1562 lines · plain
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, half %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.f16 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, #816; CHECK-NEXT: vadd.f16 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 <8 x half> undef, half %B, i32 030 %broadcast.splat11 = shufflevector <8 x half> %broadcast.splatinsert10, <8 x half> undef, <8 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 half, ptr %A, i32 %index36 %wide.load = load <8 x half>, ptr %i1, align 437 %i3 = fadd fast <8 x half> %wide.load, %broadcast.splat1138 %i4 = getelementptr inbounds half, ptr %C, i32 %index39 store <8 x half> %i3, ptr %i4, align 440 %index.next = add i32 %index, 841 %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, half %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.f16 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, #860; CHECK-NEXT: vadd.f16 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 <8 x half> undef, half %B, i32 074 %broadcast.splat11 = shufflevector <8 x half> %broadcast.splatinsert10, <8 x half> undef, <8 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 half, ptr %A, i32 %index80 %wide.load = load <8 x half>, ptr %i1, align 481 %i3 = fadd fast <8 x half> %broadcast.splat11, %wide.load82 %i4 = getelementptr inbounds half, ptr %C, i32 %index83 store <8 x half> %i3, ptr %i4, align 484 %index.next = add i32 %index, 885 %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, half %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.f16 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, #8104; CHECK-NEXT: vmul.f16 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 <8 x half> undef, half %B, i32 0118 %broadcast.splat11 = shufflevector <8 x half> %broadcast.splatinsert10, <8 x half> undef, <8 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 half, ptr %A, i32 %index124 %wide.load = load <8 x half>, ptr %i1, align 4125 %i3 = fmul fast <8 x half> %wide.load, %broadcast.splat11126 %i4 = getelementptr inbounds half, ptr %C, i32 %index127 store <8 x half> %i3, ptr %i4, align 4128 %index.next = add i32 %index, 8129 %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, half %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.f16 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, #8148; CHECK-NEXT: vmul.f16 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 <8 x half> undef, half %B, i32 0162 %broadcast.splat11 = shufflevector <8 x half> %broadcast.splatinsert10, <8 x half> undef, <8 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 half, ptr %A, i32 %index168 %wide.load = load <8 x half>, ptr %i1, align 4169 %i3 = fmul fast <8 x half> %broadcast.splat11, %wide.load170 %i4 = getelementptr inbounds half, ptr %C, i32 %index171 store <8 x half> %i3, ptr %i4, align 4172 %index.next = add i32 %index, 8173 %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, half %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.f16 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, #8192; CHECK-NEXT: vsub.f16 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 <8 x half> undef, half %B, i32 0206 %broadcast.splat11 = shufflevector <8 x half> %broadcast.splatinsert10, <8 x half> undef, <8 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 half, ptr %A, i32 %index212 %wide.load = load <8 x half>, ptr %i1, align 4213 %i3 = fsub fast <8 x half> %wide.load, %broadcast.splat11214 %i4 = getelementptr inbounds half, ptr %C, i32 %index215 store <8 x half> %i3, ptr %i4, align 4216 %index.next = add i32 %index, 8217 %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, half %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.f16 r3, s0232; CHECK-NEXT: vdup.16 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, #8237; CHECK-NEXT: vsub.f16 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 <8 x half> undef, half %B, i32 0251 %broadcast.splat11 = shufflevector <8 x half> %broadcast.splatinsert10, <8 x half> undef, <8 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 half, ptr %A, i32 %index257 %wide.load = load <8 x half>, ptr %i1, align 4258 %i3 = fsub fast <8 x half> %broadcast.splat11, %wide.load259 %i4 = getelementptr inbounds half, ptr %C, i32 %index260 store <8 x half> %i3, ptr %i4, align 4261 %index.next = add i32 %index, 8262 %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, half %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.f16 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, #8283; CHECK-NEXT: vfmas.f16 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 <8 x half> undef, half %C, i32 0297 %broadcast.splat14 = shufflevector <8 x half> %broadcast.splatinsert13, <8 x half> undef, <8 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 half, ptr %A, i32 %index303 %wide.load = load <8 x half>, ptr %i1, align 4304 %i3 = getelementptr inbounds half, ptr %B, i32 %index305 %wide.load12 = load <8 x half>, ptr %i3, align 4306 %i5 = fmul fast <8 x half> %wide.load12, %wide.load307 %i6 = fadd fast <8 x half> %i5, %broadcast.splat14308 %i7 = getelementptr inbounds half, ptr %D, i32 %index309 store <8 x half> %i6, ptr %i7, align 4310 %index.next = add i32 %index, 8311 %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, half %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.f16 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, #8331; CHECK-NEXT: vfmas.f16 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 <8 x half> undef, half %C, i32 0345 %broadcast.splat14 = shufflevector <8 x half> %broadcast.splatinsert13, <8 x half> undef, <8 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 half, ptr %A, i32 %index351 %wide.load = load <8 x half>, ptr %i1, align 4352 %i3 = getelementptr inbounds half, ptr %B, i32 %index353 %wide.load12 = load <8 x half>, ptr %i3, align 4354 %i5 = fmul fast <8 x half> %wide.load12, %wide.load355 %i6 = fadd fast <8 x half> %broadcast.splat14, %i5356 %i7 = getelementptr inbounds half, ptr %D, i32 %index357 store <8 x half> %i6, ptr %i7, align 4358 %index.next = add i32 %index, 8359 %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, half %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.f16 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, #8379; CHECK-NEXT: vfma.f16 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 <8 x half> undef, half %C, i32 0393 %broadcast.splat13 = shufflevector <8 x half> %broadcast.splatinsert12, <8 x half> undef, <8 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 half, ptr %A, i32 %index399 %wide.load = load <8 x half>, ptr %i1, align 4400 %i3 = fmul fast <8 x half> %wide.load, %broadcast.splat13401 %i4 = getelementptr inbounds half, ptr %B, i32 %index402 %wide.load14 = load <8 x half>, ptr %i4, align 4403 %i6 = fadd fast <8 x half> %i3, %wide.load14404 %i7 = getelementptr inbounds half, ptr %D, i32 %index405 store <8 x half> %i6, ptr %i7, align 4406 %index.next = add i32 %index, 8407 %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, half %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.f16 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, #8427; CHECK-NEXT: vfma.f16 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 <8 x half> undef, half %C, i32 0441 %broadcast.splat13 = shufflevector <8 x half> %broadcast.splatinsert12, <8 x half> undef, <8 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 half, ptr %A, i32 %index447 %wide.load = load <8 x half>, ptr %i1, align 4448 %i3 = fmul fast <8 x half> %broadcast.splat13, %wide.load449 %i4 = getelementptr inbounds half, ptr %B, i32 %index450 %wide.load14 = load <8 x half>, ptr %i4, align 4451 %i6 = fadd fast <8 x half> %i3, %wide.load14452 %i7 = getelementptr inbounds half, ptr %D, i32 %index453 store <8 x half> %i6, ptr %i7, align 4454 %index.next = add i32 %index, 8455 %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, half %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.f16 r12, s0471; CHECK-NEXT: vdup.16 q0, r12472; CHECK-NEXT: vneg.f16 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, #8479; CHECK-NEXT: vfma.f16 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 <8 x half> undef, half %C, i32 0493 %broadcast.splat14 = shufflevector <8 x half> %broadcast.splatinsert13, <8 x half> undef, <8 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 half, ptr %A, i32 %index499 %wide.load = load <8 x half>, ptr %i1, align 4500 %i3 = getelementptr inbounds half, ptr %B, i32 %index501 %wide.load12 = load <8 x half>, ptr %i3, align 4502 %i5 = fmul fast <8 x half> %wide.load12, %wide.load503 %i6 = fsub fast <8 x half> %i5, %broadcast.splat14504 %i7 = getelementptr inbounds half, ptr %D, i32 %index505 store <8 x half> %i6, ptr %i7, align 4506 %index.next = add i32 %index, 8507 %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, half %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.f16 r12, s0522; CHECK-NEXT: vdup.16 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, #8529; CHECK-NEXT: vfms.f16 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 <8 x half> undef, half %C, i32 0543 %broadcast.splat14 = shufflevector <8 x half> %broadcast.splatinsert13, <8 x half> undef, <8 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 half, ptr %A, i32 %index549 %wide.load = load <8 x half>, ptr %i1, align 4550 %i3 = getelementptr inbounds half, ptr %B, i32 %index551 %wide.load12 = load <8 x half>, ptr %i3, align 4552 %i5 = fmul fast <8 x half> %wide.load12, %wide.load553 %i6 = fsub fast <8 x half> %broadcast.splat14, %i5554 %i7 = getelementptr inbounds half, ptr %D, i32 %index555 store <8 x half> %i6, ptr %i7, align 4556 %index.next = add i32 %index, 8557 %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, half %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.f16 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, #8577; CHECK-NEXT: vneg.f16 q0, q0578; CHECK-NEXT: vfma.f16 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 <8 x half> undef, half %C, i32 0592 %broadcast.splat13 = shufflevector <8 x half> %broadcast.splatinsert12, <8 x half> undef, <8 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 half, ptr %A, i32 %index598 %wide.load = load <8 x half>, ptr %i1, align 4599 %i3 = fmul fast <8 x half> %wide.load, %broadcast.splat13600 %i4 = getelementptr inbounds half, ptr %B, i32 %index601 %wide.load14 = load <8 x half>, ptr %i4, align 4602 %i6 = fsub fast <8 x half> %i3, %wide.load14603 %i7 = getelementptr inbounds half, ptr %D, i32 %index604 store <8 x half> %i6, ptr %i7, align 4605 %index.next = add i32 %index, 8606 %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, half %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.f16 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, #8626; CHECK-NEXT: vneg.f16 q0, q0627; CHECK-NEXT: vfma.f16 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 <8 x half> undef, half %C, i32 0641 %broadcast.splat13 = shufflevector <8 x half> %broadcast.splatinsert12, <8 x half> undef, <8 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 half, ptr %A, i32 %index647 %wide.load = load <8 x half>, ptr %i1, align 4648 %i3 = fmul fast <8 x half> %broadcast.splat13, %wide.load649 %i4 = getelementptr inbounds half, ptr %B, i32 %index650 %wide.load14 = load <8 x half>, ptr %i4, align 4651 %i6 = fsub fast <8 x half> %i3, %wide.load14652 %i7 = getelementptr inbounds half, ptr %D, i32 %index653 store <8 x half> %i6, ptr %i7, align 4654 %index.next = add i32 %index, 8655 %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, #1670; CHECK-NEXT: .LBB14_1: @ %for.body.us671; CHECK-NEXT: @ =>This Loop Header: Depth=1672; CHECK-NEXT: @ Child Loop BB14_2 Depth 2673; CHECK-NEXT: ldrh r4, [r1]674; CHECK-NEXT: mov r5, r2675; CHECK-NEXT: mov r6, r12676; CHECK-NEXT: vdup.16 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, #8684; CHECK-NEXT: vfms.f16 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, #2692; 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 half, ptr %pPRT_in.addr.035.us, i32 %numCols713 %i = load half, ptr %pOutT1.addr.036.us, align 4714 %broadcast.splatinsert47 = insertelement <8 x half> undef, half %i, i32 0715 %broadcast.splat48 = shufflevector <8 x half> %broadcast.splatinsert47, <8 x half> undef, <8 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 half, ptr %pInT1.addr.038.us, i32 %index721 %next.gep45 = getelementptr half, ptr %pPRT_in.addr.035.us, i32 %index722 %wide.load = load <8 x half>, ptr %next.gep, align 4723 %wide.load46 = load <8 x half>, ptr %next.gep45, align 4724 %i3 = fmul fast <8 x half> %wide.load46, %broadcast.splat48725 %i4 = fsub fast <8 x half> %wide.load, %i3726 store <8 x half> %i4, ptr %next.gep, align 4727 %index.next = add i32 %index, 8728 %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 half, ptr %pOutT1.addr.036.us, i32 1733 %scevgep40 = getelementptr half, 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 #16749; CHECK-NEXT: sub sp, #16750; CHECK-NEXT: ldrh.w r9, [r0]751; CHECK-NEXT: ldr.w r10, [r0, #4]752; CHECK-NEXT: sub.w r6, r9, #1753; CHECK-NEXT: cmp r6, #3754; CHECK-NEXT: bhi .LBB15_6755; CHECK-NEXT: @ %bb.1: @ %if.then756; CHECK-NEXT: ldr r7, [r0, #8]757; CHECK-NEXT: add.w r4, r10, r6, lsl #1758; CHECK-NEXT: lsrs r5, r3, #2759; CHECK-NEXT: ldrh.w r8, [r7, #6]760; CHECK-NEXT: ldrh.w r12, [r7, #4]761; CHECK-NEXT: ldrh r6, [r7, #2]762; CHECK-NEXT: ldrh r7, [r7]763; CHECK-NEXT: wls lr, r5, .LBB15_5764; CHECK-NEXT: @ %bb.2: @ %while.body.lr.ph765; CHECK-NEXT: str.w r9, [sp, #12] @ 4-byte Spill766; CHECK-NEXT: bic r5, r3, #3767; CHECK-NEXT: add.w r9, r10, #2768; CHECK-NEXT: str r5, [sp] @ 4-byte Spill769; CHECK-NEXT: add.w r5, r2, r5, lsl #1770; CHECK-NEXT: str r5, [sp, #4] @ 4-byte Spill771; CHECK-NEXT: str r1, [sp, #8] @ 4-byte Spill772; CHECK-NEXT: .LBB15_3: @ %while.body773; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1774; CHECK-NEXT: vldrw.u32 q0, [r1], #8775; CHECK-NEXT: sub.w r11, r9, #2776; CHECK-NEXT: add.w r5, r9, #2777; CHECK-NEXT: vstrb.8 q0, [r4], #8778; CHECK-NEXT: vldrw.u32 q0, [r11]779; CHECK-NEXT: vldrw.u32 q1, [r9]780; CHECK-NEXT: vmul.f16 q0, q0, r7781; CHECK-NEXT: vfma.f16 q0, q1, r6782; CHECK-NEXT: vldrw.u32 q1, [r5]783; CHECK-NEXT: vfma.f16 q0, q1, r12784; CHECK-NEXT: vldrw.u32 q1, [r9, #4]785; CHECK-NEXT: add.w r9, r9, #8786; CHECK-NEXT: vfma.f16 q0, q1, r8787; CHECK-NEXT: vstrb.8 q0, [r2], #8788; CHECK-NEXT: le lr, .LBB15_3789; CHECK-NEXT: @ %bb.4: @ %while.end.loopexit790; CHECK-NEXT: ldr r2, [sp] @ 4-byte Reload791; CHECK-NEXT: ldr r1, [sp, #8] @ 4-byte Reload792; CHECK-NEXT: ldr.w r9, [sp, #12] @ 4-byte Reload793; CHECK-NEXT: add.w r10, r10, r2, lsl #1794; CHECK-NEXT: add.w r1, r1, r2, lsl #1795; CHECK-NEXT: ldr r2, [sp, #4] @ 4-byte Reload796; CHECK-NEXT: .LBB15_5: @ %while.end797; CHECK-NEXT: and r5, r3, #3798; CHECK-NEXT: vldrw.u32 q0, [r1]799; CHECK-NEXT: vctp.16 r5800; CHECK-NEXT: add.w r1, r10, #2801; CHECK-NEXT: vpst802; CHECK-NEXT: vstrht.16 q0, [r4]803; CHECK-NEXT: vldrw.u32 q0, [r10]804; CHECK-NEXT: vldrw.u32 q1, [r1]805; CHECK-NEXT: add.w r1, r10, #6806; CHECK-NEXT: vmul.f16 q0, q0, r7807; CHECK-NEXT: vfma.f16 q0, q1, r6808; CHECK-NEXT: vldrw.u32 q1, [r10, #4]809; CHECK-NEXT: vfma.f16 q0, q1, r12810; CHECK-NEXT: vldrw.u32 q1, [r1]811; CHECK-NEXT: vfma.f16 q0, q1, r8812; CHECK-NEXT: vpst813; CHECK-NEXT: vstrht.16 q0, [r2]814; CHECK-NEXT: ldr.w r10, [r0, #4]815; CHECK-NEXT: .LBB15_6: @ %if.end816; CHECK-NEXT: add.w r0, r10, r3, lsl #1817; CHECK-NEXT: lsr.w r1, r9, #2818; CHECK-NEXT: wls lr, r1, .LBB15_10819; CHECK-NEXT: @ %bb.7: @ %while.body51.preheader820; CHECK-NEXT: bic r2, r9, #3821; CHECK-NEXT: adds r1, r2, r3822; CHECK-NEXT: mov r3, r10823; CHECK-NEXT: add.w r1, r10, r1, lsl #1824; CHECK-NEXT: .LBB15_8: @ %while.body51825; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1826; CHECK-NEXT: vldrw.u32 q0, [r0], #8827; CHECK-NEXT: vstrb.8 q0, [r3], #8828; CHECK-NEXT: le lr, .LBB15_8829; CHECK-NEXT: @ %bb.9: @ %while.end55.loopexit830; CHECK-NEXT: add.w r10, r10, r2, lsl #1831; CHECK-NEXT: mov r0, r1832; CHECK-NEXT: .LBB15_10: @ %while.end55833; CHECK-NEXT: ands r1, r9, #3834; CHECK-NEXT: beq .LBB15_12835; CHECK-NEXT: @ %bb.11: @ %if.then59836; CHECK-NEXT: vldrw.u32 q0, [r0]837; CHECK-NEXT: vctp.16 r1838; CHECK-NEXT: vpst839; CHECK-NEXT: vstrht.16 q0, [r10]840; CHECK-NEXT: .LBB15_12: @ %if.end61841; CHECK-NEXT: add sp, #16842; CHECK-NEXT: pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}843entry:844 %pState1 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 1845 %i = load ptr, ptr %pState1, align 4846 %pCoeffs2 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 2847 %i1 = load ptr, ptr %pCoeffs2, align 4848 %numTaps3 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 0849 %i2 = load i16, ptr %numTaps3, align 4850 %conv = zext i16 %i2 to i32851 %sub = add nsw i32 %conv, -1852 %cmp = icmp ult i32 %sub, 4853 br i1 %cmp, label %if.then, label %if.end854 855if.then: ; preds = %entry856 %arrayidx = getelementptr inbounds half, ptr %i, i32 %sub857 %incdec.ptr = getelementptr inbounds half, ptr %i1, i32 1858 %i3 = load half, ptr %i1, align 4859 %incdec.ptr6 = getelementptr inbounds half, ptr %i1, i32 2860 %i4 = load half, ptr %incdec.ptr, align 4861 %incdec.ptr7 = getelementptr inbounds half, ptr %i1, i32 3862 %i5 = load half, ptr %incdec.ptr6, align 4863 %i6 = load half, ptr %incdec.ptr7, align 4864 %shr = lshr i32 %blockSize, 2865 %cmp9146 = icmp eq i32 %shr, 0866 %.pre161 = insertelement <8 x half> undef, half %i3, i32 0867 %.pre162 = shufflevector <8 x half> %.pre161, <8 x half> undef, <8 x i32> zeroinitializer868 %.pre163 = insertelement <8 x half> undef, half %i4, i32 0869 %.pre164 = shufflevector <8 x half> %.pre163, <8 x half> undef, <8 x i32> zeroinitializer870 %.pre165 = insertelement <8 x half> undef, half %i5, i32 0871 %.pre166 = shufflevector <8 x half> %.pre165, <8 x half> undef, <8 x i32> zeroinitializer872 %.pre167 = insertelement <8 x half> undef, half %i6, i32 0873 %.pre168 = shufflevector <8 x half> %.pre167, <8 x half> undef, <8 x i32> zeroinitializer874 br i1 %cmp9146, label %while.end, label %while.body.lr.ph875 876while.body.lr.ph: ; preds = %if.then877 %i7 = and i32 %blockSize, -4878 %scevgep158 = getelementptr half, ptr %pDst, i32 %i7879 br label %while.body880 881while.body: ; preds = %while.body, %while.body.lr.ph882 %pStateCur.0151 = phi ptr [ %arrayidx, %while.body.lr.ph ], [ %add.ptr, %while.body ]883 %pSamples.0150 = phi ptr [ %i, %while.body.lr.ph ], [ %add.ptr24, %while.body ]884 %pOutput.0149 = phi ptr [ %pDst, %while.body.lr.ph ], [ %add.ptr23, %while.body ]885 %pTempSrc.0148 = phi ptr [ %pSrc, %while.body.lr.ph ], [ %add.ptr11, %while.body ]886 %blkCnt.0147 = phi i32 [ %shr, %while.body.lr.ph ], [ %dec, %while.body ]887 %i9 = load <8 x half>, ptr %pTempSrc.0148, align 4888 store <8 x half> %i9, ptr %pStateCur.0151, align 4889 %add.ptr = getelementptr inbounds half, ptr %pStateCur.0151, i32 4890 %add.ptr11 = getelementptr inbounds half, ptr %pTempSrc.0148, i32 4891 %i12 = load <8 x half>, ptr %pSamples.0150, align 4892 %i13 = fmul fast <8 x half> %i12, %.pre162893 %arrayidx12 = getelementptr inbounds half, ptr %pSamples.0150, i32 1894 %i15 = load <8 x half>, ptr %arrayidx12, align 4895 %mul = fmul fast <8 x half> %i15, %.pre164896 %add = fadd fast <8 x half> %mul, %i13897 %arrayidx13 = getelementptr inbounds half, ptr %pSamples.0150, i32 2898 %i17 = load <8 x half>, ptr %arrayidx13, align 4899 %mul16 = fmul fast <8 x half> %i17, %.pre166900 %add17 = fadd fast <8 x half> %add, %mul16901 %arrayidx18 = getelementptr inbounds half, ptr %pSamples.0150, i32 3902 %i19 = load <8 x half>, ptr %arrayidx18, align 4903 %mul21 = fmul fast <8 x half> %i19, %.pre168904 %add22 = fadd fast <8 x half> %add17, %mul21905 store <8 x half> %add22, ptr %pOutput.0149, align 4906 %add.ptr23 = getelementptr inbounds half, ptr %pOutput.0149, i32 4907 %add.ptr24 = getelementptr inbounds half, ptr %pSamples.0150, i32 4908 %dec = add nsw i32 %blkCnt.0147, -1909 %cmp9 = icmp eq i32 %dec, 0910 br i1 %cmp9, label %while.end.loopexit, label %while.body911 912while.end.loopexit: ; preds = %while.body913 %scevgep157 = getelementptr half, ptr %pSrc, i32 %i7914 %scevgep159 = getelementptr half, ptr %i, i32 %i7915 br label %while.end916 917while.end: ; preds = %while.end.loopexit, %if.then918 %pTempSrc.0.lcssa = phi ptr [ %scevgep157, %while.end.loopexit ], [ %pSrc, %if.then ]919 %pOutput.0.lcssa = phi ptr [ %scevgep158, %while.end.loopexit ], [ %pDst, %if.then ]920 %pSamples.0.lcssa = phi ptr [ %scevgep159, %while.end.loopexit ], [ %i, %if.then ]921 %pStateCur.0.lcssa = phi ptr [ %add.ptr, %while.end.loopexit ], [ %arrayidx, %if.then ]922 %and = and i32 %blockSize, 3923 %i21 = tail call <8 x i1> @llvm.arm.mve.vctp16(i32 %and)924 %i23 = load <8 x half>, ptr %pTempSrc.0.lcssa, align 4925 tail call void @llvm.masked.store.v8f16.p0(<8 x half> %i23, ptr %pStateCur.0.lcssa, i32 4, <8 x i1> %i21)926 %i26 = load <8 x half>, ptr %pSamples.0.lcssa, align 4927 %i27 = fmul fast <8 x half> %i26, %.pre162928 %arrayidx29 = getelementptr inbounds half, ptr %pSamples.0.lcssa, i32 1929 %i29 = load <8 x half>, ptr %arrayidx29, align 4930 %mul32 = fmul fast <8 x half> %i29, %.pre164931 %add33 = fadd fast <8 x half> %mul32, %i27932 %arrayidx34 = getelementptr inbounds half, ptr %pSamples.0.lcssa, i32 2933 %i31 = load <8 x half>, ptr %arrayidx34, align 4934 %mul37 = fmul fast <8 x half> %i31, %.pre166935 %add38 = fadd fast <8 x half> %add33, %mul37936 %arrayidx39 = getelementptr inbounds half, ptr %pSamples.0.lcssa, i32 3937 %i33 = load <8 x half>, ptr %arrayidx39, align 4938 %mul42 = fmul fast <8 x half> %i33, %.pre168939 %add43 = fadd fast <8 x half> %add38, %mul42940 tail call void @llvm.masked.store.v8f16.p0(<8 x half> %add43, ptr %pOutput.0.lcssa, i32 4, <8 x i1> %i21)941 %.pre = load ptr, ptr %pState1, align 4942 br label %if.end943 944if.end: ; preds = %while.end, %entry945 %i35 = phi ptr [ %.pre, %while.end ], [ %i, %entry ]946 %arrayidx45 = getelementptr inbounds half, ptr %i35, i32 %blockSize947 %shr47 = lshr i32 %conv, 2948 %cmp49141 = icmp eq i32 %shr47, 0949 br i1 %cmp49141, label %while.end55, label %while.body51.preheader950 951while.body51.preheader: ; preds = %if.end952 %i36 = and i32 %conv, 65532953 %i37 = add i32 %i36, %blockSize954 %scevgep = getelementptr half, ptr %i35, i32 %i37955 br label %while.body51956 957while.body51: ; preds = %while.body51, %while.body51.preheader958 %pTempSrc.1144 = phi ptr [ %add.ptr52, %while.body51 ], [ %arrayidx45, %while.body51.preheader ]959 %pTempDest.0143 = phi ptr [ %add.ptr53, %while.body51 ], [ %i35, %while.body51.preheader ]960 %blkCnt.1142 = phi i32 [ %dec54, %while.body51 ], [ %shr47, %while.body51.preheader ]961 %i39 = load <8 x half>, ptr %pTempSrc.1144, align 4962 store <8 x half> %i39, ptr %pTempDest.0143, align 4963 %add.ptr52 = getelementptr inbounds half, ptr %pTempSrc.1144, i32 4964 %add.ptr53 = getelementptr inbounds half, ptr %pTempDest.0143, i32 4965 %dec54 = add nsw i32 %blkCnt.1142, -1966 %cmp49 = icmp eq i32 %dec54, 0967 br i1 %cmp49, label %while.end55.loopexit, label %while.body51968 969while.end55.loopexit: ; preds = %while.body51970 %scevgep156 = getelementptr half, ptr %i35, i32 %i36971 br label %while.end55972 973while.end55: ; preds = %while.end55.loopexit, %if.end974 %pTempDest.0.lcssa = phi ptr [ %i35, %if.end ], [ %scevgep156, %while.end55.loopexit ]975 %pTempSrc.1.lcssa = phi ptr [ %arrayidx45, %if.end ], [ %scevgep, %while.end55.loopexit ]976 %and56 = and i32 %conv, 3977 %cmp57 = icmp eq i32 %and56, 0978 br i1 %cmp57, label %if.end61, label %if.then59979 980if.then59: ; preds = %while.end55981 %i41 = tail call <8 x i1> @llvm.arm.mve.vctp16(i32 %and56)982 %i43 = load <8 x half>, ptr %pTempSrc.1.lcssa, align 4983 tail call void @llvm.masked.store.v8f16.p0(<8 x half> %i43, ptr %pTempDest.0.lcssa, i32 4, <8 x i1> %i41)984 br label %if.end61985 986if.end61: ; preds = %if.then59, %while.end55987 ret void988}989 990 991define void @fir(ptr nocapture readonly %S, ptr nocapture readonly %pSrc, ptr nocapture %pDst, i32 %blockSize) {992; CHECK-LABEL: fir:993; CHECK: @ %bb.0: @ %entry994; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, r10, r11, lr}995; CHECK-NEXT: push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}996; CHECK-NEXT: .pad #20997; CHECK-NEXT: sub sp, #20998; CHECK-NEXT: cmp r3, #8999; CHECK-NEXT: str r1, [sp, #16] @ 4-byte Spill1000; CHECK-NEXT: blo.w .LBB16_121001; CHECK-NEXT: @ %bb.1: @ %if.then1002; CHECK-NEXT: lsrs.w r12, r3, #21003; CHECK-NEXT: beq.w .LBB16_121004; CHECK-NEXT: @ %bb.2: @ %while.body.lr.ph1005; CHECK-NEXT: ldrh r4, [r0]1006; CHECK-NEXT: movs r1, #11007; CHECK-NEXT: ldrd r5, r3, [r0, #4]1008; CHECK-NEXT: sub.w r0, r4, #81009; CHECK-NEXT: add.w r7, r0, r0, lsr #291010; CHECK-NEXT: and r0, r0, #71011; CHECK-NEXT: asrs r6, r7, #31012; CHECK-NEXT: cmp r6, #11013; CHECK-NEXT: it gt1014; CHECK-NEXT: asrgt r1, r7, #31015; CHECK-NEXT: add.w r7, r5, r4, lsl #11016; CHECK-NEXT: str r1, [sp] @ 4-byte Spill1017; CHECK-NEXT: subs r1, r7, #21018; CHECK-NEXT: rsbs r7, r4, #01019; CHECK-NEXT: str r4, [sp, #8] @ 4-byte Spill1020; CHECK-NEXT: str r7, [sp, #4] @ 4-byte Spill1021; CHECK-NEXT: str r0, [sp, #12] @ 4-byte Spill1022; CHECK-NEXT: b .LBB16_61023; CHECK-NEXT: .LBB16_3: @ %while.end.loopexit1024; CHECK-NEXT: @ in Loop: Header=BB16_6 Depth=11025; CHECK-NEXT: ldr r0, [sp, #12] @ 4-byte Reload1026; CHECK-NEXT: add.w r6, r6, r0, lsl #11027; CHECK-NEXT: b .LBB16_51028; CHECK-NEXT: .LBB16_4: @ %for.end1029; CHECK-NEXT: @ in Loop: Header=BB16_6 Depth=11030; CHECK-NEXT: ldr r0, [sp, #12] @ 4-byte Reload1031; CHECK-NEXT: wls lr, r0, .LBB16_51032; CHECK-NEXT: b .LBB16_101033; CHECK-NEXT: .LBB16_5: @ %while.end1034; CHECK-NEXT: @ in Loop: Header=BB16_6 Depth=11035; CHECK-NEXT: ldr r0, [sp, #4] @ 4-byte Reload1036; CHECK-NEXT: subs.w r12, r12, #11037; CHECK-NEXT: vstrb.8 q0, [r2], #81038; CHECK-NEXT: add.w r0, r6, r0, lsl #11039; CHECK-NEXT: add.w r5, r0, #81040; CHECK-NEXT: beq.w .LBB16_121041; CHECK-NEXT: .LBB16_6: @ %while.body1042; CHECK-NEXT: @ =>This Loop Header: Depth=11043; CHECK-NEXT: @ Child Loop BB16_8 Depth 21044; CHECK-NEXT: @ Child Loop BB16_11 Depth 21045; CHECK-NEXT: ldr r0, [sp, #16] @ 4-byte Reload1046; CHECK-NEXT: ldrh.w lr, [r3, #14]1047; CHECK-NEXT: vldrw.u32 q0, [r0], #81048; CHECK-NEXT: ldrh.w r10, [r3, #12]1049; CHECK-NEXT: ldrh r7, [r3, #10]1050; CHECK-NEXT: ldrh r4, [r3, #8]1051; CHECK-NEXT: ldrh r6, [r3, #6]1052; CHECK-NEXT: ldrh.w r9, [r3, #4]1053; CHECK-NEXT: ldrh.w r11, [r3, #2]1054; CHECK-NEXT: ldrh.w r8, [r3]1055; CHECK-NEXT: vstrb.8 q0, [r1], #81056; CHECK-NEXT: vldrw.u32 q0, [r5]1057; CHECK-NEXT: str r0, [sp, #16] @ 4-byte Spill1058; CHECK-NEXT: adds r0, r5, #21059; CHECK-NEXT: vldrw.u32 q1, [r0]1060; CHECK-NEXT: vmul.f16 q0, q0, r81061; CHECK-NEXT: adds r0, r5, #61062; CHECK-NEXT: vfma.f16 q0, q1, r111063; CHECK-NEXT: vldrw.u32 q1, [r5, #4]1064; CHECK-NEXT: vfma.f16 q0, q1, r91065; CHECK-NEXT: vldrw.u32 q1, [r0]1066; CHECK-NEXT: add.w r0, r5, #101067; CHECK-NEXT: vfma.f16 q0, q1, r61068; CHECK-NEXT: vldrw.u32 q1, [r5, #8]1069; CHECK-NEXT: add.w r6, r5, #161070; CHECK-NEXT: vfma.f16 q0, q1, r41071; CHECK-NEXT: vldrw.u32 q1, [r0]1072; CHECK-NEXT: add.w r0, r5, #141073; CHECK-NEXT: vfma.f16 q0, q1, r71074; CHECK-NEXT: vldrw.u32 q1, [r5, #12]1075; CHECK-NEXT: vfma.f16 q0, q1, r101076; CHECK-NEXT: vldrw.u32 q1, [r0]1077; CHECK-NEXT: ldr r0, [sp, #8] @ 4-byte Reload1078; CHECK-NEXT: vfma.f16 q0, q1, lr1079; CHECK-NEXT: cmp r0, #161080; CHECK-NEXT: blo .LBB16_91081; CHECK-NEXT: @ %bb.7: @ %for.body.preheader1082; CHECK-NEXT: @ in Loop: Header=BB16_6 Depth=11083; CHECK-NEXT: ldr r0, [sp] @ 4-byte Reload1084; CHECK-NEXT: add.w r5, r3, #161085; CHECK-NEXT: dls lr, r01086; CHECK-NEXT: .LBB16_8: @ %for.body1087; CHECK-NEXT: @ Parent Loop BB16_6 Depth=11088; CHECK-NEXT: @ => This Inner Loop Header: Depth=21089; CHECK-NEXT: ldrh r0, [r5], #161090; CHECK-NEXT: vldrw.u32 q1, [r6]1091; CHECK-NEXT: adds r4, r6, #21092; CHECK-NEXT: vfma.f16 q0, q1, r01093; CHECK-NEXT: vldrw.u32 q1, [r4]1094; CHECK-NEXT: ldrh r0, [r5, #-14]1095; CHECK-NEXT: adds r4, r6, #61096; CHECK-NEXT: vfma.f16 q0, q1, r01097; CHECK-NEXT: ldrh r0, [r5, #-12]1098; CHECK-NEXT: vldrw.u32 q1, [r6, #4]1099; CHECK-NEXT: vfma.f16 q0, q1, r01100; CHECK-NEXT: vldrw.u32 q1, [r4]1101; CHECK-NEXT: ldrh r0, [r5, #-10]1102; CHECK-NEXT: add.w r4, r6, #101103; CHECK-NEXT: vfma.f16 q0, q1, r01104; CHECK-NEXT: ldrh r0, [r5, #-8]1105; CHECK-NEXT: vldrw.u32 q1, [r6, #8]1106; CHECK-NEXT: vfma.f16 q0, q1, r01107; CHECK-NEXT: vldrw.u32 q1, [r4]1108; CHECK-NEXT: ldrh r0, [r5, #-6]1109; CHECK-NEXT: ldrh r4, [r5, #-2]1110; CHECK-NEXT: vfma.f16 q0, q1, r01111; CHECK-NEXT: ldrh r0, [r5, #-4]1112; CHECK-NEXT: vldrw.u32 q1, [r6, #12]1113; CHECK-NEXT: vfma.f16 q0, q1, r01114; CHECK-NEXT: add.w r0, r6, #141115; CHECK-NEXT: vldrw.u32 q1, [r0]1116; CHECK-NEXT: adds r6, #161117; CHECK-NEXT: vfma.f16 q0, q1, r41118; CHECK-NEXT: le lr, .LBB16_81119; CHECK-NEXT: b .LBB16_41120; CHECK-NEXT: .LBB16_9: @ in Loop: Header=BB16_6 Depth=11121; CHECK-NEXT: add.w r5, r3, #161122; CHECK-NEXT: b .LBB16_41123; CHECK-NEXT: .LBB16_10: @ %while.body76.preheader1124; CHECK-NEXT: @ in Loop: Header=BB16_6 Depth=11125; CHECK-NEXT: mov r0, r61126; CHECK-NEXT: .LBB16_11: @ %while.body761127; CHECK-NEXT: @ Parent Loop BB16_6 Depth=11128; CHECK-NEXT: @ => This Inner Loop Header: Depth=21129; CHECK-NEXT: ldrh r4, [r5], #21130; CHECK-NEXT: vldrh.u16 q1, [r0], #21131; CHECK-NEXT: vfma.f16 q0, q1, r41132; CHECK-NEXT: le lr, .LBB16_111133; CHECK-NEXT: b .LBB16_31134; CHECK-NEXT: .LBB16_12: @ %if.end1135; CHECK-NEXT: add sp, #201136; CHECK-NEXT: pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}1137entry:1138 %pState1 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 11139 %i = load ptr, ptr %pState1, align 41140 %pCoeffs2 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 21141 %i1 = load ptr, ptr %pCoeffs2, align 41142 %numTaps3 = getelementptr inbounds %struct.arm_fir_instance_f32, ptr %S, i32 0, i32 01143 %i2 = load i16, ptr %numTaps3, align 41144 %conv = zext i16 %i2 to i321145 %cmp = icmp ugt i32 %blockSize, 71146 br i1 %cmp, label %if.then, label %if.end1147 1148if.then: ; preds = %entry1149 %shr = lshr i32 %blockSize, 21150 %cmp5217 = icmp eq i32 %shr, 01151 br i1 %cmp5217, label %if.end, label %while.body.lr.ph1152 1153while.body.lr.ph: ; preds = %if.then1154 %sub = add nsw i32 %conv, -11155 %arrayidx = getelementptr inbounds half, ptr %i, i32 %sub1156 %incdec.ptr = getelementptr inbounds half, ptr %i1, i32 11157 %incdec.ptr7 = getelementptr inbounds half, ptr %i1, i32 21158 %incdec.ptr8 = getelementptr inbounds half, ptr %i1, i32 31159 %incdec.ptr9 = getelementptr inbounds half, ptr %i1, i32 41160 %incdec.ptr10 = getelementptr inbounds half, ptr %i1, i32 51161 %incdec.ptr11 = getelementptr inbounds half, ptr %i1, i32 61162 %incdec.ptr12 = getelementptr inbounds half, ptr %i1, i32 71163 %sub37 = add nsw i32 %conv, -81164 %div = sdiv i32 %sub37, 81165 %pCoeffsCur.0199 = getelementptr inbounds half, ptr %i1, i32 81166 %cmp38201 = icmp ugt i16 %i2, 151167 %and = and i32 %sub37, 71168 %cmp74210 = icmp eq i32 %and, 01169 %idx.neg = sub nsw i32 0, %conv1170 %i3 = icmp sgt i32 %div, 11171 %smax = select i1 %i3, i32 %div, i32 11172 br label %while.body1173 1174while.body: ; preds = %while.end, %while.body.lr.ph1175 %blkCnt.0222 = phi i32 [ %shr, %while.body.lr.ph ], [ %dec84, %while.end ]1176 %pStateCur.0221 = phi ptr [ %arrayidx, %while.body.lr.ph ], [ %add.ptr, %while.end ]1177 %pSamples.0220 = phi ptr [ %i, %while.body.lr.ph ], [ %add.ptr83, %while.end ]1178 %pTempSrc.0219 = phi ptr [ %pSrc, %while.body.lr.ph ], [ %add.ptr14, %while.end ]1179 %pOutput.0218 = phi ptr [ %pDst, %while.body.lr.ph ], [ %add.ptr81, %while.end ]1180 %i4 = load half, ptr %i1, align 41181 %i5 = load half, ptr %incdec.ptr, align 41182 %i6 = load half, ptr %incdec.ptr7, align 41183 %i7 = load half, ptr %incdec.ptr8, align 41184 %i8 = load half, ptr %incdec.ptr9, align 41185 %i9 = load half, ptr %incdec.ptr10, align 41186 %i10 = load half, ptr %incdec.ptr11, align 41187 %i11 = load half, ptr %incdec.ptr12, align 41188 %i13 = load <8 x half>, ptr %pTempSrc.0219, align 41189 store <8 x half> %i13, ptr %pStateCur.0221, align 41190 %add.ptr = getelementptr inbounds half, ptr %pStateCur.0221, i32 41191 %add.ptr14 = getelementptr inbounds half, ptr %pTempSrc.0219, i32 41192 %i16 = load <8 x half>, ptr %pSamples.0220, align 41193 %.splatinsert = insertelement <8 x half> undef, half %i4, i32 01194 %.splat = shufflevector <8 x half> %.splatinsert, <8 x half> undef, <8 x i32> zeroinitializer1195 %i17 = fmul fast <8 x half> %i16, %.splat1196 %arrayidx15 = getelementptr inbounds half, ptr %pSamples.0220, i32 11197 %i19 = load <8 x half>, ptr %arrayidx15, align 41198 %.splatinsert16 = insertelement <8 x half> undef, half %i5, i32 01199 %.splat17 = shufflevector <8 x half> %.splatinsert16, <8 x half> undef, <8 x i32> zeroinitializer1200 %i20 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i19, <8 x half> %.splat17, <8 x half> %i17)1201 %arrayidx18 = getelementptr inbounds half, ptr %pSamples.0220, i32 21202 %i22 = load <8 x half>, ptr %arrayidx18, align 41203 %.splatinsert19 = insertelement <8 x half> undef, half %i6, i32 01204 %.splat20 = shufflevector <8 x half> %.splatinsert19, <8 x half> undef, <8 x i32> zeroinitializer1205 %i23 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i22, <8 x half> %.splat20, <8 x half> %i20)1206 %arrayidx21 = getelementptr inbounds half, ptr %pSamples.0220, i32 31207 %i25 = load <8 x half>, ptr %arrayidx21, align 41208 %.splatinsert22 = insertelement <8 x half> undef, half %i7, i32 01209 %.splat23 = shufflevector <8 x half> %.splatinsert22, <8 x half> undef, <8 x i32> zeroinitializer1210 %i26 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i25, <8 x half> %.splat23, <8 x half> %i23)1211 %arrayidx24 = getelementptr inbounds half, ptr %pSamples.0220, i32 41212 %i28 = load <8 x half>, ptr %arrayidx24, align 41213 %.splatinsert25 = insertelement <8 x half> undef, half %i8, i32 01214 %.splat26 = shufflevector <8 x half> %.splatinsert25, <8 x half> undef, <8 x i32> zeroinitializer1215 %i29 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i28, <8 x half> %.splat26, <8 x half> %i26)1216 %arrayidx27 = getelementptr inbounds half, ptr %pSamples.0220, i32 51217 %i31 = load <8 x half>, ptr %arrayidx27, align 41218 %.splatinsert28 = insertelement <8 x half> undef, half %i9, i32 01219 %.splat29 = shufflevector <8 x half> %.splatinsert28, <8 x half> undef, <8 x i32> zeroinitializer1220 %i32 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i31, <8 x half> %.splat29, <8 x half> %i29)1221 %arrayidx30 = getelementptr inbounds half, ptr %pSamples.0220, i32 61222 %i34 = load <8 x half>, ptr %arrayidx30, align 41223 %.splatinsert31 = insertelement <8 x half> undef, half %i10, i32 01224 %.splat32 = shufflevector <8 x half> %.splatinsert31, <8 x half> undef, <8 x i32> zeroinitializer1225 %i35 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i34, <8 x half> %.splat32, <8 x half> %i32)1226 %arrayidx33 = getelementptr inbounds half, ptr %pSamples.0220, i32 71227 %i37 = load <8 x half>, ptr %arrayidx33, align 41228 %.splatinsert34 = insertelement <8 x half> undef, half %i11, i32 01229 %.splat35 = shufflevector <8 x half> %.splatinsert34, <8 x half> undef, <8 x i32> zeroinitializer1230 %i38 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i37, <8 x half> %.splat35, <8 x half> %i35)1231 %pSamples.1200 = getelementptr inbounds half, ptr %pSamples.0220, i32 81232 br i1 %cmp38201, label %for.body, label %for.end1233 1234for.body: ; preds = %for.body, %while.body1235 %pSamples.1207 = phi ptr [ %pSamples.1, %for.body ], [ %pSamples.1200, %while.body ]1236 %pCoeffsCur.0206 = phi ptr [ %pCoeffsCur.0, %for.body ], [ %pCoeffsCur.0199, %while.body ]1237 %.pn205 = phi ptr [ %pCoeffsCur.0206, %for.body ], [ %i1, %while.body ]1238 %i.0204 = phi i32 [ %inc, %for.body ], [ 0, %while.body ]1239 %vecAcc0.0203 = phi <8 x half> [ %i70, %for.body ], [ %i38, %while.body ]1240 %pSamples.0.pn202 = phi ptr [ %pSamples.1207, %for.body ], [ %pSamples.0220, %while.body ]1241 %incdec.ptr40 = getelementptr inbounds half, ptr %.pn205, i32 91242 %i39 = load half, ptr %pCoeffsCur.0206, align 41243 %incdec.ptr41 = getelementptr inbounds half, ptr %.pn205, i32 101244 %i40 = load half, ptr %incdec.ptr40, align 41245 %incdec.ptr42 = getelementptr inbounds half, ptr %.pn205, i32 111246 %i41 = load half, ptr %incdec.ptr41, align 41247 %incdec.ptr43 = getelementptr inbounds half, ptr %.pn205, i32 121248 %i42 = load half, ptr %incdec.ptr42, align 41249 %incdec.ptr44 = getelementptr inbounds half, ptr %.pn205, i32 131250 %i43 = load half, ptr %incdec.ptr43, align 41251 %incdec.ptr45 = getelementptr inbounds half, ptr %.pn205, i32 141252 %i44 = load half, ptr %incdec.ptr44, align 41253 %incdec.ptr46 = getelementptr inbounds half, ptr %.pn205, i32 151254 %i45 = load half, ptr %incdec.ptr45, align 41255 %i46 = load half, ptr %incdec.ptr46, align 41256 %i48 = load <8 x half>, ptr %pSamples.1207, align 41257 %.splatinsert48 = insertelement <8 x half> undef, half %i39, i32 01258 %.splat49 = shufflevector <8 x half> %.splatinsert48, <8 x half> undef, <8 x i32> zeroinitializer1259 %i49 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i48, <8 x half> %.splat49, <8 x half> %vecAcc0.0203)1260 %arrayidx50 = getelementptr inbounds half, ptr %pSamples.0.pn202, i32 91261 %i51 = load <8 x half>, ptr %arrayidx50, align 41262 %.splatinsert51 = insertelement <8 x half> undef, half %i40, i32 01263 %.splat52 = shufflevector <8 x half> %.splatinsert51, <8 x half> undef, <8 x i32> zeroinitializer1264 %i52 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i51, <8 x half> %.splat52, <8 x half> %i49)1265 %arrayidx53 = getelementptr inbounds half, ptr %pSamples.0.pn202, i32 101266 %i54 = load <8 x half>, ptr %arrayidx53, align 41267 %.splatinsert54 = insertelement <8 x half> undef, half %i41, i32 01268 %.splat55 = shufflevector <8 x half> %.splatinsert54, <8 x half> undef, <8 x i32> zeroinitializer1269 %i55 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i54, <8 x half> %.splat55, <8 x half> %i52)1270 %arrayidx56 = getelementptr inbounds half, ptr %pSamples.0.pn202, i32 111271 %i57 = load <8 x half>, ptr %arrayidx56, align 41272 %.splatinsert57 = insertelement <8 x half> undef, half %i42, i32 01273 %.splat58 = shufflevector <8 x half> %.splatinsert57, <8 x half> undef, <8 x i32> zeroinitializer1274 %i58 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i57, <8 x half> %.splat58, <8 x half> %i55)1275 %arrayidx59 = getelementptr inbounds half, ptr %pSamples.0.pn202, i32 121276 %i60 = load <8 x half>, ptr %arrayidx59, align 41277 %.splatinsert60 = insertelement <8 x half> undef, half %i43, i32 01278 %.splat61 = shufflevector <8 x half> %.splatinsert60, <8 x half> undef, <8 x i32> zeroinitializer1279 %i61 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i60, <8 x half> %.splat61, <8 x half> %i58)1280 %arrayidx62 = getelementptr inbounds half, ptr %pSamples.0.pn202, i32 131281 %i63 = load <8 x half>, ptr %arrayidx62, align 41282 %.splatinsert63 = insertelement <8 x half> undef, half %i44, i32 01283 %.splat64 = shufflevector <8 x half> %.splatinsert63, <8 x half> undef, <8 x i32> zeroinitializer1284 %i64 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i63, <8 x half> %.splat64, <8 x half> %i61)1285 %arrayidx65 = getelementptr inbounds half, ptr %pSamples.0.pn202, i32 141286 %i66 = load <8 x half>, ptr %arrayidx65, align 41287 %.splatinsert66 = insertelement <8 x half> undef, half %i45, i32 01288 %.splat67 = shufflevector <8 x half> %.splatinsert66, <8 x half> undef, <8 x i32> zeroinitializer1289 %i67 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i66, <8 x half> %.splat67, <8 x half> %i64)1290 %arrayidx68 = getelementptr inbounds half, ptr %pSamples.0.pn202, i32 151291 %i69 = load <8 x half>, ptr %arrayidx68, align 41292 %.splatinsert69 = insertelement <8 x half> undef, half %i46, i32 01293 %.splat70 = shufflevector <8 x half> %.splatinsert69, <8 x half> undef, <8 x i32> zeroinitializer1294 %i70 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i69, <8 x half> %.splat70, <8 x half> %i67)1295 %inc = add nuw nsw i32 %i.0204, 11296 %pCoeffsCur.0 = getelementptr inbounds half, ptr %pCoeffsCur.0206, i32 81297 %pSamples.1 = getelementptr inbounds half, ptr %pSamples.1207, i32 81298 %exitcond = icmp eq i32 %inc, %smax1299 br i1 %exitcond, label %for.end, label %for.body1300 1301for.end: ; preds = %for.body, %while.body1302 %vecAcc0.0.lcssa = phi <8 x half> [ %i38, %while.body ], [ %i70, %for.body ]1303 %pCoeffsCur.0.lcssa = phi ptr [ %pCoeffsCur.0199, %while.body ], [ %pCoeffsCur.0, %for.body ]1304 %pSamples.1.lcssa = phi ptr [ %pSamples.1200, %while.body ], [ %pSamples.1, %for.body ]1305 br i1 %cmp74210, label %while.end, label %while.body761306 1307while.body76: ; preds = %while.body76, %for.end1308 %pCoeffsCur.1214 = phi ptr [ %incdec.ptr77, %while.body76 ], [ %pCoeffsCur.0.lcssa, %for.end ]1309 %vecAcc0.1213 = phi <8 x half> [ %i74, %while.body76 ], [ %vecAcc0.0.lcssa, %for.end ]1310 %numCnt.0212 = phi i32 [ %dec, %while.body76 ], [ %and, %for.end ]1311 %pSamples.2211 = phi ptr [ %incdec.ptr80, %while.body76 ], [ %pSamples.1.lcssa, %for.end ]1312 %incdec.ptr77 = getelementptr inbounds half, ptr %pCoeffsCur.1214, i32 11313 %i71 = load half, ptr %pCoeffsCur.1214, align 41314 %i73 = load <8 x half>, ptr %pSamples.2211, align 41315 %.splatinsert78 = insertelement <8 x half> undef, half %i71, i32 01316 %.splat79 = shufflevector <8 x half> %.splatinsert78, <8 x half> undef, <8 x i32> zeroinitializer1317 %i74 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i73, <8 x half> %.splat79, <8 x half> %vecAcc0.1213)1318 %incdec.ptr80 = getelementptr inbounds half, ptr %pSamples.2211, i32 11319 %dec = add nsw i32 %numCnt.0212, -11320 %cmp74 = icmp sgt i32 %numCnt.0212, 11321 br i1 %cmp74, label %while.body76, label %while.end.loopexit1322 1323while.end.loopexit: ; preds = %while.body761324 %scevgep = getelementptr half, ptr %pSamples.1.lcssa, i32 %and1325 br label %while.end1326 1327while.end: ; preds = %while.end.loopexit, %for.end1328 %pSamples.2.lcssa = phi ptr [ %pSamples.1.lcssa, %for.end ], [ %scevgep, %while.end.loopexit ]1329 %vecAcc0.1.lcssa = phi <8 x half> [ %vecAcc0.0.lcssa, %for.end ], [ %i74, %while.end.loopexit ]1330 store <8 x half> %vecAcc0.1.lcssa, ptr %pOutput.0218, align 41331 %add.ptr81 = getelementptr inbounds half, ptr %pOutput.0218, i32 41332 %add.ptr82 = getelementptr inbounds half, ptr %pSamples.2.lcssa, i32 41333 %add.ptr83 = getelementptr inbounds half, ptr %add.ptr82, i32 %idx.neg1334 %dec84 = add nsw i32 %blkCnt.0222, -11335 %cmp5 = icmp eq i32 %dec84, 01336 br i1 %cmp5, label %if.end, label %while.body1337 1338if.end: ; preds = %while.end, %if.then, %entry1339 ret void1340}1341 1342%struct.arm_biquad_cascade_df2T_instance_f16 = type { i8, ptr, ptr }1343define void @arm_biquad_cascade_df2T_f16(ptr nocapture readonly %S, ptr nocapture readonly %pSrc, ptr nocapture %pDst, i32 %blockSize) {1344; CHECK-LABEL: arm_biquad_cascade_df2T_f16:1345; CHECK: @ %bb.0: @ %entry1346; CHECK-NEXT: .save {r4, r5, r6, r7, r8, r9, lr}1347; CHECK-NEXT: push.w {r4, r5, r6, r7, r8, r9, lr}1348; CHECK-NEXT: .pad #41349; CHECK-NEXT: sub sp, #41350; CHECK-NEXT: .vsave {d8, d9, d10, d11}1351; CHECK-NEXT: vpush {d8, d9, d10, d11}1352; CHECK-NEXT: vmov.i32 q0, #0x01353; CHECK-NEXT: ldrd r6, r12, [r0, #4]1354; CHECK-NEXT: ldrb.w r9, [r0]1355; CHECK-NEXT: vldr.16 s0, .LCPI17_01356; CHECK-NEXT: lsr.w r8, r3, #11357; CHECK-NEXT: b .LBB17_31358; CHECK-NEXT: .LBB17_1: @ %if.else1359; CHECK-NEXT: @ in Loop: Header=BB17_3 Depth=11360; CHECK-NEXT: vmovx.f16 s5, s41361; CHECK-NEXT: vstr.16 s4, [r6]1362; CHECK-NEXT: .LBB17_2: @ %if.end1363; CHECK-NEXT: @ in Loop: Header=BB17_3 Depth=11364; CHECK-NEXT: vstr.16 s5, [r6, #2]1365; CHECK-NEXT: add.w r12, r12, #101366; CHECK-NEXT: adds r6, #41367; CHECK-NEXT: subs.w r9, r9, #11368; CHECK-NEXT: mov r1, r21369; CHECK-NEXT: beq .LBB17_81370; CHECK-NEXT: .LBB17_3: @ %do.body1371; CHECK-NEXT: @ =>This Loop Header: Depth=11372; CHECK-NEXT: @ Child Loop BB17_5 Depth 21373; CHECK-NEXT: vldrh.u16 q2, [r12]1374; CHECK-NEXT: movs r5, #01375; CHECK-NEXT: vmov q4, q21376; CHECK-NEXT: vshlc q4, r5, #161377; CHECK-NEXT: vldrh.u16 q3, [r12, #4]1378; CHECK-NEXT: vmov q5, q31379; CHECK-NEXT: vshlc q5, r5, #161380; CHECK-NEXT: vldrh.u16 q1, [r6]1381; CHECK-NEXT: vmov.f32 s5, s11382; CHECK-NEXT: mov r5, r21383; CHECK-NEXT: wls lr, r8, .LBB17_61384; CHECK-NEXT: @ %bb.4: @ %while.body.preheader1385; CHECK-NEXT: @ in Loop: Header=BB17_3 Depth=11386; CHECK-NEXT: mov r5, r21387; CHECK-NEXT: .LBB17_5: @ %while.body1388; CHECK-NEXT: @ Parent Loop BB17_3 Depth=11389; CHECK-NEXT: @ => This Inner Loop Header: Depth=21390; CHECK-NEXT: ldrh r7, [r1], #41391; CHECK-NEXT: vmov r0, s01392; CHECK-NEXT: vfma.f16 q1, q2, r71393; CHECK-NEXT: ldrh r4, [r1, #-2]1394; CHECK-NEXT: vmov.u16 r7, q1[0]1395; CHECK-NEXT: vfma.f16 q1, q3, r71396; CHECK-NEXT: vins.f16 s5, s01397; CHECK-NEXT: vfma.f16 q1, q4, r41398; CHECK-NEXT: vmov.u16 r4, q1[1]1399; CHECK-NEXT: vfma.f16 q1, q5, r41400; CHECK-NEXT: strh r4, [r5, #2]1401; CHECK-NEXT: vmov.f32 s4, s51402; CHECK-NEXT: strh r7, [r5], #41403; CHECK-NEXT: vmov.16 q1[2], r01404; CHECK-NEXT: le lr, .LBB17_51405; CHECK-NEXT: .LBB17_6: @ %while.end1406; CHECK-NEXT: @ in Loop: Header=BB17_3 Depth=11407; CHECK-NEXT: lsls r0, r3, #311408; CHECK-NEXT: beq .LBB17_11409; CHECK-NEXT: @ %bb.7: @ %if.then1410; CHECK-NEXT: @ in Loop: Header=BB17_3 Depth=11411; CHECK-NEXT: ldrh r0, [r1]1412; CHECK-NEXT: vfma.f16 q1, q2, r01413; CHECK-NEXT: vmov.u16 r0, q1[0]1414; CHECK-NEXT: vfma.f16 q1, q3, r01415; CHECK-NEXT: strh r0, [r5]1416; CHECK-NEXT: vmovx.f16 s2, s41417; CHECK-NEXT: vstr.16 s2, [r6]1418; CHECK-NEXT: b .LBB17_21419; CHECK-NEXT: .LBB17_8: @ %do.end1420; CHECK-NEXT: vpop {d8, d9, d10, d11}1421; CHECK-NEXT: add sp, #41422; CHECK-NEXT: pop.w {r4, r5, r6, r7, r8, r9, pc}1423; CHECK-NEXT: .p2align 11424; CHECK-NEXT: @ %bb.9:1425; CHECK-NEXT: .LCPI17_0:1426; CHECK-NEXT: .short 0x0000 @ half 01427entry:1428 %pState1 = getelementptr inbounds %struct.arm_biquad_cascade_df2T_instance_f16, ptr %S, i32 0, i32 11429 %i = load ptr, ptr %pState1, align 41430 %numStages = getelementptr inbounds %struct.arm_biquad_cascade_df2T_instance_f16, ptr %S, i32 0, i32 01431 %i1 = load i8, ptr %numStages, align 41432 %conv = zext i8 %i1 to i321433 %pCoeffs = getelementptr inbounds %struct.arm_biquad_cascade_df2T_instance_f16, ptr %S, i32 0, i32 21434 %i2 = load ptr, ptr %pCoeffs, align 41435 %div = lshr i32 %blockSize, 11436 %cmp.not90 = icmp eq i32 %div, 01437 %and = and i32 %blockSize, 11438 %tobool.not = icmp eq i32 %and, 01439 br label %do.body1440 1441do.body: ; preds = %if.end, %entry1442 %stage.0 = phi i32 [ %conv, %entry ], [ %dec23, %if.end ]1443 %pCurCoeffs.0 = phi ptr [ %i2, %entry ], [ %add.ptr2, %if.end ]1444 %pState.0 = phi ptr [ %i, %entry ], [ %pState.1, %if.end ]1445 %pIn.0 = phi ptr [ %pSrc, %entry ], [ %pDst, %if.end ]1446 %i4 = load <8 x half>, ptr %pCurCoeffs.0, align 21447 %add.ptr = getelementptr inbounds half, ptr %pCurCoeffs.0, i32 21448 %i6 = load <8 x half>, ptr %add.ptr, align 21449 %add.ptr2 = getelementptr inbounds half, ptr %pCurCoeffs.0, i32 51450 %i8 = load <8 x half>, ptr %pState.0, align 21451 %i9 = shufflevector <8 x half> %i8, <8 x half> <half poison, half poison, half 0xH0000, half 0xH0000, half poison, half poison, half poison, half poison>, <8 x i32> <i32 0, i32 1, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7>1452 %i10 = bitcast <8 x half> %i4 to <8 x i16>1453 %i11 = tail call { i32, <8 x i16> } @llvm.arm.mve.vshlc.v8i16(<8 x i16> %i10, i32 0, i32 16)1454 %i12 = extractvalue { i32, <8 x i16> } %i11, 01455 %i13 = extractvalue { i32, <8 x i16> } %i11, 11456 %i14 = bitcast <8 x i16> %i13 to <8 x half>1457 %i15 = bitcast <8 x half> %i6 to <8 x i16>1458 %i16 = tail call { i32, <8 x i16> } @llvm.arm.mve.vshlc.v8i16(<8 x i16> %i15, i32 %i12, i32 16)1459 %i17 = extractvalue { i32, <8 x i16> } %i16, 11460 %i18 = bitcast <8 x i16> %i17 to <8 x half>1461 br i1 %cmp.not90, label %while.end, label %while.body1462 1463while.body: ; preds = %while.body, %do.body1464 %pIn.194 = phi ptr [ %incdec.ptr4, %while.body ], [ %pIn.0, %do.body ]1465 %state.093 = phi <8 x half> [ %i30, %while.body ], [ %i9, %do.body ]1466 %pOut.192 = phi ptr [ %incdec.ptr12, %while.body ], [ %pDst, %do.body ]1467 %sample.091 = phi i32 [ %dec, %while.body ], [ %div, %do.body ]1468 %incdec.ptr = getelementptr inbounds half, ptr %pIn.194, i32 11469 %i19 = load half, ptr %pIn.194, align 21470 %incdec.ptr4 = getelementptr inbounds half, ptr %pIn.194, i32 21471 %i20 = load half, ptr %incdec.ptr, align 21472 %.splatinsert = insertelement <8 x half> poison, half %i19, i32 01473 %.splat = shufflevector <8 x half> %.splatinsert, <8 x half> poison, <8 x i32> zeroinitializer1474 %i21 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i4, <8 x half> %.splat, <8 x half> %state.093)1475 %i22 = extractelement <8 x half> %i21, i32 01476 %.splat6 = shufflevector <8 x half> %i21, <8 x half> poison, <8 x i32> zeroinitializer1477 %i23 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i6, <8 x half> %.splat6, <8 x half> %i21)1478 %i24 = insertelement <8 x half> %i23, half 0xH0000, i32 31479 %.splatinsert7 = insertelement <8 x half> poison, half %i20, i32 01480 %.splat8 = shufflevector <8 x half> %.splatinsert7, <8 x half> poison, <8 x i32> zeroinitializer1481 %i25 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i14, <8 x half> %.splat8, <8 x half> %i24)1482 %i26 = extractelement <8 x half> %i25, i32 11483 %.splat10 = shufflevector <8 x half> %i25, <8 x half> undef, <8 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>1484 %i27 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i18, <8 x half> %.splat10, <8 x half> %i25)1485 %i28 = shufflevector <8 x half> %i27, <8 x half> undef, <8 x i32> <i32 2, i32 undef, i32 undef, i32 3, i32 4, i32 5, i32 6, i32 7>1486 %i29 = insertelement <8 x half> %i28, half 0xH0000, i32 21487 %i30 = shufflevector <8 x half> %i29, <8 x half> %i27, <8 x i32> <i32 0, i32 11, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>1488 %incdec.ptr11 = getelementptr inbounds half, ptr %pOut.192, i32 11489 store half %i22, ptr %pOut.192, align 21490 %incdec.ptr12 = getelementptr inbounds half, ptr %pOut.192, i32 21491 store half %i26, ptr %incdec.ptr11, align 21492 %dec = add nsw i32 %sample.091, -11493 %cmp.not = icmp eq i32 %dec, 01494 br i1 %cmp.not, label %while.end, label %while.body1495 1496while.end: ; preds = %while.body, %do.body1497 %pOut.1.lcssa = phi ptr [ %pDst, %do.body ], [ %incdec.ptr12, %while.body ]1498 %state.0.lcssa = phi <8 x half> [ %i9, %do.body ], [ %i30, %while.body ]1499 %pIn.1.lcssa = phi ptr [ %pIn.0, %do.body ], [ %incdec.ptr4, %while.body ]1500 br i1 %tobool.not, label %if.else, label %if.then1501 1502if.then: ; preds = %while.end1503 %i31 = load half, ptr %pIn.1.lcssa, align 21504 %.splatinsert14 = insertelement <8 x half> poison, half %i31, i32 01505 %.splat15 = shufflevector <8 x half> %.splatinsert14, <8 x half> poison, <8 x i32> zeroinitializer1506 %i32 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i4, <8 x half> %.splat15, <8 x half> %state.0.lcssa)1507 %i33 = extractelement <8 x half> %i32, i32 01508 %.splat17 = shufflevector <8 x half> %i32, <8 x half> poison, <8 x i32> zeroinitializer1509 %i34 = tail call fast <8 x half> @llvm.fma.v8f16(<8 x half> %i6, <8 x half> %.splat17, <8 x half> %i32)1510 store half %i33, ptr %pOut.1.lcssa, align 21511 %i35 = extractelement <8 x half> %i34, i32 11512 store half %i35, ptr %pState.0, align 21513 %i36 = extractelement <8 x half> %i34, i32 21514 br label %if.end1515 1516if.else: ; preds = %while.end1517 %i37 = extractelement <8 x half> %state.0.lcssa, i32 01518 store half %i37, ptr %pState.0, align 21519 %i38 = extractelement <8 x half> %state.0.lcssa, i32 11520 br label %if.end1521 1522if.end: ; preds = %if.else, %if.then1523 %.sink = phi half [ %i38, %if.else ], [ %i36, %if.then ]1524 %i39 = getelementptr inbounds half, ptr %pState.0, i32 11525 store half %.sink, ptr %i39, align 21526 %pState.1 = getelementptr inbounds half, ptr %pState.0, i32 21527 %dec23 = add i32 %stage.0, -11528 %cmp24.not = icmp eq i32 %dec23, 01529 br i1 %cmp24.not, label %do.end, label %do.body1530 1531do.end: ; preds = %if.end1532 ret void1533}1534 1535define arm_aapcs_vfpcc half @vecAddAcrossF16Mve(<8 x half> %in) {1536; CHECK-LABEL: vecAddAcrossF16Mve:1537; CHECK: @ %bb.0: @ %entry1538; CHECK-NEXT: vrev32.16 q1, q01539; CHECK-NEXT: vadd.f16 q0, q1, q01540; CHECK-NEXT: vrev64.32 q1, q01541; CHECK-NEXT: vadd.f16 q0, q0, q11542; CHECK-NEXT: vadd.f16 s0, s0, s21543; CHECK-NEXT: bx lr1544entry:1545 %i = shufflevector <8 x half> %in, <8 x half> poison, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>1546 %i1 = fadd fast <8 x half> %i, %in1547 %i2 = bitcast <8 x half> %i1 to <4 x i32>1548 %i3 = shufflevector <4 x i32> %i2, <4 x i32> poison, <4 x i32> <i32 1, i32 undef, i32 3, i32 undef>1549 %i4 = bitcast <4 x i32> %i3 to <8 x half>1550 %i5 = fadd fast <8 x half> %i1, %i41551 %i6 = extractelement <8 x half> %i5, i32 01552 %i7 = extractelement <8 x half> %i5, i32 41553 %add = fadd fast half %i6, %i71554 ret half %add1555}1556 1557declare { i32, <8 x i16> } @llvm.arm.mve.vshlc.v8i16(<8 x i16>, i32, i32)1558declare void @llvm.assume(i1)1559declare <8 x i1> @llvm.arm.mve.vctp16(i32)1560declare <8 x half> @llvm.fma.v8f16(<8 x half>, <8 x half>, <8 x half>)1561declare void @llvm.masked.store.v8f16.p0(<8 x half>, ptr, i32 immarg, <8 x i1>)1562