114 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 -tail-predication=force-enabled-no-reductions -o - %s | FileCheck %s3 4define arm_aapcs_vfpcc <16 x i8> @vcmp_vpst_combination(ptr %pSrc, i16 zeroext %blockSize, ptr nocapture %pResult, ptr nocapture %pIndex) {5; CHECK-LABEL: vcmp_vpst_combination:6; CHECK: @ %bb.0: @ %entry7; CHECK-NEXT: .save {r7, lr}8; CHECK-NEXT: push {r7, lr}9; CHECK-NEXT: vmov.i8 q0, #0x7f10; CHECK-NEXT: dlstp.8 lr, r111; CHECK-NEXT: .LBB0_1: @ %do.body12; CHECK-NEXT: @ =>This Inner Loop Header: Depth=113; CHECK-NEXT: vldrb.u8 q1, [r0]14; CHECK-NEXT: vpt.s8 ge, q0, q115; CHECK-NEXT: vmovt q0, q116; CHECK-NEXT: letp lr, .LBB0_117; CHECK-NEXT: @ %bb.2: @ %do.end18; CHECK-NEXT: pop {r7, pc}19entry:20 %conv = zext i16 %blockSize to i3221 %0 = tail call { <16 x i8>, i32 } @llvm.arm.mve.vidup.v16i8(i32 0, i32 1)22 %1 = extractvalue { <16 x i8>, i32 } %0, 023 br label %do.body24 25do.body: ; preds = %do.body, %entry26 %indexVec.0 = phi <16 x i8> [ %1, %entry ], [ %add, %do.body ]27 %curExtremIdxVec.0 = phi <16 x i8> [ zeroinitializer, %entry ], [ %6, %do.body ]28 %curExtremValVec.0 = phi <16 x i8> [ <i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127, i8 127>, %entry ], [ %6, %do.body ]29 %blkCnt.0 = phi i32 [ %conv, %entry ], [ %sub2, %do.body ]30 %2 = tail call <16 x i1> @llvm.arm.mve.vctp8(i32 %blkCnt.0)31 %3 = tail call <16 x i8> @llvm.masked.load.v16i8.p0(ptr %pSrc, i32 1, <16 x i1> %2, <16 x i8> zeroinitializer)32 %4 = icmp sle <16 x i8> %3, %curExtremValVec.033 %5 = and <16 x i1> %4, %234 %6 = tail call <16 x i8> @llvm.arm.mve.orr.predicated.v16i8.v16i1(<16 x i8> %3, <16 x i8> %3, <16 x i1> %5, <16 x i8> %curExtremValVec.0)35 %add = add <16 x i8> %indexVec.0, <i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16, i8 16>36 %sub2 = add nsw i32 %blkCnt.0, -1637 %cmp = icmp sgt i32 %blkCnt.0, 1638 br i1 %cmp, label %do.body, label %do.end39 40do.end: ; preds = %do.body41 ret <16 x i8> %642}43 44define i32 @vcmp_new_vpst_combination(i32 %len, ptr nocapture readonly %arr) {45; CHECK-LABEL: vcmp_new_vpst_combination:46; CHECK: @ %bb.0: @ %entry47; CHECK-NEXT: .save {r7, lr}48; CHECK-NEXT: push {r7, lr}49; CHECK-NEXT: cmp r0, #150; CHECK-NEXT: blt .LBB1_451; CHECK-NEXT: @ %bb.1: @ %vector.ph52; CHECK-NEXT: vmov.i32 q0, #0x153; CHECK-NEXT: movs r2, #054; CHECK-NEXT: dlstp.32 lr, r055; CHECK-NEXT: .LBB1_2: @ %vector.body56; CHECK-NEXT: @ =>This Inner Loop Header: Depth=157; CHECK-NEXT: vldrw.u32 q1, [r1], #1658; CHECK-NEXT: vcmp.i32 ne, q1, zr59; CHECK-NEXT: vmov.i32 q1, #0x060; CHECK-NEXT: vpst61; CHECK-NEXT: vmovt q1, q062; CHECK-NEXT: vaddva.u32 r2, q163; CHECK-NEXT: letp lr, .LBB1_264; CHECK-NEXT: @ %bb.3: @ %for.cond.cleanup65; CHECK-NEXT: mov r0, r266; CHECK-NEXT: pop {r7, pc}67; CHECK-NEXT: .LBB1_4:68; CHECK-NEXT: movs r2, #069; CHECK-NEXT: mov r0, r270; CHECK-NEXT: pop {r7, pc}71entry:72 %cmp7 = icmp sgt i32 %len, 073 br i1 %cmp7, label %vector.ph, label %for.cond.cleanup74 75vector.ph: ; preds = %entry76 %n.rnd.up = add i32 %len, 377 %n.vec = and i32 %n.rnd.up, -478 br label %vector.body79 80vector.body: ; preds = %vector.body, %vector.ph81 %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]82 %vec.phi = phi i32 [ 0, %vector.ph ], [ %5, %vector.body ]83 %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 %len)84 %0 = getelementptr inbounds i32, ptr %arr, i32 %index85 %1 = bitcast ptr %0 to ptr86 %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr %1, i32 4, <4 x i1> %active.lane.mask, <4 x i32> undef)87 %2 = icmp ne <4 x i32> %wide.masked.load, zeroinitializer88 %narrow = and <4 x i1> %active.lane.mask, %289 %3 = zext <4 x i1> %narrow to <4 x i32>90 %4 = call i32 @llvm.experimental.vector.reduce.add.v4i32(<4 x i32> %3)91 %5 = add i32 %4, %vec.phi92 %index.next = add i32 %index, 493 %6 = icmp eq i32 %index.next, %n.vec94 br i1 %6, label %for.cond.cleanup, label %vector.body95 96for.cond.cleanup: ; preds = %vector.body, %entry97 %count.0.lcssa = phi i32 [ 0, %entry ], [ %5, %vector.body ]98 ret i32 %count.0.lcssa99}100 101declare <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32, i32)102 103declare <4 x i32> @llvm.masked.load.v4i32.p0(ptr, i32 immarg, <4 x i1>, <4 x i32>)104 105declare i32 @llvm.experimental.vector.reduce.add.v4i32(<4 x i32>)106 107declare { <16 x i8>, i32 } @llvm.arm.mve.vidup.v16i8(i32, i32)108 109declare <16 x i1> @llvm.arm.mve.vctp8(i32)110 111declare <16 x i8> @llvm.masked.load.v16i8.p0(ptr, i32 immarg, <16 x i1>, <16 x i8>)112 113declare <16 x i8> @llvm.arm.mve.orr.predicated.v16i8.v16i1(<16 x i8>, <16 x i8>, <16 x i1>, <16 x i8>)114