[PATCH] D115973: [llvm-profgen] Support symbol loading for debug fission

Wed Jan 5 10:09:52 PST 2022

hoy added inline comments.

================
Comment at: llvm/test/tools/llvm-profgen/split-dwarf.test:12
+; RUN: llvm-profgen --format=text --unsymbolized-profile=%t --dwp=%S/Inputs/split-dwarf-split-inlining.dwo --binary=%S/Inputs/split-dwarf-split-inlining.perfbin --output=%t4 --fill-zero-for-all-funcs
+; RUN: FileCheck %s --input-file %t4 --check-prefix=CHECK-SPLIT-DWARF
+
----------------
ayermolo wrote:
> hoy wrote:
> > wlei wrote:
> > > hoy wrote:
> > > > Add a test case for no --dwp?
> > > Good point, added. Without -dwp, it will only load the inliner function.
> > Thanks for adding the test. Per the dwo switch description, should the dwo file be loaded if it is not specified and placed alongside with the binary?
> > 
> > `When not specified, it will be <binary>.dwp in the same directory as the main binary.` 
> > 
> > I guess it'll look for `split-dwarf-split-inlining.perfbin.dwo` instead of `split-dwarf-split-inlining.dwo`?
> .dwo/.o files don't need to be next to the binary, and there is no assumption that they will be. In fact in production environment they are not. The fully resolved path is specified in DWARF CU. If .dwo/.o files are not there then it's a warning/error. Depending on context.
Yeah, but in this case we have them side by side. I guess it's an issue of the file naming that resulted in the dwo file not loaded. Should we change the arg description to not support the automatic load?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D115973/new/

https://reviews.llvm.org/D115973