regression: dead heap allocations aren't optimized out anymore

[oli-obk]

#22159 was closed after a llvm update (#22526). It used to work (I remember I tried it out). Now it doesn't work anymore.

fn main() {
    let _ = Box::new(42);
}

http://is.gd/Wekr7w

LLVM-IR:

; ModuleID = 'rust_out.0.rs'
target datalayout = "e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

; Function Attrs: uwtable
define internal void @_ZN4main20h657e6a8d1dc11120eaaE() unnamed_addr #0 {
entry-block:
  %0 = tail call i8* @je_mallocx(i64 4, i32 0)
  %1 = icmp eq i8* %0, null
  br i1 %1, label %then-block-57-.i.i, label %"_ZN5boxed12Box$LT$T$GT$3new19h823444268625886800E.exit"

then-block-57-.i.i:                               ; preds = %entry-block
  tail call void @_ZN3oom20he7076b57c17ed7c6HYaE()
  unreachable

"_ZN5boxed12Box$LT$T$GT$3new19h823444268625886800E.exit": ; preds = %entry-block
  %2 = bitcast i8* %0 to i32*
  store i32 42, i32* %2, align 4
  %3 = icmp eq i8* %0, inttoptr (i64 2097865012304223517 to i8*)
  br i1 %3, label %"_ZN14Box$LT$i32$GT$8drop.86517h1e7c6ecb62969b35E.exit", label %cond.i

cond.i:                                           ; preds = %"_ZN5boxed12Box$LT$T$GT$3new19h823444268625886800E.exit"
  tail call void @je_sdallocx(i8* %0, i64 4, i32 0)
  br label %"_ZN14Box$LT$i32$GT$8drop.86517h1e7c6ecb62969b35E.exit"

"_ZN14Box$LT$i32$GT$8drop.86517h1e7c6ecb62969b35E.exit": ; preds = %"_ZN5boxed12Box$LT$T$GT$3new19h823444268625886800E.exit", %cond.i
  ret void
}

define i64 @main(i64, i8**) unnamed_addr #1 {
top:
  %2 = tail call i64 @_ZN2rt10lang_start20he050f8de3bcc02b7VRIE(i8* bitcast (void ()* @_ZN4main20h657e6a8d1dc11120eaaE to i8*), i64 %0, i8** %1)
  ret i64 %2
}

declare i64 @_ZN2rt10lang_start20he050f8de3bcc02b7VRIE(i8*, i64, i8**) unnamed_addr #1

declare noalias i8* @je_mallocx(i64, i32) unnamed_addr #1

; Function Attrs: cold noinline noreturn
declare void @_ZN3oom20he7076b57c17ed7c6HYaE() unnamed_addr #2

declare void @je_sdallocx(i8*, i64, i32) unnamed_addr #1

attributes #0 = { uwtable "split-stack" }
attributes #1 = { "split-stack" }
attributes #2 = { cold noinline noreturn "split-stack" }

[oli-obk]

hmmm, I'm not so sure if it's a regression. It still works for vectors: http://is.gd/yNAFF6 . Maybe it never worked for boxes?

[steveklabnik]

So, where do we go with this ticket? Was/is this a regression? is it worth tracking?

[dotdash]

@steveklabnik Probably a regression from when we moved from actual zeroing-drop to using the non-zero drop pattern. The check for that pattern basically says "leak this if the adress matches the drop pattern", so LLVM cannot remove the allocation because that would change semantics. Dynamic drop will fix this, and I think we should keep this open to track that, just to make sure.

[wesleywiser]

This appears to be fixed: https://is.gd/kT0Gs9 Is this worth creating some kind of regression test for?

[oli-obk]

We could test it by adding a run pass test that uses a custom allocator (which doesn't allocate, but panics).

[bluss]

tentatively needstest, then. Maybe a src/test/codegen test?

[nagisa]

Seems to not work again on nightly.

[est31]

I've used @Mark-Simulacrum 's bisection tool to find out the regressing commit. It was the LLVM 4.0 upgrade, commit 0777c75 .