[llvm-readobj][test] - Add a test for --elf-cg-profile option.

[grimar]

Summary

This adds a test to document --elf-cg-profile option we have.
I am going to refactor this area, and this patch is mostly to
create a base for a follow-up change.

[jhenderson]

Aside: why is this switch prefixed with "elf"? I wouldn't think this section is ELF-specific in principle...

llvm/test/tools/llvm-readobj/ELF/call-graph-profile.test

2

handle the --elf-cg-profile...

33

Maybe refer to the symbols by name for clarity/robustness?

[grimar]

In D74115#1863351, @jhenderson wrote:

Aside: why is this switch prefixed with "elf"? I wouldn't think this section is ELF-specific in principle...

Hmmm. Should we rename to --llvm-cg-profile? Its a LLVM extention first of all.
And if we want to implement the GNU version (that is what I am working on), perhaps it is good idea to add this suffix to emphasise that is not a common GNU option?

[jhenderson]

In D74115#1863367, @grimar wrote:

In D74115#1863351, @jhenderson wrote:

Aside: why is this switch prefixed with "elf"? I wouldn't think this section is ELF-specific in principle...

Hmmm. Should we rename to --llvm-cg-profile? Its a LLVM extention first of all.
And if we want to implement the GNU version (that is what I am working on), perhaps it is good idea to all this suffix to emphasise that is not a common GNU option?

Seems reasonable to me. If the section does get adopted by GNU, it likely will need to go through some standardisation (e.g. getting a new SHT_* type), so the more generic "--cg-profile" switch can be saved for then.

[grimar]

In D74115#1863400, @jhenderson wrote:

In D74115#1863367, @grimar wrote:

In D74115#1863351, @jhenderson wrote:

Aside: why is this switch prefixed with "elf"? I wouldn't think this section is ELF-specific in principle...

Hmmm. Should we rename to --llvm-cg-profile? Its a LLVM extention first of all.
And if we want to implement the GNU version (that is what I am working on), perhaps it is good idea to all this suffix to emphasise that is not a common GNU option?

Seems reasonable to me. If the section does get adopted by GNU, it likely will need to go through some standardisation (e.g. getting a new SHT_* type), so the more generic "--cg-profile" switch can be saved for then.

OK. I'll leave it for a one more follow-up then (+ lets see what others think too).

[grimar]

• Addressed review comments.

llvm/test/tools/llvm-readobj/ELF/call-graph-profile.test

33

Done.

[jhenderson]

LGTM.

[MaskRay]

--llvm-cg-profile sounds good.

(I heard that .llvm.call-graph-profile is not that useful.. It also conflicts with lld --symbol-ordering-file=, which may be more useful..

llvm/test/tools/llvm-readobj/ELF/call-graph-profile.test

+## This test checks how we handle the --elf-cg-profile option.
+
+# RUN: yaml2obj %s -o %t.o
+# RUN: llvm-readobj %t.o --elf-cg-profile | FileCheck %s --check-prefix=LLVM
+# RUN: llvm-readelf %t.o --elf-cg-profile | FileCheck %s --check-prefix=GNU
+
+# LLVM:      CGProfile [
+# LLVM-NEXT:  CGProfileEntry {
+# LLVM-NEXT:    From: foo (1)
+# LLVM-NEXT:    To: bar (2)
+# LLVM-NEXT:    Weight: 89
+# LLVM-NEXT:  }
+# LLVM-NEXT:  CGProfileEntry {
+# LLVM-NEXT:    From: bar (2)
+# LLVM-NEXT:    To: foo (1)
+# LLVM-NEXT:    Weight: 98
+# LLVM-NEXT:  }
+# LLVM-NEXT: ]
+
+# GNU: GNUStyle::printCGProfile not implemented
+
+--- !ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2LSB
+  Type:    ET_DYN
+  Machine: EM_X86_64
+Sections:
+  - Name: .llvm.call-graph-profile
+    Type: SHT_LLVM_CALL_GRAPH_PROFILE
+    Entries:
+      - From:   foo
+        To:     bar
+        Weight: 89
+      - From:   bar
+        To:     foo
+        Weight: 98
+Symbols:
+  - Name: foo
+  - Name: bar