https://bugs.llvm.org/show_bug.cgi?id=37788
Bug ID: 37788
Summary: `llvm.experimental.stackmap` is erroneously marked
`Throws`?
Product: new-bugs
Version: unspecified
Hardware: All
OS: All
Status: NEW
Severity: normal
Priority: P
Component: new bugs
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Over the last few days I have been experimenting with inserting stackmaps from
the Rust compiler.
After inserting some `call`s (not `invoke`s) to llvm.experimental.stackmap into
rustc's llvm
codegen, and then allowing the compiler to build itself using this change, I
was surprised to see:
```
Cannot invoke an intrinsic other than donothing, patchpoint, statepoint,
coro_resume or coro_destroy
invoke void (i64, i32, ...%) @llvm.experimental.stackmap(i64 1, i32 0)
to label 13
```
As I understand, this is LLVM's way of saying, you can't `invoke` (expecting a
possible exception) an intrinsic which cannot raise an exception.
Having spoke to one of the Rust developers about this, and after trying
a few things, it seems that LLVM's inlining pass is itself allowing `call`s to
the stackmap intrinsic to be translated to `invoke`s. Then the verifier rejects
the resulting code.
We suspect that the fix is as follows:
```
diff --git a/include/llvm/IR/Intrinsics.td b/include/llvm/IR/Intrinsics.td
index a2a1f26292c..9bbe22645ac 100644
--- a/include/llvm/IR/Intrinsics.td
+++ b/include/llvm/IR/Intrinsics.td
@@ -720,8 +720,7 @@ def int_invariant_group_barrier :
Intrinsic<[llvm_anyptr_ty],
//===------------------------ Stackmap Intrinsics
-------------------------===//
//
def int_experimental_stackmap : Intrinsic<[],
- [llvm_i64_ty, llvm_i32_ty, llvm_vararg_ty],
- [Throws]>;
+ [llvm_i64_ty, llvm_i32_ty, llvm_vararg_ty],
[]>;
def int_experimental_patchpoint_void : Intrinsic<[],
[llvm_i64_ty, llvm_i32_ty,
llvm_ptr_ty, llvm_i32_ty,
```
In other words, remove `Throws` from the signature of
`int_experimental_stackmap`.
With this patch applied, I managed to run a stage 2 rustc build without the
verifier getting upset.
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
[email protected]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
