74 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 62; RUN: opt -p loop-vectorize -force-vector-width=4 -enable-epilogue-vectorization -epilogue-vectorization-force-VF=4 -S %s | FileCheck %s3 4@end = external global [128 x i8]5 6; Test case for https://github.com/llvm/llvm-project/issues/162128.7define void @test_epilogue_step_scev_expansion(ptr %dst) {8; CHECK-LABEL: define void @test_epilogue_step_scev_expansion(9; CHECK-SAME: ptr [[DST:%.*]]) {10; CHECK-NEXT: [[ITER_CHECK:.*]]:11; CHECK-NEXT: br i1 false, label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]12; CHECK: [[VECTOR_MAIN_LOOP_ITER_CHECK]]:13; CHECK-NEXT: br i1 false, label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]14; CHECK: [[VECTOR_PH]]:15; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 sub (i64 0, i64 ptrtoint (ptr @end to i64)), 416; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 sub (i64 0, i64 ptrtoint (ptr @end to i64)), [[N_MOD_VF]]17; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]18; CHECK: [[VECTOR_BODY]]:19; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]20; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[DST]], i64 [[INDEX]]21; CHECK-NEXT: store <4 x i8> zeroinitializer, ptr [[TMP0]], align 122; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 423; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]24; CHECK-NEXT: br i1 [[TMP1]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]25; CHECK: [[MIDDLE_BLOCK]]:26; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 sub (i64 0, i64 ptrtoint (ptr @end to i64)), [[N_VEC]]27; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]]28; CHECK: [[VEC_EPILOG_ITER_CHECK]]:29; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 sub (i64 0, i64 ptrtoint (ptr @end to i64)), [[N_VEC]]30; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 431; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]32; CHECK: [[VEC_EPILOG_PH]]:33; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]34; CHECK-NEXT: [[N_MOD_VF1:%.*]] = urem i64 sub (i64 0, i64 ptrtoint (ptr @end to i64)), 435; CHECK-NEXT: [[N_VEC2:%.*]] = sub i64 sub (i64 0, i64 ptrtoint (ptr @end to i64)), [[N_MOD_VF1]]36; CHECK-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]37; CHECK: [[VEC_EPILOG_VECTOR_BODY]]:38; CHECK-NEXT: [[INDEX3:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT4:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]39; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[DST]], i64 [[INDEX3]]40; CHECK-NEXT: store <4 x i8> zeroinitializer, ptr [[TMP2]], align 141; CHECK-NEXT: [[INDEX_NEXT4]] = add nuw i64 [[INDEX3]], 442; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT4]], [[N_VEC2]]43; CHECK-NEXT: br i1 [[TMP3]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]44; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]:45; CHECK-NEXT: [[CMP_N5:%.*]] = icmp eq i64 sub (i64 0, i64 ptrtoint (ptr @end to i64)), [[N_VEC2]]46; CHECK-NEXT: br i1 [[CMP_N5]], label %[[EXIT]], label %[[VEC_EPILOG_SCALAR_PH]]47; CHECK: [[VEC_EPILOG_SCALAR_PH]]:48; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC2]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[ITER_CHECK]] ]49; CHECK-NEXT: br label %[[LOOP:.*]]50; CHECK: [[LOOP]]:51; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]52; CHECK-NEXT: [[GEP_DST:%.*]] = getelementptr i8, ptr [[DST]], i64 [[IV]]53; CHECK-NEXT: store i8 0, ptr [[GEP_DST]], align 154; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 155; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], sub (i64 0, i64 ptrtoint (ptr getelementptr inbounds nuw (i8, ptr @end, i64 1) to i64))56; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]57; CHECK: [[EXIT]]:58; CHECK-NEXT: ret void59;60entry:61 br label %loop62 63loop:64 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]65 %gep.dst = getelementptr i8, ptr %dst, i64 %iv66 store i8 0, ptr %gep.dst, align 167 %iv.next = add i64 %iv, 168 %ec = icmp eq i64 %iv, sub (i64 0, i64 ptrtoint (ptr getelementptr inbounds nuw (i8, ptr @end, i64 1) to i64))69 br i1 %ec, label %exit, label %loop70 71exit:72 ret void73}74