@_assemblyVision doesn't seem to see allocations from "partial apply forwarder for reabstraction thunk helper"

[weissi]

Description

The NIOThreadPool._blockingWaitForWork() function in SwiftNIO is supposed to be @inlinable, the desired type is this:

public typealias WorkItem = @Sendable (WorkItemState) -> Void
public enum WorkItemState: Sendable {
    case active
    case cancelled
}

@inline(never)  // <-- adding this adds an allocation
@inlinable
internal func _blockingWaitForWork(identifier: Int) -> (item: WorkItem, state: WorkItemState)? {

Unfortunately, this makes the function allocate each time on its return because it creates a
partial apply forwarder for reabstraction thunk helper from @escaping @callee_guaranteed @Sendable (@unowned NIOPosix.NIOThreadPool.WorkItemState) -> () to @escaping @callee_guaranteed @Sendable (@in_guaranteed NIOPosix.NIOThreadPool.WorkItemState) -> (@out ()) at <compiler-generated>.

Therefore, the function currently doesn't have the @inline(never) but it should because it's a "marker function" that we want to always see in stack traces, even ones that can't properly deal with inlined functions.

Anyway, @_assemblyVision doesn't seem to show the allocation for the partial apply forwarder for reabstraction thunk helper.

Problematic assembly:

test_read_10000_chunks_from_file`NIOThreadPool._blockingWaitForWork(identifier:):
test_read_10000_chunks_from_file[0x10015d7c0] <+0>:   sub    sp, sp, #0x50
test_read_10000_chunks_from_file[0x10015d7c4] <+4>:   stp    x24, x23, [sp, #0x10]
test_read_10000_chunks_from_file[0x10015d7c8] <+8>:   stp    x22, x21, [sp, #0x20]
test_read_10000_chunks_from_file[0x10015d7cc] <+12>:  stp    x20, x19, [sp, #0x30]
test_read_10000_chunks_from_file[0x10015d7d0] <+16>:  stp    x29, x30, [sp, #0x40]
test_read_10000_chunks_from_file[0x10015d7d4] <+20>:  add    x29, sp, #0x40
test_read_10000_chunks_from_file[0x10015d7d8] <+24>:  mov    x19, x20
test_read_10000_chunks_from_file[0x10015d7dc] <+28>:  mov    x21, x0
test_read_10000_chunks_from_file[0x10015d7e0] <+32>:  ldr    x20, [x20, #0x10]
test_read_10000_chunks_from_file[0x10015d7e4] <+36>:  strb   wzr, [sp, #0xf]
test_read_10000_chunks_from_file[0x10015d7e8] <+40>:  add    x0, sp, #0xf
test_read_10000_chunks_from_file[0x10015d7ec] <+44>:  bl     0x10002d900    ; NIOConcurrencyHelpers.ConditionLock.lock(whenValue: T) -> () at lock.swift:193
test_read_10000_chunks_from_file[0x10015d7f0] <+48>:  mov    x0, x19
test_read_10000_chunks_from_file[0x10015d7f4] <+52>:  mov    x1, x21
test_read_10000_chunks_from_file[0x10015d7f8] <+56>:  bl     0x10015d89c    ; closure #1 () -> (unlockWith: NIOPosix.NIOThreadPool._WorkState, result: Swift.Optional<(item: @Sendable (NIOPosix.NIOThreadPool.WorkItemState) -> (), state: NIOPosix.NIOThreadPool.WorkItemState)>) in NIOPosix.NIOThreadPool._blockingWaitForWork(identifier: Swift.Int) -> Swift.Optional<(item: @Sendable (NIOPosix.NIOThreadPool.WorkItemState) -> (), state: NIOPosix.NIOThreadPool.WorkItemState)> at NIOThreadPool.swift:258
test_read_10000_chunks_from_file[0x10015d7fc] <+60>:  mov    x19, x1
test_read_10000_chunks_from_file[0x10015d800] <+64>:  and    w24, w0, #0x1
test_read_10000_chunks_from_file[0x10015d804] <+68>:  cbz    x1, 0x10015d86c ; <+172> [inlined] generic specialization <serialized, NIOPosix.NIOThreadPool._WorkState> of NIOConcurrencyHelpers.ConditionLock._unlock(with: Swift.Optional<τ_0_0>) -> () at NIOThreadPool.swift:533:9
test_read_10000_chunks_from_file[0x10015d808] <+72>:  mov    x21, x2
test_read_10000_chunks_from_file[0x10015d80c] <+76>:  mov    x22, x3
test_read_10000_chunks_from_file[0x10015d810] <+80>:  adrp   x0, 209
test_read_10000_chunks_from_file[0x10015d814] <+84>:  add    x0, x0, #0x188 ; block_descriptor.40 + 480
test_read_10000_chunks_from_file[0x10015d818] <+88>:  mov    w1, #0x20 ; =32 
test_read_10000_chunks_from_file[0x10015d81c] <+92>:  mov    w2, #0x7 ; =7 
test_read_10000_chunks_from_file[0x10015d820] <+96>:  bl     0x1001bd5e8    ; symbol stub for: swift_allocObject
test_read_10000_chunks_from_file[0x10015d824] <+100>: mov    x23, x0
test_read_10000_chunks_from_file[0x10015d828] <+104>: stp    x19, x21, [x0, #0x10]
test_read_10000_chunks_from_file[0x10015d82c] <+108>: and    w19, w22, #0x1
test_read_10000_chunks_from_file[0x10015d830] <+112>: strb   w24, [sp, #0xe]
test_read_10000_chunks_from_file[0x10015d834] <+116>: add    x0, sp, #0xe
test_read_10000_chunks_from_file[0x10015d838] <+120>: bl     0x10002dd64    ; NIOConcurrencyHelpers.ConditionLock.unlock(withValue: T) -> () at lock.swift:293
test_read_10000_chunks_from_file[0x10015d83c] <+124>: adrp   x0, 209
test_read_10000_chunks_from_file[0x10015d840] <+128>: add    x0, x0, #0x160 ; block_descriptor.40 + 440
test_read_10000_chunks_from_file[0x10015d844] <+132>: mov    w1, #0x20 ; =32 
test_read_10000_chunks_from_file[0x10015d848] <+136>: mov    w2, #0x7 ; =7 
test_read_10000_chunks_from_file[0x10015d84c] <+140>: bl     0x1001bd5e8    ; symbol stub for: swift_allocObject  <<<<<<< JW: PROBLEMATIC ALLOC
test_read_10000_chunks_from_file[0x10015d850] <+144>: mov    x1, x0
test_read_10000_chunks_from_file[0x10015d854] <+148>: adrp   x8, 2
test_read_10000_chunks_from_file[0x10015d858] <+152>: add    x8, x8, #0x680 ; partial apply forwarder for reabstraction thunk helper from @escaping @callee_guaranteed @Sendable (@unowned NIOPosix.NIOThreadPool.WorkItemState) -> () to @escaping @callee_guaranteed @Sendable (@in_guaranteed NIOPosix.NIOThreadPool.WorkItemState) -> (@out ()) at <compiler-generated>
test_read_10000_chunks_from_file[0x10015d85c] <+156>: stp    x8, x23, [x0, #0x10]
test_read_10000_chunks_from_file[0x10015d860] <+160>: adrp   x0, 2
test_read_10000_chunks_from_file[0x10015d864] <+164>: add    x0, x0, #0x64c ; partial apply forwarder for reabstraction thunk helper from @escaping @callee_guaranteed @Sendable (@in_guaranteed NIOPosix.NIOThreadPool.WorkItemState) -> (@out ()) to @escaping @callee_guaranteed @Sendable (@unowned NIOPosix.NIOThreadPool.WorkItemState) -> () at <compiler-generated>
test_read_10000_chunks_from_file[0x10015d868] <+168>: b      0x10015d880    ; <+192> at NIOThreadPool.swift:300:5
test_read_10000_chunks_from_file[0x10015d86c] <+172>: strb   w24, [sp, #0xe]
test_read_10000_chunks_from_file[0x10015d870] <+176>: add    x0, sp, #0xe
test_read_10000_chunks_from_file[0x10015d874] <+180>: bl     0x10002dd64    ; NIOConcurrencyHelpers.ConditionLock.unlock(withValue: T) -> () at lock.swift:293
test_read_10000_chunks_from_file[0x10015d878] <+184>: mov    x0, #0x0 ; =0 
test_read_10000_chunks_from_file[0x10015d87c] <+188>: mov    x1, #0x0 ; =0 
test_read_10000_chunks_from_file[0x10015d880] <+192>: mov    x2, x19
test_read_10000_chunks_from_file[0x10015d884] <+196>: ldp    x29, x30, [sp, #0x40]
test_read_10000_chunks_from_file[0x10015d888] <+200>: ldp    x20, x19, [sp, #0x30]
test_read_10000_chunks_from_file[0x10015d88c] <+204>: ldp    x22, x21, [sp, #0x20]
test_read_10000_chunks_from_file[0x10015d890] <+208>: ldp    x24, x23, [sp, #0x10]
test_read_10000_chunks_from_file[0x10015d894] <+212>: add    sp, sp, #0x50
test_read_10000_chunks_from_file[0x10015d898] <+216>: ret    

Reproduction

git clone https://github.com/apple/swift-nio && cd swift-nio
git reset --hard 70dfce82b0c892c06e5730a43dc3fd717d4c1fbd

## add the `@inline(never)`
git apply <<"EOF"
diff --git a/Sources/NIOPosix/NIOThreadPool.swift b/Sources/NIOPosix/NIOThreadPool.swift
index 6d3bc08f7..524809a2c 100644
--- a/Sources/NIOPosix/NIOThreadPool.swift
+++ b/Sources/NIOPosix/NIOThreadPool.swift
@@ -253,6 +253,7 @@ public final class NIOThreadPool {
     // This function is one of those and giving it a consistent name makes it much easier to remove from the profiles
     // when only interested in on-CPU work.
     @inlinable
+    @inline(never)
     internal func _blockingWaitForWork(identifier: Int) -> (item: WorkItem, state: WorkItemState)? {
         self._conditionLock.withLock(when: .hasWork) {
             () -> (unlockWith: _WorkState, result: (WorkItem, WorkItemState)?) in
EOF

swift build -c release

Expected behavior

Does not add a swift_allocObject

Environment

Swift 5.9 through 6.0 on Linux and macOS

Additional information

No response

[weissi]

CC @gottesmm, I assume @_assemblyVision is supposed to show the partial apply forwarder for reabstraction thunk helper, no?