NewSigma wrote:
After further consideration, I think this is not a satisfactory solution to the
issue. I may revisit it if I develop a concrete plan. Apologies for any
inconvenience.
https://github.com/llvm/llvm-project/pull/140548
___
cfe-commits ma
https://github.com/NewSigma closed
https://github.com/llvm/llvm-project/pull/140548
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
@@ -1343,6 +1343,11 @@ llvm::Value
*CodeGenFunction::EmitLifetimeStart(llvm::TypeSize Size,
if (!ShouldEmitLifetimeMarkers)
return nullptr;
+ // No lifetimes on promise alloca, or middle end passes will assume promise
+ // dead after lifetime.end, leading to mis-optim
https://github.com/NewSigma updated
https://github.com/llvm/llvm-project/pull/140548
>From d2e1a8d8a650f2a38387c500f942a0c6722c8a84 Mon Sep 17 00:00:00 2001
From: NewSigma
Date: Mon, 19 May 2025 22:12:45 +0800
Subject: [PATCH 1/2] Do not emit lifetime intrinsics for coro promise alloca
---
cl
NewSigma wrote:
Here is a example
``` LLVM
define i32 @fn() {
entry:
%__promise = alloca i32, align 4
%id = call token @llvm.coro.id(i32 16, ptr nonnull %__promise, ptr nonnull
@fn, ptr null)
%hdl = call ptr @llvm.coro.begin(token %id, ptr null) #14
%promise.addr = call ptr @llvm.coro.p
https://github.com/NewSigma created
https://github.com/llvm/llvm-project/pull/140548
Coro promise has same lifetime as coro frame. It do not need explicit lifetime
guarding. If we add lifetimes to it, middle end passes may assume promise dead
after lifetime.end, leading to mis-optimizations.
