Differential D74507
[llvm-objdump] Print method name from debug info in disassembly output. Authored by rupprecht on Feb 12 2020, 1:57 PM.
Details GNU objdump prints the method name in disassembly output, and upon further investigation this seems to come from debug info, not the symbol table. Some additional refactoring is necessary to make this work even when the line number is 0/the filename is unknown. The added test case includes a note for this scenario. See http://llvm.org/PR41341 for more info.
Diff Detail
Event Timeline
Comment Actions
Comment Actions @rupprecht, could you post a before/after + GNU output example please, as I'm struggling purely from the tests to understand what change has actually been made.
Comment Actions Sure, here's an example you can play around with: $ echo 'extern "C" int foo() { return 5; }; namespace xyz { int bar() { return 10; } }' > /tmp/src.cc; clang++ -g -c /tmp/src.cc -o /tmp/obj.o $ objdump -ldC /tmp/obj.o # GNU objdump ... Disassembly of section .text: 0000000000000000 <foo>: foo(): /tmp/src.cc:1 0: 55 push %rbp 1: 48 89 e5 mov %rsp,%rbp 4: b8 05 00 00 00 mov $0x5,%eax 9: 5d pop %rbp a: c3 retq b: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) 0000000000000010 <xyz::bar()>: _ZN3xyz3barEv(): # <-- Note, see https://sourceware.org/ml/binutils/2020-02/msg00369.html to demangle this too /tmp/src.cc:1 10: 55 push %rbp 11: 48 89 e5 mov %rsp,%rbp 14: b8 0a 00 00 00 mov $0xa,%eax 19: 5d pop %rbp 1a: c3 retq $ llvm-objdump -ldC /tmp/obj.o # llvm-objdump, w/o this patch ... 0000000000000000 foo: ; /tmp/src.cc:1 0: 55 pushq %rbp 1: 48 89 e5 movq %rsp, %rbp 4: b8 05 00 00 00 movl $5, %eax 9: 5d popq %rbp a: c3 retq b: 0f 1f 44 00 00 nopl (%rax,%rax) 0000000000000010 xyz::bar(): 10: 55 pushq %rbp 11: 48 89 e5 movq %rsp, %rbp 14: b8 0a 00 00 00 movl $10, %eax 19: 5d popq %rbp 1a: c3 retq $ llvm-objdump -ldC /tmp/obj.o # llvm-objdump, w/ this patch ... 0000000000000000 foo: ; foo(): # <-- this is new ; /tmp/src.cc:1 0: 55 pushq %rbp 1: 48 89 e5 movq %rsp, %rbp 4: b8 05 00 00 00 movl $5, %eax 9: 5d popq %rbp a: c3 retq ; /tmp/src.cc:1 # <-- just noticed this unexpected diff, I'll see what's going on here.... b: 0f 1f 44 00 00 nopl (%rax,%rax) 0000000000000010 xyz::bar(): ; _ZN3xyz3barEv(): # <-- this is new ; /tmp/src.cc:1 # <-- this is new and correct (although wasn't the intention of this patch to fix) 10: 55 pushq %rbp 11: 48 89 e5 movq %rsp, %rbp 14: b8 0a 00 00 00 movl $10, %eax 19: 5d popq %rbp 1a: c3 retq http://llvm.org/PR41341 also has some examples. It seems superfluous with the trivial examples, but helps in situations where the object file has debugging information but no (or limited) symbols. Note this patch doesn't address the syntactic difference of the leading ; delimiter, which is still a difference between llvm-objdump and GNU objdump. Comment Actions
Comment Actions FYI, I usually run: arcfilter () {
git log -1 --pretty=%B | awk '/Reviewers:|Subscribers:/{p=1} /Reviewed By:|Differential Revision:/{p=0} !p && !/^Summary:/' | git commit --amend -F -
}to remove unneeded metadata tags from the git description before committing. Not sure about the most appropriate way. (I don't know enough about Phabricator... Maybe there is some arc magic doing this.) Comment Actions Hopefully fixed by d17123b2577e610b2a19de1f530cecea353c8c7a (I feel so sad about the canonical path separator on Windows... So many times a DWARF test accidentally failed on Windows. I asked whether we can use / everywhere but the answer was no.) | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Clearly there's prior art here, but testing objdump with llc seems a bit heavyweight/circular (no doubt some llc tests use llvm-objdump to verify their result) - I'd expect this to use raw assembly + llvm-mc at most.
not a requirement for change, just a thought