//! This crate provides a fully written in Rust memory allocator for Vulkan, DirectX 12 and Metal.
//!
//! # Setting up the Vulkan memory allocator
//!
//! ```no_run
//! # #[cfg(feature = "vulkan")]
//! # fn main() {
//! use gpu_allocator::vulkan::*;
//! # use ash::vk;
//! # let device = todo!();
//! # let instance = todo!();
//! # let physical_device = todo!();
//!
//! let mut allocator = Allocator::new(&AllocatorCreateDesc {
//!     instance,
//!     device,
//!     physical_device,
//!     debug_settings: Default::default(),
//!     buffer_device_address: true,  // Ideally, check the BufferDeviceAddressFeatures struct.
//!     allocation_sizes: Default::default(),
//! });
//! # }
//! # #[cfg(not(feature = "vulkan"))]
//! # fn main() {}
//! ```
//!
//! # Simple Vulkan allocation example
//!
//! ```no_run
//! # #[cfg(feature = "vulkan")]
//! # fn main() {
//! use gpu_allocator::vulkan::*;
//! use gpu_allocator::MemoryLocation;
//! # use ash::vk;
//! # let device = todo!();
//! # let instance = todo!();
//! # let physical_device = todo!();
//! # let mut allocator = Allocator::new(&AllocatorCreateDesc {
//! #     instance,
//! #     device,
//! #     physical_device,
//! #     debug_settings: Default::default(),
//! #     buffer_device_address: true,  // Ideally, check the BufferDeviceAddressFeatures struct.
//! #     allocation_sizes: Default::default(),
//! # }).unwrap();
//!
//! // Setup vulkan info
//! let vk_info = vk::BufferCreateInfo::default()
//!     .size(512)
//!     .usage(vk::BufferUsageFlags::STORAGE_BUFFER);
//!
//! let buffer = unsafe { device.create_buffer(&vk_info, None) }.unwrap();
//! let requirements = unsafe { device.get_buffer_memory_requirements(buffer) };
//!
//! let allocation = allocator
//!     .allocate(&AllocationCreateDesc {
//!         name: "Example allocation",
//!         requirements,
//!         location: MemoryLocation::CpuToGpu,
//!         linear: true, // Buffers are always linear
//!         allocation_scheme: AllocationScheme::GpuAllocatorManaged,
//!     }).unwrap();
//!
//! // Bind memory to the buffer
//! unsafe { device.bind_buffer_memory(buffer, allocation.memory(), allocation.offset()).unwrap() };
//!
//! // Cleanup
//! allocator.free(allocation).unwrap();
//! unsafe { device.destroy_buffer(buffer, None) };
//! # }
//! # #[cfg(not(feature = "vulkan"))]
//! # fn main() {}
//! ```
//!
//! # Setting up the D3D12 memory allocator
//!
//! ```no_run
//! # #[cfg(feature = "d3d12")]
//! # fn main() {
//! use gpu_allocator::d3d12::*;
//! # let device = todo!();
//!
//! let mut allocator = Allocator::new(&AllocatorCreateDesc {
//!     device: ID3D12DeviceVersion::Device(device),
//!     debug_settings: Default::default(),
//!     allocation_sizes: Default::default(),
//! });
//! # }
//! # #[cfg(not(feature = "d3d12"))]
//! # fn main() {}
//! ```
//!
//! # Simple d3d12 allocation example
//!
//! ```no_run
//! # #[cfg(feature = "d3d12")]
//! # fn main() -> windows::core::Result<()> {
//! use gpu_allocator::d3d12::*;
//! use gpu_allocator::MemoryLocation;
//! # use windows::Win32::Graphics::{Dxgi, Direct3D12};
//! # let device = todo!();
//!
//! # let mut allocator = Allocator::new(&AllocatorCreateDesc {
//! #     device: ID3D12DeviceVersion::Device(device),
//! #     debug_settings: Default::default(),
//! #     allocation_sizes: Default::default(),
//! # }).unwrap();
//!
//! let buffer_desc = Direct3D12::D3D12_RESOURCE_DESC {
//!     Dimension: Direct3D12::D3D12_RESOURCE_DIMENSION_BUFFER,
//!     Alignment: 0,
//!     Width: 512,
//!     Height: 1,
//!     DepthOrArraySize: 1,
//!     MipLevels: 1,
//!     Format: Dxgi::Common::DXGI_FORMAT_UNKNOWN,
//!     SampleDesc: Dxgi::Common::DXGI_SAMPLE_DESC {
//!         Count: 1,
//!         Quality: 0,
//!     },
//!     Layout: Direct3D12::D3D12_TEXTURE_LAYOUT_ROW_MAJOR,
//!     Flags: Direct3D12::D3D12_RESOURCE_FLAG_NONE,
//! };
//! let allocation_desc = AllocationCreateDesc::from_d3d12_resource_desc(
//!     &allocator.device(),
//!     &buffer_desc,
//!     "Example allocation",
//!     MemoryLocation::GpuOnly,
//! );
//! let allocation = allocator.allocate(&allocation_desc).unwrap();
//! let mut resource: Option<Direct3D12::ID3D12Resource> = None;
//! let hr = unsafe {
//!     device.CreatePlacedResource(
//!         allocation.heap(),
//!         allocation.offset(),
//!         &buffer_desc,
//!         Direct3D12::D3D12_RESOURCE_STATE_COMMON,
//!         None,
//!         &mut resource,
//!     )
//! }?;
//!
//! // Cleanup
//! drop(resource);
//! allocator.free(allocation).unwrap();
//! # Ok(())
//! # }
//! # #[cfg(not(feature = "d3d12"))]
//! # fn main() {}
//! ```
//!
//! # Setting up the Metal memory allocator
//!
//! ```no_run
//! # #[cfg(feature = "metal")]
//! # fn main() {
//! use gpu_allocator::metal::*;
//! # let device = objc2_metal::MTLCreateSystemDefaultDevice().expect("No MTLDevice found");
//! let mut allocator = Allocator::new(&AllocatorCreateDesc {
//!     device: device.clone(),
//!     debug_settings: Default::default(),
//!     allocation_sizes: Default::default(),
//!     create_residency_set: false,
//! });
//! # }
//! # #[cfg(not(feature = "metal"))]
//! # fn main() {}
//! ```
//!
//! # Simple Metal allocation example
//!
//! ```no_run
//! # #[cfg(feature = "metal")]
//! # fn main() {
//! use gpu_allocator::metal::*;
//! use gpu_allocator::MemoryLocation;
//! # let device = objc2_metal::MTLCreateSystemDefaultDevice().expect("No MTLDevice found");
//! # let mut allocator = Allocator::new(&AllocatorCreateDesc {
//! #     device: device.clone(),
//! #     debug_settings: Default::default(),
//! #     allocation_sizes: Default::default(),
//! #    create_residency_set: false,
//! # })
//! # .unwrap();
//! let allocation_desc = AllocationCreateDesc::buffer(
//!     &device,
//!     "Example allocation",
//!     512, // size in bytes
//!     MemoryLocation::GpuOnly,
//! );
//! let allocation = allocator.allocate(&allocation_desc).unwrap();
//! # use objc2_metal::MTLHeap;
//! let heap = unsafe { allocation.heap() };
//! let resource = unsafe {
//!     heap.newBufferWithLength_options_offset(
//!         allocation.size() as usize,
//!         heap.resourceOptions(),
//!         allocation.offset() as usize,
//!     )
//! }
//! .unwrap();
//!
//! // Cleanup
//! drop(resource);
//! allocator.free(&allocation).unwrap();
//! # }
//! # #[cfg(not(feature = "metal"))]
//! # fn main() {}
//! ```
#![deny(clippy::unimplemented, clippy::unwrap_used, clippy::ok_expect)]
#![warn(
    clippy::alloc_instead_of_core,
    clippy::std_instead_of_alloc,
    clippy::std_instead_of_core
)]
#![cfg_attr(not(feature = "std"), no_std)]

#[macro_use]
extern crate alloc;

#[cfg(all(not(feature = "std"), feature = "visualizer"))]
compile_error!("Cannot enable `visualizer` feature in `no_std` environment.");

#[cfg(not(any(feature = "std", feature = "hashbrown")))]
compile_error!("Either `std` or `hashbrown` feature must be enabled");

mod result;
pub use result::*;

pub(crate) mod allocator;

pub use allocator::{AllocationReport, AllocatorReport, MemoryBlockReport};

#[cfg(feature = "visualizer")]
pub mod visualizer;

#[cfg(feature = "vulkan")]
pub mod vulkan;

#[cfg(all(windows, feature = "d3d12"))]
pub mod d3d12;

#[cfg(all(target_vendor = "apple", feature = "metal"))]
pub mod metal;

#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
pub enum MemoryLocation {
    /// The allocated resource is stored at an unknown memory location; let the driver decide what's the best location
    Unknown,
    /// Store the allocation in GPU only accessible memory - typically this is the faster GPU resource and this should be
    /// where most of the allocations live.
    GpuOnly,
    /// Memory useful for uploading data to the GPU and potentially for constant buffers
    CpuToGpu,
    /// Memory useful for CPU readback of data
    GpuToCpu,
}

#[non_exhaustive]
#[derive(Copy, Clone, Debug)]
pub struct AllocatorDebugSettings {
    /// Logs out debugging information about the various heaps the current device has on startup
    pub log_memory_information: bool,
    /// Logs out all memory leaks on shutdown with log level Warn
    pub log_leaks_on_shutdown: bool,
    /// Stores a copy of the full backtrace for every allocation made, this makes it easier to debug leaks
    /// or other memory allocations, but storing stack traces has a RAM overhead so should be disabled
    /// in shipping applications.
    #[cfg(feature = "std")]
    pub store_stack_traces: bool,
    /// Log out every allocation as it's being made with log level Debug, rather spammy so off by default
    pub log_allocations: bool,
    /// Log out every free that is being called with log level Debug, rather spammy so off by default
    pub log_frees: bool,
    /// Log out stack traces when either `log_allocations` or `log_frees` is enabled.
    #[cfg(feature = "std")]
    pub log_stack_traces: bool,
}

impl Default for AllocatorDebugSettings {
    fn default() -> Self {
        Self {
            log_memory_information: false,
            log_leaks_on_shutdown: true,
            #[cfg(feature = "std")]
            store_stack_traces: false,
            log_allocations: false,
            log_frees: false,
            #[cfg(feature = "std")]
            log_stack_traces: false,
        }
    }
}

/// The sizes of the memory blocks that the allocator will create.
///
/// Useful for tuning the allocator to your application's needs. For example most games will be fine with the default
/// values, but eg. an app might want to use smaller block sizes to reduce the amount of memory used.
///
/// Clamped between 4MB and 256MB, and rounds up to the nearest multiple of 4MB for alignment reasons.
///
/// Note that these limits only apply to shared memory blocks that can hold multiple allocations.
/// If an allocation does not fit within the corresponding maximum block size, it will be placed
/// in a dedicated memory block holding only this allocation, without limitations other than what
/// the underlying hardware and driver are able to provide.
///
/// # Fixed or growable block size
///
/// This structure represents ranges of allowed sizes for shared memory blocks.
/// By default, if the upper bounds are not extended using `with_max_*_memblock_size`,
/// the allocator will be configured to use a fixed memory block size for shared
/// allocations.
///
/// Otherwise, the allocator will pick a memory block size within the specifed
/// range, depending on the number of existing allocations for the memory
/// type.
///
/// As a rule of thumb, the allocator will start with the minimum block size
/// and double the size with each new allocation, up to the specified maximum
/// block size. This growth is tracked independently for each memory type.
/// The block size also decreases when blocks are deallocated.
///
/// # Example
///
/// ```
/// use gpu_allocator::AllocationSizes;
/// const MB: u64 = 1024 * 1024;
/// // This configuration uses fixed memory block sizes.
/// let fixed = AllocationSizes::new(256 * MB, 64 * MB);
///
/// // This configuration starts with 8MB memory blocks
/// // and grows the block size of a given memory type each
/// // time a new allocation is needed, up to a limit of
/// // 256MB for device memory and 64MB for host memory.
/// let growing = AllocationSizes::new(8 * MB, 8 * MB)
///     .with_max_device_memblock_size(256 * MB)
///     .with_max_host_memblock_size(64 * MB);
/// ```
#[derive(Clone, Copy, Debug)]
pub struct AllocationSizes {
    /// The initial size for device memory blocks.
    ///
    /// The size of new device memory blocks doubles each time a new block is needed, up to
    /// [`AllocationSizes::max_device_memblock_size`].
    ///
    /// Defaults to 256MB.
    min_device_memblock_size: u64,
    /// The maximum size for device memory blocks.
    ///
    /// Defaults to the value of [`AllocationSizes::min_device_memblock_size`].
    max_device_memblock_size: u64,
    /// The initial size for host memory blocks.
    ///
    /// The size of new host memory blocks doubles each time a new block is needed, up to
    /// [`AllocationSizes::max_host_memblock_size`].
    ///
    /// Defaults to 64MB.
    min_host_memblock_size: u64,
    /// The maximum size for host memory blocks.
    ///
    /// Defaults to the value of [`AllocationSizes::min_host_memblock_size`].
    max_host_memblock_size: u64,
}

impl AllocationSizes {
    /// Sets the minimum device and host memory block sizes.
    ///
    /// The maximum block sizes are initialized to the minimum sizes and
    /// can be increased using [`AllocationSizes::with_max_device_memblock_size`] and
    /// [`AllocationSizes::with_max_host_memblock_size`].
    pub fn new(device_memblock_size: u64, host_memblock_size: u64) -> Self {
        let device_memblock_size = Self::adjust_memblock_size(device_memblock_size, "Device");
        let host_memblock_size = Self::adjust_memblock_size(host_memblock_size, "Host");

        Self {
            min_device_memblock_size: device_memblock_size,
            max_device_memblock_size: device_memblock_size,
            min_host_memblock_size: host_memblock_size,
            max_host_memblock_size: host_memblock_size,
        }
    }

    /// Sets the maximum device memblock size, in bytes.
    pub fn with_max_device_memblock_size(mut self, size: u64) -> Self {
        self.max_device_memblock_size =
            Self::adjust_memblock_size(size, "Device").max(self.min_device_memblock_size);

        self
    }

    /// Sets the maximum host memblock size, in bytes.
    pub fn with_max_host_memblock_size(mut self, size: u64) -> Self {
        self.max_host_memblock_size =
            Self::adjust_memblock_size(size, "Host").max(self.min_host_memblock_size);

        self
    }

    fn adjust_memblock_size(size: u64, kind: &str) -> u64 {
        const MB: u64 = 1024 * 1024;

        let size = size.clamp(4 * MB, 256 * MB);

        if size % (4 * MB) == 0 {
            return size;
        }

        let val = size / (4 * MB) + 1;
        let new_size = val * 4 * MB;
        log::warn!(
            "{kind} memory block size must be a multiple of 4MB, clamping to {}MB",
            new_size / MB
        );

        new_size
    }

    /// Used internally to decide the size of a shared memory block
    /// based within the allowed range, based on the number of
    /// existing allocations. The more blocks there already are
    /// (where the requested allocation didn't fit), the larger
    /// the returned memory block size is going to be (up to
    /// `max_*_memblock_size`).
    pub(crate) fn get_memblock_size(&self, is_host: bool, count: usize) -> u64 {
        let (min_size, max_size) = if is_host {
            (self.min_host_memblock_size, self.max_host_memblock_size)
        } else {
            (self.min_device_memblock_size, self.max_device_memblock_size)
        };

        // The ranges are clamped to 4MB..256MB so we never need to
        // shift by more than 7 bits. Clamping here to avoid having
        // to worry about overflows.
        let shift = count.min(7) as u64;
        (min_size << shift).min(max_size)
    }
}

impl Default for AllocationSizes {
    fn default() -> Self {
        const MB: u64 = 1024 * 1024;
        Self {
            min_device_memblock_size: 256 * MB,
            max_device_memblock_size: 256 * MB,
            min_host_memblock_size: 64 * MB,
            max_host_memblock_size: 64 * MB,
        }
    }
}