020; CHECK-NEXT:    ldr r0, [sp, #12] @ 4-byte Reload1021; CHECK-NEXT:    add.w r7, r7, r0, lsl #21022; CHECK-NEXT:    b .LBB16_51023; CHECK-NEXT:  .LBB16_4: @ %for.end1024; CHECK-NEXT:    @ in Loop: Header=BB16_6 Depth=11025; CHECK-NEXT:    ldr r1, [sp, #20] @ 4-byte Reload1026; CHECK-NEXT:    ldrd r0, r9, [sp, #12] @ 8-byte Folded Reload1027; CHECK-NEXT:    wls lr, r0, .LBB16_51028; CHECK-NEXT:    b .LBB16_101029; CHECK-NEXT:  .LBB16_5: @ %while.end1030; CHECK-NEXT:    @ in Loop: Header=BB16_6 Depth=11031; CHECK-NEXT:    ldr r0, [sp, #4] @ 4-byte Reload1032; CHECK-NEXT:    subs.w r12, r12, #11033; CHECK-NEXT:    vstrb.8 q0, [r2], #161034; CHECK-NEXT:    add.w r0, r7, r0, lsl #21035; CHECK-NEXT:    add.w r7, r0, #161036; CHECK-NEXT:    beq .LBB16_121037; CHECK-NEXT:  .LBB16_6: @ %while.body1038; CHECK-NEXT:    @ =>This Loop Header: Depth=11039; CHECK-NEXT:    @ Child Loop BB16_8 Depth 21040; CHECK-NEXT:    @ Child Loop BB16_11 Depth 21041; CHECK-NEXT:    add.w lr, r10, #81042; CHECK-NEXT:    vldrw.u32 q0, [r1], #161043; CHECK-NEXT:    ldrd r3, r4, [r10]1044; CHECK-NEXT:    ldm.w lr, {r0, r5, r6, lr}1045; CHECK-NEXT:    ldrd r11, r8, [r10, #24]1046; CHECK-NEXT:    vstrb.8 q0, [r9], #161047; CHECK-NEXT:    vldrw.u32 q0, [r7], #321048; CHECK-NEXT:    str r1, [sp, #20] @ 4-byte Spill1049; CHECK-NEXT:    str.w r9, [sp, #16] @ 4-byte Spill1050; CHECK-NEXT:    vldrw.u32 q1, [r7, #-28]1051; CHECK-NEXT:    vmul.f32 q0, q0, r31052; CHECK-NEXT:    vldrw.u32 q6, [r7, #-24]1053; CHECK-NEXT:    vldrw.u32 q4, [r7, #-20]1054; CHECK-NEXT:    vfma.f32 q0, q1, r41055; CHECK-NEXT:    vldrw.u32 q5, [r7, #-16]1056; CHECK-NEXT:    vfma.f32 q0, q6, r01057; CHECK-NEXT:    vldrw.u32 q2, [r7, #-12]1058; CHECK-NEXT:    vfma.f32 q0, q4, r51059; CHECK-NEXT:    vldrw.u32 q3, [r7, #-8]1060; CHECK-NEXT:    vfma.f32 q0, q5, r61061; CHECK-NEXT:    vldrw.u32 q1, [r7, #-4]1062; CHECK-NEXT:    vfma.f32 q0, q2, lr1063; CHECK-NEXT:    ldr r0, [sp, #8] @ 4-byte Reload1064; CHECK-NEXT:    vfma.f32 q0, q3, r111065; CHECK-NEXT:    vfma.f32 q0, q1, r81066; CHECK-NEXT:    cmp r0, #161067; CHECK-NEXT:    blo .LBB16_91068; CHECK-NEXT:  @ %bb.7: @ %for.body.preheader1069; CHECK-NEXT:    @ in Loop: Header=BB16_6 Depth=11070; CHECK-NEXT:    ldr r0, [sp] @ 4-byte Reload1071; CHECK-NEXT:    add.w r4, r10, #321072; CHECK-NEXT:    dls lr, r01073; CHECK-NEXT:  .LBB16_8: @ %for.body1074; CHECK-NEXT:    @ Parent Loop BB16_6 Depth=11075; CHECK-NEXT:    @ => This Inner Loop Header: Depth=21076; CHECK-NEXT:    ldm.w r4, {r0, r3, r5, r6, r8, r11}1077; CHECK-NEXT:    vldrw.u32 q1, [r7], #321078; CHECK-NEXT:    vldrw.u32 q6, [r7, #-24]1079; CHECK-NEXT:    vldrw.u32 q4, [r7, #-20]1080; CHECK-NEXT:    vfma.f32 q0, q1, r01081; CHECK-NEXT:    vldrw.u32 q1, [r7, #-28]1082; CHECK-NEXT:    vldrw.u32 q5, [r7, #-16]1083; CHECK-NEXT:    vldrw.u32 q2, [r7, #-12]1084; CHECK-NEXT:    vfma.f32 q0, q1, r31085; CHECK-NEXT:    ldrd r9, r1, [r4, #24]1086; CHECK-NEXT:    vfma.f32 q0, q6, r51087; CHECK-NEXT:    vldrw.u32 q3, [r7, #-8]1088; CHECK-NEXT:    vfma.f32 q0, q4, r61089; CHECK-NEXT:    vldrw.u32 q1, [r7, #-4]1090; CHECK-NEXT:    vfma.f32 q0, q5, r81091; CHECK-NEXT:    adds r4, #321092; CHECK-NEXT:    vfma.f32 q0, q2, r111093; CHECK-NEXT:    vfma.f32 q0, q3, r91094; CHECK-NEXT:    vfma.f32 q0, q1, r11095; CHECK-NEXT:    le lr, .LBB16_81096; CHECK-NEXT:    b .LBB16_41097; CHECK-NEXT:  .LBB16_9: @ in Loop: Header=BB16_6 Depth=11098; CHECK-NEXT:    add.w r4, r10, #321099; CHECK-NEXT:    b .LBB16_41100; CHECK-NEXT:  .LBB16_10: @ %while.body76.preheader1101; CHECK-NEXT:    @ in Loop: Header=BB16_6 Depth=11102; CHECK-NEXT:    mov r3, r71103; CHECK-NEXT:  .LBB16_11: @ %while.body761104; CHECK-NEXT:    @ Parent Loop BB16_6 Depth=11105; CHECK-NEXT:    @ => This Inner Loop Header: Depth=21106; CHECK-NEXT:    ldr r0, [r4], #41107; CHECK-NEXT:    vldrw.u32 q1, [r3], #41108; CHECK-NEXT:    vfma.f32 q0, q1, r01109; CHECK-NEXT:    le lr, .LBB16_111110; CHECK-NEXT:    b .LBB16_31111; CHECK-NEXT:  .LBB16_12:1112; CHECK-NEXT:    add sp, #241113; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13}1114; CHECK-NEXT:    add sp, #41115; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}1116; CHECK-NEXT:  .LBB16_13: @ %if.end1117; CHECK-NEXT:    bx lr1118entry:1119  %pState1 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 11120  %i = load ptr, ptr %pState1, align 41121  %pCoeffs2 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 21122  %i1 = load ptr, ptr %pCoeffs2, align 41123  %numTaps3 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 01124  %i2 = load i16, ptr %numTaps3, align 41125  %conv = zext i16 %i2 to i321126  %cmp = icmp ugt i32 %blockSize, 71127  br i1 %cmp, label %if.then, label %if.end1128 1129if.then:                                          ; preds = %entry1130  %shr = lshr i32 %blockSize, 21131  %cmp5217 = icmp eq i32 %shr, 01132  br i1 %cmp5217, label %if.end, label %while.body.lr.ph1133 1134while.body.lr.ph:                                 ; preds = %if.then1135  %sub = add nsw i32 %conv, -11136  %arrayidx = getelementptr inbounds float, ptr %i, i32 %sub1137  %incdec.ptr = getelementptr inbounds float, ptr %i1, i32 11138  %incdec.ptr7 = getelementptr inbounds float, ptr %i1, i32 21139  %incdec.ptr8 = getelementptr inbounds float, ptr %i1, i32 31140  %incdec.ptr9 = getelementptr inbounds float, ptr %i1, i32 41141  %incdec.ptr10 = getelementptr inbounds float, ptr %i1, i32 51142  %incdec.ptr11 = getelementptr inbounds float, ptr %i1, i32 61143  %incdec.ptr12 = getelementptr inbounds float, ptr %i1, i32 71144  %sub37 = add nsw i32 %conv, -81145  %div = sdiv i32 %sub37, 81146  %pCoeffsCur.0199 = getelementptr inbounds float, ptr %i1, i32 81147  %cmp38201 = icmp ugt i16 %i2, 151148  %and = and i32 %sub37, 71149  %cmp74210 = icmp eq i32 %and, 01150  %idx.neg = sub nsw i32 0, %conv1151  %i3 = icmp sgt i32 %div, 11152  %smax = select i1 %i3, i32 %div, i32 11153  br label %while.body1154 1155while.body:                                       ; preds = %while.end, %while.body.lr.ph1156  %blkCnt.0222 = phi i32 [ %shr, %while.body.lr.ph ], [ %dec84, %while.end ]1157  %pStateCur.0221 = phi ptr [ %arrayidx, %while.body.lr.ph ], [ %add.ptr, %while.end ]1158  %pSamples.0220 = phi ptr [ %i, %while.body.lr.ph ], [ %add.ptr83, %while.end ]1159  %pTempSrc.0219 = phi ptr [ %pSrc, %while.body.lr.ph ], [ %add.ptr14, %while.end ]1160  %pOutput.0218 = phi ptr [ %pDst, %while.body.lr.ph ], [ %add.ptr81, %while.end ]1161  %i4 = load float, ptr %i1, align 41162  %i5 = load float, ptr %incdec.ptr, align 41163  %i6 = load float, ptr %incdec.ptr7, align 41164  %i7 = load float, ptr %incdec.ptr8, align 41165  %i8 = load float, ptr %incdec.ptr9, align 41166  %i9 = load float, ptr %incdec.ptr10, align 41167  %i10 = load float, ptr %incdec.ptr11, align 41168  %i11 = load float, ptr %incdec.ptr12, align 41169  %i13 = load <4 x float>, ptr %pTempSrc.0219, align 41170  store <4 x float> %i13, ptr %pStateCur.0221, align 41171  %add.ptr = getelementptr inbounds float, ptr %pStateCur.0221, i32 41172  %add.ptr14 = getelementptr inbounds float, ptr %pTempSrc.0219, i32 41173  %i16 = load <4 x float>, ptr %pSamples.0220, align 41174  %.splatinsert = insertelement <4 x float> undef, float %i4, i32 01175  %.splat = shufflevector <4 x float> %.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer1176  %i17 = fmul fast <4 x float> %i16, %.splat1177  %arrayidx15 = getelementptr inbounds float, ptr %pSamples.0220, i32 11178  %i19 = load <4 x float>, ptr %arrayidx15, align 41179  %.splatinsert16 = insertelement <4 x float> undef, float %i5, i32 01180  %.splat17 = shufflevector <4 x float> %.splatinsert16, <4 x float> undef, <4 x i32> zeroinitializer1181  %i20 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i19, <4 x float> %.splat17, <4 x float> %i17)1182  %arrayidx18 = getelementptr inbounds float, ptr %pSamples.0220, i32 21183  %i22 = load <4 x float>, ptr %arrayidx18, align 41184  %.splatinsert19 = insertelement <4 x float> undef, float %i6, i32 01185  %.splat20 = shufflevector <4 x float> %.splatinsert19, <4 x float> undef, <4 x i32> zeroinitializer1186  %i23 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i22, <4 x float> %.splat20, <4 x float> %i20)1187  %arrayidx21 = getelementptr inbounds float, ptr %pSamples.0220, i32 31188  %i25 = load <4 x float>, ptr %arrayidx21, align 41189  %.splatinsert22 = insertelement <4 x float> undef, float %i7, i32 01190  %.splat23 = shufflevector <4 x float> %.splatinsert22, <4 x float> undef, <4 x i32> zeroinitializer1191  %i26 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i25, <4 x float> %.splat23, <4 x float> %i23)1192  %arrayidx24 = getelementptr inbounds float, ptr %pSamples.0220, i32 41193  %i28 = load <4 x float>, ptr %arrayidx24, align 41194  %.splatinsert25 = insertelement <4 x float> undef, float %i8, i32 01195  %.splat26 = shufflevector <4 x float> %.splatinsert25, <4 x float> undef, <4 x i32> zeroinitializer1196  %i29 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i28, <4 x float> %.splat26, <4 x float> %i26)1197  %arrayidx27 = getelementptr inbounds float, ptr %pSamples.0220, i32 51198  %i31 = load <4 x float>, ptr %arrayidx27, align 41199  %.splatinsert28 = insertelement <4 x float> undef, float %i9, i32 01200  %.splat29 = shufflevector <4 x float> %.splatinsert28, <4 x float> undef, <4 x i32> zeroinitializer1201  %i32 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i31, <4 x float> %.splat29, <4 x float> %i29)1202  %arrayidx30 = getelementptr inbounds float, ptr %pSamples.0220, i32 61203  %i34 = load <4 x float>, ptr %arrayidx30, align 41204  %.splatinsert31 = insertelement <4 x float> undef, float %i10, i32 01205  %.splat32 = shufflevector <4 x float> %.splatinsert31, <4 x float> undef, <4 x i32> zeroinitializer1206  %i35 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i34, <4 x float> %.splat32, <4 x float> %i32)1207  %arrayidx33 = getelementptr inbounds float, ptr %pSamples.0220, i32 71208  %i37 = load <4 x float>, ptr %arrayidx33, align 41209  %.splatinsert34 = insertelement <4 x float> undef, float %i11, i32 01210  %.splat35 = shufflevector <4 x float> %.splatinsert34, <4 x float> undef, <4 x i32> zeroinitializer1211  %i38 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i37, <4 x float> %.splat35, <4 x float> %i35)1212  %pSamples.1200 = getelementptr inbounds float, ptr %pSamples.0220, i32 81213  br i1 %cmp38201, label %for.body, label %for.end1214 1215for.body:                                         ; preds = %for.body, %while.body1216  %pSamples.1207 = phi ptr [ %pSamples.1, %for.body ], [ %pSamples.1200, %while.body ]1217  %pCoeffsCur.0206 = phi ptr [ %pCoeffsCur.0, %for.body ], [ %pCoeffsCur.0199, %while.body ]1218  %.pn205 = phi ptr [ %pCoeffsCur.0206, %for.body ], [ %i1, %while.body ]1219  %i.0204 = phi i32 [ %inc, %for.body ], [ 0, %while.body ]1220  %vecAcc0.0203 = phi <4 x float> [ %i70, %for.body ], [ %i38, %while.body ]1221  %pSamples.0.pn202 = phi ptr [ %pSamples.1207, %for.body ], [ %pSamples.0220, %while.body ]1222  %incdec.ptr40 = getelementptr inbounds float, ptr %.pn205, i32 91223  %i39 = load float, ptr %pCoeffsCur.0206, align 41224  %incdec.ptr41 = getelementptr inbounds float, ptr %.pn205, i32 101225  %i40 = load float, ptr %incdec.ptr40, align 41226  %incdec.ptr42 = getelementptr inbounds float, ptr %.pn205, i32 111227  %i41 = load float, ptr %incdec.ptr41, align 41228  %incdec.ptr43 = getelementptr inbounds float, ptr %.pn205, i32 121229  %i42 = load float, ptr %incdec.ptr42, align 41230  %incdec.ptr44 = getelementptr inbounds float, ptr %.pn205, i32 131231  %i43 = load float, ptr %incdec.ptr43, align 41232  %incdec.ptr45 = getelementptr inbounds float, ptr %.pn205, i32 141233  %i44 = load float, ptr %incdec.ptr44, align 41234  %incdec.ptr46 = getelementptr inbounds float, ptr %.pn205, i32 151235  %i45 = load float, ptr %incdec.ptr45, align 41236  %i46 = load float, ptr %incdec.ptr46, align 41237  %i48 = load <4 x float>, ptr %pSamples.1207, align 41238  %.splatinsert48 = insertelement <4 x float> undef, float %i39, i32 01239  %.splat49 = shufflevector <4 x float> %.splatinsert48, <4 x float> undef, <4 x i32> zeroinitializer1240  %i49 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i48, <4 x float> %.splat49, <4 x float> %vecAcc0.0203)1241  %arrayidx50 = getelementptr inbounds float, ptr %pSamples.0.pn202, i32 91242  %i51 = load <4 x float>, ptr %arrayidx50, align 41243  %.splatinsert51 = insertelement <4 x float> undef, float %i40, i32 01244  %.splat52 = shufflevector <4 x float> %.splatinsert51, <4 x float> undef, <4 x i32> zeroinitializer1245  %i52 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i51, <4 x float> %.splat52, <4 x float> %i49)1246  %arrayidx53 = getelementptr inbounds float, ptr %pSamples.0.pn202, i32 101247  %i54 = load <4 x float>, ptr %arrayidx53, align 41248  %.splatinsert54 = insertelement <4 x float> undef, float %i41, i32 01249  %.splat55 = shufflevector <4 x float> %.splatinsert54, <4 x float> undef, <4 x i32> zeroinitializer1250  %i55 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i54, <4 x float> %.splat55, <4 x float> %i52)1251  %arrayidx56 = getelementptr inbounds float, ptr %pSamples.0.pn202, i32 111252  %i57 = load <4 x float>, ptr %arrayidx56, align 41253  %.splatinsert57 = insertelement <4 x float> undef, float %i42, i32 01254  %.splat58 = shufflevector <4 x float> %.splatinsert57, <4 x float> undef, <4 x i32> zeroinitializer1255  %i58 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i57, <4 x float> %.splat58, <4 x float> %i55)1256  %arrayidx59 = getelementptr inbounds float, ptr %pSamples.0.pn202, i32 121257  %i60 = load <4 x float>, ptr %arrayidx59, align 41258  %.splatinsert60 = insertelement <4 x float> undef, float %i43, i32 01259  %.splat61 = shufflevector <4 x float> %.splatinsert60, <4 x float> undef, <4 x i32> zeroinitializer1260  %i61 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i60, <4 x float> %.splat61, <4 x float> %i58)1261  %arrayidx62 = getelementptr inbounds float, ptr %pSamples.0.pn202, i32 131262  %i63 = load <4 x float>, ptr %arrayidx62, align 41263  %.splatinsert63 = insertelement <4 x float> undef, float %i44, i32 01264  %.splat64 = shufflevector <4 x float> %.splatinsert63, <4 x float> undef, <4 x i32> zeroinitializer1265  %i64 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i63, <4 x float> %.splat64, <4 x float> %i61)1266  %arrayidx65 = getelementptr inbounds float, ptr %pSamples.0.pn202, i32 141267  %i66 = load <4 x float>, ptr %arrayidx65, align 41268  %.splatinsert66 = insertelement <4 x float> undef, float %i45, i32 01269  %.splat67 = shufflevector <4 x float> %.splatinsert66, <4 x float> undef, <4 x i32> zeroinitializer1270  %i67 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i66, <4 x float> %.splat67, <4 x float> %i64)1271  %arrayidx68 = getelementptr inbounds float, ptr %pSamples.0.pn202, i32 151272  %i69 = load <4 x float>, ptr %arrayidx68, align 41273  %.splatinsert69 = insertelement <4 x float> undef, float %i46, i32 01274  %.splat70 = shufflevector <4 x float> %.splatinsert69, <4 x float> undef, <4 x i32> zeroinitializer1275  %i70 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i69, <4 x float> %.splat70, <4 x float> %i67)1276  %inc = add nuw nsw i32 %i.0204, 11277  %pCoeffsCur.0 = getelementptr inbounds float, ptr %pCoeffsCur.0206, i32 81278  %pSamples.1 = getelementptr inbounds float, ptr %pSamples.1207, i32 81279  %exitcond = icmp eq i32 %inc, %smax1280  br i1 %exitcond, label %for.end, label %for.body1281 1282for.end:                                          ; preds = %for.body, %while.body1283  %vecAcc0.0.lcssa = phi <4 x float> [ %i38, %while.body ], [ %i70, %for.body ]1284  %pCoeffsCur.0.lcssa = phi ptr [ %pCoeffsCur.0199, %while.body ], [ %pCoeffsCur.0, %for.body ]1285  %pSamples.1.lcssa = phi ptr [ %pSamples.1200, %while.body ], [ %pSamples.1, %for.body ]1286  br i1 %cmp74210, label %while.end, label %while.body761287 1288while.body76:                                     ; preds = %while.body76, %for.end1289  %pCoeffsCur.1214 = phi ptr [ %incdec.ptr77, %while.body76 ], [ %pCoeffsCur.0.lcssa, %for.end ]1290  %vecAcc0.1213 = phi <4 x float> [ %i74, %while.body76 ], [ %vecAcc0.0.lcssa, %for.end ]1291  %numCnt.0212 = phi i32 [ %dec, %while.body76 ], [ %and, %for.end ]1292  %pSamples.2211 = phi ptr [ %incdec.ptr80, %while.body76 ], [ %pSamples.1.lcssa, %for.end ]1293  %incdec.ptr77 = getelementptr inbounds float, ptr %pCoeffsCur.1214, i32 11294  %i71 = load float, ptr %pCoeffsCur.1214, align 41295  %i73 = load <4 x float>, ptr %pSamples.2211, align 41296  %.splatinsert78 = insertelement <4 x float> undef, float %i71, i32 01297  %.splat79 = shufflevector <4 x float> %.splatinsert78, <4 x float> undef, <4 x i32> zeroinitializer1298  %i74 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i73, <4 x float> %.splat79, <4 x float> %vecAcc0.1213)1299  %incdec.ptr80 = getelementptr inbounds float, ptr %pSamples.2211, i32 11300  %dec = add nsw i32 %numCnt.0212, -11301  %cmp74 = icmp sgt i32 %numCnt.0212, 11302  br i1 %cmp74, label %while.body76, label %while.end.loopexit1303 1304while.end.loopexit:                               ; preds = %while.body761305  %scevgep = getelementptr float, ptr %pSamples.1.lcssa, i32 %and1306  br label %while.end1307 1308while.end:                                        ; preds = %while.end.loopexit, %for.end1309  %pSamples.2.lcssa = phi ptr [ %pSamples.1.lcssa, %for.end ], [ %scevgep, %while.end.loopexit ]1310  %vecAcc0.1.lcssa = phi <4 x float> [ %vecAcc0.0.lcssa, %for.end ], [ %i74, %while.end.loopexit ]1311  store <4 x float> %vecAcc0.1.lcssa, ptr %pOutput.0218, align 41312  %add.ptr81 = getelementptr inbounds float, ptr %pOutput.0218, i32 41313  %add.ptr82 = getelementptr inbounds float, ptr %pSamples.2.lcssa, i32 41314  %add.ptr83 = getelementptr inbounds float, ptr %add.ptr82, i32 %idx.neg1315  %dec84 = add nsw i32 %blkCnt.0222, -11316  %cmp5 = icmp eq i32 %dec84, 01317  br i1 %cmp5, label %if.end, label %while.body1318 1319if.end:                                           ; preds = %while.end, %if.then, %entry1320  ret void1321}1322 1323%struct.arm_biquad_cascade_stereo_df2T_instance_f32 = type { i8, ptr, ptr }1324define arm_aapcs_vfpcc void @arm_biquad_cascade_stereo_df2T_f32(ptr nocapture readonly %arg, ptr %arg1, ptr %arg2, i32 %arg3) {1325; CHECK-LABEL: arm_biquad_cascade_stereo_df2T_f32:1326; CHECK:       @ %bb.0: @ %bb1327; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, lr}1328; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, lr}1329; CHECK-NEXT:    .vsave {d8, d9, d10, d11}1330; CHECK-NEXT:    vpush {d8, d9, d10, d11}1331; CHECK-NEXT:    .pad #241332; CHECK-NEXT:    sub sp, #241333; CHECK-NEXT:    mov r8, r31334; CHECK-NEXT:    ldrb.w r12, [r0]1335; CHECK-NEXT:    ldrd r0, r3, [r0, #4]1336; CHECK-NEXT:    movs r4, #01337; CHECK-NEXT:    cmp.w r8, #01338; CHECK-NEXT:    strd r4, r4, [sp, #16]1339; CHECK-NEXT:    beq .LBB17_51340; CHECK-NEXT:  @ %bb.1:1341; CHECK-NEXT:    movs r5, #21342; CHECK-NEXT:    viwdup.u32 q0, r4, r5, #11343; CHECK-NEXT:    mov r4, sp1344; CHECK-NEXT:  .LBB17_2: @ %bb291345; CHECK-NEXT:    @ =>This Loop Header: Depth=11346; CHECK-NEXT:    @ Child Loop BB17_3 Depth 21347; CHECK-NEXT:    ldrd r5, r7, [r3]1348; CHECK-NEXT:    vldrw.u32 q1, [r0]1349; CHECK-NEXT:    ldr r6, [r3, #12]1350; CHECK-NEXT:    vldr s8, [r3, #8]1351; CHECK-NEXT:    vstrw.32 q1, [r4]1352; CHECK-NEXT:    vdup.32 q1, r71353; CHECK-NEXT:    vldr s12, [r3, #16]1354; CHECK-NEXT:    vmov.f32 s6, s81355; CHECK-NEXT:    dls lr, r81356; CHECK-NEXT:    vmov.f32 s7, s81357; CHECK-NEXT:    vdup.32 q2, r61358; CHECK-NEXT:    vmov.f32 s10, s121359; CHECK-NEXT:    mov r7, r21360; CHECK-NEXT:    vmov.f32 s11, s121361; CHECK-NEXT:  .LBB17_3: @ %bb551362; CHECK-NEXT:    @ Parent Loop BB17_2 Depth=11363; CHECK-NEXT:    @ => This Inner Loop Header: Depth=21364; CHECK-NEXT:    vldrw.u32 q4, [r1, q0, uxtw #2]1365; CHECK-NEXT:    vldrw.u32 q5, [r4, q0, uxtw #2]1366; CHECK-NEXT:    vldrw.u32 q3, [sp, #8]1367; CHECK-NEXT:    adds r1, #81368; CHECK-NEXT:    vfma.f32 q5, q4, r51369; CHECK-NEXT:    vfma.f32 q3, q5, q21370; CHECK-NEXT:    vstmia r7!, {s20, s21}1371; CHECK-NEXT:    vfma.f32 q3, q4, q11372; CHECK-NEXT:    vstrw.32 q3, [r4]1373; CHECK-NEXT:    le lr, .LBB17_31374; CHECK-NEXT:  @ %bb.4: @ %bb751375; CHECK-NEXT:    @ in Loop: Header=BB17_2 Depth=11376; CHECK-NEXT:    adds r3, #201377; CHECK-NEXT:    subs.w r12, r12, #11378; CHECK-NEXT:    vstrb.8 q3, [r0], #161379; CHECK-NEXT:    mov r1, r21380; CHECK-NEXT:    bne .LBB17_21381; CHECK-NEXT:    b .LBB17_71382; CHECK-NEXT:  .LBB17_5: @ %bb21.preheader1383; CHECK-NEXT:    dls lr, r121384; CHECK-NEXT:    mov r1, sp1385; CHECK-NEXT:  .LBB17_6: @ %bb211386; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=11387; CHECK-NEXT:    vldrw.u32 q0, [r0], #161388; CHECK-NEXT:    vstrw.32 q0, [r1]1389; CHECK-NEXT:    le lr, .LBB17_61390; CHECK-NEXT:  .LBB17_7: @ %bb801391; CHECK-NEXT:    add sp, #241392; CHECK-NEXT:    vpop {d8, d9, d10, d11}1393; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, pc}1394bb:1395  %i = alloca [6 x float], align 41396  %i4 = getelementptr inbounds %struct.arm_biquad_cascade_stereo_df2T_instance_f32, ptr %arg, i32 0, i32 11397  %i5 = load ptr, ptr %i4, align 41398  %i6 = getelementptr inbounds %struct.arm_biquad_cascade_stereo_df2T_instance_f32, ptr %arg, i32 0, i32 21399  %i7 = load ptr, ptr %i6, align 41400  %i8 = getelementptr inbounds %struct.arm_biquad_cascade_stereo_df2T_instance_f32, ptr %arg, i32 0, i32 01401  %i9 = load i8, ptr %i8, align 41402  %i10 = zext i8 %i9 to i321403  call void @llvm.lifetime.start.p0(i64 24, ptr nonnull %i)1404  %i12 = tail call { <4 x i32>, i32 } @llvm.arm.mve.viwdup.v4i32(i32 0, i32 2, i32 1)1405  %i13 = extractvalue { <4 x i32>, i32 } %i12, 01406  %i14 = getelementptr inbounds [6 x float], ptr %i, i32 0, i32 41407  store float 0.000000e+00, ptr %i14, align 41408  %i15 = getelementptr inbounds [6 x float], ptr %i, i32 0, i32 51409  store float 0.000000e+00, ptr %i15, align 41410  %i17 = icmp eq i32 %arg3, 01411  %i19 = getelementptr inbounds [6 x float], ptr %i, i32 0, i32 21412  br i1 %i17, label %bb21, label %bb291413 1414bb21:                                             ; preds = %bb21, %bb1415  %i22 = phi i32 [ %i27, %bb21 ], [ %i10, %bb ]1416  %i23 = phi ptr [ %i26, %bb21 ], [ %i5, %bb ]1417  %i25 = load <4 x float>, ptr %i23, align 81418  store <4 x float> %i25, ptr %i, align 41419  %i26 = getelementptr inbounds float, ptr %i23, i32 41420  %i27 = add i32 %i22, -11421  %i28 = icmp eq i32 %i27, 01422  br i1 %i28, label %bb80, label %bb211423 1424bb29:                                             ; preds = %bb75, %bb1425  %i30 = phi i32 [ %i78, %bb75 ], [ %i10, %bb ]1426  %i31 = phi ptr [ %i76, %bb75 ], [ %i7, %bb ]1427  %i32 = phi ptr [ %i77, %bb75 ], [ %i5, %bb ]1428  %i33 = phi ptr [ %arg2, %bb75 ], [ %arg1, %bb ]1429  %i34 = getelementptr inbounds float, ptr %i31, i32 11430  %i35 = load float, ptr %i31, align 41431  %i36 = getelementptr inbounds float, ptr %i31, i32 21432  %i37 = load float, ptr %i34, align 41433  %i38 = getelementptr inbounds float, ptr %i31, i32 31434  %i39 = load float, ptr %i36, align 41435  %i40 = getelementptr inbounds float, ptr %i31, i32 41436  %i41 = load float, ptr %i38, align 41437  %i42 = load float, ptr %i40, align 41438  %i43 = insertelement <4 x float> undef, float %i41, i32 01439  %i44 = shufflevector <4 x float> %i43, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 undef, i32 undef>1440  %i45 = insertelement <4 x float> %i44, float %i42, i32 21441  %i46 = insertelement <4 x float> %i45, float %i42, i32 31442  %i47 = insertelement <4 x float> undef, float %i37, i32 01443  %i48 = shufflevector <4 x float> %i47, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 undef, i32 undef>1444  %i49 = insertelement <4 x float> %i48, float %i39, i32 21445  %i50 = insertelement <4 x float> %i49, float %i39, i32 31446  %i52 = load <4 x float>, ptr %i32, align 81447  store <4 x float> %i52, ptr %i, align 41448  %i53 = insertelement <4 x float> undef, float %i35, i32 01449  %i54 = shufflevector <4 x float> %i53, <4 x float> undef, <4 x i32> zeroinitializer1450  br label %bb551451 1452bb55:                                             ; preds = %bb55, %bb291453  %i56 = phi ptr [ %i33, %bb29 ], [ %i72, %bb55 ]1454  %i57 = phi ptr [ %arg2, %bb29 ], [ %i68, %bb55 ]1455  %i58 = phi i32 [ %arg3, %bb29 ], [ %i73, %bb55 ]1456  %i59 = call <4 x i32> @llvm.arm.mve.vldr.gather.offset.v4i32.p0.v4i32(ptr nonnull %i, <4 x i32> %i13, i32 32, i32 2, i32 1)1457  %i60 = bitcast <4 x i32> %i59 to <4 x float>1458  %i62 = call <4 x i32> @llvm.arm.mve.vldr.gather.offset.v4i32.p0.v4i32(ptr %i56, <4 x i32> %i13, i32 32, i32 2, i32 1)1459  %i63 = bitcast <4 x i32> %i62 to <4 x float>1460  %i64 = call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i63, <4 x float> %i54, <4 x float> %i60)1461  %i65 = extractelement <4 x float> %i64, i32 01462  %i66 = getelementptr inbounds float, ptr %i57, i32 11463  store float %i65, ptr %i57, align 41464  %i67 = extractelement <4 x float> %i64, i32 11465  %i68 = getelementptr inbounds float, ptr %i57, i32 21466  store float %i67, ptr %i66, align 41467  %i69 = load <4 x float>, ptr %i19, align 41468  %i70 = call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i64, <4 x float> %i46, <4 x float> %i69)1469  %i71 = call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i63, <4 x float> %i50, <4 x float> %i70)1470  store <4 x float> %i71, ptr %i, align 41471  %i72 = getelementptr inbounds float, ptr %i56, i32 21472  %i73 = add i32 %i58, -11473  %i74 = icmp eq i32 %i73, 01474  br i1 %i74, label %bb75, label %bb551475 1476bb75:                                             ; preds = %bb551477  %i76 = getelementptr inbounds float, ptr %i31, i32 51478  store <4 x float> %i71, ptr %i32, align 41479  %i77 = getelementptr inbounds float, ptr %i32, i32 41480  %i78 = add i32 %i30, -11481  %i79 = icmp eq i32 %i78, 01482  br i1 %i79, label %bb80, label %bb291483 1484bb80:                                             ; preds = %bb75, %bb211485  call void @llvm.lifetime.end.p0(i64 24, ptr nonnull %i)1486  ret void1487}1488 1489define arm_aapcs_vfpcc void @fms(ptr nocapture readonly %pSrc1, ptr nocapture readonly %pSrc2, ptr nocapture readonly %pSrc3, ptr nocapture %pDst, i32 %N, i32 %M) {1490; CHECK-LABEL: fms:1491; CHECK:       @ %bb.0: @ %entry1492; CHECK-NEXT:    .save {r4, r5, r7, lr}1493; CHECK-NEXT:    push {r4, r5, r7, lr}1494; CHECK-NEXT:    ldr r4, [sp, #16]1495; CHECK-NEXT:    lsrs r5, r4, #21496; CHECK-NEXT:    beq .LBB18_51497; CHECK-NEXT:  @ %bb.1: @ %do.body.preheader1498; CHECK-NEXT:    ldr.w r12, [sp, #20]1499; CHECK-NEXT:  .LBB18_2: @ %do.body1500; CHECK-NEXT:    @ =>This Loop Header: Depth=11501; CHECK-NEXT:    @ Child Loop BB18_3 Depth 21502; CHECK-NEXT:    ldr r4, [r2]1503; CHECK-NEXT:    dls lr, r51504; CHECK-NEXT:    vdup.32 q0, r41505; CHECK-NEXT:  .LBB18_3: @ %while.body1506; CHECK-NEXT:    @ Parent Loop BB18_2 Depth=11507; CHECK-NEXT:    @ => This Inner Loop Header: Depth=21508; CHECK-NEXT:    vldrw.u32 q1, [r1], #161509; CHECK-NEXT:    vldrw.u32 q2, [r0], #161510; CHECK-NEXT:    vfms.f32 q2, q1, q01511; CHECK-NEXT:    vstrb.8 q2, [r3], #161512; CHECK-NEXT:    le lr, .LBB18_31513; CHECK-NEXT:  @ %bb.4: @ %while.end1514; CHECK-NEXT:    @ in Loop: Header=BB18_2 Depth=11515; CHECK-NEXT:    adds r2, #41516; CHECK-NEXT:    subs.w r12, r12, #11517; CHECK-NEXT:    bne .LBB18_21518; CHECK-NEXT:  .LBB18_5: @ %do.end1519; CHECK-NEXT:    pop {r4, r5, r7, pc}1520entry:1521  %shr = lshr i32 %N, 21522  %cmp15 = icmp eq i32 %shr, 01523  br i1 %cmp15, label %do.end, label %do.body1524 1525do.body:                                          ; preds = %while.end, %entry1526  %pDst.addr.0 = phi ptr [ %add.ptr2, %while.end ], [ %pDst, %entry ]1527  %M.addr.0 = phi i32 [ %dec3, %while.end ], [ %M, %entry ]1528  %pSrc3.addr.0 = phi ptr [ %incdec.ptr, %while.end ], [ %pSrc3, %entry ]1529  %pSrc2.addr.0 = phi ptr [ %add.ptr1, %while.end ], [ %pSrc2, %entry ]1530  %pSrc1.addr.0 = phi ptr [ %add.ptr, %while.end ], [ %pSrc1, %entry ]1531  %i = load float, ptr %pSrc3.addr.0, align 41532  %.splatinsert = insertelement <4 x float> undef, float %i, i32 01533  %.splat = shufflevector <4 x float> %.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer1534  br label %while.body1535 1536while.body:                                       ; preds = %while.body, %do.body1537  %pSrc1.addr.119 = phi ptr [ %pSrc1.addr.0, %do.body ], [ %add.ptr, %while.body ]1538  %pSrc2.addr.118 = phi ptr [ %pSrc2.addr.0, %do.body ], [ %add.ptr1, %while.body ]1539  %blkCnt.017 = phi i32 [ %shr, %do.body ], [ %dec, %while.body ]1540  %pDst.addr.116 = phi ptr [ %pDst.addr.0, %do.body ], [ %add.ptr2, %while.body ]1541  %i2 = load <4 x float>, ptr %pSrc1.addr.119, align 41542  %i4 = load <4 x float>, ptr %pSrc2.addr.118, align 41543  %i5 = fneg fast <4 x float> %i41544  %i6 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %.splat, <4 x float> %i5, <4 x float> %i2)1545  store <4 x float> %i6, ptr %pDst.addr.116, align 41546  %add.ptr = getelementptr inbounds float, ptr %pSrc1.addr.119, i32 41547  %add.ptr1 = getelementptr inbounds float, ptr %pSrc2.addr.118, i32 41548  %add.ptr2 = getelementptr inbounds float, ptr %pDst.addr.116, i32 41549  %dec = add nsw i32 %blkCnt.017, -11550  %cmp = icmp eq i32 %dec, 01551  br i1 %cmp, label %while.end, label %while.body1552 1553while.end:                                        ; preds = %while.body1554  %incdec.ptr = getelementptr inbounds float, ptr %pSrc3.addr.0, i32 11555  %dec3 = add i32 %M.addr.0, -11556  %cmp4 = icmp eq i32 %dec3, 01557  br i1 %cmp4, label %do.end, label %do.body1558 1559do.end:                                           ; preds = %while.end, %entry1560  ret void1561}1562 1563 1564%struct.arm_biquad_casd_df1_inst_f32 = type { i32, ptr, ptr }1565define arm_aapcs_vfpcc void @arm_biquad_cascade_df1_f32(ptr nocapture readonly %S, ptr nocapture readonly %pSrc, ptr nocapture %pDst, i32 %blockSize) {1566; CHECK-LABEL: arm_biquad_cascade_df1_f32:1567; CHECK:       @ %bb.0: @ %entry1568; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}1569; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}1570; CHECK-NEXT:    .pad #41571; CHECK-NEXT:    sub sp, #41572; CHECK-NEXT:    .vsave {d8, d9, d10, d11, d12, d13, d14, d15}1573; CHECK-NEXT:    vpush {d8, d9, d10, d11, d12, d13, d14, d15}1574; CHECK-NEXT:    .pad #161575; CHECK-NEXT:    sub sp, #161576; CHECK-NEXT:    ldm.w r0, {r7, r9, r11}1577; CHECK-NEXT:    and r0, r3, #31578; CHECK-NEXT:    @ implicit-def: $r51579; CHECK-NEXT:    str r0, [sp, #8] @ 4-byte Spill1580; CHECK-NEXT:    lsrs r0, r3, #21581; CHECK-NEXT:    str r0, [sp, #4] @ 4-byte Spill1582; CHECK-NEXT:    b .LBB19_31583; CHECK-NEXT:  .LBB19_1: @ in Loop: Header=BB19_3 Depth=11584; CHECK-NEXT:    mov r8, r31585; CHECK-NEXT:    mov r3, r121586; CHECK-NEXT:    mov r0, r41587; CHECK-NEXT:    mov r12, r101588; CHECK-NEXT:  .LBB19_2: @ %if.end691589; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11590; CHECK-NEXT:    ldr r7, [sp, #12] @ 4-byte Reload1591; CHECK-NEXT:    add.w r11, r11, #1281592; CHECK-NEXT:    strd r8, r0, [r9]1593; CHECK-NEXT:    mov r1, r21594; CHECK-NEXT:    strd r3, r12, [r9, #8]1595; CHECK-NEXT:    add.w r9, r9, #161596; CHECK-NEXT:    subs r7, #11597; CHECK-NEXT:    beq.w .LBB19_131598; CHECK-NEXT:  .LBB19_3: @ %do.body1599; CHECK-NEXT:    @ =>This Loop Header: Depth=11600; CHECK-NEXT:    @ Child Loop BB19_5 Depth 21601; CHECK-NEXT:    ldr.w r10, [r9, #12]1602; CHECK-NEXT:    mov r6, r21603; CHECK-NEXT:    ldm.w r9, {r3, r4, r12}1604; CHECK-NEXT:    ldr r0, [sp, #4] @ 4-byte Reload1605; CHECK-NEXT:    str r7, [sp, #12] @ 4-byte Spill1606; CHECK-NEXT:    wls lr, r0, .LBB19_61607; CHECK-NEXT:  @ %bb.4: @ %while.body.lr.ph1608; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11609; CHECK-NEXT:    mov r6, r21610; CHECK-NEXT:  .LBB19_5: @ %while.body1611; CHECK-NEXT:    @ Parent Loop BB19_3 Depth=11612; CHECK-NEXT:    @ => This Inner Loop Header: Depth=21613; CHECK-NEXT:    mov r5, r31614; CHECK-NEXT:    mov r8, r41615; CHECK-NEXT:    ldrd r4, r3, [r1, #8]1616; CHECK-NEXT:    vldrw.u32 q2, [r11]1617; CHECK-NEXT:    vldrw.u32 q6, [r11, #16]1618; CHECK-NEXT:    ldrd r0, r7, [r1]1619; CHECK-NEXT:    vmul.f32 q2, q2, r31620; CHECK-NEXT:    vldrw.u32 q7, [r11, #32]1621; CHECK-NEXT:    vfma.f32 q2, q6, r41622; CHECK-NEXT:    vldrw.u32 q4, [r11, #48]1623; CHECK-NEXT:    vfma.f32 q2, q7, r71624; CHECK-NEXT:    vldrw.u32 q5, [r11, #64]1625; CHECK-NEXT:    vfma.f32 q2, q4, r01626; CHECK-NEXT:    vldrw.u32 q3, [r11, #80]1627; CHECK-NEXT:    vfma.f32 q2, q5, r51628; CHECK-NEXT:    vldrw.u32 q1, [r11, #96]1629; CHECK-NEXT:    vfma.f32 q2, q3, r81630; CHECK-NEXT:    vldrw.u32 q0, [r11, #112]1631; CHECK-NEXT:    vfma.f32 q2, q1, r121632; CHECK-NEXT:    adds r1, #161633; CHECK-NEXT:    vfma.f32 q2, q0, r101634; CHECK-NEXT:    mov r5, r31635; CHECK-NEXT:    vmov r10, r12, d51636; CHECK-NEXT:    vstrb.8 q2, [r6], #161637; CHECK-NEXT:    le lr, .LBB19_51638; CHECK-NEXT:  .LBB19_6: @ %while.end1639; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11640; CHECK-NEXT:    ldr r7, [sp, #8] @ 4-byte Reload1641; CHECK-NEXT:    cmp r7, #01642; CHECK-NEXT:    beq .LBB19_11643; CHECK-NEXT:  @ %bb.7: @ %if.then1644; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11645; CHECK-NEXT:    ldrd lr, r0, [r1]1646; CHECK-NEXT:    vldrw.u32 q0, [r11]1647; CHECK-NEXT:    ldrd r8, r1, [r1, #8]1648; CHECK-NEXT:    vldrw.u32 q6, [r11, #16]1649; CHECK-NEXT:    vldrw.u32 q7, [r11, #32]1650; CHECK-NEXT:    vldrw.u32 q4, [r11, #48]1651; CHECK-NEXT:    vmul.f32 q0, q0, r11652; CHECK-NEXT:    vldrw.u32 q5, [r11, #64]1653; CHECK-NEXT:    vfma.f32 q0, q6, r81654; CHECK-NEXT:    vldrw.u32 q3, [r11, #80]1655; CHECK-NEXT:    vfma.f32 q0, q7, r01656; CHECK-NEXT:    vldrw.u32 q2, [r11, #96]1657; CHECK-NEXT:    vfma.f32 q0, q4, lr1658; CHECK-NEXT:    vldrw.u32 q1, [r11, #112]1659; CHECK-NEXT:    vfma.f32 q0, q5, r31660; CHECK-NEXT:    cmp r7, #11661; CHECK-NEXT:    vfma.f32 q0, q3, r41662; CHECK-NEXT:    vfma.f32 q0, q2, r121663; CHECK-NEXT:    vfma.f32 q0, q1, r101664; CHECK-NEXT:    vmov r4, s01665; CHECK-NEXT:    bne .LBB19_91666; CHECK-NEXT:  @ %bb.8: @ %if.then581667; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11668; CHECK-NEXT:    str r4, [r6]1669; CHECK-NEXT:    mov r8, lr1670; CHECK-NEXT:    mov r0, r51671; CHECK-NEXT:    mov r3, r41672; CHECK-NEXT:    b .LBB19_121673; CHECK-NEXT:  .LBB19_9: @ %if.else1674; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11675; CHECK-NEXT:    vmov r12, s11676; CHECK-NEXT:    cmp r7, #21677; CHECK-NEXT:    vstr s1, [r6, #4]1678; CHECK-NEXT:    str r4, [r6]1679; CHECK-NEXT:    bne .LBB19_111680; CHECK-NEXT:  @ %bb.10: @ in Loop: Header=BB19_3 Depth=11681; CHECK-NEXT:    mov r8, r01682; CHECK-NEXT:    mov r3, r121683; CHECK-NEXT:    mov r0, lr1684; CHECK-NEXT:    mov r12, r41685; CHECK-NEXT:    b .LBB19_121686; CHECK-NEXT:  .LBB19_11: @ %if.else641687; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11688; CHECK-NEXT:    vmov r3, s21689; CHECK-NEXT:    vstr s2, [r6, #8]1690; CHECK-NEXT:  .LBB19_12: @ %if.end691691; CHECK-NEXT:    @ in Loop: Header=BB19_3 Depth=11692; CHECK-NEXT:    mov r5, r11693; CHECK-NEXT:    b .LBB19_21694; CHECK-NEXT:  .LBB19_13: @ %do.end1695; CHECK-NEXT:    add sp, #161696; CHECK-NEXT:    vpop {d8, d9, d10, d11, d12, d13, d14, d15}1697; CHECK-NEXT:    add sp, #41698; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}1699entry:1700  %pState1 = getelementptr inbounds %struct.arm_biquad_casd_df1_inst_f32, ptr %S, i32 0, i32 11701  %i = load ptr, ptr %pState1, align 41702  %pCoeffs2 = getelementptr inbounds %struct.arm_biquad_casd_df1_inst_f32, ptr %S, i32 0, i32 21703  %i1 = load ptr, ptr %pCoeffs2, align 41704  %numStages = getelementptr inbounds %struct.arm_biquad_casd_df1_inst_f32, ptr %S, i32 0, i32 01705  %i2 = load i32, ptr %numStages, align 41706  %shr = lshr i32 %blockSize, 21707  %cmp201 = icmp eq i32 %shr, 01708  %and = and i32 %blockSize, 31709  %tobool = icmp eq i32 %and, 01710  %cmp57 = icmp eq i32 %and, 11711  %cmp60 = icmp eq i32 %and, 21712  br label %do.body1713 1714do.body:                                          ; preds = %if.end69, %entry1715  %pState.0 = phi ptr [ %i, %entry ], [ %incdec.ptr73, %if.end69 ]1716  %pCoeffs.0 = phi ptr [ %i1, %entry ], [ %add.ptr74, %if.end69 ]1717  %pIn.0 = phi ptr [ %pSrc, %entry ], [ %pDst, %if.end69 ]1718  %X3.0 = phi float [ undef, %entry ], [ %X3.2, %if.end69 ]1719  %stage.0 = phi i32 [ %i2, %entry ], [ %dec75, %if.end69 ]1720  %i3 = load float, ptr %pState.0, align 41721  %arrayidx3 = getelementptr inbounds float, ptr %pState.0, i32 11722  %i4 = load float, ptr %arrayidx3, align 41723  %arrayidx4 = getelementptr inbounds float, ptr %pState.0, i32 21724  %i5 = load float, ptr %arrayidx4, align 41725  %arrayidx5 = getelementptr inbounds float, ptr %pState.0, i32 31726  %i6 = load float, ptr %arrayidx5, align 41727  br i1 %cmp201, label %while.end, label %while.body.lr.ph1728 1729while.body.lr.ph:                                 ; preds = %do.body1730  %arrayidx9 = getelementptr inbounds float, ptr %pCoeffs.0, i32 41731  %arrayidx12 = getelementptr inbounds float, ptr %pCoeffs.0, i32 81732  %arrayidx15 = getelementptr inbounds float, ptr %pCoeffs.0, i32 121733  %arrayidx18 = getelementptr inbounds float, ptr %pCoeffs.0, i32 161734  %arrayidx21 = getelementptr inbounds float, ptr %pCoeffs.0, i32 201735  %arrayidx24 = getelementptr inbounds float, ptr %pCoeffs.0, i32 241736  %arrayidx27 = getelementptr inbounds float, ptr %pCoeffs.0, i32 281737  br label %while.body1738 1739while.body:                                       ; preds = %while.body, %while.body.lr.ph1740  %sample.0208 = phi i32 [ %shr, %while.body.lr.ph ], [ %dec, %while.body ]1741  %pIn.1207 = phi ptr [ %pIn.0, %while.body.lr.ph ], [ %incdec.ptr8, %while.body ]1742  %pOut.1206 = phi ptr [ %pDst, %while.body.lr.ph ], [ %add.ptr, %while.body ]1743  %Yn2.0205 = phi float [ %i6, %while.body.lr.ph ], [ %i37, %while.body ]1744  %Yn1.0204 = phi float [ %i5, %while.body.lr.ph ], [ %i36, %while.body ]1745  %Xn2.0203 = phi float [ %i4, %while.body.lr.ph ], [ %i17, %while.body ]1746  %Xn1.0202 = phi float [ %i3, %while.body.lr.ph ], [ %i18, %while.body ]1747  %incdec.ptr = getelementptr inbounds float, ptr %pIn.1207, i32 11748  %i15 = load float, ptr %pIn.1207, align 41749  %incdec.ptr6 = getelementptr inbounds float, ptr %pIn.1207, i32 21750  %i16 = load float, ptr %incdec.ptr, align 41751  %incdec.ptr7 = getelementptr inbounds float, ptr %pIn.1207, i32 31752  %i17 = load float, ptr %incdec.ptr6, align 41753  %incdec.ptr8 = getelementptr inbounds float, ptr %pIn.1207, i32 41754  %i18 = load float, ptr %incdec.ptr7, align 41755  %i19 = load <4 x float>, ptr %pCoeffs.0, align 41756  %.splatinsert = insertelement <4 x float> undef, float %i18, i32 01757  %.splat = shufflevector <4 x float> %.splatinsert, <4 x float> undef, <4 x i32> zeroinitializer1758  %i20 = fmul fast <4 x float> %.splat, %i191759  %i21 = load <4 x float>, ptr %arrayidx9, align 41760  %.splatinsert10 = insertelement <4 x float> undef, float %i17, i32 01761  %.splat11 = shufflevector <4 x float> %.splatinsert10, <4 x float> undef, <4 x i32> zeroinitializer1762  %i22 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i21, <4 x float> %.splat11, <4 x float> %i20)1763  %i23 = load <4 x float>, ptr %arrayidx12, align 41764  %.splatinsert13 = insertelement <4 x float> undef, float %i16, i32 01765  %.splat14 = shufflevector <4 x float> %.splatinsert13, <4 x float> undef, <4 x i32> zeroinitializer1766  %i24 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i23, <4 x float> %.splat14, <4 x float> %i22)1767  %i25 = load <4 x float>, ptr %arrayidx15, align 41768  %.splatinsert16 = insertelement <4 x float> undef, float %i15, i32 01769  %.splat17 = shufflevector <4 x float> %.splatinsert16, <4 x float> undef, <4 x i32> zeroinitializer1770  %i26 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i25, <4 x float> %.splat17, <4 x float> %i24)1771  %i27 = load <4 x float>, ptr %arrayidx18, align 41772  %.splatinsert19 = insertelement <4 x float> undef, float %Xn1.0202, i32 01773  %.splat20 = shufflevector <4 x float> %.splatinsert19, <4 x float> undef, <4 x i32> zeroinitializer1774  %i28 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i27, <4 x float> %.splat20, <4 x float> %i26)1775  %i29 = load <4 x float>, ptr %arrayidx21, align 41776  %.splatinsert22 = insertelement <4 x float> undef, float %Xn2.0203, i32 01777  %.splat23 = shufflevector <4 x float> %.splatinsert22, <4 x float> undef, <4 x i32> zeroinitializer1778  %i30 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i29, <4 x float> %.splat23, <4 x float> %i28)1779  %i31 = load <4 x float>, ptr %arrayidx24, align 41780  %.splatinsert25 = insertelement <4 x float> undef, float %Yn1.0204, i32 01781  %.splat26 = shufflevector <4 x float> %.splatinsert25, <4 x float> undef, <4 x i32> zeroinitializer1782  %i32 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i31, <4 x float> %.splat26, <4 x float> %i30)1783  %i33 = load <4 x float>, ptr %arrayidx27, align 41784  %.splatinsert28 = insertelement <4 x float> undef, float %Yn2.0205, i32 01785  %.splat29 = shufflevector <4 x float> %.splatinsert28, <4 x float> undef, <4 x i32> zeroinitializer1786  %i34 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i33, <4 x float> %.splat29, <4 x float> %i32)1787  store <4 x float> %i34, ptr %pOut.1206, align 41788  %add.ptr = getelementptr inbounds float, ptr %pOut.1206, i32 41789  %i36 = extractelement <4 x float> %i34, i32 31790  %i37 = extractelement <4 x float> %i34, i32 21791  %dec = add nsw i32 %sample.0208, -11792  %cmp = icmp eq i32 %dec, 01793  br i1 %cmp, label %while.end, label %while.body1794 1795while.end:                                        ; preds = %while.body, %do.body1796  %Xn1.0.lcssa = phi float [ %i3, %do.body ], [ %i18, %while.body ]1797  %Xn2.0.lcssa = phi float [ %i4, %do.body ], [ %i17, %while.body ]1798  %Yn1.0.lcssa = phi float [ %i5, %do.body ], [ %i36, %while.body ]1799  %Yn2.0.lcssa = phi float [ %i6, %do.body ], [ %i37, %while.body ]1800  %pOut.1.lcssa = phi ptr [ %pDst, %do.body ], [ %add.ptr, %while.body ]1801  %pIn.1.lcssa = phi ptr [ %pIn.0, %do.body ], [ %incdec.ptr8, %while.body ]1802  %X3.1.lcssa = phi float [ %X3.0, %do.body ], [ %i18, %while.body ]1803  br i1 %tobool, label %if.end69, label %if.then1804 1805if.then:                                          ; preds = %while.end1806  %incdec.ptr30 = getelementptr inbounds float, ptr %pIn.1.lcssa, i32 11807  %i38 = load float, ptr %pIn.1.lcssa, align 41808  %incdec.ptr31 = getelementptr inbounds float, ptr %pIn.1.lcssa, i32 21809  %i39 = load float, ptr %incdec.ptr30, align 41810  %incdec.ptr32 = getelementptr inbounds float, ptr %pIn.1.lcssa, i32 31811  %i40 = load float, ptr %incdec.ptr31, align 41812  %i41 = load float, ptr %incdec.ptr32, align 41813  %i43 = load <4 x float>, ptr %pCoeffs.0, align 41814  %.splatinsert34 = insertelement <4 x float> undef, float %i41, i32 01815  %.splat35 = shufflevector <4 x float> %.splatinsert34, <4 x float> undef, <4 x i32> zeroinitializer1816  %i44 = fmul fast <4 x float> %.splat35, %i431817  %arrayidx36 = getelementptr inbounds float, ptr %pCoeffs.0, i32 41818  %i46 = load <4 x float>, ptr %arrayidx36, align 41819  %.splatinsert37 = insertelement <4 x float> undef, float %i40, i32 01820  %.splat38 = shufflevector <4 x float> %.splatinsert37, <4 x float> undef, <4 x i32> zeroinitializer1821  %i47 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i46, <4 x float> %.splat38, <4 x float> %i44)1822  %arrayidx39 = getelementptr inbounds float, ptr %pCoeffs.0, i32 81823  %i49 = load <4 x float>, ptr %arrayidx39, align 41824  %.splatinsert40 = insertelement <4 x float> undef, float %i39, i32 01825  %.splat41 = shufflevector <4 x float> %.splatinsert40, <4 x float> undef, <4 x i32> zeroinitializer1826  %i50 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i49, <4 x float> %.splat41, <4 x float> %i47)1827  %arrayidx42 = getelementptr inbounds float, ptr %pCoeffs.0, i32 121828  %i52 = load <4 x float>, ptr %arrayidx42, align 41829  %.splatinsert43 = insertelement <4 x float> undef, float %i38, i32 01830  %.splat44 = shufflevector <4 x float> %.splatinsert43, <4 x float> undef, <4 x i32> zeroinitializer1831  %i53 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i52, <4 x float> %.splat44, <4 x float> %i50)1832  %arrayidx45 = getelementptr inbounds float, ptr %pCoeffs.0, i32 161833  %i55 = load <4 x float>, ptr %arrayidx45, align 41834  %.splatinsert46 = insertelement <4 x float> undef, float %Xn1.0.lcssa, i32 01835  %.splat47 = shufflevector <4 x float> %.splatinsert46, <4 x float> undef, <4 x i32> zeroinitializer1836  %i56 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i55, <4 x float> %.splat47, <4 x float> %i53)1837  %arrayidx48 = getelementptr inbounds float, ptr %pCoeffs.0, i32 201838  %i58 = load <4 x float>, ptr %arrayidx48, align 41839  %.splatinsert49 = insertelement <4 x float> undef, float %Xn2.0.lcssa, i32 01840  %.splat50 = shufflevector <4 x float> %.splatinsert49, <4 x float> undef, <4 x i32> zeroinitializer1841  %i59 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i58, <4 x float> %.splat50, <4 x float> %i56)1842  %arrayidx51 = getelementptr inbounds float, ptr %pCoeffs.0, i32 241843  %i61 = load <4 x float>, ptr %arrayidx51, align 41844  %.splatinsert52 = insertelement <4 x float> undef, float %Yn1.0.lcssa, i32 01845  %.splat53 = shufflevector <4 x float> %.splatinsert52, <4 x float> undef, <4 x i32> zeroinitializer1846  %i62 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i61, <4 x float> %.splat53, <4 x float> %i59)1847  %arrayidx54 = getelementptr inbounds float, ptr %pCoeffs.0, i32 281848  %i64 = load <4 x float>, ptr %arrayidx54, align 41849  %.splatinsert55 = insertelement <4 x float> undef, float %Yn2.0.lcssa, i32 01850  %.splat56 = shufflevector <4 x float> %.splatinsert55, <4 x float> undef, <4 x i32> zeroinitializer1851  %i65 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i64, <4 x float> %.splat56, <4 x float> %i62)1852  %i66 = extractelement <4 x float> %i65, i32 01853  br i1 %cmp57, label %if.then58, label %if.else1854 1855if.then58:                                        ; preds = %if.then1856  store float %i66, ptr %pOut.1.lcssa, align 41857  br label %if.end691858 1859if.else:                                          ; preds = %if.then1860  %incdec.ptr62 = getelementptr inbounds float, ptr %pOut.1.lcssa, i32 11861  store float %i66, ptr %pOut.1.lcssa, align 41862  %i67 = extractelement <4 x float> %i65, i32 11863  store float %i67, ptr %incdec.ptr62, align 41864  br i1 %cmp60, label %if.end69, label %if.else641865 1866if.else64:                                        ; preds = %if.else1867  %incdec.ptr63 = getelementptr inbounds float, ptr %pOut.1.lcssa, i32 21868  %i68 = extractelement <4 x float> %i65, i32 21869  store float %i68, ptr %incdec.ptr63, align 41870  br label %if.end691871 1872if.end69:                                         ; preds = %if.else64, %if.else, %if.then58, %while.end1873  %Xn1.1 = phi float [ %i38, %if.then58 ], [ %i40, %if.else64 ], [ %Xn1.0.lcssa, %while.end ], [ %i39, %if.else ]1874  %Xn2.1 = phi float [ %X3.1.lcssa, %if.then58 ], [ %i39, %if.else64 ], [ %Xn2.0.lcssa, %while.end ], [ %i38, %if.else ]1875  %Yn1.1 = phi float [ %i66, %if.then58 ], [ %i68, %if.else64 ], [ %Yn1.0.lcssa, %while.end ], [ %i67, %if.else ]1876  %Yn2.1 = phi float [ %Yn1.0.lcssa, %if.then58 ], [ %i67, %if.else64 ], [ %Yn2.0.lcssa, %while.end ], [ %i66, %if.else ]1877  %X3.2 = phi float [ %i41, %if.then58 ], [ %i41, %if.else64 ], [ %X3.1.lcssa, %while.end ], [ %i41, %if.else ]1878  store float %Xn1.1, ptr %pState.0, align 41879  store float %Xn2.1, ptr %arrayidx3, align 41880  store float %Yn1.1, ptr %arrayidx4, align 41881  %incdec.ptr73 = getelementptr inbounds float, ptr %pState.0, i32 41882  store float %Yn2.1, ptr %arrayidx5, align 41883  %add.ptr74 = getelementptr inbounds float, ptr %pCoeffs.0, i32 321884  %dec75 = add i32 %stage.0, -11885  %cmp76 = icmp eq i32 %dec75, 01886  br i1 %cmp76, label %do.end, label %do.body1887 1888do.end:                                           ; preds = %if.end691889  ret void1890}1891 1892 1893%struct.arm_biquad_cascade_df2T_instance_f32 = type { i8, ptr, ptr }1894define void @arm_biquad_cascade_df2T_f32(ptr nocapture readonly %S, ptr nocapture readonly %pSrc, ptr nocapture %pDst, i32 %blockSize) {1895; CHECK-LABEL: arm_biquad_cascade_df2T_f32:1896; CHECK:       @ %bb.0: @ %entry1897; CHECK-NEXT:    .save {r4, r5, r6, r7, r8, lr}1898; CHECK-NEXT:    push.w {r4, r5, r6, r7, r8, lr}1899; CHECK-NEXT:    .vsave {d8, d9, d10, d11}1900; CHECK-NEXT:    vpush {d8, d9, d10, d11}1901; CHECK-NEXT:    ldrd r6, r12, [r0, #4]1902; CHECK-NEXT:    lsr.w r8, r3, #11903; CHECK-NEXT:    ldrb r0, [r0]1904; CHECK-NEXT:    vldr s0, .LCPI20_01905; CHECK-NEXT:    b .LBB20_31906; CHECK-NEXT:  .LBB20_1: @ %if.else1907; CHECK-NEXT:    @ in Loop: Header=BB20_3 Depth=11908; CHECK-NEXT:    vmov.f32 s6, s51909; CHECK-NEXT:    vstr s4, [r6]1910; CHECK-NEXT:  .LBB20_2: @ %if.end1911; CHECK-NEXT:    @ in Loop: Header=BB20_3 Depth=11912; CHECK-NEXT:    vstr s6, [r6, #4]1913; CHECK-NEXT:    add.w r12, r12, #201914; CHECK-NEXT:    adds r6, #81915; CHECK-NEXT:    subs r0, #11916; CHECK-NEXT:    mov r1, r21917; CHECK-NEXT:    beq .LBB20_81918; CHECK-NEXT:  .LBB20_3: @ %do.body1919; CHECK-NEXT:    @ =>This Loop Header: Depth=11920; CHECK-NEXT:    @ Child Loop BB20_5 Depth 21921; CHECK-NEXT:    vldrw.u32 q3, [r12]1922; CHECK-NEXT:    movs r5, #01923; CHECK-NEXT:    vmov q4, q31924; CHECK-NEXT:    vshlc q4, r5, #321925; CHECK-NEXT:    vldrw.u32 q2, [r12, #8]1926; CHECK-NEXT:    vmov q5, q21927; CHECK-NEXT:    vshlc q5, r5, #321928; CHECK-NEXT:    vldrw.u32 q1, [r6]1929; CHECK-NEXT:    vmov.f32 s6, s01930; CHECK-NEXT:    mov r5, r21931; CHECK-NEXT:    vmov.f32 s7, s01932; CHECK-NEXT:    wls lr, r8, .LBB20_61933; CHECK-NEXT:  @ %bb.4: @ %while.body.preheader1934; CHECK-NEXT:    @ in Loop: Header=BB20_3 Depth=11935; CHECK-NEXT:    mov r5, r21936; CHECK-NEXT:  .LBB20_5: @ %while.body1937; CHECK-NEXT:    @ Parent Loop BB20_3 Depth=11938; CHECK-NEXT:    @ => This Inner Loop Header: Depth=21939; CHECK-NEXT:    ldrd r7, r4, [r1], #81940; CHECK-NEXT:    vfma.f32 q1, q3, r71941; CHECK-NEXT:    vmov r7, s41942; CHECK-NEXT:    vmov.f32 s2, s41943; CHECK-NEXT:    vfma.f32 q1, q2, r71944; CHECK-NEXT:    vmov.f32 s7, s01945; CHECK-NEXT:    vfma.f32 q1, q4, r41946; CHECK-NEXT:    vmov r4, s51947; CHECK-NEXT:    vstr s5, [r5, #4]1948; CHECK-NEXT:    vfma.f32 q1, q5, r41949; CHECK-NEXT:    vmov.f32 s4, s61950; CHECK-NEXT:    vmov.f32 s5, s71951; CHECK-NEXT:    vmov.f32 s6, s01952; CHECK-NEXT:    vstr s2, [r5]1953; CHECK-NEXT:    adds r5, #81954; CHECK-NEXT:    le lr, .LBB20_51955; CHECK-NEXT:  .LBB20_6: @ %while.end1956; CHECK-NEXT:    @ in Loop: Header=BB20_3 Depth=11957; CHECK-NEXT:    lsls r7, r3, #311958; CHECK-NEXT:    beq .LBB20_11959; CHECK-NEXT:  @ %bb.7: @ %if.then1960; CHECK-NEXT:    @ in Loop: Header=BB20_3 Depth=11961; CHECK-NEXT:    ldr r1, [r1]1962; CHECK-NEXT:    vfma.f32 q1, q3, r11963; CHECK-NEXT:    vmov r1, s41964; CHECK-NEXT:    vstr s4, [r5]1965; CHECK-NEXT:    vfma.f32 q1, q2, r11966; CHECK-NEXT:    vstr s5, [r6]1967; CHECK-NEXT:    b .LBB20_21968; CHECK-NEXT:  .LBB20_8: @ %do.end1969; CHECK-NEXT:    vpop {d8, d9, d10, d11}1970; CHECK-NEXT:    pop.w {r4, r5, r6, r7, r8, pc}1971; CHECK-NEXT:    .p2align 21972; CHECK-NEXT:  @ %bb.9:1973; CHECK-NEXT:  .LCPI20_0:1974; CHECK-NEXT:    .long 0x00000000 @ float 01975entry:1976  %pState1 = getelementptr inbounds %struct.arm_biquad_cascade_df2T_instance_f32, ptr %S, i32 0, i32 11977  %i = load ptr, ptr %pState1, align 41978  %numStages = getelementptr inbounds %struct.arm_biquad_cascade_df2T_instance_f32, ptr %S, i32 0, i32 01979  %i1 = load i8, ptr %numStages, align 41980  %conv = zext i8 %i1 to i321981  %pCoeffs = getelementptr inbounds %struct.arm_biquad_cascade_df2T_instance_f32, ptr %S, i32 0, i32 21982  %i2 = load ptr, ptr %pCoeffs, align 41983  %div = lshr i32 %blockSize, 11984  %cmp.not90 = icmp eq i32 %div, 01985  %and = and i32 %blockSize, 11986  %tobool.not = icmp eq i32 %and, 01987  br label %do.body1988 1989do.body:                                          ; preds = %if.end, %entry1990  %stage.0 = phi i32 [ %conv, %entry ], [ %dec23, %if.end ]1991  %pCurCoeffs.0 = phi ptr [ %i2, %entry ], [ %add.ptr2, %if.end ]1992  %pState.0 = phi ptr [ %i, %entry ], [ %pState.1, %if.end ]1993  %pIn.0 = phi ptr [ %pSrc, %entry ], [ %pDst, %if.end ]1994  %i4 = load <4 x float>, ptr %pCurCoeffs.0, align 41995  %add.ptr = getelementptr inbounds float, ptr %pCurCoeffs.0, i32 21996  %i6 = load <4 x float>, ptr %add.ptr, align 41997  %add.ptr2 = getelementptr inbounds float, ptr %pCurCoeffs.0, i32 51998  %i8 = load <4 x float>, ptr %pState.0, align 81999  %i9 = shufflevector <4 x float> %i8, <4 x float> <float poison, float poison, float 0.000000e+00, float 0.000000e+00>, <4 x i32> <i32 0, i32 1, i32 6, i32 7>2000  %i10 = bitcast <4 x float> %i4 to <4 x i32>2001  %i11 = tail call { i32, <4 x i32> } @llvm.arm.mve.vshlc.v4i32(<4 x i32> %i10, i32 0, i32 32)2002  %i12 = extractvalue { i32, <4 x i32> } %i11, 02003  %i13 = extractvalue { i32, <4 x i32> } %i11, 12004  %i14 = bitcast <4 x i32> %i13 to <4 x float>2005  %i15 = bitcast <4 x float> %i6 to <4 x i32>2006  %i16 = tail call { i32, <4 x i32> } @llvm.arm.mve.vshlc.v4i32(<4 x i32> %i15, i32 %i12, i32 32)2007  %i17 = extractvalue { i32, <4 x i32> } %i16, 12008  %i18 = bitcast <4 x i32> %i17 to <4 x float>2009  br i1 %cmp.not90, label %while.end, label %while.body2010 2011while.body:                                       ; preds = %while.body, %do.body2012  %pIn.194 = phi ptr [ %incdec.ptr4, %while.body ], [ %pIn.0, %do.body ]2013  %state.093 = phi <4 x float> [ %i30, %while.body ], [ %i9, %do.body ]2014  %pOut.192 = phi ptr [ %incdec.ptr12, %while.body ], [ %pDst, %do.body ]2015  %sample.091 = phi i32 [ %dec, %while.body ], [ %div, %do.body ]2016  %incdec.ptr = getelementptr inbounds float, ptr %pIn.194, i32 12017  %i19 = load float, ptr %pIn.194, align 42018  %incdec.ptr4 = getelementptr inbounds float, ptr %pIn.194, i32 22019  %i20 = load float, ptr %incdec.ptr, align 42020  %.splatinsert = insertelement <4 x float> poison, float %i19, i32 02021  %.splat = shufflevector <4 x float> %.splatinsert, <4 x float> poison, <4 x i32> zeroinitializer2022  %i21 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i4, <4 x float> %.splat, <4 x float> %state.093)2023  %i22 = extractelement <4 x float> %i21, i32 02024  %.splat6 = shufflevector <4 x float> %i21, <4 x float> poison, <4 x i32> zeroinitializer2025  %i23 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i6, <4 x float> %.splat6, <4 x float> %i21)2026  %i24 = insertelement <4 x float> %i23, float 0.000000e+00, i32 32027  %.splatinsert7 = insertelement <4 x float> poison, float %i20, i32 02028  %.splat8 = shufflevector <4 x float> %.splatinsert7, <4 x float> poison, <4 x i32> zeroinitializer2029  %i25 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i14, <4 x float> %.splat8, <4 x float> %i24)2030  %i26 = extractelement <4 x float> %i25, i32 12031  %.splat10 = shufflevector <4 x float> %i25, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>2032  %i27 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i18, <4 x float> %.splat10, <4 x float> %i25)2033  %i28 = shufflevector <4 x float> %i27, <4 x float> undef, <4 x i32> <i32 2, i32 undef, i32 undef, i32 3>2034  %i29 = insertelement <4 x float> %i28, float 0.000000e+00, i32 22035  %i30 = shufflevector <4 x float> %i29, <4 x float> %i27, <4 x i32> <i32 0, i32 7, i32 2, i32 3>2036  %incdec.ptr11 = getelementptr inbounds float, ptr %pOut.192, i32 12037  store float %i22, ptr %pOut.192, align 42038  %incdec.ptr12 = getelementptr inbounds float, ptr %pOut.192, i32 22039  store float %i26, ptr %incdec.ptr11, align 42040  %dec = add nsw i32 %sample.091, -12041  %cmp.not = icmp eq i32 %dec, 02042  br i1 %cmp.not, label %while.end, label %while.body2043 2044while.end:                                        ; preds = %while.body, %do.body2045  %pOut.1.lcssa = phi ptr [ %pDst, %do.body ], [ %incdec.ptr12, %while.body ]2046  %state.0.lcssa = phi <4 x float> [ %i9, %do.body ], [ %i30, %while.body ]2047  %pIn.1.lcssa = phi ptr [ %pIn.0, %do.body ], [ %incdec.ptr4, %while.body ]2048  br i1 %tobool.not, label %if.else, label %if.then2049 2050if.then:                                          ; preds = %while.end2051  %i31 = load float, ptr %pIn.1.lcssa, align 42052  %.splatinsert14 = insertelement <4 x float> poison, float %i31, i32 02053  %.splat15 = shufflevector <4 x float> %.splatinsert14, <4 x float> poison, <4 x i32> zeroinitializer2054  %i32 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i4, <4 x float> %.splat15, <4 x float> %state.0.lcssa)2055  %i33 = extractelement <4 x float> %i32, i32 02056  %.splat17 = shufflevector <4 x float> %i32, <4 x float> poison, <4 x i32> zeroinitializer2057  %i34 = tail call fast <4 x float> @llvm.fma.v4f32(<4 x float> %i6, <4 x float> %.splat17, <4 x float> %i32)2058  store float %i33, ptr %pOut.1.lcssa, align 42059  %i35 = extractelement <4 x float> %i34, i32 12060  store float %i35, ptr %pState.0, align 42061  %i36 = extractelement <4 x float> %i34, i32 22062  br label %if.end2063 2064if.else:                                          ; preds = %while.end2065  %i37 = extractelement <4 x float> %state.0.lcssa, i32 02066  store float %i37, ptr %pState.0, align 42067  %i38 = extractelement <4 x float> %state.0.lcssa, i32 12068  br label %if.end2069 2070if.end:                                           ; preds = %if.else, %if.then2071  %.sink = phi float [ %i38, %if.else ], [ %i36, %if.then ]2072  %i39 = getelementptr inbounds float, ptr %pState.0, i32 12073  store float %.sink, ptr %i39, align 42074  %pState.1 = getelementptr inbounds float, ptr %pState.0, i32 22075  %dec23 = add i32 %stage.0, -12076  %cmp24.not = icmp eq i32 %dec23, 02077  br i1 %cmp24.not, label %do.end, label %do.body2078 2079do.end:                                           ; preds = %if.end2080  ret void2081}2082 2083define arm_aapcs_vfpcc float @vecAddAcrossF32Mve(<4 x float> %in) {2084; CHECK-LABEL: vecAddAcrossF32Mve:2085; CHECK:       @ %bb.0: @ %entry2086; CHECK-NEXT:    vadd.f32 s0, s0, s12087; CHECK-NEXT:    vadd.f32 s0, s0, s22088; CHECK-NEXT:    vadd.f32 s0, s0, s32089; CHECK-NEXT:    bx lr2090entry:2091  %i = extractelement <4 x float> %in, i32 02092  %i1 = extractelement <4 x float> %in, i32 12093  %add = fadd fast float %i, %i12094  %i2 = extractelement <4 x float> %in, i32 22095  %add1 = fadd fast float %add, %i22096  %i3 = extractelement <4 x float> %in, i32 32097  %add2 = fadd fast float %add1, %i32098  ret float %add22099}2100 2101 2102declare { i32, <4 x i32> } @llvm.arm.mve.vshlc.v4i32(<4 x i32>, i32, i32) #12103declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture)2104declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture)2105declare { <4 x i32>, i32 } @llvm.arm.mve.viwdup.v4i32(i32, i32, i32)2106declare <4 x i32> @llvm.arm.mve.vldr.gather.offset.v4i32.p0.v4i32(ptr, <4 x i32>, i32, i32, i32)2107declare void @llvm.assume(i1)2108declare <4 x i1> @llvm.arm.mve.vctp32(i32)2109declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>)2110declare void @llvm.masked.store.v4f32.p0(<4 x float>, ptr, i32 immarg, <4 x i1>)2111