[PATCH] D72489: [DWARF] Emit DW_AT_call_return_pc as an address

[David Blaikie via Phabricator via llvm-commits]

dblaikie added a comment.

I ran some size analysis (& hadn't realized that this special case was only done in the DWARFv5 codepath... that leads me to wonder about size impact in DWARFv4, so maybe I'll need to go back and run some general size analysis cost of these call sites).

The .rela.debug_addr for a clang self host optimized (-O3) build, split DWARF, increased from 62.2MB to 70.5MB, which isn't great - but worse in the context of the further address reduction like the ones I've discussed on llvm-dev recently. "ranges everywhere" gets that 62.2MB down to 4.4MB, so growing it by a further 8MB would be pretty unfortunate.



================
Comment at: llvm/lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:984-991
   // from one function to another.
   if (DD->getDwarfVersion() == 4 && DD->tuneForGDB()) {
     assert(PCAddr && "Missing PC information for a call");
     addLabelAddress(CallSiteDIE, dwarf::DW_AT_low_pc, PCAddr);
   } else if (!IsTail || DD->tuneForGDB()) {
-    assert(PCOffset && "Missing return PC information for a call");
-    addAddressExpr(CallSiteDIE, dwarf::DW_AT_call_return_pc, PCOffset);
+    assert(PCAddr && "Missing return PC information for a call");
+    addLabelAddress(CallSiteDIE, dwarf::DW_AT_call_return_pc, PCAddr);
----------------
Side question: How'd this end up like this? Why all these GDB tuning checks? Seems like it'd add another layer of complexity/variety that'll make it harder for us to all be evaluating the same things. 


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D72489/new/

https://reviews.llvm.org/D72489