================
@@ -143,6 +143,10 @@ def CIR_RecordLayoutAttr : CIR_Attr<"RecordLayout",
"record_layout", [
- `record_align_in_bytes`: from `ASTRecordLayout::getAlignment()`.
Needed because CIR's DataLayout cannot account for
`__attribute__((aligned(N)))`.
+ - `is_empty`: whether the record carries no data for argument passing
----------------
adams381 wrote:
It cannot be calculated from the CIR type as it stands. `struct { int : 3; }`
and `struct { unsigned char c; }` lower to `!cir.struct<"UnnamedBits" {!u8i}>`
and `!cir.struct<"OneByte" {!u8i}>`, identical apart from the name and neither
marked `padded`, yet classic CodeGen drops the first from the signature and
passes the second in a register. A structural test answers both the same way,
and they need opposite answers. Both cases are in `record-type-metadata.cpp`.
I debated with myself on the shape of this PR as I was implementing it. "All
fields are padding" was the first version of this patch, matching `padded` plus
a single padding-byte member. It handles the common empty class and quietly
mispasses the unnamed bit-field, which is what moved me to the AST.
CIR does not record a bit-field's storage unit, so that byte is
indistinguishable from a byte of data. Recording it would make emptiness
calculable, but it is a CIRGen change of a different size than this patch.
`is_empty` is implemented the same way as `arg_passing_kind`,
`has_trivial_dtor` and `record_align` on `cir.record_layout`. They are all AST
facts the type does not preserve.
https://github.com/llvm/llvm-project/pull/214742
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
