76 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 62; RUN: opt -passes=loop-vectorize -S %s | FileCheck %s3 4; This tests exercises the out-of-band fixing-up of scalar resume values.5 6target triple = "aarch64"7 8define void @epilogue_vectorization_fix_scalar_resume_values(ptr %dst, i64 %n) {9; CHECK-LABEL: define void @epilogue_vectorization_fix_scalar_resume_values(10; CHECK-SAME: ptr [[DST:%.*]], i64 [[N:%.*]]) {11; CHECK-NEXT: [[ITER_CHECK:.*]]:12; CHECK-NEXT: [[REM:%.*]] = urem i64 [[N]], 313; CHECK-NEXT: br i1 true, label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]14; CHECK: [[VECTOR_MAIN_LOOP_ITER_CHECK]]:15; CHECK-NEXT: br i1 true, label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]16; CHECK: [[VECTOR_PH]]:17; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[REM]], 3218; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[REM]], [[N_MOD_VF]]19; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]20; CHECK: [[VECTOR_BODY]]:21; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]22; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[DST]], i64 [[INDEX]]23; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP0]], i64 1624; CHECK-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP0]], align 125; CHECK-NEXT: store <16 x i8> zeroinitializer, ptr [[TMP1]], align 126; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 3227; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]28; CHECK-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]29; CHECK: [[MIDDLE_BLOCK]]:30; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[REM]], [[N_VEC]]31; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]]32; CHECK: [[VEC_EPILOG_ITER_CHECK]]:33; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 834; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]35; CHECK: [[VEC_EPILOG_PH]]:36; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]37; CHECK-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]38; CHECK: [[VEC_EPILOG_VECTOR_BODY]]:39; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]40; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[DST]], i64 [[INDEX1]]41; CHECK-NEXT: store <8 x i8> zeroinitializer, ptr [[TMP3]], align 142; CHECK-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 843; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 044; CHECK-NEXT: br i1 [[TMP4]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]45; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]:46; CHECK-NEXT: [[CMP_N3:%.*]] = icmp eq i64 [[REM]], 047; CHECK-NEXT: br i1 [[CMP_N3]], label %[[EXIT]], label %[[VEC_EPILOG_SCALAR_PH]]48; CHECK: [[VEC_EPILOG_SCALAR_PH]]:49; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 0, %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[ITER_CHECK]] ]50; CHECK-NEXT: br label %[[LOOP:.*]]51; CHECK: [[LOOP]]:52; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]53; CHECK-NEXT: [[GEP_DST_IV:%.*]] = getelementptr i8, ptr [[DST]], i64 [[IV]]54; CHECK-NEXT: store i8 0, ptr [[GEP_DST_IV]], align 155; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 156; CHECK-NEXT: [[EXIT_COND:%.*]] = icmp eq i64 [[IV_NEXT]], [[REM]]57; CHECK-NEXT: br i1 [[EXIT_COND]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]58; CHECK: [[EXIT]]:59; CHECK-NEXT: ret void60;61entry:62 %rem = urem i64 %n, 363 br label %loop64 65loop:66 %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]67 %gep.dst.iv = getelementptr i8, ptr %dst, i64 %iv68 store i8 0, ptr %gep.dst.iv, align 169 %iv.next = add i64 %iv, 170 %exit.cond = icmp eq i64 %iv.next, %rem71 br i1 %exit.cond, label %exit, label %loop72 73exit:74 ret void75}76