http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54343
Bug #: 54343
Summary: RTL postreload leaks DF memory
Classification: Unclassified
Product: gcc
Version: 4.8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: rtl-optimization
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: rgue...@gcc.gnu.org
With valgrind on the testcase from PR54146 you can see leaks of the kind
==28803== 3,984 bytes in 249 blocks are definitely lost in loss record 28,125
of 29,883
==28803== at 0x4C29ADD: malloc (in
/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==28803== by 0xCE28D7: xmalloc (xmalloc.c:147)
==28803== by 0x6B064D: df_refs_add_to_chains(df_collection_rec*,
basic_block_def*, rtx_def*) (df-scan.c:2669)
==28803== by 0x6B5144: df_bb_refs_record(int, bool) (df-scan.c:3644)
==28803== by 0x6B524C: df_scan_blocks() (df-scan.c:678)
==28803== by 0x7B6B7D: rest_of_handle_reload() (ira.c:4390)
==28803== by 0x818C4A: execute_one_pass(opt_pass*) (passes.c:2157)
==28803== by 0x818F94: execute_pass_list(opt_pass*) (passes.c:2212)
==28803== by 0x818FA6: execute_pass_list(opt_pass*) (passes.c:2213)
==28803== by 0x68F583: expand_function(cgraph_node*) (cgraphunit.c:1609)
==28803== by 0x6909E8: compile() (cgraphunit.c:1714)
==28803== by 0x690F64: finalize_compilation_unit() (cgraphunit.c:2089)
do_reload re-adds the live problem and calls run_fast_dce. Either that
or subsequent passes that end up removing instructions do not bother
with freeing associated DF info. At least I don't see that delete_insn
calls df_insn_delete or so. Thus keeping DF initialized throughout
postreload may be not the best idea ...
