|
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4 |
| 2 | +; RUN: opt -p loop-vectorize -mattr="+v" -S %s | FileCheck %s |
| 3 | + |
| 4 | +target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128" |
| 5 | +target triple = "riscv64-unknown-linux-gnu" |
| 6 | + |
| 7 | +; Test case for https://github.com/llvm/llvm-project/issues/87378. |
| 8 | +define void @pr87378_vpinstruction_or_drop_poison_generating_flags(ptr %arg, i64 %a, i64 %b, i64 %c) { |
| 9 | +; CHECK-LABEL: define void @pr87378_vpinstruction_or_drop_poison_generating_flags( |
| 10 | +; CHECK-SAME: ptr [[ARG:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR0:[0-9]+]] { |
| 11 | +; CHECK-NEXT: entry: |
| 12 | +; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() |
| 13 | +; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 8 |
| 14 | +; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 1001, [[TMP1]] |
| 15 | +; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| 16 | +; CHECK: vector.ph: |
| 17 | +; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() |
| 18 | +; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 8 |
| 19 | +; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1001, [[TMP3]] |
| 20 | +; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1001, [[N_MOD_VF]] |
| 21 | +; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64() |
| 22 | +; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8 |
| 23 | +; CHECK-NEXT: [[TMP6:%.*]] = call <vscale x 8 x i64> @llvm.experimental.stepvector.nxv8i64() |
| 24 | +; CHECK-NEXT: [[TMP7:%.*]] = add <vscale x 8 x i64> [[TMP6]], zeroinitializer |
| 25 | +; CHECK-NEXT: [[TMP8:%.*]] = mul <vscale x 8 x i64> [[TMP7]], shufflevector (<vscale x 8 x i64> insertelement (<vscale x 8 x i64> poison, i64 1, i64 0), <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer) |
| 26 | +; CHECK-NEXT: [[INDUCTION:%.*]] = add <vscale x 8 x i64> zeroinitializer, [[TMP8]] |
| 27 | +; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64() |
| 28 | +; CHECK-NEXT: [[TMP10:%.*]] = mul i64 [[TMP9]], 8 |
| 29 | +; CHECK-NEXT: [[TMP11:%.*]] = mul i64 1, [[TMP10]] |
| 30 | +; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[TMP11]], i64 0 |
| 31 | +; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x i64> [[DOTSPLATINSERT]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 32 | +; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[A]], i64 0 |
| 33 | +; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 8 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 34 | +; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[B]], i64 0 |
| 35 | +; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 8 x i64> [[BROADCAST_SPLATINSERT1]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 36 | +; CHECK-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[C]], i64 0 |
| 37 | +; CHECK-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 8 x i64> [[BROADCAST_SPLATINSERT3]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 38 | +; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| 39 | +; CHECK: vector.body: |
| 40 | +; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| 41 | +; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 8 x i64> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] |
| 42 | +; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 0 |
| 43 | +; CHECK-NEXT: [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <vscale x 8 x i64> poison, i64 [[TMP12]], i64 0 |
| 44 | +; CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 8 x i64> [[BROADCAST_SPLATINSERT5]], <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer |
| 45 | +; CHECK-NEXT: [[TMP13:%.*]] = icmp ule <vscale x 8 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] |
| 46 | +; CHECK-NEXT: [[TMP14:%.*]] = icmp ule <vscale x 8 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]] |
| 47 | +; CHECK-NEXT: [[TMP15:%.*]] = select <vscale x 8 x i1> [[TMP13]], <vscale x 8 x i1> [[TMP14]], <vscale x 8 x i1> zeroinitializer |
| 48 | +; CHECK-NEXT: [[TMP16:%.*]] = xor <vscale x 8 x i1> [[TMP13]], shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i64 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer) |
| 49 | +; CHECK-NEXT: [[TMP17:%.*]] = or <vscale x 8 x i1> [[TMP15]], [[TMP16]] |
| 50 | +; CHECK-NEXT: [[TMP18:%.*]] = icmp ule <vscale x 8 x i64> [[VEC_IND]], [[BROADCAST_SPLAT4]] |
| 51 | +; CHECK-NEXT: [[TMP19:%.*]] = select <vscale x 8 x i1> [[TMP17]], <vscale x 8 x i1> [[TMP18]], <vscale x 8 x i1> zeroinitializer |
| 52 | +; CHECK-NEXT: [[TMP20:%.*]] = xor <vscale x 8 x i1> [[TMP14]], shufflevector (<vscale x 8 x i1> insertelement (<vscale x 8 x i1> poison, i1 true, i64 0), <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer) |
| 53 | +; CHECK-NEXT: [[TMP21:%.*]] = select <vscale x 8 x i1> [[TMP13]], <vscale x 8 x i1> [[TMP20]], <vscale x 8 x i1> zeroinitializer |
| 54 | +; CHECK-NEXT: [[TMP22:%.*]] = or <vscale x 8 x i1> [[TMP19]], [[TMP21]] |
| 55 | +; CHECK-NEXT: [[PREDPHI:%.*]] = select <vscale x 8 x i1> [[TMP19]], <vscale x 8 x i64> [[BROADCAST_SPLAT6]], <vscale x 8 x i64> shufflevector (<vscale x 8 x i64> insertelement (<vscale x 8 x i64> poison, i64 poison, i64 0), <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer) |
| 56 | +; CHECK-NEXT: [[TMP23:%.*]] = extractelement <vscale x 8 x i64> [[PREDPHI]], i32 0 |
| 57 | +; CHECK-NEXT: [[TMP24:%.*]] = getelementptr i16, ptr [[ARG]], i64 [[TMP23]] |
| 58 | +; CHECK-NEXT: [[TMP25:%.*]] = getelementptr i16, ptr [[TMP24]], i32 0 |
| 59 | +; CHECK-NEXT: call void @llvm.masked.store.nxv8i16.p0(<vscale x 8 x i16> zeroinitializer, ptr [[TMP25]], i32 2, <vscale x 8 x i1> [[TMP22]]) |
| 60 | +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]] |
| 61 | +; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 8 x i64> [[VEC_IND]], [[DOTSPLAT]] |
| 62 | +; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] |
| 63 | +; CHECK-NEXT: br i1 [[TMP26]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| 64 | +; CHECK: middle.block: |
| 65 | +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 1001, [[N_VEC]] |
| 66 | +; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] |
| 67 | +; CHECK: scalar.ph: |
| 68 | +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| 69 | +; CHECK-NEXT: br label [[LOOP_HEADER:%.*]] |
| 70 | +; CHECK: loop.header: |
| 71 | +; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ] |
| 72 | +; CHECK-NEXT: [[C_1:%.*]] = icmp ule i64 [[IV]], [[A]] |
| 73 | +; CHECK-NEXT: br i1 [[C_1]], label [[THEN_1:%.*]], label [[ELSE_1:%.*]] |
| 74 | +; CHECK: then.1: |
| 75 | +; CHECK-NEXT: [[C_2:%.*]] = icmp ule i64 [[IV]], [[B]] |
| 76 | +; CHECK-NEXT: br i1 [[C_2]], label [[ELSE_1]], label [[MERGE:%.*]] |
| 77 | +; CHECK: else.1: |
| 78 | +; CHECK-NEXT: [[C_3:%.*]] = icmp ule i64 [[IV]], [[C]] |
| 79 | +; CHECK-NEXT: br i1 [[C_3]], label [[THEN_2:%.*]], label [[LOOP_LATCH]] |
| 80 | +; CHECK: then.2: |
| 81 | +; CHECK-NEXT: br label [[MERGE]] |
| 82 | +; CHECK: merge: |
| 83 | +; CHECK-NEXT: [[IDX:%.*]] = phi i64 [ poison, [[THEN_1]] ], [ [[IV]], [[THEN_2]] ] |
| 84 | +; CHECK-NEXT: [[GETELEMENTPTR:%.*]] = getelementptr i16, ptr [[ARG]], i64 [[IDX]] |
| 85 | +; CHECK-NEXT: store i16 0, ptr [[GETELEMENTPTR]], align 2 |
| 86 | +; CHECK-NEXT: br label [[LOOP_LATCH]] |
| 87 | +; CHECK: loop.latch: |
| 88 | +; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1 |
| 89 | +; CHECK-NEXT: [[ICMP:%.*]] = icmp eq i64 [[IV]], 1000 |
| 90 | +; CHECK-NEXT: br i1 [[ICMP]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP3:![0-9]+]] |
| 91 | +; CHECK: exit: |
| 92 | +; CHECK-NEXT: ret void |
| 93 | +; |
| 94 | +entry: |
| 95 | + br label %loop.header |
| 96 | + |
| 97 | +loop.header: |
| 98 | + %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ] |
| 99 | + %c.1 = icmp ule i64 %iv, %a |
| 100 | + br i1 %c.1, label %then.1, label %else.1 |
| 101 | + |
| 102 | +then.1: |
| 103 | + %c.2 = icmp ule i64 %iv, %b |
| 104 | + br i1 %c.2, label %else.1, label %merge |
| 105 | + |
| 106 | +else.1: |
| 107 | + %c.3 = icmp ule i64 %iv, %c |
| 108 | + br i1 %c.3, label %then.2, label %loop.latch |
| 109 | + |
| 110 | +then.2: |
| 111 | + br label %merge |
| 112 | + |
| 113 | +merge: |
| 114 | + %idx = phi i64 [ poison, %then.1 ], [ %iv, %then.2 ] |
| 115 | + %getelementptr = getelementptr i16, ptr %arg, i64 %idx |
| 116 | + store i16 0, ptr %getelementptr, align 2 |
| 117 | + br label %loop.latch |
| 118 | + |
| 119 | +loop.latch: |
| 120 | + %iv.next = add i64 %iv, 1 |
| 121 | + %icmp = icmp eq i64 %iv, 1000 |
| 122 | + br i1 %icmp, label %exit, label %loop.header |
| 123 | + |
| 124 | +exit: |
| 125 | + ret void |
| 126 | +} |
| 127 | +;. |
| 128 | +; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]} |
| 129 | +; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1} |
| 130 | +; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"} |
| 131 | +; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]} |
| 132 | +;. |
0 commit comments