Crash using Metal buffers - multithreading

Question

I use a background thread to create Metal buffers:
var vertices:UnsafeMutablePointer<Float>?=nil
var indexes:UnsafeMutablePointer<UInt32>?=nil
var vertexCount: UInt32=UInt32()
var indexCount: UInt32=UInt32()
triangulatePoints(hull, &vertices, &indexes, &vertexCount, &indexCount)
defer {
free(vertices)
free(indexes)
}
if let vertexBuffer=device.makeBuffer(bytes: vertices!,
length: Int(vertexCount)*MemoryLayout<Float>.size*3,
options: .storageModeShared),
let indexBuffer=device.makeBuffer(bytes: indexes!,
length: Int(indexCount)*MemoryLayout<UInt32>.size,
options: .storageModeShared)
{
metalObject=MetalObject(vertices: vertexBuffer, indices: indexBuffer)
}
The MetalObjects are temporarily stored in a dictionary and later accumulated in arrays and the array is returned. This happens from within a dispatch queue (number #5 in the Debug stack trace below):
CustomOperationQueue.sharedInstance.addSerialOperation {
//This calls the device.makeBuffers() and stores them in a dictionary owned by the image
image.constructMetalBuffers(withPrecision: precision, device: device)
//The MetalObjects are accumulated in arrays. Sorted by their draw type, such as instanced or not. One array of each type.
let metalLayer:MetalImageLayer=image.layerWithDelegate(device: device, delegate: self)
OperationQueue.main.addOperation {
//The MTKView delegate observes this property and triggers a redraw when it changes
self.metalLayer=metalLayer
}
The MetalObjects are simple containers, which are also stored in other containers:
class MetalObject: NSObject {
var vertexBuffer: MTLBuffer
var indexBuffer: MTLBuffer
init(vertices: MTLBuffer, indices:MTLBuffer) {
vertexBuffer=vertices
indexBuffer=indices
}
}
class MetalImageLayer: NSObject {
var layers=[MetalPolarityLayer]()
}
class MetalPolarityLayer: NSObject {
var elements=[MetalObject]() //For general objects that need to be drawn piece by piece
var regions=[MetalObject]() //For region objects that can be instanced
}
In Debug mode, I often get crashes on the line metalObject=MetalObject(vertices: vertexBuffer, indices: indexBuffer) with a EXC_BAD_ACCESS (code=1, address=0x40dedeadbec8). The stack trace looks like:
#0 0x000000010166a6e2 in swift_unknownRetain ()
#1 0x000000010005bf5f in StrokeLine.constructMetalBuffers(withPrecision:device:) at Image+Metal.swift:445
#2 0x000000010005c8b2 in #objc StrokeLine.constructMetalBuffers(withPrecision:device:) ()
#3 0x0000000100050b4d in Image.constructMetalBuffers(withPrecision:device:) at Image+Metal.swift:40
#4 0x0000000100051082 in #objc Image.constructMetalBuffers(withPrecision:device:) ()
#5 0x0000000100067eef in closure #1 in WindowLayer.observeValue(forKeyPath:of:change:context:) at WindowLayer.swift:181
Inspecting the vertexBuffer from the debugger then gives: po vertexBuffer
error: Execution was interrupted, reason: EXC_BAD_ACCESS (code=1, address=0x2e7466697770).
On the other hand in Release mode, I sometimes get crashes on
Crashed Thread: 0 Dispatch queue: com.apple.main-thread
Exception Type: EXC_CRASH (SIGABRT)
Exception Codes: 0x0000000000000000, 0x0000000000000000
Exception Note: EXC_CORPSE_NOTIFY
Application Specific Information:
abort() called
Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0 libsystem_kernel.dylib 0x00007fff60084fce __pthread_kill + 10
1 libsystem_pthread.dylib 0x00007fff601c2150 pthread_kill + 333
2 libsystem_c.dylib 0x00007fff5ffe132a abort + 127
3 com.apple.IOAccelerator 0x00007fff5279a1ea ioAccelResourceListAddNewGroupAndResource + 221
4 com.apple.IOAccelerator 0x00007fff52799d19 IOAccelResourceListAddResource + 304
5 com.apple.driver.AppleIntelKBLGraphicsMTLDriver 0x00007fff32aa02ef IGBufferMemory<31ul>::listId(unsigned int) + 85
6 com.apple.driver.AppleIntelKBLGraphicsMTLDriver 0x00007fff32aa654b ShaderBindingStage::updateUAVTable(unsigned int, unsigned int, bool) + 459
7 com.apple.driver.AppleIntelKBLGraphicsMTLDriver 0x00007fff32a9fe0d ShaderBindingStage::updateBindingTableData() + 1631
8 com.apple.driver.AppleIntelKBLGraphicsMTLDriver 0x00007fff32a9f70b ShaderBindingStage::writeTableIf() + 55
9 com.apple.driver.AppleIntelKBLGraphicsMTLDriver 0x00007fff32a9cc11 IGAccelRenderCommandEncoder::programPipeline(sPrimitiveData const&) + 4685
10 com.apple.driver.AppleIntelKBLGraphicsMTLDriver 0x00007fff32a9b7b4 IGAccelRenderCommandEncoder::drawIndexedPrimitives(MTLPrimitiveType, unsigned int, MTLIndexType, MTLIGAccelBuffer*, unsigned int, unsigned int, unsigned int, unsigned int) + 710
11 com.apple.driver.AppleIntelKBLGraphicsMTLDriver 0x00007fff32a9b40c -[MTLIGAccelRenderCommandEncoder drawIndexedPrimitives:indexCount:indexType:indexBuffer:indexBufferOffset:] + 324
No further information is logged about what might cause the crash.
Using Instruments to find Zombies gives no results.
How should I go about finding out what goes wrong here?