1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993
#![deny(clippy::unimplemented, clippy::unwrap_used, clippy::ok_expect)]
#[cfg(feature = "visualizer")]
mod visualizer;
#[cfg(feature = "visualizer")]
pub use visualizer::AllocatorVisualizer;
use std::{backtrace::Backtrace, fmt, marker::PhantomData, sync::Arc};
use ash::vk;
use log::{debug, Level};
use super::allocator::{self, AllocationType};
use crate::{
allocator::fmt_bytes, AllocationError, AllocationSizes, AllocatorDebugSettings, MemoryLocation,
Result,
};
#[derive(Copy, Clone, Debug, Eq, PartialEq)]
pub enum AllocationScheme {
/// Perform a dedicated, driver-managed allocation for the given buffer, allowing
/// it to perform optimizations on this type of allocation.
DedicatedBuffer(vk::Buffer),
/// Perform a dedicated, driver-managed allocation for the given image, allowing
/// it to perform optimizations on this type of allocation.
DedicatedImage(vk::Image),
/// The memory for this resource will be allocated and managed by gpu-allocator.
GpuAllocatorManaged,
}
#[derive(Clone, Debug)]
pub struct AllocationCreateDesc<'a> {
/// Name of the allocation, for tracking and debugging purposes
pub name: &'a str,
/// Vulkan memory requirements for an allocation
pub requirements: vk::MemoryRequirements,
/// Location where the memory allocation should be stored
pub location: MemoryLocation,
/// If the resource is linear (buffer / linear texture) or a regular (tiled) texture.
pub linear: bool,
/// Determines how this allocation should be managed.
pub allocation_scheme: AllocationScheme,
}
/// Wrapper type to only mark a raw pointer [`Send`] + [`Sync`] without having to
/// mark the entire [`Allocation`] as such, instead relying on the compiler to
/// auto-implement this or fail if fields are added that violate this constraint
#[derive(Clone, Copy, Debug)]
pub(crate) struct SendSyncPtr(std::ptr::NonNull<std::ffi::c_void>);
// Sending is fine because mapped_ptr does not change based on the thread we are in
unsafe impl Send for SendSyncPtr {}
// Sync is also okay because Sending &Allocation is safe: a mutable reference
// to the data in mapped_ptr is never exposed while `self` is immutably borrowed.
// In order to break safety guarantees, the user needs to `unsafe`ly dereference
// `mapped_ptr` themselves.
unsafe impl Sync for SendSyncPtr {}
pub struct AllocatorCreateDesc {
pub instance: ash::Instance,
pub device: ash::Device,
pub physical_device: ash::vk::PhysicalDevice,
pub debug_settings: AllocatorDebugSettings,
pub buffer_device_address: bool,
pub allocation_sizes: AllocationSizes,
}
/// A piece of allocated memory.
///
/// Could be contained in its own individual underlying memory object or as a sub-region
/// of a larger allocation.
///
/// # Copying data into a CPU-mapped [`Allocation`]
///
/// You'll very likely want to copy data into CPU-mapped [`Allocation`]s in order to send that data to the GPU.
/// Doing this data transfer correctly without invoking undefined behavior can be quite fraught and non-obvious<sup>[\[1\]]</sup>.
///
/// To help you do this correctly, [`Allocation`] implements [`presser::Slab`], which means you can directly
/// pass it in to many of `presser`'s [helper functions] (for example, [`copy_from_slice_to_offset`]).
///
/// In most cases, this will work perfectly. However, note that if you try to use an [`Allocation`] as a
/// [`Slab`] and it is not valid to do so (if it is not CPU-mapped or if its `size > isize::MAX`),
/// you will cause a panic. If you aren't sure about these conditions, you may use [`Allocation::try_as_mapped_slab`].
///
/// ## Example
///
/// Say we've created an [`Allocation`] called `my_allocation`, which is CPU-mapped.
/// ```ignore
/// let mut my_allocation: Allocation = my_allocator.allocate(...)?;
/// ```
///
/// And we want to fill it with some data in the form of a `my_gpu_data: Vec<MyGpuVector>`, defined as such:
///
/// ```ignore
/// // note that this is size(12) but align(16), thus we have 4 padding bytes.
/// // this would mean a `&[MyGpuVector]` is invalid to cast as a `&[u8]`, but
/// // we can still use `presser` to copy it directly in a valid manner.
/// #[repr(C, align(16))]
/// #[derive(Clone, Copy)]
/// struct MyGpuVertex {
/// x: f32,
/// y: f32,
/// z: f32,
/// }
///
/// let my_gpu_data: Vec<MyGpuData> = make_vertex_data();
/// ```
///
/// Depending on how the data we're copying will be used, the vulkan device may have a minimum
/// alignment requirement for that data:
///
/// ```ignore
/// let min_gpu_align = my_vulkan_device_specifications.min_alignment_thing;
/// ```
///
/// Finally, we can use [`presser::copy_from_slice_to_offset_with_align`] to perform the copy,
/// simply passing `&mut my_allocation` since [`Allocation`] implements [`Slab`].
///
/// ```ignore
/// let copy_record = presser::copy_from_slice_to_offset_with_align(
/// &my_gpu_data[..], // a slice containing all elements of my_gpu_data
/// &mut my_allocation, // our Allocation
/// 0, // start as close to the beginning of the allocation as possible
/// min_gpu_align, // the minimum alignment we queried previously
/// )?;
/// ```
///
/// It's important to note that the data may not have actually been copied starting at the requested
/// `start_offset` (0 in the example above) depending on the alignment of the underlying allocation
/// as well as the alignment requirements of `MyGpuVector` and the `min_gpu_align` we passed in. Thus,
/// we can query the `copy_record` for the actual starting offset:
///
/// ```ignore
/// let actual_data_start_offset = copy_record.copy_start_offset;
/// ```
///
/// ## Safety
///
/// It is technically not fully safe to use an [`Allocation`] as a [`presser::Slab`] because we can't validate that the
/// GPU is not using the data in the buffer while `self` is borrowed. However, trying
/// to validate this statically is really hard and the community has basically decided that
/// requiring `unsafe` for functions like this creates too much "unsafe-noise", ultimately making it
/// harder to debug more insidious unsafety that is unrelated to GPU-CPU sync issues.
///
/// So, as would always be the case, you must ensure the GPU
/// is not using the data in `self` for the duration that you hold the returned [`MappedAllocationSlab`].
///
/// [`Slab`]: presser::Slab
/// [`copy_from_slice_to_offset`]: presser::copy_from_slice_to_offset
/// [helper functions]: presser#functions
/// [\[1\]]: presser#motivation
#[derive(Debug)]
pub struct Allocation {
chunk_id: Option<std::num::NonZeroU64>,
offset: u64,
size: u64,
memory_block_index: usize,
memory_type_index: usize,
device_memory: vk::DeviceMemory,
mapped_ptr: Option<SendSyncPtr>,
dedicated_allocation: bool,
memory_properties: vk::MemoryPropertyFlags,
name: Option<Box<str>>,
}
impl Allocation {
/// Tries to borrow the CPU-mapped memory that backs this allocation as a [`presser::Slab`], which you can then
/// use to safely copy data into the raw, potentially-uninitialized buffer.
/// See [the documentation of Allocation][Allocation#example] for an example of this.
///
/// Returns [`None`] if `self.mapped_ptr()` is `None`, or if `self.size()` is greater than `isize::MAX` because
/// this could lead to undefined behavior.
///
/// Note that [`Allocation`] implements [`Slab`] natively, so you can actually pass this allocation as a [`Slab`]
/// directly. However, if `self` is not actually a valid [`Slab`] (this function would return `None` as described above),
/// then trying to use it as a [`Slab`] will panic.
///
/// # Safety
///
/// See the note about safety in [the documentation of Allocation][Allocation#safety]
///
/// [`Slab`]: presser::Slab
// best to be explicit where the lifetime is coming from since we're doing unsafe things
// and relying on an inferred liftime type in the PhantomData below
#[allow(clippy::needless_lifetimes)]
pub fn try_as_mapped_slab<'a>(&'a mut self) -> Option<MappedAllocationSlab<'a>> {
let mapped_ptr = self.mapped_ptr()?.cast().as_ptr();
if self.size > isize::MAX as _ {
return None;
}
// this will always succeed since size is <= isize::MAX which is < usize::MAX
let size = self.size as usize;
Some(MappedAllocationSlab {
_borrowed_alloc: PhantomData,
mapped_ptr,
size,
})
}
pub fn chunk_id(&self) -> Option<std::num::NonZeroU64> {
self.chunk_id
}
///Returns the [`vk::MemoryPropertyFlags`] of this allocation.
pub fn memory_properties(&self) -> vk::MemoryPropertyFlags {
self.memory_properties
}
/// Returns the [`vk::DeviceMemory`] object that is backing this allocation.
/// This memory object can be shared with multiple other allocations and shouldn't be freed (or allocated from)
/// without this library, because that will lead to undefined behavior.
///
/// # Safety
/// The result of this function can safely be used to pass into [`ash::Device::bind_buffer_memory()`],
/// [`ash::Device::bind_image_memory()`] etc. It is exposed for this reason. Keep in mind to also
/// pass [`Self::offset()`] along to those.
pub unsafe fn memory(&self) -> vk::DeviceMemory {
self.device_memory
}
/// Returns [`true`] if this allocation is using a dedicated underlying allocation.
pub fn is_dedicated(&self) -> bool {
self.dedicated_allocation
}
/// Returns the offset of the allocation on the [`vk::DeviceMemory`].
/// When binding the memory to a buffer or image, this offset needs to be supplied as well.
pub fn offset(&self) -> u64 {
self.offset
}
/// Returns the size of the allocation
pub fn size(&self) -> u64 {
self.size
}
/// Returns a valid mapped pointer if the memory is host visible, otherwise it will return None.
/// The pointer already points to the exact memory region of the suballocation, so no offset needs to be applied.
pub fn mapped_ptr(&self) -> Option<std::ptr::NonNull<std::ffi::c_void>> {
self.mapped_ptr.map(|SendSyncPtr(p)| p)
}
/// Returns a valid mapped slice if the memory is host visible, otherwise it will return None.
/// The slice already references the exact memory region of the allocation, so no offset needs to be applied.
pub fn mapped_slice(&self) -> Option<&[u8]> {
self.mapped_ptr().map(|ptr| unsafe {
std::slice::from_raw_parts(ptr.cast().as_ptr(), self.size as usize)
})
}
/// Returns a valid mapped mutable slice if the memory is host visible, otherwise it will return None.
/// The slice already references the exact memory region of the allocation, so no offset needs to be applied.
pub fn mapped_slice_mut(&mut self) -> Option<&mut [u8]> {
self.mapped_ptr().map(|ptr| unsafe {
std::slice::from_raw_parts_mut(ptr.cast().as_ptr(), self.size as usize)
})
}
pub fn is_null(&self) -> bool {
self.chunk_id.is_none()
}
}
impl Default for Allocation {
fn default() -> Self {
Self {
chunk_id: None,
offset: 0,
size: 0,
memory_block_index: !0,
memory_type_index: !0,
device_memory: vk::DeviceMemory::null(),
mapped_ptr: None,
memory_properties: vk::MemoryPropertyFlags::empty(),
name: None,
dedicated_allocation: false,
}
}
}
/// A wrapper struct over a borrowed [`Allocation`] that infallibly implements [`presser::Slab`].
///
/// This type should be acquired by calling [`Allocation::try_as_mapped_slab`].
pub struct MappedAllocationSlab<'a> {
_borrowed_alloc: PhantomData<&'a mut Allocation>,
mapped_ptr: *mut u8,
size: usize,
}
// SAFETY: See the safety comment of Allocation::as_mapped_slab above.
unsafe impl<'a> presser::Slab for MappedAllocationSlab<'a> {
fn base_ptr(&self) -> *const u8 {
self.mapped_ptr
}
fn base_ptr_mut(&mut self) -> *mut u8 {
self.mapped_ptr
}
fn size(&self) -> usize {
self.size
}
}
// SAFETY: See the safety comment of Allocation::as_mapped_slab above.
unsafe impl presser::Slab for Allocation {
fn base_ptr(&self) -> *const u8 {
self.mapped_ptr
.expect("tried to use a non-mapped Allocation as a Slab")
.0
.as_ptr()
.cast()
}
fn base_ptr_mut(&mut self) -> *mut u8 {
self.mapped_ptr
.expect("tried to use a non-mapped Allocation as a Slab")
.0
.as_ptr()
.cast()
}
fn size(&self) -> usize {
if self.size > isize::MAX as _ {
panic!("tried to use an Allocation with size > isize::MAX as a Slab")
}
// this will always work if the above passed
self.size as usize
}
}
#[derive(Debug)]
pub(crate) struct MemoryBlock {
pub(crate) device_memory: vk::DeviceMemory,
pub(crate) size: u64,
pub(crate) mapped_ptr: Option<SendSyncPtr>,
pub(crate) sub_allocator: Box<dyn allocator::SubAllocator>,
#[cfg(feature = "visualizer")]
pub(crate) dedicated_allocation: bool,
}
impl MemoryBlock {
fn new(
device: &ash::Device,
size: u64,
mem_type_index: usize,
mapped: bool,
buffer_device_address: bool,
allocation_scheme: AllocationScheme,
requires_personal_block: bool,
) -> Result<Self> {
let device_memory = {
let alloc_info = vk::MemoryAllocateInfo::builder()
.allocation_size(size)
.memory_type_index(mem_type_index as u32);
let allocation_flags = vk::MemoryAllocateFlags::DEVICE_ADDRESS;
let mut flags_info = vk::MemoryAllocateFlagsInfo::builder().flags(allocation_flags);
// TODO(manon): Test this based on if the device has this feature enabled or not
let alloc_info = if buffer_device_address {
alloc_info.push_next(&mut flags_info)
} else {
alloc_info
};
// Flag the memory as dedicated if required.
let mut dedicated_memory_info = vk::MemoryDedicatedAllocateInfo::builder();
let alloc_info = match allocation_scheme {
AllocationScheme::DedicatedBuffer(buffer) => {
dedicated_memory_info = dedicated_memory_info.buffer(buffer);
alloc_info.push_next(&mut dedicated_memory_info)
}
AllocationScheme::DedicatedImage(image) => {
dedicated_memory_info = dedicated_memory_info.image(image);
alloc_info.push_next(&mut dedicated_memory_info)
}
AllocationScheme::GpuAllocatorManaged => alloc_info,
};
unsafe { device.allocate_memory(&alloc_info, None) }.map_err(|e| match e {
vk::Result::ERROR_OUT_OF_DEVICE_MEMORY => AllocationError::OutOfMemory,
e => AllocationError::Internal(format!(
"Unexpected error in vkAllocateMemory: {:?}",
e
)),
})?
};
let mapped_ptr = mapped
.then(|| {
unsafe {
device.map_memory(
device_memory,
0,
vk::WHOLE_SIZE,
vk::MemoryMapFlags::empty(),
)
}
.map_err(|e| {
unsafe { device.free_memory(device_memory, None) };
AllocationError::FailedToMap(e.to_string())
})
.and_then(|p| {
std::ptr::NonNull::new(p).map(SendSyncPtr).ok_or_else(|| {
AllocationError::FailedToMap("Returned mapped pointer is null".to_owned())
})
})
})
.transpose()?;
let sub_allocator: Box<dyn allocator::SubAllocator> = if allocation_scheme
!= AllocationScheme::GpuAllocatorManaged
|| requires_personal_block
{
Box::new(allocator::DedicatedBlockAllocator::new(size))
} else {
Box::new(allocator::FreeListAllocator::new(size))
};
Ok(Self {
device_memory,
size,
mapped_ptr,
sub_allocator,
#[cfg(feature = "visualizer")]
dedicated_allocation: allocation_scheme != AllocationScheme::GpuAllocatorManaged,
})
}
fn destroy(self, device: &ash::Device) {
if self.mapped_ptr.is_some() {
unsafe { device.unmap_memory(self.device_memory) };
}
unsafe { device.free_memory(self.device_memory, None) };
}
}
#[derive(Debug)]
pub(crate) struct MemoryType {
pub(crate) memory_blocks: Vec<Option<MemoryBlock>>,
pub(crate) memory_properties: vk::MemoryPropertyFlags,
pub(crate) memory_type_index: usize,
pub(crate) heap_index: usize,
pub(crate) mappable: bool,
pub(crate) active_general_blocks: usize,
pub(crate) buffer_device_address: bool,
}
impl MemoryType {
fn allocate(
&mut self,
device: &ash::Device,
desc: &AllocationCreateDesc<'_>,
granularity: u64,
backtrace: Arc<Backtrace>,
allocation_sizes: &AllocationSizes,
) -> Result<Allocation> {
let allocation_type = if desc.linear {
AllocationType::Linear
} else {
AllocationType::NonLinear
};
let memblock_size = if self
.memory_properties
.contains(vk::MemoryPropertyFlags::HOST_VISIBLE)
{
allocation_sizes.host_memblock_size
} else {
allocation_sizes.device_memblock_size
};
let size = desc.requirements.size;
let alignment = desc.requirements.alignment;
let dedicated_allocation = desc.allocation_scheme != AllocationScheme::GpuAllocatorManaged;
let requires_personal_block = size > memblock_size;
// Create a dedicated block for large memory allocations or allocations that require dedicated memory allocations.
if dedicated_allocation || requires_personal_block {
let mem_block = MemoryBlock::new(
device,
size,
self.memory_type_index,
self.mappable,
self.buffer_device_address,
desc.allocation_scheme,
requires_personal_block,
)?;
let mut block_index = None;
for (i, block) in self.memory_blocks.iter().enumerate() {
if block.is_none() {
block_index = Some(i);
break;
}
}
let block_index = match block_index {
Some(i) => {
self.memory_blocks[i].replace(mem_block);
i
}
None => {
self.memory_blocks.push(Some(mem_block));
self.memory_blocks.len() - 1
}
};
let mem_block = self.memory_blocks[block_index]
.as_mut()
.ok_or_else(|| AllocationError::Internal("Memory block must be Some".into()))?;
let (offset, chunk_id) = mem_block.sub_allocator.allocate(
size,
alignment,
allocation_type,
granularity,
desc.name,
backtrace,
)?;
return Ok(Allocation {
chunk_id: Some(chunk_id),
offset,
size,
memory_block_index: block_index,
memory_type_index: self.memory_type_index,
device_memory: mem_block.device_memory,
mapped_ptr: mem_block.mapped_ptr,
memory_properties: self.memory_properties,
name: Some(desc.name.into()),
dedicated_allocation,
});
}
let mut empty_block_index = None;
for (mem_block_i, mem_block) in self.memory_blocks.iter_mut().enumerate().rev() {
if let Some(mem_block) = mem_block {
let allocation = mem_block.sub_allocator.allocate(
size,
alignment,
allocation_type,
granularity,
desc.name,
backtrace.clone(),
);
match allocation {
Ok((offset, chunk_id)) => {
let mapped_ptr = if let Some(SendSyncPtr(mapped_ptr)) = mem_block.mapped_ptr
{
let offset_ptr = unsafe { mapped_ptr.as_ptr().add(offset as usize) };
std::ptr::NonNull::new(offset_ptr).map(SendSyncPtr)
} else {
None
};
return Ok(Allocation {
chunk_id: Some(chunk_id),
offset,
size,
memory_block_index: mem_block_i,
memory_type_index: self.memory_type_index,
device_memory: mem_block.device_memory,
memory_properties: self.memory_properties,
mapped_ptr,
dedicated_allocation: false,
name: Some(desc.name.into()),
});
}
Err(err) => match err {
AllocationError::OutOfMemory => {} // Block is full, continue search.
_ => return Err(err), // Unhandled error, return.
},
}
} else if empty_block_index.is_none() {
empty_block_index = Some(mem_block_i);
}
}
let new_memory_block = MemoryBlock::new(
device,
memblock_size,
self.memory_type_index,
self.mappable,
self.buffer_device_address,
desc.allocation_scheme,
false,
)?;
let new_block_index = if let Some(block_index) = empty_block_index {
self.memory_blocks[block_index] = Some(new_memory_block);
block_index
} else {
self.memory_blocks.push(Some(new_memory_block));
self.memory_blocks.len() - 1
};
self.active_general_blocks += 1;
let mem_block = self.memory_blocks[new_block_index]
.as_mut()
.ok_or_else(|| AllocationError::Internal("Memory block must be Some".into()))?;
let allocation = mem_block.sub_allocator.allocate(
size,
alignment,
allocation_type,
granularity,
desc.name,
backtrace,
);
let (offset, chunk_id) = match allocation {
Ok(value) => value,
Err(err) => match err {
AllocationError::OutOfMemory => {
return Err(AllocationError::Internal(
"Allocation that must succeed failed. This is a bug in the allocator."
.into(),
))
}
_ => return Err(err),
},
};
let mapped_ptr = if let Some(SendSyncPtr(mapped_ptr)) = mem_block.mapped_ptr {
let offset_ptr = unsafe { mapped_ptr.as_ptr().add(offset as usize) };
std::ptr::NonNull::new(offset_ptr).map(SendSyncPtr)
} else {
None
};
Ok(Allocation {
chunk_id: Some(chunk_id),
offset,
size,
memory_block_index: new_block_index,
memory_type_index: self.memory_type_index,
device_memory: mem_block.device_memory,
mapped_ptr,
memory_properties: self.memory_properties,
name: Some(desc.name.into()),
dedicated_allocation: false,
})
}
#[allow(clippy::needless_pass_by_value)]
fn free(&mut self, allocation: Allocation, device: &ash::Device) -> Result<()> {
let block_idx = allocation.memory_block_index;
let mem_block = self.memory_blocks[block_idx]
.as_mut()
.ok_or_else(|| AllocationError::Internal("Memory block must be Some.".into()))?;
mem_block.sub_allocator.free(allocation.chunk_id)?;
if mem_block.sub_allocator.is_empty() {
if mem_block.sub_allocator.supports_general_allocations() {
if self.active_general_blocks > 1 {
let block = self.memory_blocks[block_idx].take();
let block = block.ok_or_else(|| {
AllocationError::Internal("Memory block must be Some.".into())
})?;
block.destroy(device);
self.active_general_blocks -= 1;
}
} else {
let block = self.memory_blocks[block_idx].take();
let block = block.ok_or_else(|| {
AllocationError::Internal("Memory block must be Some.".into())
})?;
block.destroy(device);
}
}
Ok(())
}
}
pub struct Allocator {
pub(crate) memory_types: Vec<MemoryType>,
pub(crate) memory_heaps: Vec<vk::MemoryHeap>,
device: ash::Device,
pub(crate) buffer_image_granularity: u64,
pub(crate) debug_settings: AllocatorDebugSettings,
allocation_sizes: AllocationSizes,
}
impl fmt::Debug for Allocator {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
let mut allocation_report = vec![];
let mut total_reserved_size_in_bytes = 0;
for memory_type in &self.memory_types {
for block in memory_type.memory_blocks.iter().flatten() {
total_reserved_size_in_bytes += block.size;
allocation_report.extend(block.sub_allocator.report_allocations())
}
}
let total_used_size_in_bytes = allocation_report.iter().map(|report| report.size).sum();
allocation_report.sort_by_key(|alloc| std::cmp::Reverse(alloc.size));
writeln!(
f,
"================================================================",
)?;
writeln!(
f,
"ALLOCATION BREAKDOWN ({} / {})",
fmt_bytes(total_used_size_in_bytes),
fmt_bytes(total_reserved_size_in_bytes),
)?;
let max_num_allocations_to_print = f.precision().map_or(usize::MAX, |n| n);
for (idx, alloc) in allocation_report.iter().enumerate() {
if idx >= max_num_allocations_to_print {
break;
}
writeln!(
f,
"{:max_len$.max_len$}\t- {}",
alloc.name,
fmt_bytes(alloc.size),
max_len = allocator::VISUALIZER_TABLE_MAX_ENTRY_NAME_LEN,
)?;
}
Ok(())
}
}
impl Allocator {
pub fn new(desc: &AllocatorCreateDesc) -> Result<Self> {
if desc.physical_device == ash::vk::PhysicalDevice::null() {
return Err(AllocationError::InvalidAllocatorCreateDesc(
"AllocatorCreateDesc field `physical_device` is null.".into(),
));
}
let mem_props = unsafe {
desc.instance
.get_physical_device_memory_properties(desc.physical_device)
};
let memory_types = &mem_props.memory_types[..mem_props.memory_type_count as _];
let memory_heaps = mem_props.memory_heaps[..mem_props.memory_heap_count as _].to_vec();
if desc.debug_settings.log_memory_information {
debug!("memory type count: {}", mem_props.memory_type_count);
debug!("memory heap count: {}", mem_props.memory_heap_count);
for (i, mem_type) in memory_types.iter().enumerate() {
let flags = mem_type.property_flags;
debug!(
"memory type[{}]: prop flags: 0x{:x}, heap[{}]",
i,
flags.as_raw(),
mem_type.heap_index,
);
}
for (i, heap) in memory_heaps.iter().enumerate() {
debug!(
"heap[{}] flags: 0x{:x}, size: {} MiB",
i,
heap.flags.as_raw(),
heap.size / (1024 * 1024)
);
}
}
let memory_types = memory_types
.iter()
.enumerate()
.map(|(i, mem_type)| MemoryType {
memory_blocks: Vec::default(),
memory_properties: mem_type.property_flags,
memory_type_index: i,
heap_index: mem_type.heap_index as usize,
mappable: mem_type
.property_flags
.contains(vk::MemoryPropertyFlags::HOST_VISIBLE),
active_general_blocks: 0,
buffer_device_address: desc.buffer_device_address,
})
.collect::<Vec<_>>();
let physical_device_properties = unsafe {
desc.instance
.get_physical_device_properties(desc.physical_device)
};
let granularity = physical_device_properties.limits.buffer_image_granularity;
Ok(Self {
memory_types,
memory_heaps,
device: desc.device.clone(),
buffer_image_granularity: granularity,
debug_settings: desc.debug_settings,
allocation_sizes: AllocationSizes::default(),
})
}
pub fn allocate(&mut self, desc: &AllocationCreateDesc<'_>) -> Result<Allocation> {
let size = desc.requirements.size;
let alignment = desc.requirements.alignment;
let backtrace = Arc::new(if self.debug_settings.store_stack_traces {
Backtrace::force_capture()
} else {
Backtrace::disabled()
});
if self.debug_settings.log_allocations {
debug!(
"Allocating `{}` of {} bytes with an alignment of {}.",
&desc.name, size, alignment
);
if self.debug_settings.log_stack_traces {
let backtrace = Backtrace::force_capture();
debug!("Allocation stack trace: {}", backtrace);
}
}
if size == 0 || !alignment.is_power_of_two() {
return Err(AllocationError::InvalidAllocationCreateDesc);
}
let mem_loc_preferred_bits = match desc.location {
MemoryLocation::GpuOnly => vk::MemoryPropertyFlags::DEVICE_LOCAL,
MemoryLocation::CpuToGpu => {
vk::MemoryPropertyFlags::HOST_VISIBLE
| vk::MemoryPropertyFlags::HOST_COHERENT
| vk::MemoryPropertyFlags::DEVICE_LOCAL
}
MemoryLocation::GpuToCpu => {
vk::MemoryPropertyFlags::HOST_VISIBLE
| vk::MemoryPropertyFlags::HOST_COHERENT
| vk::MemoryPropertyFlags::HOST_CACHED
}
MemoryLocation::Unknown => vk::MemoryPropertyFlags::empty(),
};
let mut memory_type_index_opt =
self.find_memorytype_index(&desc.requirements, mem_loc_preferred_bits);
if memory_type_index_opt.is_none() {
let mem_loc_required_bits = match desc.location {
MemoryLocation::GpuOnly => vk::MemoryPropertyFlags::DEVICE_LOCAL,
MemoryLocation::CpuToGpu | MemoryLocation::GpuToCpu => {
vk::MemoryPropertyFlags::HOST_VISIBLE | vk::MemoryPropertyFlags::HOST_COHERENT
}
MemoryLocation::Unknown => vk::MemoryPropertyFlags::empty(),
};
memory_type_index_opt =
self.find_memorytype_index(&desc.requirements, mem_loc_required_bits);
}
let memory_type_index = match memory_type_index_opt {
Some(x) => x as usize,
None => return Err(AllocationError::NoCompatibleMemoryTypeFound),
};
//Do not try to create a block if the heap is smaller than the required size (avoids validation warnings).
let memory_type = &mut self.memory_types[memory_type_index];
let allocation = if size > self.memory_heaps[memory_type.heap_index].size {
Err(AllocationError::OutOfMemory)
} else {
memory_type.allocate(
&self.device,
desc,
self.buffer_image_granularity,
backtrace.clone(),
&self.allocation_sizes,
)
};
if desc.location == MemoryLocation::CpuToGpu {
if allocation.is_err() {
let mem_loc_preferred_bits =
vk::MemoryPropertyFlags::HOST_VISIBLE | vk::MemoryPropertyFlags::HOST_COHERENT;
let memory_type_index_opt =
self.find_memorytype_index(&desc.requirements, mem_loc_preferred_bits);
let memory_type_index = match memory_type_index_opt {
Some(x) => x as usize,
None => return Err(AllocationError::NoCompatibleMemoryTypeFound),
};
self.memory_types[memory_type_index].allocate(
&self.device,
desc,
self.buffer_image_granularity,
backtrace,
&self.allocation_sizes,
)
} else {
allocation
}
} else {
allocation
}
}
pub fn free(&mut self, allocation: Allocation) -> Result<()> {
if self.debug_settings.log_frees {
let name = allocation.name.as_deref().unwrap_or("<null>");
debug!("Freeing `{}`.", name);
if self.debug_settings.log_stack_traces {
let backtrace = Backtrace::force_capture();
debug!("Free stack trace: {}", backtrace);
}
}
if allocation.is_null() {
return Ok(());
}
self.memory_types[allocation.memory_type_index].free(allocation, &self.device)?;
Ok(())
}
pub fn rename_allocation(&mut self, allocation: &mut Allocation, name: &str) -> Result<()> {
allocation.name = Some(name.into());
if allocation.is_null() {
return Ok(());
}
let mem_type = &mut self.memory_types[allocation.memory_type_index];
let mem_block = mem_type.memory_blocks[allocation.memory_block_index]
.as_mut()
.ok_or_else(|| AllocationError::Internal("Memory block must be Some.".into()))?;
mem_block
.sub_allocator
.rename_allocation(allocation.chunk_id, name)?;
Ok(())
}
pub fn report_memory_leaks(&self, log_level: Level) {
for (mem_type_i, mem_type) in self.memory_types.iter().enumerate() {
for (block_i, mem_block) in mem_type.memory_blocks.iter().enumerate() {
if let Some(mem_block) = mem_block {
mem_block
.sub_allocator
.report_memory_leaks(log_level, mem_type_i, block_i);
}
}
}
}
fn find_memorytype_index(
&self,
memory_req: &vk::MemoryRequirements,
flags: vk::MemoryPropertyFlags,
) -> Option<u32> {
self.memory_types
.iter()
.find(|memory_type| {
(1 << memory_type.memory_type_index) & memory_req.memory_type_bits != 0
&& memory_type.memory_properties.contains(flags)
})
.map(|memory_type| memory_type.memory_type_index as _)
}
}
impl Drop for Allocator {
fn drop(&mut self) {
if self.debug_settings.log_leaks_on_shutdown {
self.report_memory_leaks(Level::Warn);
}
// Free all remaining memory blocks
for mem_type in self.memory_types.iter_mut() {
for mem_block in mem_type.memory_blocks.iter_mut() {
let block = mem_block.take();
if let Some(block) = block {
block.destroy(&self.device);
}
}
}
}
}