pre-commit: test PR76458 by dtcxzyw · Pull Request #55 · dtcxzyw/llvm-opt-benchmark

dtcxzyw · 2023-12-27T17:50:52Z

github-actions · 2023-12-27T18:08:00Z

baseline: llvm/llvm-project@fe94ae8
patch: llvm/llvm-project#76458
sha256: 41e6a325a9c1b1fdd9794e6a641d2a150e02df3f8cf0999086a3885213806a66
commit: ac88e3d
ac88e3d pre-commit: Update
bench/abseil-cpp/optimized/algorithm_test.cc.ll
bench/abseil-cpp/optimized/bernoulli_distribution_test.cc.ll
bench/abseil-cpp/optimized/beta_distribution_test.cc.ll
bench/abseil-cpp/optimized/bit_gen_ref_test.cc.ll
bench/abseil-cpp/optimized/chi_square_test.cc.ll
bench/abseil-cpp/optimized/common_policy_traits_test.cc.ll
bench/abseil-cpp/optimized/container_test.cc.ll
bench/abseil-cpp/optimized/convert_test.cc.ll
bench/abseil-cpp/optimized/cord.cc.ll
bench/abseil-cpp/optimized/cord_rep_btree_navigator_test.cc.ll
bench/abseil-cpp/optimized/cord_rep_btree_reader_test.cc.ll
bench/abseil-cpp/optimized/cord_rep_btree_test.cc.ll
bench/abseil-cpp/optimized/cord_rep_crc_test.cc.ll
bench/abseil-cpp/optimized/cord_test.cc.ll
bench/abseil-cpp/optimized/cordz_handle.cc.ll
bench/abseil-cpp/optimized/cordz_handle_test.cc.ll
bench/abseil-cpp/optimized/cordz_info_statistics_test.cc.ll
bench/abseil-cpp/optimized/cordz_info_test.cc.ll
bench/abseil-cpp/optimized/cordz_sample_token_test.cc.ll
bench/abseil-cpp/optimized/cordz_test.cc.ll
bench/abseil-cpp/optimized/crc_cord_state.cc.ll
bench/abseil-cpp/optimized/crc_cord_state_test.cc.ll
bench/abseil-cpp/optimized/crc_memcpy_test.cc.ll
bench/abseil-cpp/optimized/discrete_distribution.cc.ll
bench/abseil-cpp/optimized/discrete_distribution_test.cc.ll
bench/abseil-cpp/optimized/duration_test.cc.ll
bench/abseil-cpp/optimized/endian_test.cc.ll
bench/abseil-cpp/optimized/examples_test.cc.ll
bench/abseil-cpp/optimized/exception_safety_testing_test.cc.ll
bench/abseil-cpp/optimized/explicit_seed_seq_test.cc.ll
bench/abseil-cpp/optimized/exponential_biased_test.cc.ll
bench/abseil-cpp/optimized/exponential_distribution_test.cc.ll
bench/abseil-cpp/optimized/failure_signal_handler_test.cc.ll
bench/abseil-cpp/optimized/fixed_array_test.cc.ll
bench/abseil-cpp/optimized/flat_hash_map_test.cc.ll
bench/abseil-cpp/optimized/flat_hash_set_test.cc.ll
bench/abseil-cpp/optimized/gaussian_distribution_test.cc.ll
bench/abseil-cpp/optimized/graphcycles.cc.ll
bench/abseil-cpp/optimized/graphcycles_test.cc.ll
bench/abseil-cpp/optimized/hash_generator_testing.cc.ll
bench/abseil-cpp/optimized/hash_instantiated_test.cc.ll
bench/abseil-cpp/optimized/hash_policy_traits_test.cc.ll
bench/abseil-cpp/optimized/hash_test.cc.ll
bench/abseil-cpp/optimized/hashtablez_sampler_test.cc.ll
bench/abseil-cpp/optimized/inlined_vector_exception_safety_test.cc.ll
bench/abseil-cpp/optimized/inlined_vector_test.cc.ll
bench/abseil-cpp/optimized/int128_stream_test.cc.ll
bench/abseil-cpp/optimized/int128_test.cc.ll
bench/abseil-cpp/optimized/layout_test.cc.ll
bench/abseil-cpp/optimized/log_basic_test.cc.ll
bench/abseil-cpp/optimized/log_modifier_methods_test.cc.ll
bench/abseil-cpp/optimized/log_severity_test.cc.ll
bench/abseil-cpp/optimized/log_sink_set.cc.ll
bench/abseil-cpp/optimized/log_sink_test.cc.ll
bench/abseil-cpp/optimized/log_streamer_test.cc.ll
bench/abseil-cpp/optimized/log_uniform_int_distribution_test.cc.ll
bench/abseil-cpp/optimized/marshalling.cc.ll
bench/abseil-cpp/optimized/marshalling_test.cc.ll
bench/abseil-cpp/optimized/mock_distributions_test.cc.ll
bench/abseil-cpp/optimized/mocking_bit_gen_test.cc.ll
bench/abseil-cpp/optimized/mutex_test.cc.ll
bench/abseil-cpp/optimized/no_destructor_test.cc.ll
bench/abseil-cpp/optimized/node_hash_map_test.cc.ll
bench/abseil-cpp/optimized/node_hash_set_test.cc.ll
bench/abseil-cpp/optimized/non_temporal_memcpy_test.cc.ll
bench/abseil-cpp/optimized/nonsecure_base_test.cc.ll
bench/abseil-cpp/optimized/numbers_test.cc.ll
bench/abseil-cpp/optimized/parse.cc.ll
bench/abseil-cpp/optimized/parse_test.cc.ll
bench/abseil-cpp/optimized/parser_test.cc.ll
bench/abseil-cpp/optimized/pcg_engine_test.cc.ll
bench/abseil-cpp/optimized/periodic_sampler_test.cc.ll
bench/abseil-cpp/optimized/poisson_distribution_test.cc.ll
bench/abseil-cpp/optimized/randen_engine_test.cc.ll
bench/abseil-cpp/optimized/raw_hash_set_test.cc.ll
bench/abseil-cpp/optimized/reflection.cc.ll
bench/abseil-cpp/optimized/reflection_test.cc.ll
bench/abseil-cpp/optimized/salted_seed_seq_test.cc.ll
bench/abseil-cpp/optimized/sample_recorder_test.cc.ll
bench/abseil-cpp/optimized/seed_sequences.cc.ll
bench/abseil-cpp/optimized/seed_sequences_test.cc.ll
bench/abseil-cpp/optimized/sequence_lock_test.cc.ll
bench/abseil-cpp/optimized/span_test.cc.ll
bench/abseil-cpp/optimized/status_internal.cc.ll
bench/abseil-cpp/optimized/status_test.cc.ll
bench/abseil-cpp/optimized/statusor_test.cc.ll
bench/abseil-cpp/optimized/stderr_log_sink_test.cc.ll
bench/abseil-cpp/optimized/str_cat_test.cc.ll
bench/abseil-cpp/optimized/str_join_test.cc.ll
bench/abseil-cpp/optimized/str_replace.cc.ll
bench/abseil-cpp/optimized/str_replace_test.cc.ll
bench/abseil-cpp/optimized/str_split_test.cc.ll
bench/abseil-cpp/optimized/time_test.cc.ll
bench/abseil-cpp/optimized/time_zone_impl.cc.ll
bench/abseil-cpp/optimized/time_zone_info.cc.ll
bench/abseil-cpp/optimized/unordered_map_test.cc.ll
bench/abseil-cpp/optimized/unordered_set_test.cc.ll
bench/abseil-cpp/optimized/usage.cc.ll
bench/abseil-cpp/optimized/vlog_config.cc.ll

dtcxzyw · 2023-12-27T19:24:03Z

Regression:

diff --git a/bench/proxygen/optimized/HTTPTransactionEgressSM.cpp.ll b/bench/proxygen/optimized/HTTPTransactionEgressSM.cpp.ll
index 94a76d09..b01dce6a 100644
--- a/bench/proxygen/optimized/HTTPTransactionEgressSM.cpp.ll
+++ b/bench/proxygen/optimized/HTTPTransactionEgressSM.cpp.ll
@@ -243,7 +243,13 @@ if.then.i.i:                                      ; preds = %entry
 
 _ZNSt6vectorISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE17_S_check_init_lenEmRKS7_.exit.i: ; preds = %entry
   %cmp.not.i.i = icmp eq i64 %__l.coerce1, 0
-  br i1 %cmp.not.i.i, label %invoke.cont, label %for.body.i.i.i.i.preheader.i
+  br i1 %cmp.not.i.i, label %_ZNSt12_Vector_baseISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE11_M_allocateEm.exit.thread.i, label %for.body.i.i.i.i.preheader.i
+
+_ZNSt12_Vector_baseISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE11_M_allocateEm.exit.thread.i: ; preds = %_ZNSt6vectorISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE17_S_check_init_lenEmRKS7_.exit.i
+  %add.ptr8.i = getelementptr inbounds i8, ptr null, i64 %add.ptr.i.idx
+  %_M_end_of_storage9.i = getelementptr inbounds %"struct.std::_Vector_base<std::pair<std::pair<proxygen::HTTPTransactionEgressSMData::State, proxygen::HTTPTransactionEgressSMData::Event>, proxygen::HTTPTransactionEgressSMData::State>, std::allocator<std::pair<std::pair<proxygen::HTTPTransactionEgressSMData::State, proxygen::HTTPTransactionEgressSMData::Event>, proxygen::HTTPTransactionEgressSMData::State>>>::_Vector_impl_data", ptr %this, i64 0, i32 2
+  store ptr %add.ptr8.i, ptr %_M_end_of_storage9.i, align 8
+  br label %invoke.cont
 
 for.body.i.i.i.i.preheader.i:                     ; preds = %_ZNSt6vectorISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE17_S_check_init_lenEmRKS7_.exit.i
   %call5.i.i.i.i2 = invoke noalias noundef nonnull ptr @_Znwm(i64 noundef %add.ptr.i.idx) #15
@@ -251,7 +257,7 @@ for.body.i.i.i.i.preheader.i:                     ; preds = %_ZNSt6vectorISt4pai
 
 call5.i.i.i.i.noexc:                              ; preds = %for.body.i.i.i.i.preheader.i
   store ptr %call5.i.i.i.i2, ptr %this, align 8
-  %add.ptr.i1 = getelementptr inbounds %"struct.std::pair.5", ptr %call5.i.i.i.i2, i64 %__l.coerce1
+  %add.ptr.i1 = getelementptr inbounds i8, ptr %call5.i.i.i.i2, i64 %add.ptr.i.idx
   %_M_end_of_storage.i = getelementptr inbounds %"struct.std::_Vector_base<std::pair<std::pair<proxygen::HTTPTransactionEgressSMData::State, proxygen::HTTPTransactionEgressSMData::Event>, proxygen::HTTPTransactionEgressSMData::State>, std::allocator<std::pair<std::pair<proxygen::HTTPTransactionEgressSMData::State, proxygen::HTTPTransactionEgressSMData::Event>, proxygen::HTTPTransactionEgressSMData::State>>>::_Vector_impl_data", ptr %this, i64 0, i32 2
   store ptr %add.ptr.i1, ptr %_M_end_of_storage.i, align 8
   %0 = add nsw i64 %add.ptr.i.idx, -3
@@ -262,8 +268,8 @@ call5.i.i.i.i.noexc:                              ; preds = %for.body.i.i.i.i.pr
   %scevgep.i = getelementptr i8, ptr %call5.i.i.i.i2, i64 %3
   br label %invoke.cont
 
-invoke.cont:                                      ; preds = %_ZNSt6vectorISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE17_S_check_init_lenEmRKS7_.exit.i, %call5.i.i.i.i.noexc
-  %__cur.0.lcssa.i.i.i.i.i = phi ptr [ %scevgep.i, %call5.i.i.i.i.noexc ], [ null, %_ZNSt6vectorISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE17_S_check_init_lenEmRKS7_.exit.i ]
+invoke.cont:                                      ; preds = %call5.i.i.i.i.noexc, %_ZNSt12_Vector_baseISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE11_M_allocateEm.exit.thread.i
+  %__cur.0.lcssa.i.i.i.i.i = phi ptr [ %scevgep.i, %call5.i.i.i.i.noexc ], [ null, %_ZNSt12_Vector_baseISt4pairIS0_IN8proxygen27HTTPTransactionEgressSMData5StateENS2_5EventEES3_ESaIS6_EE11_M_allocateEm.exit.thread.i ]
   %_M_finish.i = getelementptr inbounds %"struct.std::_Vector_base<std::pair<std::pair<proxygen::HTTPTransactionEgressSMData::State, proxygen::HTTPTransactionEgressSMData::Event>, proxygen::HTTPTransactionEgressSMData::State>, std::allocator<std::pair<std::pair<proxygen::HTTPTransactionEgressSMData::State, proxygen::HTTPTransactionEgressSMData::Event>, proxygen::HTTPTransactionEgressSMData::State>>>::_Vector_impl_data", ptr %this, i64 0, i32 1
   store ptr %__cur.0.lcssa.i.i.i.i.i, ptr %_M_finish.i, align 8
   ret void

nikic · 2023-12-28T09:07:13Z

@dtcxzyw It's possible to fold getelementptr inbounds i8, ptr null, i64 %add.ptr.i.idx to null (assuming this is not compiled with -fno-delete-null-pointer-checks).

dtcxzyw · 2023-12-28T13:06:53Z

@dtcxzyw It's possible to fold getelementptr inbounds i8, ptr null, i64 %add.ptr.i.idx to null (assuming this is not compiled with -fno-delete-null-pointer-checks).

Alive2: https://alive2.llvm.org/ce/z/Zy_4Dz
My concern is that it would break some incorrect offsetof idioms like #define offsetof(st, m) ((size_t)&(((st *)0)->m)), which is UB according to the C/C++ standard.
Example: https://godbolt.org/z/PvT57z895

I will post a patch later.

nikic · 2023-12-28T13:54:20Z

@dtcxzyw We could avoid doing the transform if the offset is also constant. Or change clang frontend to not emit inbounds for this specific pattern. I agree that it would be best not to break it.

dtcxzyw · 2023-12-28T15:30:40Z

A possible miscompilation after simplifying gep inbounds null, idx -> null:
https://godbolt.org/z/Gshj9rz6e

#include <vector>

std::vector<std::vector<int>> test(std::vector<int> y) {
	std::vector<std::vector<int>> x;
	x.push_back(y);
	return x;
}

diff:

148,151c148
<   %add.ptr.i.i.i.i.i70 = getelementptr inbounds i32, ptr null, i64 %sub.ptr.div.i.i.i.i
<   %_M_end_of_storage.i.i.i.i.i71 = getelementptr inbounds %"struct.std::_Vector_base<int, std::allocator<int>>::_Vector_impl_data", ptr %add.ptr, i64 0, i32 2
<   tail call void @llvm.memset.p0.i64(ptr noundef nonnull align 8 dereferenceable(16) %add.ptr, i8 0, i64 16, i1 false)
<   store ptr %add.ptr.i.i.i.i.i70, ptr %_M_end_of_storage.i.i.i.i.i71, align 8, !tbaa !16
---
>   tail call void @llvm.memset.p0.i64(ptr noundef nonnull align 8 dereferenceable(24) %add.ptr, i8 0, i64 24, i1 false)

But this BB seems to be unreachable.

…se pointers (#130734) In the LLVM middle-end we want to fold `gep inbounds null, idx -> null`: https://alive2.llvm.org/ce/z/5ZkPx- This pattern is common in real-world programs (dtcxzyw/llvm-opt-benchmark#55 (comment)). Generally, it exists in some (actually) unreachable blocks, which is introduced by JumpThreading. However, some old-style offsetof macros are still widely used in real-world C/C++ code (e.g., hwloc/slurm/luajit). To avoid breaking existing code and inconvenience to downstream users, this patch removes the inbounds flag from the struct gep if the base pointer is null.

…ith null base pointers (#130734) In the LLVM middle-end we want to fold `gep inbounds null, idx -> null`: https://alive2.llvm.org/ce/z/5ZkPx- This pattern is common in real-world programs (dtcxzyw/llvm-opt-benchmark#55 (comment)). Generally, it exists in some (actually) unreachable blocks, which is introduced by JumpThreading. However, some old-style offsetof macros are still widely used in real-world C/C++ code (e.g., hwloc/slurm/luajit). To avoid breaking existing code and inconvenience to downstream users, this patch removes the inbounds flag from the struct gep if the base pointer is null.

dtcxzyw and others added 2 commits December 28, 2023 01:50

pre-commit: test PR76458

5227db2

PR Link: llvm/llvm-project#76458

pre-commit: Update

ac88e3d

dtcxzyw added the regression label Dec 27, 2023

dtcxzyw force-pushed the main branch 9 times, most recently from 296115f to 7225cd8 Compare September 17, 2024 07:23

dtcxzyw closed this Oct 22, 2024

dtcxzyw deleted the test-pr76458 branch October 22, 2024 12:01

dtcxzyw mentioned this pull request Mar 11, 2025

[Clang][CodeGen] Do not set inbounds flag for struct GEP with null base pointers llvm/llvm-project#130734

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

pre-commit: test PR76458#55

pre-commit: test PR76458#55
dtcxzyw wants to merge 2 commits intomainfrom
test-pr76458

dtcxzyw commented Dec 27, 2023

Uh oh!

github-actions bot commented Dec 27, 2023

Uh oh!

dtcxzyw commented Dec 27, 2023

Uh oh!

nikic commented Dec 28, 2023

Uh oh!

dtcxzyw commented Dec 28, 2023 •

edited

Loading

Uh oh!

nikic commented Dec 28, 2023

Uh oh!

dtcxzyw commented Dec 28, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

dtcxzyw commented Dec 27, 2023

Uh oh!

github-actions bot commented Dec 27, 2023

Uh oh!

dtcxzyw commented Dec 27, 2023

Uh oh!

nikic commented Dec 28, 2023

Uh oh!

dtcxzyw commented Dec 28, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nikic commented Dec 28, 2023

Uh oh!

dtcxzyw commented Dec 28, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dtcxzyw commented Dec 28, 2023 •

edited

Loading