brintos

brintos / llvm-project-archived public Read only

0
0
Text · 9.7 KiB · 4445141 Raw
154 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter "br " --filter "^.*:" --filter "icmp" --version 52; RUN: opt < %s -S -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4  -enable-epilogue-vectorization \3; RUN:   -epilogue-vectorization-force-VF=4 | FileCheck %s --check-prefix=MAINVF4IC1_EPI44; RUN: opt < %s -S -passes=loop-vectorize -force-vector-interleave=2 -force-vector-width=4  -enable-epilogue-vectorization \5; RUN:   -epilogue-vectorization-force-VF=4 | FileCheck %s --check-prefix=MAINVF4IC2_EPI46 7define void @f0(i8 %n, i32 %len, ptr %p) !prof !0 {8; MAINVF4IC1_EPI4-LABEL: define void @f0(9; MAINVF4IC1_EPI4-SAME: i8 [[N:%.*]], i32 [[LEN:%.*]], ptr [[P:%.*]]) !prof [[PROF0:![0-9]+]] {10; MAINVF4IC1_EPI4:  [[ENTRY:.*:]]11; MAINVF4IC1_EPI4:    [[CMP_ENTRY:%.*]] = icmp sgt i32 [[LEN]], 012; MAINVF4IC1_EPI4:    br i1 [[CMP_ENTRY]], label %[[ITER_CHECK:.*]], label %[[EXIT:.*]], !prof [[PROF1:![0-9]+]]13; MAINVF4IC1_EPI4:  [[ITER_CHECK]]:14; MAINVF4IC1_EPI4:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0:%.*]], 415; MAINVF4IC1_EPI4:    br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]], !prof [[PROF2:![0-9]+]]16; MAINVF4IC1_EPI4:  [[VECTOR_SCEVCHECK]]:17; MAINVF4IC1_EPI4:    [[TMP2:%.*]] = icmp slt i8 [[TMP1:%.*]], 018; MAINVF4IC1_EPI4:    [[TMP3:%.*]] = icmp ugt i32 [[LEN]], 25519; MAINVF4IC1_EPI4:    br i1 [[TMP4:%.*]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]], !prof [[PROF2]]20; MAINVF4IC1_EPI4:  [[VECTOR_MAIN_LOOP_ITER_CHECK]]:21; MAINVF4IC1_EPI4:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i32 [[TMP0]], 422; MAINVF4IC1_EPI4:    br i1 [[MIN_ITERS_CHECK1]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]], !prof [[PROF2]]23; MAINVF4IC1_EPI4:  [[VECTOR_PH]]:24; MAINVF4IC1_EPI4:    br label %[[VECTOR_BODY:.*]]25; MAINVF4IC1_EPI4:  [[VECTOR_BODY]]:26; MAINVF4IC1_EPI4:    [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT:%.*]], [[N_VEC:%.*]]27; MAINVF4IC1_EPI4:    br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !prof [[PROF3:![0-9]+]], !llvm.loop [[LOOP4:![0-9]+]]28; MAINVF4IC1_EPI4:  [[MIDDLE_BLOCK]]:29; MAINVF4IC1_EPI4:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]30; MAINVF4IC1_EPI4:    br i1 [[CMP_N]], label %[[EXIT_LOOPEXIT:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]], !prof [[PROF8:![0-9]+]]31; MAINVF4IC1_EPI4:  [[VEC_EPILOG_ITER_CHECK]]:32; MAINVF4IC1_EPI4:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i32 [[N_VEC_REMAINING:%.*]], 433; MAINVF4IC1_EPI4:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF9:![0-9]+]]34; MAINVF4IC1_EPI4:  [[VEC_EPILOG_PH]]:35; MAINVF4IC1_EPI4:    br label %[[VEC_EPILOG_VECTOR_BODY:.*]]36; MAINVF4IC1_EPI4:  [[VEC_EPILOG_VECTOR_BODY]]:37; MAINVF4IC1_EPI4:    [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT6:%.*]], [[N_VEC3:%.*]]38; MAINVF4IC1_EPI4:    br i1 [[TMP12]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !prof [[PROF10:![0-9]+]], !llvm.loop [[LOOP11:![0-9]+]]39; MAINVF4IC1_EPI4:  [[VEC_EPILOG_MIDDLE_BLOCK]]:40; MAINVF4IC1_EPI4:    [[CMP_N8:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC3]]41; MAINVF4IC1_EPI4:    br i1 [[CMP_N8]], label %[[EXIT_LOOPEXIT]], label %[[VEC_EPILOG_SCALAR_PH]], !prof [[PROF8]]42; MAINVF4IC1_EPI4:  [[VEC_EPILOG_SCALAR_PH]]:43; MAINVF4IC1_EPI4:    br label %[[LOOP:.*]]44; MAINVF4IC1_EPI4:  [[LOOP]]:45; MAINVF4IC1_EPI4:    [[CMP_LOOP:%.*]] = icmp ult i32 [[I32:%.*]], [[LEN]]46; MAINVF4IC1_EPI4:    br i1 [[CMP_LOOP]], label %[[LOOP]], label %[[EXIT_LOOPEXIT]], !prof [[PROF13:![0-9]+]], !llvm.loop [[LOOP14:![0-9]+]]47; MAINVF4IC1_EPI4:  [[EXIT_LOOPEXIT]]:48; MAINVF4IC1_EPI4:    br label %[[EXIT]]49; MAINVF4IC1_EPI4:  [[EXIT]]:50;51; MAINVF4IC2_EPI4-LABEL: define void @f0(52; MAINVF4IC2_EPI4-SAME: i8 [[N:%.*]], i32 [[LEN:%.*]], ptr [[P:%.*]]) !prof [[PROF0:![0-9]+]] {53; MAINVF4IC2_EPI4:  [[ENTRY:.*:]]54; MAINVF4IC2_EPI4:    [[CMP_ENTRY:%.*]] = icmp sgt i32 [[LEN]], 055; MAINVF4IC2_EPI4:    br i1 [[CMP_ENTRY]], label %[[ITER_CHECK:.*]], label %[[EXIT:.*]], !prof [[PROF1:![0-9]+]]56; MAINVF4IC2_EPI4:  [[ITER_CHECK]]:57; MAINVF4IC2_EPI4:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0:%.*]], 458; MAINVF4IC2_EPI4:    br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]], !prof [[PROF2:![0-9]+]]59; MAINVF4IC2_EPI4:  [[VECTOR_SCEVCHECK]]:60; MAINVF4IC2_EPI4:    [[TMP2:%.*]] = icmp slt i8 [[TMP1:%.*]], 061; MAINVF4IC2_EPI4:    [[TMP3:%.*]] = icmp ugt i32 [[LEN]], 25562; MAINVF4IC2_EPI4:    br i1 [[TMP4:%.*]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]], !prof [[PROF2]]63; MAINVF4IC2_EPI4:  [[VECTOR_MAIN_LOOP_ITER_CHECK]]:64; MAINVF4IC2_EPI4:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i32 [[TMP0]], 865; MAINVF4IC2_EPI4:    br i1 [[MIN_ITERS_CHECK1]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]], !prof [[PROF2]]66; MAINVF4IC2_EPI4:  [[VECTOR_PH]]:67; MAINVF4IC2_EPI4:    br label %[[VECTOR_BODY:.*]]68; MAINVF4IC2_EPI4:  [[VECTOR_BODY]]:69; MAINVF4IC2_EPI4:    [[TMP9:%.*]] = icmp eq i32 [[INDEX_NEXT:%.*]], [[N_VEC:%.*]]70; MAINVF4IC2_EPI4:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !prof [[PROF3:![0-9]+]], !llvm.loop [[LOOP4:![0-9]+]]71; MAINVF4IC2_EPI4:  [[MIDDLE_BLOCK]]:72; MAINVF4IC2_EPI4:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]73; MAINVF4IC2_EPI4:    br i1 [[CMP_N]], label %[[EXIT_LOOPEXIT:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]], !prof [[PROF8:![0-9]+]]74; MAINVF4IC2_EPI4:  [[VEC_EPILOG_ITER_CHECK]]:75; MAINVF4IC2_EPI4:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i32 [[N_VEC_REMAINING:%.*]], 476; MAINVF4IC2_EPI4:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF9:![0-9]+]]77; MAINVF4IC2_EPI4:  [[VEC_EPILOG_PH]]:78; MAINVF4IC2_EPI4:    br label %[[VEC_EPILOG_VECTOR_BODY:.*]]79; MAINVF4IC2_EPI4:  [[VEC_EPILOG_VECTOR_BODY]]:80; MAINVF4IC2_EPI4:    [[TMP13:%.*]] = icmp eq i32 [[INDEX_NEXT6:%.*]], [[N_VEC3:%.*]]81; MAINVF4IC2_EPI4:    br i1 [[TMP13]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !prof [[PROF10:![0-9]+]], !llvm.loop [[LOOP11:![0-9]+]]82; MAINVF4IC2_EPI4:  [[VEC_EPILOG_MIDDLE_BLOCK]]:83; MAINVF4IC2_EPI4:    [[CMP_N8:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC3]]84; MAINVF4IC2_EPI4:    br i1 [[CMP_N8]], label %[[EXIT_LOOPEXIT]], label %[[VEC_EPILOG_SCALAR_PH]], !prof [[PROF13:![0-9]+]]85; MAINVF4IC2_EPI4:  [[VEC_EPILOG_SCALAR_PH]]:86; MAINVF4IC2_EPI4:    br label %[[LOOP:.*]]87; MAINVF4IC2_EPI4:  [[LOOP]]:88; MAINVF4IC2_EPI4:    [[CMP_LOOP:%.*]] = icmp ult i32 [[I32:%.*]], [[LEN]]89; MAINVF4IC2_EPI4:    br i1 [[CMP_LOOP]], label %[[LOOP]], label %[[EXIT_LOOPEXIT]], !prof [[PROF14:![0-9]+]], !llvm.loop [[LOOP15:![0-9]+]]90; MAINVF4IC2_EPI4:  [[EXIT_LOOPEXIT]]:91; MAINVF4IC2_EPI4:    br label %[[EXIT]]92; MAINVF4IC2_EPI4:  [[EXIT]]:93;94entry:95  %cmp.entry = icmp sgt i32 %len, 096  br i1 %cmp.entry, label %loop, label %exit, !prof !197 98loop:99  %i8 = phi i8 [0, %entry], [%i8.inc, %loop]100  %i32 = phi i32 [0, %entry], [%i32.inc, %loop]101 102  %ptr = getelementptr inbounds i32, ptr %p, i8 %i8103  store i32 %i32, ptr %ptr104 105  %i8.inc = add i8 %i8, 1106  %i32.inc = add i32 %i32, 1107 108  %cmp.loop = icmp ult i32 %i32, %len109  br i1 %cmp.loop, label %loop, label %exit, !prof !2110 111exit:112  ret void113}114 115!0 = !{!"function_entry_count", i64 13}116!1 = !{!"branch_weights", i32 12, i32 1}117!2 = !{!"branch_weights", i32 1234, i32 1}118;.119; MAINVF4IC1_EPI4: [[PROF0]] = !{!"function_entry_count", i64 13}120; MAINVF4IC1_EPI4: [[PROF1]] = !{!"branch_weights", i32 12, i32 1}121; MAINVF4IC1_EPI4: [[PROF2]] = !{!"branch_weights", i32 1, i32 127}122; MAINVF4IC1_EPI4: [[PROF3]] = !{!"branch_weights", i32 1, i32 307}123; MAINVF4IC1_EPI4: [[LOOP4]] = distinct !{[[LOOP4]], [[META5:![0-9]+]], [[META6:![0-9]+]], [[META7:![0-9]+]]}124; MAINVF4IC1_EPI4: [[META5]] = !{!"llvm.loop.isvectorized", i32 1}125; MAINVF4IC1_EPI4: [[META6]] = !{!"llvm.loop.unroll.runtime.disable"}126; MAINVF4IC1_EPI4: [[META7]] = !{!"llvm.loop.estimated_trip_count", i32 308}127; MAINVF4IC1_EPI4: [[PROF8]] = !{!"branch_weights", i32 1, i32 3}128; MAINVF4IC1_EPI4: [[PROF9]] = !{!"branch_weights", i32 4, i32 0}129; MAINVF4IC1_EPI4: [[PROF10]] = !{!"branch_weights", i32 0, i32 0}130; MAINVF4IC1_EPI4: [[LOOP11]] = distinct !{[[LOOP11]], [[META5]], [[META6]], [[META12:![0-9]+]]}131; MAINVF4IC1_EPI4: [[META12]] = !{!"llvm.loop.estimated_trip_count", i32 0}132; MAINVF4IC1_EPI4: [[PROF13]] = !{!"branch_weights", i32 2, i32 1}133; MAINVF4IC1_EPI4: [[LOOP14]] = distinct !{[[LOOP14]], [[META5]], [[META15:![0-9]+]]}134; MAINVF4IC1_EPI4: [[META15]] = !{!"llvm.loop.estimated_trip_count", i32 3}135;.136; MAINVF4IC2_EPI4: [[PROF0]] = !{!"function_entry_count", i64 13}137; MAINVF4IC2_EPI4: [[PROF1]] = !{!"branch_weights", i32 12, i32 1}138; MAINVF4IC2_EPI4: [[PROF2]] = !{!"branch_weights", i32 1, i32 127}139; MAINVF4IC2_EPI4: [[PROF3]] = !{!"branch_weights", i32 1, i32 153}140; MAINVF4IC2_EPI4: [[LOOP4]] = distinct !{[[LOOP4]], [[META5:![0-9]+]], [[META6:![0-9]+]], [[META7:![0-9]+]]}141; MAINVF4IC2_EPI4: [[META5]] = !{!"llvm.loop.isvectorized", i32 1}142; MAINVF4IC2_EPI4: [[META6]] = !{!"llvm.loop.unroll.runtime.disable"}143; MAINVF4IC2_EPI4: [[META7]] = !{!"llvm.loop.estimated_trip_count", i32 154}144; MAINVF4IC2_EPI4: [[PROF8]] = !{!"branch_weights", i32 1, i32 7}145; MAINVF4IC2_EPI4: [[PROF9]] = !{!"branch_weights", i32 4, i32 4}146; MAINVF4IC2_EPI4: [[PROF10]] = !{!"branch_weights", i32 0, i32 0}147; MAINVF4IC2_EPI4: [[LOOP11]] = distinct !{[[LOOP11]], [[META5]], [[META6]], [[META12:![0-9]+]]}148; MAINVF4IC2_EPI4: [[META12]] = !{!"llvm.loop.estimated_trip_count", i32 0}149; MAINVF4IC2_EPI4: [[PROF13]] = !{!"branch_weights", i32 1, i32 3}150; MAINVF4IC2_EPI4: [[PROF14]] = !{!"branch_weights", i32 2, i32 1}151; MAINVF4IC2_EPI4: [[LOOP15]] = distinct !{[[LOOP15]], [[META5]], [[META16:![0-9]+]]}152; MAINVF4IC2_EPI4: [[META16]] = !{!"llvm.loop.estimated_trip_count", i32 3}153;.154