[PATCH] D135488: [codegen][WIP] Display stack layouts in console

Nick Desaulniers via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Tue Oct 11 15:10:41 PDT 2022


nickdesaulniers added a comment.

Nice! Diff 466919 fixes up my confusion regarding Slot vs SlotID.  Though I do like how the Slots are sorted, since it keeps the Offsets in order FWICT. Maybe we could drop the SlotID column, then print out Slot for the variables, rather than the SlotID? I mean, I only really care about the Offset in the order of Slot, so maybe an aggressive suggestion would be to drop Slot and SlotID columns. (Yes, that's more work for the pretty printer to do, but it makes it clearer to the developer debugging the warning). Or print the variable identifier as a column in the table (thought that might upset the nice table alignment).  Do we ever re-use stack slots for different variables with non-overlapping live ranges? If so, then a stack slot is potentially one-to-many variables, so putting it in a table might not look great.

For the next potential improvement: I think DWARF (when present) contains information about which variables where spilled where on the stack.  It looks like they are not part of `MF.getVariableDbgInfo()`. Maybe we can print that information, too?  If a particular variable is being spilled often, then the developer may be able to shorten the live range manually somehow, like re-materialization.

In D135488#3844204 <https://reviews.llvm.org/D135488#3844204>, @paulkirth wrote:

> In D135488#3844117 <https://reviews.llvm.org/D135488#3844117>, @nickdesaulniers wrote:
>
>> What does `Fixed` mean?
>
> FrameInfo has a notion of a "Fixed" slot.  From `MachineFrameInfo::CreateFixedObject`'s docstring `"All fixed objects should be created before other objects are created for efficiency. By default, fixed objects are not pointed to by LLVM IR values. This returns an index with a negative value."`
>
> I figured that may be important for people to know/see, so I left it in. It doesn't mean much to me though, and I've thought about dropping it.

I'm definitely going to forget that. Perhaps it makes sense to document it somewhere?  Otherwise it might just be adding noise.

Seems like our initial slots are contiguous with Fixed allocations.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D135488/new/

https://reviews.llvm.org/D135488



More information about the cfe-commits mailing list