169 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 62; RUN: opt -passes=loop-vectorize -mcpu=skylake-avx512 -S %s | FileCheck %s3 4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:10:11:12:13"5target triple = "x86_64-unknown-linux-gnu"6 7@jlplt_ijl_alloc_array_1d_10294_got = external dso_local local_unnamed_addr global ptr8 9define ptr addrspace(10) @japi1_vect_42283(ptr nocapture readonly %0, i32 %1) local_unnamed_addr #0 {10; CHECK-LABEL: define ptr addrspace(10) @japi1_vect_42283(11; CHECK-SAME: ptr readonly captures(none) [[TMP0:%.*]], i32 [[TMP1:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {12; CHECK-NEXT: [[ITER_CHECK:.*]]:13; CHECK-NEXT: [[TMP2:%.*]] = sext i32 [[TMP1]] to i6414; CHECK-NEXT: [[TMP3:%.*]] = load atomic ptr, ptr @jlplt_ijl_alloc_array_1d_10294_got unordered, align 815; CHECK-NEXT: [[TMP4:%.*]] = tail call ptr addrspace(10) [[TMP3]](ptr addrspace(10) null, i64 0)16; CHECK-NEXT: [[TMP5:%.*]] = load ptr addrspace(10), ptr [[TMP0]], align 8, !tbaa [[JTBAA_VALUE_TBAA0:![0-9]+]]17; CHECK-NEXT: [[TMP6:%.*]] = addrspacecast ptr addrspace(10) [[TMP4]] to ptr addrspace(11)18; CHECK-NEXT: [[TMP7:%.*]] = load ptr addrspace(13), ptr addrspace(11) [[TMP6]], align 8, !tbaa [[JTBAA_ARRAYPTR_TBAA5:![0-9]+]]19; CHECK-NEXT: [[DOTELT:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(10) [[TMP5]], i64 0, i32 020; CHECK-NEXT: [[DOTUNPACK:%.*]] = load ptr addrspace(10), ptr addrspace(10) [[DOTELT]], align 8, !tbaa [[JTBAA_IMMUT_TBAA8:![0-9]+]]21; CHECK-NEXT: [[DOTELT1:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(10) [[TMP5]], i64 0, i32 122; CHECK-NEXT: [[DOTUNPACK2:%.*]] = load i64, ptr addrspace(10) [[DOTELT1]], align 8, !tbaa [[JTBAA_IMMUT_TBAA8]]23; CHECK-NEXT: [[TMP8:%.*]] = add nsw i64 [[TMP2]], 124; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP8]], 425; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]26; CHECK: [[VECTOR_MAIN_LOOP_ITER_CHECK]]:27; CHECK-NEXT: [[TMP17:%.*]] = icmp ult i64 [[TMP8]], 1628; CHECK-NEXT: br i1 [[TMP17]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]29; CHECK: [[VECTOR_PH]]:30; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP8]], 1631; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP8]], [[N_MOD_VF]]32; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x ptr addrspace(10)> poison, ptr addrspace(10) [[DOTUNPACK]], i64 033; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x ptr addrspace(10)> [[BROADCAST_SPLATINSERT]], <4 x ptr addrspace(10)> poison, <4 x i32> zeroinitializer34; CHECK-NEXT: [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <4 x i64> poison, i64 [[DOTUNPACK2]], i64 035; CHECK-NEXT: [[BROADCAST_SPLAT8:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT7]], <4 x i64> poison, <4 x i32> zeroinitializer36; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]37; CHECK: [[VECTOR_BODY]]:38; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]39; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]40; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)41; CHECK-NEXT: [[STEP_ADD4:%.*]] = add <4 x i64> [[STEP_ADD]], splat (i64 4)42; CHECK-NEXT: [[STEP_ADD5:%.*]] = add <4 x i64> [[STEP_ADD4]], splat (i64 4)43; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[VEC_IND]], i32 044; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[STEP_ADD]], i32 045; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[STEP_ADD4]], i32 046; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[STEP_ADD5]], i32 047; CHECK-NEXT: call void @llvm.masked.scatter.v4p10.v4p13(<4 x ptr addrspace(10)> [[BROADCAST_SPLAT]], <4 x ptr addrspace(13)> align 8 [[TMP18]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10:![0-9]+]]48; CHECK-NEXT: call void @llvm.masked.scatter.v4p10.v4p13(<4 x ptr addrspace(10)> [[BROADCAST_SPLAT]], <4 x ptr addrspace(13)> align 8 [[TMP19]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]49; CHECK-NEXT: call void @llvm.masked.scatter.v4p10.v4p13(<4 x ptr addrspace(10)> [[BROADCAST_SPLAT]], <4 x ptr addrspace(13)> align 8 [[TMP20]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]50; CHECK-NEXT: call void @llvm.masked.scatter.v4p10.v4p13(<4 x ptr addrspace(10)> [[BROADCAST_SPLAT]], <4 x ptr addrspace(13)> align 8 [[TMP21]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]51; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[VEC_IND]], i32 152; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[STEP_ADD]], i32 153; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[STEP_ADD4]], i32 154; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[STEP_ADD5]], i32 155; CHECK-NEXT: call void @llvm.masked.scatter.v4i64.v4p13(<4 x i64> [[BROADCAST_SPLAT8]], <4 x ptr addrspace(13)> align 8 [[TMP22]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]56; CHECK-NEXT: call void @llvm.masked.scatter.v4i64.v4p13(<4 x i64> [[BROADCAST_SPLAT8]], <4 x ptr addrspace(13)> align 8 [[TMP23]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]57; CHECK-NEXT: call void @llvm.masked.scatter.v4i64.v4p13(<4 x i64> [[BROADCAST_SPLAT8]], <4 x ptr addrspace(13)> align 8 [[TMP24]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]58; CHECK-NEXT: call void @llvm.masked.scatter.v4i64.v4p13(<4 x i64> [[BROADCAST_SPLAT8]], <4 x ptr addrspace(13)> align 8 [[TMP25]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]59; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 1660; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i64> [[STEP_ADD5]], splat (i64 4)61; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]62; CHECK-NEXT: br i1 [[TMP26]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]63; CHECK: [[MIDDLE_BLOCK]]:64; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP8]], [[N_VEC]]65; CHECK-NEXT: br i1 [[CMP_N]], label %[[L44:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]]66; CHECK: [[VEC_EPILOG_ITER_CHECK]]:67; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 468; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF15:![0-9]+]]69; CHECK: [[VEC_EPILOG_PH]]:70; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]71; CHECK-NEXT: [[N_MOD_VF4:%.*]] = urem i64 [[TMP8]], 472; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[TMP8]], [[N_MOD_VF4]]73; CHECK-NEXT: [[BROADCAST_SPLATINSERT10:%.*]] = insertelement <4 x ptr addrspace(10)> poison, ptr addrspace(10) [[DOTUNPACK]], i64 074; CHECK-NEXT: [[BROADCAST_SPLAT11:%.*]] = shufflevector <4 x ptr addrspace(10)> [[BROADCAST_SPLATINSERT10]], <4 x ptr addrspace(10)> poison, <4 x i32> zeroinitializer75; CHECK-NEXT: [[BROADCAST_SPLATINSERT12:%.*]] = insertelement <4 x i64> poison, i64 [[DOTUNPACK2]], i64 076; CHECK-NEXT: [[BROADCAST_SPLAT13:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT12]], <4 x i64> poison, <4 x i32> zeroinitializer77; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[VEC_EPILOG_RESUME_VAL]], i64 078; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i64> [[DOTSPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer79; CHECK-NEXT: [[INDUCTION:%.*]] = add <4 x i64> [[DOTSPLAT]], <i64 0, i64 1, i64 2, i64 3>80; CHECK-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]81; CHECK: [[VEC_EPILOG_VECTOR_BODY]]:82; CHECK-NEXT: [[INDEX7:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT14:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]83; CHECK-NEXT: [[VEC_IND8:%.*]] = phi <4 x i64> [ [[INDUCTION]], %[[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT9:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]84; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[VEC_IND8]], i32 085; CHECK-NEXT: call void @llvm.masked.scatter.v4p10.v4p13(<4 x ptr addrspace(10)> [[BROADCAST_SPLAT11]], <4 x ptr addrspace(13)> align 8 [[TMP28]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]86; CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], <4 x i64> [[VEC_IND8]], i32 187; CHECK-NEXT: call void @llvm.masked.scatter.v4i64.v4p13(<4 x i64> [[BROADCAST_SPLAT13]], <4 x ptr addrspace(13)> align 8 [[TMP29]], <4 x i1> splat (i1 true)), !tbaa [[JTBAA_ARRAYBUF_TBAA10]]88; CHECK-NEXT: [[INDEX_NEXT14]] = add nuw i64 [[INDEX7]], 489; CHECK-NEXT: [[VEC_IND_NEXT9]] = add <4 x i64> [[VEC_IND8]], splat (i64 4)90; CHECK-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT14]], [[N_VEC5]]91; CHECK-NEXT: br i1 [[TMP30]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]92; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]:93; CHECK-NEXT: [[CMP_N15:%.*]] = icmp eq i64 [[TMP8]], [[N_VEC5]]94; CHECK-NEXT: br i1 [[CMP_N15]], label %[[L44]], label %[[VEC_EPILOG_SCALAR_PH]]95; CHECK: [[VEC_EPILOG_SCALAR_PH]]:96; CHECK-NEXT: [[BC_RESUME_VAL17:%.*]] = phi i64 [ [[N_VEC5]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[ITER_CHECK]] ]97; CHECK-NEXT: br label %[[L26:.*]]98; CHECK: [[L26]]:99; CHECK-NEXT: [[VALUE_PHI5:%.*]] = phi i64 [ [[BC_RESUME_VAL17]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[TMP27:%.*]], %[[L26]] ]100; CHECK-NEXT: [[DOTREPACK:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], i64 [[VALUE_PHI5]], i32 0101; CHECK-NEXT: store ptr addrspace(10) [[DOTUNPACK]], ptr addrspace(13) [[DOTREPACK]], align 8, !tbaa [[JTBAA_ARRAYBUF_TBAA10]]102; CHECK-NEXT: [[DOTREPACK4:%.*]] = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) [[TMP7]], i64 [[VALUE_PHI5]], i32 1103; CHECK-NEXT: store i64 [[DOTUNPACK2]], ptr addrspace(13) [[DOTREPACK4]], align 8, !tbaa [[JTBAA_ARRAYBUF_TBAA10]]104; CHECK-NEXT: [[TMP27]] = add i64 [[VALUE_PHI5]], 1105; CHECK-NEXT: [[DOTNOT:%.*]] = icmp eq i64 [[VALUE_PHI5]], [[TMP2]]106; CHECK-NEXT: br i1 [[DOTNOT]], label %[[L44]], label %[[L26]], !llvm.loop [[LOOP17:![0-9]+]]107; CHECK: [[L44]]:108; CHECK-NEXT: ret ptr addrspace(10) null109;110top:111 %2 = sext i32 %1 to i64112 %3 = load atomic ptr addrspace(10) (ptr addrspace(10), i64)*, ptr addrspace(10) (ptr addrspace(10), i64)** bitcast (ptr @jlplt_ijl_alloc_array_1d_10294_got to ptr addrspace(10) (ptr addrspace(10), i64)**) unordered, align 8113 %4 = tail call ptr addrspace(10) %3(ptr addrspace(10) null, i64 0)114 %5 = load ptr addrspace(10), ptr %0, align 8, !tbaa !0115 %6 = addrspacecast ptr addrspace(10) %4 to ptr addrspace(11)116 %7 = load ptr addrspace(13), ptr addrspace(11) %6, align 8, !tbaa !5117 %.elt = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(10) %5, i64 0, i32 0118 %.unpack = load ptr addrspace(10), ptr addrspace(10) %.elt, align 8, !tbaa !8119 %.elt1 = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(10) %5, i64 0, i32 1120 %.unpack2 = load i64, ptr addrspace(10) %.elt1, align 8, !tbaa !8121 br label %L26122 123L26: ; preds = %L26, %top124 %value_phi5 = phi i64 [ 0, %top ], [ %8, %L26 ]125 %.repack = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) %7, i64 %value_phi5, i32 0126 store ptr addrspace(10) %.unpack, ptr addrspace(13) %.repack, align 8, !tbaa !10127 %.repack4 = getelementptr inbounds { ptr addrspace(10), i64 }, ptr addrspace(13) %7, i64 %value_phi5, i32 1128 store i64 %.unpack2, ptr addrspace(13) %.repack4, align 8, !tbaa !10129 %8 = add i64 %value_phi5, 1130 %.not = icmp eq i64 %value_phi5, %2131 br i1 %.not, label %L44, label %L26132 133L44: ; preds = %L26134 ret ptr addrspace(10) null135}136 137!0 = !{!1, !1, i64 0}138!1 = !{!"jtbaa_value", !2, i64 0}139!2 = !{!"jtbaa_data", !3, i64 0}140!3 = !{!"jtbaa", !4, i64 0}141!4 = !{!"jtbaa"}142!5 = !{!6, !6, i64 0}143!6 = !{!"jtbaa_arrayptr", !7, i64 0}144!7 = !{!"jtbaa_array", !3, i64 0}145!8 = !{!9, !9, i64 0}146!9 = !{!"jtbaa_immut", !1, i64 0}147!10 = !{!11, !11, i64 0}148!11 = !{!"jtbaa_arraybuf", !2, i64 0}149;.150; CHECK: [[JTBAA_VALUE_TBAA0]] = !{[[META1:![0-9]+]], [[META1]], i64 0}151; CHECK: [[META1]] = !{!"jtbaa_value", [[META2:![0-9]+]], i64 0}152; CHECK: [[META2]] = !{!"jtbaa_data", [[META3:![0-9]+]], i64 0}153; CHECK: [[META3]] = !{!"jtbaa", [[META4:![0-9]+]], i64 0}154; CHECK: [[META4]] = !{!"jtbaa"}155; CHECK: [[JTBAA_ARRAYPTR_TBAA5]] = !{[[META6:![0-9]+]], [[META6]], i64 0}156; CHECK: [[META6]] = !{!"jtbaa_arrayptr", [[META7:![0-9]+]], i64 0}157; CHECK: [[META7]] = !{!"jtbaa_array", [[META3]], i64 0}158; CHECK: [[JTBAA_IMMUT_TBAA8]] = !{[[META9:![0-9]+]], [[META9]], i64 0}159; CHECK: [[META9]] = !{!"jtbaa_immut", [[META1]], i64 0}160; CHECK: [[JTBAA_ARRAYBUF_TBAA10]] = !{[[META11:![0-9]+]], [[META11]], i64 0}161; CHECK: [[META11]] = !{!"jtbaa_arraybuf", [[META2]], i64 0}162; CHECK: [[LOOP12]] = distinct !{[[LOOP12]], [[META13:![0-9]+]], [[META14:![0-9]+]]}163; CHECK: [[META13]] = !{!"llvm.loop.isvectorized", i32 1}164; CHECK: [[META14]] = !{!"llvm.loop.unroll.runtime.disable"}165; CHECK: [[PROF15]] = !{!"branch_weights", i32 4, i32 12}166; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META13]], [[META14]]}167; CHECK: [[LOOP17]] = distinct !{[[LOOP17]], [[META14]], [[META13]]}168;.169