|
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4 |
| 2 | +; RUN: opt -p loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S %s | FileCheck %s |
| 3 | + |
| 4 | +target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128-ni:1-p2:32:8:8:32-ni:2" |
| 5 | + |
| 6 | +; Test with blend recipe in header VPBB, from |
| 7 | +; https://github.com/llvm/llvm-project/issues/88297. |
| 8 | +define i64 @pr88297() { |
| 9 | +; CHECK-LABEL: define i64 @pr88297() { |
| 10 | +; CHECK-NEXT: entry: |
| 11 | +; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| 12 | +; CHECK: vector.ph: |
| 13 | +; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| 14 | +; CHECK: vector.body: |
| 15 | +; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| 16 | +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| 17 | +; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000 |
| 18 | +; CHECK-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] |
| 19 | +; CHECK: middle.block: |
| 20 | +; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| 21 | +; CHECK: scalar.ph: |
| 22 | +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| 23 | +; CHECK-NEXT: br label [[LOOP_HEADER:%.*]] |
| 24 | +; CHECK: loop.header: |
| 25 | +; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ] |
| 26 | +; CHECK-NEXT: br i1 false, label [[LOOP_LATCH]], label [[THEN:%.*]] |
| 27 | +; CHECK: then: |
| 28 | +; CHECK-NEXT: br label [[LOOP_LATCH]] |
| 29 | +; CHECK: loop.latch: |
| 30 | +; CHECK-NEXT: [[R:%.*]] = phi i64 [ 1, [[THEN]] ], [ 0, [[LOOP_HEADER]] ] |
| 31 | +; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1 |
| 32 | +; CHECK-NEXT: [[ICMP:%.*]] = icmp sgt i32 [[IV]], 1000 |
| 33 | +; CHECK-NEXT: br i1 [[ICMP]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP3:![0-9]+]] |
| 34 | +; CHECK: exit: |
| 35 | +; CHECK-NEXT: [[R_LCSSA:%.*]] = phi i64 [ [[R]], [[LOOP_LATCH]] ], [ 1, [[MIDDLE_BLOCK]] ] |
| 36 | +; CHECK-NEXT: ret i64 [[R_LCSSA]] |
| 37 | +; |
| 38 | +entry: |
| 39 | + br label %loop.header |
| 40 | + |
| 41 | +loop.header: |
| 42 | + %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ] |
| 43 | + br i1 false, label %loop.latch, label %then |
| 44 | + |
| 45 | +then: |
| 46 | + br label %loop.latch |
| 47 | + |
| 48 | +loop.latch: |
| 49 | + %r = phi i64 [ 1, %then ], [ 0, %loop.header ] |
| 50 | + %iv.next = add i32 %iv, 1 |
| 51 | + %icmp = icmp sgt i32 %iv, 1000 |
| 52 | + br i1 %icmp, label %exit, label %loop.header |
| 53 | + |
| 54 | +exit: |
| 55 | + %r.lcssa = phi i64 [ %r, %loop.latch ] |
| 56 | + ret i64 %r.lcssa |
| 57 | +} |
| 58 | + |
| 59 | +define i64 @pr88297_incoming_ops_reordered() { |
| 60 | +; CHECK-LABEL: define i64 @pr88297_incoming_ops_reordered() { |
| 61 | +; CHECK-NEXT: entry: |
| 62 | +; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| 63 | +; CHECK: vector.ph: |
| 64 | +; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| 65 | +; CHECK: vector.body: |
| 66 | +; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| 67 | +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| 68 | +; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000 |
| 69 | +; CHECK-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] |
| 70 | +; CHECK: middle.block: |
| 71 | +; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| 72 | +; CHECK: scalar.ph: |
| 73 | +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| 74 | +; CHECK-NEXT: br label [[LOOP_HEADER:%.*]] |
| 75 | +; CHECK: loop.header: |
| 76 | +; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ] |
| 77 | +; CHECK-NEXT: br i1 false, label [[LOOP_LATCH]], label [[THEN:%.*]] |
| 78 | +; CHECK: then: |
| 79 | +; CHECK-NEXT: br label [[LOOP_LATCH]] |
| 80 | +; CHECK: loop.latch: |
| 81 | +; CHECK-NEXT: [[R:%.*]] = phi i64 [ 0, [[LOOP_HEADER]] ], [ 1, [[THEN]] ] |
| 82 | +; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1 |
| 83 | +; CHECK-NEXT: [[ICMP:%.*]] = icmp sgt i32 [[IV]], 1000 |
| 84 | +; CHECK-NEXT: br i1 [[ICMP]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP5:![0-9]+]] |
| 85 | +; CHECK: exit: |
| 86 | +; CHECK-NEXT: [[R_LCSSA:%.*]] = phi i64 [ [[R]], [[LOOP_LATCH]] ], [ 1, [[MIDDLE_BLOCK]] ] |
| 87 | +; CHECK-NEXT: ret i64 [[R_LCSSA]] |
| 88 | +; |
| 89 | +entry: |
| 90 | + br label %loop.header |
| 91 | + |
| 92 | +loop.header: |
| 93 | + %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ] |
| 94 | + br i1 false, label %loop.latch, label %then |
| 95 | + |
| 96 | +then: |
| 97 | + br label %loop.latch |
| 98 | + |
| 99 | +loop.latch: |
| 100 | + %r = phi i64 [ 0, %loop.header ], [ 1, %then ] |
| 101 | + %iv.next = add i32 %iv, 1 |
| 102 | + %icmp = icmp sgt i32 %iv, 1000 |
| 103 | + br i1 %icmp, label %exit, label %loop.header |
| 104 | + |
| 105 | +exit: |
| 106 | + %r.lcssa = phi i64 [ %r, %loop.latch ] |
| 107 | + ret i64 %r.lcssa |
| 108 | +} |
| 109 | + |
| 110 | +define i64 @invar_cond(i1 %c) { |
| 111 | +; CHECK-LABEL: define i64 @invar_cond( |
| 112 | +; CHECK-SAME: i1 [[C:%.*]]) { |
| 113 | +; CHECK-NEXT: entry: |
| 114 | +; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| 115 | +; CHECK: vector.ph: |
| 116 | +; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C]], i64 0 |
| 117 | +; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer |
| 118 | +; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| 119 | +; CHECK: vector.body: |
| 120 | +; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| 121 | +; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[BROADCAST_SPLAT]], <4 x i64> zeroinitializer, <4 x i64> <i64 1, i64 1, i64 1, i64 1> |
| 122 | +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| 123 | +; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000 |
| 124 | +; CHECK-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] |
| 125 | +; CHECK: middle.block: |
| 126 | +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <4 x i64> [[PREDPHI]], i32 3 |
| 127 | +; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| 128 | +; CHECK: scalar.ph: |
| 129 | +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| 130 | +; CHECK-NEXT: br label [[LOOP_HEADER:%.*]] |
| 131 | +; CHECK: loop.header: |
| 132 | +; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ] |
| 133 | +; CHECK-NEXT: br i1 [[C]], label [[LOOP_LATCH]], label [[THEN:%.*]] |
| 134 | +; CHECK: then: |
| 135 | +; CHECK-NEXT: br label [[LOOP_LATCH]] |
| 136 | +; CHECK: loop.latch: |
| 137 | +; CHECK-NEXT: [[R:%.*]] = phi i64 [ 1, [[THEN]] ], [ 0, [[LOOP_HEADER]] ] |
| 138 | +; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1 |
| 139 | +; CHECK-NEXT: [[ICMP:%.*]] = icmp sgt i32 [[IV]], 1000 |
| 140 | +; CHECK-NEXT: br i1 [[ICMP]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP7:![0-9]+]] |
| 141 | +; CHECK: exit: |
| 142 | +; CHECK-NEXT: [[R_LCSSA:%.*]] = phi i64 [ [[R]], [[LOOP_LATCH]] ], [ [[TMP1]], [[MIDDLE_BLOCK]] ] |
| 143 | +; CHECK-NEXT: ret i64 [[R_LCSSA]] |
| 144 | +; |
| 145 | +entry: |
| 146 | + br label %loop.header |
| 147 | + |
| 148 | +loop.header: |
| 149 | + %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ] |
| 150 | + br i1 %c, label %loop.latch, label %then |
| 151 | + |
| 152 | +then: |
| 153 | + br label %loop.latch |
| 154 | + |
| 155 | +loop.latch: |
| 156 | + %r = phi i64 [ 1, %then ], [ 0, %loop.header ] |
| 157 | + %iv.next = add i32 %iv, 1 |
| 158 | + %icmp = icmp sgt i32 %iv, 1000 |
| 159 | + br i1 %icmp, label %exit, label %loop.header |
| 160 | + |
| 161 | +exit: |
| 162 | + %r.lcssa = phi i64 [ %r, %loop.latch ] |
| 163 | + ret i64 %r.lcssa |
| 164 | +} |
| 165 | + |
| 166 | +define i64 @invar_cond_incoming_ops_reordered(i1 %c) { |
| 167 | +; CHECK-LABEL: define i64 @invar_cond_incoming_ops_reordered( |
| 168 | +; CHECK-SAME: i1 [[C:%.*]]) { |
| 169 | +; CHECK-NEXT: entry: |
| 170 | +; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] |
| 171 | +; CHECK: vector.ph: |
| 172 | +; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C]], i64 0 |
| 173 | +; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer |
| 174 | +; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] |
| 175 | +; CHECK: vector.body: |
| 176 | +; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] |
| 177 | +; CHECK-NEXT: [[TMP1:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true, i1 true, i1 true> |
| 178 | +; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP1]], <4 x i64> <i64 1, i64 1, i64 1, i64 1>, <4 x i64> zeroinitializer |
| 179 | +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 |
| 180 | +; CHECK-NEXT: [[TMP0:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000 |
| 181 | +; CHECK-NEXT: br i1 [[TMP0]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] |
| 182 | +; CHECK: middle.block: |
| 183 | +; CHECK-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[PREDPHI]], i32 3 |
| 184 | +; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]] |
| 185 | +; CHECK: scalar.ph: |
| 186 | +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] |
| 187 | +; CHECK-NEXT: br label [[LOOP_HEADER:%.*]] |
| 188 | +; CHECK: loop.header: |
| 189 | +; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ] |
| 190 | +; CHECK-NEXT: br i1 [[C]], label [[LOOP_LATCH]], label [[THEN:%.*]] |
| 191 | +; CHECK: then: |
| 192 | +; CHECK-NEXT: br label [[LOOP_LATCH]] |
| 193 | +; CHECK: loop.latch: |
| 194 | +; CHECK-NEXT: [[R:%.*]] = phi i64 [ 0, [[LOOP_HEADER]] ], [ 1, [[THEN]] ] |
| 195 | +; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1 |
| 196 | +; CHECK-NEXT: [[ICMP:%.*]] = icmp sgt i32 [[IV]], 1000 |
| 197 | +; CHECK-NEXT: br i1 [[ICMP]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP9:![0-9]+]] |
| 198 | +; CHECK: exit: |
| 199 | +; CHECK-NEXT: [[R_LCSSA:%.*]] = phi i64 [ [[R]], [[LOOP_LATCH]] ], [ [[TMP2]], [[MIDDLE_BLOCK]] ] |
| 200 | +; CHECK-NEXT: ret i64 [[R_LCSSA]] |
| 201 | +; |
| 202 | +entry: |
| 203 | + br label %loop.header |
| 204 | + |
| 205 | +loop.header: |
| 206 | + %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ] |
| 207 | + br i1 %c, label %loop.latch, label %then |
| 208 | + |
| 209 | +then: |
| 210 | + br label %loop.latch |
| 211 | + |
| 212 | +loop.latch: |
| 213 | + %r = phi i64 [ 0, %loop.header ], [ 1, %then ] |
| 214 | + %iv.next = add i32 %iv, 1 |
| 215 | + %icmp = icmp sgt i32 %iv, 1000 |
| 216 | + br i1 %icmp, label %exit, label %loop.header |
| 217 | + |
| 218 | +exit: |
| 219 | + %r.lcssa = phi i64 [ %r, %loop.latch ] |
| 220 | + ret i64 %r.lcssa |
| 221 | +} |
| 222 | +;. |
| 223 | +; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]} |
| 224 | +; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1} |
| 225 | +; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"} |
| 226 | +; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]} |
| 227 | +; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]} |
| 228 | +; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]} |
| 229 | +; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]} |
| 230 | +; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META2]], [[META1]]} |
| 231 | +; CHECK: [[LOOP8]] = distinct !{[[LOOP8]], [[META1]], [[META2]]} |
| 232 | +; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META2]], [[META1]]} |
| 233 | +;. |
0 commit comments