[CIR][CIRGen] Support Lambda capturing this object
#1213
Conversation
ghehg
changed the title
this object
Seems like a unrecheable NYI case to me, if you disagree please elaborate |
I was thinking about to create an issue to see if anybody from community wants to work on it. :-) |
Good intent, thanks! I don't understand why being NYI would preclude open source contribution or filing issues though. Missing features are usually used for things that we can't afford to NYI at a certain moment because otherwise everything would fail on that path. This isn't the case for a C++23 we don't yet support.
NYI |
Sure, just removed the missing feature. Thanks for clarification about NYI and Missing feature. |
|
Made changes according to review comments. I didn't change some of the |
clang/test/CIR/CodeGen/lambda.cpp
Outdated
| // LLVM: [[THIS_ADDR:%.*]] = alloca ptr, i64 1, align 8 | ||
| // LLVM: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 | ||
| // LLVM: store ptr {{%.*}}, ptr [[THIS_ADDR]], align 8 |
There was a problem hiding this comment.
I see now, you were not using LABEL because you were matching the argument, my bad!
There was a problem hiding this comment.
You can keep THIS for the case were you are matching the arg and use LABEL for the other ones (like I see you did below). Sorry for the churn
There was a problem hiding this comment.
I don't have the full context here, but if the problem is that CHECK-LABEL can't contain variable references, a common pattern I've seen for that, assuming the variable references occur after a label bit, is to use CHECK-LABEL for the label bit and then CHECK-SAME for the variable references. May or may not be worth it depending on the situation though.
There was a problem hiding this comment.
Sure, I'll use the SAME directive then, Another reason I want to check the argument is (in future) we should check attributes to argument(that's what the COM: line is about), This test seems to be a good case for that.
clang/test/CIR/CodeGen/lambda.cpp
Outdated
| // CHECK: [[CLS_ANNO7:%.*]] = cir.load [[ARG_ADDR]] : !cir.ptr<!cir.ptr<!ty_anon2E7_>>, !cir.ptr<!ty_anon2E7_> | ||
| // CHECK: [[STRUCT_A:%.*]] = cir.get_member [[CLS_ANNO7]][0] {name = "this"} : !cir.ptr<!ty_anon2E7_> -> !cir.ptr<!ty_A> | ||
| // CHECK: [[a:%.*]] = cir.get_member [[STRUCT_A]][0] {name = "a"} : !cir.ptr<!ty_A> -> !cir.ptr<!s32i> loc(#loc70) | ||
| // CHECK: cir.return {{%.*}} : !s32i |
There was a problem hiding this comment.
The return value here is important, please check it
clang/test/CIR/CodeGen/lambda.cpp
Outdated
| // CHECK: [[STRUCT_A_PTR:%.*]] = cir.get_member [[CLS_ANNO8]][0] {name = "this"} : !cir.ptr<!ty_anon2E8_> -> !cir.ptr<!cir.ptr<!ty_A>> | ||
| // CHECK: [[STRUCT_A:%.*]] = cir.load [[STRUCT_A_PTR]] : !cir.ptr<!cir.ptr<!ty_A>>, !cir.ptr<!ty_A> | ||
| // CHECK: [[a:%.*]] = cir.get_member [[STRUCT_A]][0] {name = "a"} : !cir.ptr<!ty_A> -> !cir.ptr<!s32i> loc(#loc70) | ||
| // CHECK: cir.return {{%.*}} : !s32i |
There was a problem hiding this comment.
The return value here is important, please check it
The PR should help us to get rid of NYI
NYI UNREACHABLE executed at clang/lib/CIR/CodeGen/CIRGenExprAgg.cpp:899Relevant OG code here:
I put
HasExplicitObjectParametersupport as a missing feature, which is a new C++23 feature.