nikic wrote:
After much suffering, I managed to do a stage2+thinlto build in a Windows VM. I
believe this is the root cause:
```llvm
; RUN: opt -S -passes=memcpyopt < %s
declare void @init(ptr)
declare void @update(ptr)
define i32 @test() {
%a = alloca i32
%b = alloca i32
call void @init(ptr captures(none) %a)
call void @llvm.memcpy(ptr %b, ptr %a, i64 4, i1 false)
call void @update(ptr captures(ret: address, provenance) %b)
%v = load i32, ptr %a
ret i32 %v
}
```
The `%a` and `%b` allocas get combined and memcpy optimized out.
https://github.com/llvm/llvm-project/pull/125880
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
