<div dir="ltr"><br><div class="gmail_extra"><br><br><div class="gmail_quote">On Mon, Oct 7, 2013 at 9:53 AM, Eric Christopher <span dir="ltr"><<a href="mailto:echristo@gmail.com" target="_blank">echristo@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="im">On Sat, Oct 5, 2013 at 6:07 PM, David Blaikie <<a href="mailto:dblaikie@gmail.com">dblaikie@gmail.com</a>> wrote:<br>
><br>
><br>
><br>
> On Sat, Oct 5, 2013 at 5:45 PM, Richard Mitton <<a href="mailto:richard@codersnotes.com">richard@codersnotes.com</a>><br>
> wrote:<br>
>><br>
>> It's better without the *user in the switch, as the code that calls it<br>
>> will then just print 'invalid opcode %x' or whatever, as opposed to right<br>
>> now, where it knows about DW_OP_lo_user+0, but not any other user OPs<br>
>> (DW_OP_lo_user+N), which is silly.<br>
>><br>
>> I don't think there's anything that really uses this much already, but<br>
>> I'll see if I can figure something out to test it with and the update the<br>
>> patch<br>
><br>
><br>
> When we print assembly (rather than the usual debug info of emitting an<br>
> object file and then using llvm-dwarfdump) we annotate the assembly with<br>
> comments - but thinking about it, I don't think we annotate DWARF<br>
> expressions/operators/etc, so this probably isn't visible. I wonder if we<br>
> call the OperationEncodingString at all.<br>
><br>
<br>
</div>We may not, but it'll be nice to do so. It's handy to use when debugging anyhow.<br></blockquote><div><br></div><div>Guessing: I suspect we can't really do it easily at the moment because the dumping code would have to get a lot smarter. The form doesn't imply whether or not the bytes are a DWARF expression or not, does it? (ie: data1 may or may not be a DWARF expression, it might instead be a file/line number, etc... ) so we might end up annotating things that aren't DWARF expressions as though they are. I guess maybe we could have a table of attribute kind + form to decide whether to annotate the bytes as DWARF expressions or not.</div>
<div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<span class="HOEnZb"><font color="#888888"><br>
-eric<br>
</font></span><div class="HOEnZb"><div class="h5"><br>
>><br>
>> .<br>
>><br>
>><br>
>> Eric Christopher wrote:<br>
>>><br>
>>> Should dump differently I'd have thought. Also don't want to delete<br>
>>> the DW_OP_*_user text I'd think.<br>
>>><br>
>>> -eric<br>
>>><br>
>>> On Sat, Oct 5, 2013 at 2:33 PM, Richard Mitton<<a href="mailto:richard@codersnotes.com">richard@codersnotes.com</a>><br>
>>> wrote:<br>
>>><br>
>>>>><br>
>>>>> No functionality change.<br>
>>>>><br>
>>>><br>
>>>> Nothing to test! :)<br>
>>>><br>
>>>><br>
>>>><br>
>>>> Eric Christopher wrote:<br>
>>>><br>
>>>>><br>
>>>>> Testcase? :)<br>
>>>>><br>
>>>>> -eric<br>
>>>>><br>
>>>>> On Sat, Oct 5, 2013 at 11:19 AM, Richard<br>
>>>>> Mitton<<a href="mailto:richard@codersnotes.com">richard@codersnotes.com</a>><br>
>>>>> wrote:<br>
>>>>><br>
>>>>><br>
>>>>>><br>
>>>>>> Formally added an explicit enum for DWARF TLS support. No<br>
>>>>>> functionality<br>
>>>>>> change.<br>
>>>>>><br>
>>>>>><br>
>>>>>> <a href="http://llvm-reviews.chandlerc.com/D1845" target="_blank">http://llvm-reviews.chandlerc.com/D1845</a><br>
>>>>>><br>
>>>>>> Files:<br>
>>>>>> include/llvm/Support/Dwarf.h<br>
>>>>>> lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp<br>
>>>>>> lib/Support/Dwarf.cpp<br>
>>>>>><br>
>>>>>> Index: include/llvm/Support/Dwarf.h<br>
>>>>>> ===================================================================<br>
>>>>>> --- include/llvm/Support/Dwarf.h<br>
>>>>>> +++ include/llvm/Support/Dwarf.h<br>
>>>>>> @@ -486,6 +486,9 @@<br>
>>>>>> DW_OP_lo_user = 0xe0,<br>
>>>>>> DW_OP_hi_user = 0xff,<br>
>>>>>><br>
>>>>>> + // Extensions for GNU-style thread-local storage.<br>
>>>>>> + DW_OP_GNU_push_tls_address = 0xe0,<br>
>>>>>> +<br>
>>>>>> // Extensions for Fission proposal.<br>
>>>>>> DW_OP_GNU_addr_index = 0xfb,<br>
>>>>>> DW_OP_GNU_const_index = 0xfc,<br>
>>>>>> Index: lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp<br>
>>>>>> ===================================================================<br>
>>>>>> --- lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp<br>
>>>>>> +++ lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp<br>
>>>>>> @@ -1513,14 +1513,15 @@<br>
>>>>>> // 1) Start with a constNu of the appropriate pointer size<br>
>>>>>> addUInt(Block, 0, dwarf::DW_FORM_data1,<br>
>>>>>> PointerSize == 4 ? dwarf::DW_OP_const4u :<br>
>>>>>> dwarf::DW_OP_const8u);<br>
>>>>>> - // 2) containing the (relocated) address of the TLS variable<br>
>>>>>> + // 2) containing the (relocated) offset of the TLS variable<br>
>>>>>> + // within the module's TLS block.<br>
>>>>>> addExpr(Block, 0, dwarf::DW_FORM_udata, Expr);<br>
>>>>>> } else {<br>
>>>>>> addUInt(Block, 0, dwarf::DW_FORM_data1,<br>
>>>>>> dwarf::DW_OP_GNU_const_index);<br>
>>>>>> addUInt(Block, 0, dwarf::DW_FORM_udata,<br>
>>>>>> DU->getAddrPoolIndex(Expr));<br>
>>>>>> }<br>
>>>>>> - // 3) followed by a custom OP to tell the debugger about TLS<br>
>>>>>> (presumably)<br>
>>>>>> - addUInt(Block, 0, dwarf::DW_FORM_data1, dwarf::DW_OP_lo_user);<br>
>>>>>> + // 3) followed by a custom OP to make the debugger do a TLS<br>
>>>>>> lookup.<br>
>>>>>> + addUInt(Block, 0, dwarf::DW_FORM_data1,<br>
>>>>>> dwarf::DW_OP_GNU_push_tls_address);<br>
>>>>>> } else<br>
>>>>>> addOpAddress(Block, Sym);<br>
>>>>>> // Do not create specification DIE if context is either compile<br>
>>>>>> unit<br>
>>>>>> Index: lib/Support/Dwarf.cpp<br>
>>>>>> ===================================================================<br>
>>>>>> --- lib/Support/Dwarf.cpp<br>
>>>>>> +++ lib/Support/Dwarf.cpp<br>
>>>>>> @@ -456,10 +456,11 @@<br>
>>>>>> case DW_OP_bit_piece: return "DW_OP_bit_piece";<br>
>>>>>> case DW_OP_implicit_value: return<br>
>>>>>> "DW_OP_implicit_value";<br>
>>>>>> case DW_OP_stack_value: return<br>
>>>>>> "DW_OP_stack_value";<br>
>>>>>> - case DW_OP_lo_user: return "DW_OP_lo_user";<br>
>>>>>> - case DW_OP_hi_user: return "DW_OP_hi_user";<br>
>>>>>><br>
>>>>>> - // DWARF5 Fission Proposal Op Extensions<br>
>>>>>> + // GNU thread-local storage<br>
>>>>>> + case DW_OP_GNU_push_tls_address: return<br>
>>>>>> "DW_OP_GNU_push_tls_address";<br>
>>>>>> +<br>
>>>>>> + // DWARF5 Fission Proposal Op Extensions<br>
>>>>>> case DW_OP_GNU_addr_index: return<br>
>>>>>> "DW_OP_GNU_addr_index";<br>
>>>>>> case DW_OP_GNU_const_index: return<br>
>>>>>> "DW_OP_GNU_const_index";<br>
>>>>>> }<br>
>>>>>><br>
>>>>>> _______________________________________________<br>
>>>>>> llvm-commits mailing list<br>
>>>>>> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>>>>>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
>>>>>><br>
>>>>>><br>
>>>>>><br>
>>>><br>
>>>> _______________________________________________<br>
>>>> llvm-commits mailing list<br>
>>>> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>>>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
>>>><br>
>><br>
>> _______________________________________________<br>
>> llvm-commits mailing list<br>
>> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
>> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
><br>
><br>
><br>
> _______________________________________________<br>
> llvm-commits mailing list<br>
> <a href="mailto:llvm-commits@cs.uiuc.edu">llvm-commits@cs.uiuc.edu</a><br>
> <a href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits" target="_blank">http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits</a><br>
><br>
</div></div></blockquote></div><br></div></div>