[DebugInfo][1/N] Instruction referencing test changes: DBG_VALUE to DBG_INSTR_REF

[jmorse]

Summary

This is the first patch in a series of test updates, to change X86 tests to look for instruction referencing variable locations. This patch contains a bunch of replacements of:

DBG_VALUE $somereg

with,

SOMEINST debug-instr-number1
DBG_INSTR_REF 1, 0, ...

It's mostly SelectionDAG tests that are making sure that the variable location assignment is placed in the correct position in the instructions.

[jmorse]

llvm/test/DebugInfo/X86/pr34545.ll

44

Added this because various parts of instruction referencing try to skip doing any work if there are no instructions in scope -- and no "real" instructions in this function are in scope.

[djtodoro]

llvm/test/DebugInfo/X86/pr40427.ll

24

Are we losing the test coverage for the DBG_VALUEs by doing this? I guess it is OK for the targets that will use this feature ON by default, right?

[jmorse]

llvm/test/DebugInfo/X86/pr40427.ll

24

Yes and no -- the debug instruction placement code is shared between instruction referencing and DBG_VALUEs, so it's sort-of being covered. On the other hand, that might change, and I don't think any other architectures cover these parts of SelectionDAG.

I suppose I can leave the CHECK lines in from DBG_VALUE mode, and it's not going to harm anything / over-test, and that'll ensure that SelectionDAG is doing the right thing in DBG_VALUE mode, does that sound alright? (I'd prefer to not do that for the tests in the other patches)

[jmorse]

llvm/test/DebugInfo/X86/pr40427.ll

24

(... and to be clear, that preference is because as far as I'm aware, SelectionDAG is the only thing really shared between DBG_VALUE and instr-ref)

[djtodoro]

llvm/test/DebugInfo/X86/pr40427.ll

24

OK, sounds good to me. Let's add two checks for both modes in SelectionDAG.

[jmorse]

Updated these tests to keep the DBG_VALUE outputs for SelectionDAG outputs.