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 -enable-epilogue-vectorization -epilogue-vectorization-force-VF=4 -S %s | FileCheck %s 3 4target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128" 5 6define void @trunc_iv_steps_with_epilogue(ptr %A, i64 %N) { 7; CHECK-LABEL: define void @trunc_iv_steps_with_epilogue( 8; CHECK-SAME: ptr [[A:%.*]], i64 [[N:%.*]]) { 9; CHECK-NEXT: iter.check: 10; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], 1 11; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4 12; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_SCEVCHECK:%.*]] 13; CHECK: vector.scevcheck: 14; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[N]] to i32 15; CHECK-NEXT: [[TMP2:%.*]] = icmp slt i32 [[TMP1]], 0 16; CHECK-NEXT: [[TMP3:%.*]] = icmp ugt i64 [[N]], 4294967295 17; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP2]], [[TMP3]] 18; CHECK-NEXT: br i1 [[TMP4]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 19; CHECK: vector.main.loop.iter.check: 20; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[TMP0]], 4 21; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 22; CHECK: vector.ph: 23; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4 24; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] 25; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 26; CHECK: vector.body: 27; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 28; CHECK-NEXT: [[TMP5:%.*]] = trunc i64 [[INDEX]] to i32 29; CHECK-NEXT: [[TMP6:%.*]] = add i32 [[TMP5]], 0 30; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i32 [[TMP6]] 31; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[TMP7]], i32 0 32; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP8]], align 1 33; CHECK-NEXT: [[TMP9:%.*]] = add <4 x i8> [[WIDE_LOAD]], splat (i8 2) 34; CHECK-NEXT: store <4 x i8> [[TMP9]], ptr [[TMP8]], align 1 35; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 36; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 37; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 38; CHECK: middle.block: 39; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] 40; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 41; CHECK: vec.epilog.iter.check: 42; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[TMP0]], [[N_VEC]] 43; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4 44; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 45; CHECK: vec.epilog.ph: 46; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 47; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[TMP0]], 4 48; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF2]] 49; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 50; CHECK: vec.epilog.vector.body: 51; CHECK-NEXT: [[INDEX5:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT7:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 52; CHECK-NEXT: [[TMP11:%.*]] = trunc i64 [[INDEX5]] to i32 53; CHECK-NEXT: [[TMP12:%.*]] = add i32 [[TMP11]], 0 54; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[A]], i32 [[TMP12]] 55; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP13]], i32 0 56; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <4 x i8>, ptr [[TMP14]], align 1 57; CHECK-NEXT: [[TMP15:%.*]] = add <4 x i8> [[WIDE_LOAD6]], splat (i8 2) 58; CHECK-NEXT: store <4 x i8> [[TMP15]], ptr [[TMP14]], align 1 59; CHECK-NEXT: [[INDEX_NEXT7]] = add nuw i64 [[INDEX5]], 4 60; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT7]], [[N_VEC3]] 61; CHECK-NEXT: br i1 [[TMP16]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 62; CHECK: vec.epilog.middle.block: 63; CHECK-NEXT: [[CMP_N4:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC3]] 64; CHECK-NEXT: br i1 [[CMP_N4]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]] 65; CHECK: vec.epilog.scalar.ph: 66; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 0, [[VECTOR_SCEVCHECK]] ], [ 0, [[ITER_CHECK:%.*]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ] 67; CHECK-NEXT: br label [[LOOP:%.*]] 68; CHECK: loop: 69; CHECK-NEXT: [[IV_I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ] 70; CHECK-NEXT: [[T_IV:%.*]] = trunc i64 [[IV_I]] to i32 71; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, ptr [[A]], i32 [[T_IV]] 72; CHECK-NEXT: [[L:%.*]] = load i8, ptr [[GEP]], align 1 73; CHECK-NEXT: [[ADD:%.*]] = add i8 [[L]], 2 74; CHECK-NEXT: store i8 [[ADD]], ptr [[GEP]], align 1 75; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV_I]], 1 76; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_I]], [[N]] 77; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP4:![0-9]+]] 78; CHECK: exit: 79; CHECK-NEXT: ret void 80; 81entry: 82 br label %loop 83 84loop: 85 %iv.i = phi i64 [ 0, %entry ], [ %iv.next, %loop ] 86 %t.iv = trunc i64 %iv.i to i32 87 %gep = getelementptr i8, ptr %A, i32 %t.iv 88 %l = load i8, ptr %gep, align 1 89 %add = add i8 %l, 2 90 store i8 %add, ptr %gep 91 %iv.next = add i64 %iv.i, 1 92 %ec = icmp eq i64 %iv.i, %N 93 br i1 %ec, label %exit, label %loop 94 95exit: 96 ret void 97} 98;. 99; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]} 100; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1} 101; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"} 102; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]], [[META2]]} 103; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]]} 104;. 105