[CIR][CIRGen] Achieve union's bitfields additionally.

[566hub]

Achieve union's bitfields additionally.

[bcardosolopes]

Thanks for the PR.

When I try to build the testcase, I get a crash in a place that doesn't seem related to the changes here:

for (const auto *Field : recordDecl->fields()) {
    if (Field->isBitField()) {
      if (Field->isZeroLengthBitField(astContext))
        continue;
      llvm_unreachable("NYI");
    }

Can you elaborate a bit more on the approach and what are you trying to achieve? @gitoleg can you also review this PR?

[DukeWJ]

Thanks for the PR.

When I try to build the testcase, I get a crash in a place that doesn't seem related to the changes here:

for (const auto *Field : recordDecl->fields()) {
    if (Field->isBitField()) {
      if (Field->isZeroLengthBitField(astContext))
        continue;
      llvm_unreachable("NYI");
    }

Can you elaborate a bit more on the approach and what are you trying to achieve? @gitoleg can you also review this PR?

Hi, I have met the scene that codegen union with bitfields with clangir once, and it reported 'NYI' yet. I think this PR just reuse infrastructure of struct bitfields to explain union bitfiels type. I try locally and it seems work, and new union type has carried bitfield info I think.
input

union demo {
    int x : 16;
    int y;
};

int main() {
    union demo a;

    a.x = 0;
    a.y = 1;
}

and the output

!ty_22demo22 = !cir.struct<union "demo" {!cir.int<u, 16>, !cir.int<s, 32>}>
#bfi_x = #cir.bitfield_info<name = "x", storage_type = !u16i, size = 16, offset = 0, is_signed = true>

func {
  %1 = cir.alloca !ty_22demo22, !cir.ptr<!ty_22demo22>, ["a"] {alignment = 4 : i64} loc(#loc17)
  %3 = cir.const #cir.int<0> : !s32i loc(#loc7)
  %4 = cir.cast(bitcast, %1 : !cir.ptr<!ty_22demo22>), !cir.ptr<!u16i> loc(#loc8)
  %5 = cir.set_bitfield(#bfi_x, %4 : !cir.ptr<!u16i>, %3 : !s32i) -> !s32i loc(#loc8)

  %6 = cir.const #cir.int<1> : !s32i loc(#loc9)
  %7 = cir.get_member %1[1] {name = "y"} : !cir.ptr<!ty_22demo22> -> !cir.ptr<!s32i> loc(#loc10)
  cir.store %6, %7 : !s32i, !cir.ptr<!s32i> loc(#loc19)
}

[gitoleg]

looks reasonable ... though frankly speaking I didn't know that it's somehow possible (and useful!) to use bit fields in unions. But anyways, looks ok so far.

[gitoleg]

sorry, my bad, pushed wrong button

[bcardosolopes]

Apologies on the delay, this LGTM. Thanks a bunch for working on this