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
use {
crate::{
align_down,
block::{MemoryBlock, MemoryBlockFlavor},
buddy::{BuddyAllocator, BuddyBlock},
config::Config,
error::AllocationError,
freelist::{FreeListAllocator, FreeListBlock},
heap::Heap,
usage::{MemoryForUsage, UsageFlags},
MemoryBounds, Request,
},
alloc::boxed::Box,
core::convert::TryFrom as _,
gpu_alloc_types::{
AllocationFlags, DeviceProperties, MemoryDevice, MemoryPropertyFlags, MemoryType,
OutOfMemory,
},
};
/// Memory allocator for Vulkan-like APIs.
#[derive(Debug)]
pub struct GpuAllocator<M> {
dedicated_threshold: u64,
preferred_dedicated_threshold: u64,
transient_dedicated_threshold: u64,
max_memory_allocation_size: u64,
memory_for_usage: MemoryForUsage,
memory_types: Box<[MemoryType]>,
memory_heaps: Box<[Heap]>,
allocations_remains: u32,
non_coherent_atom_mask: u64,
starting_free_list_chunk: u64,
final_free_list_chunk: u64,
minimal_buddy_size: u64,
initial_buddy_dedicated_size: u64,
buffer_device_address: bool,
buddy_allocators: Box<[Option<BuddyAllocator<M>>]>,
freelist_allocators: Box<[Option<FreeListAllocator<M>>]>,
}
/// Hints for allocator to decide on allocation strategy.
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
#[non_exhaustive]
pub enum Dedicated {
/// Allocation directly from device.\
/// Very slow.
/// Count of allocations is limited.\
/// Use with caution.\
/// Must be used if resource has to be bound to dedicated memory object.
Required,
/// Hint for allocator that dedicated memory object is preferred.\
/// Should be used if it is known that resource placed in dedicated memory object
/// would allow for better performance.\
/// Implementation is allowed to return block to shared memory object.
Preferred,
}
impl<M> GpuAllocator<M>
where
M: MemoryBounds + 'static,
{
/// Creates new instance of `GpuAllocator`.
/// Provided `DeviceProperties` should match properties of `MemoryDevice` that will be used
/// with created `GpuAllocator` instance.
#[cfg_attr(feature = "tracing", tracing::instrument)]
pub fn new(config: Config, props: DeviceProperties<'_>) -> Self {
assert!(
props.non_coherent_atom_size.is_power_of_two(),
"`non_coherent_atom_size` must be power of two"
);
assert!(
isize::try_from(props.non_coherent_atom_size).is_ok(),
"`non_coherent_atom_size` must fit host address space"
);
GpuAllocator {
dedicated_threshold: config.dedicated_threshold,
preferred_dedicated_threshold: config
.preferred_dedicated_threshold
.min(config.dedicated_threshold),
transient_dedicated_threshold: config
.transient_dedicated_threshold
.max(config.dedicated_threshold),
max_memory_allocation_size: props.max_memory_allocation_size,
memory_for_usage: MemoryForUsage::new(props.memory_types.as_ref()),
memory_types: props.memory_types.as_ref().iter().copied().collect(),
memory_heaps: props
.memory_heaps
.as_ref()
.iter()
.map(|heap| Heap::new(heap.size))
.collect(),
buffer_device_address: props.buffer_device_address,
allocations_remains: props.max_memory_allocation_count,
non_coherent_atom_mask: props.non_coherent_atom_size - 1,
starting_free_list_chunk: config.starting_free_list_chunk,
final_free_list_chunk: config.final_free_list_chunk,
minimal_buddy_size: config.minimal_buddy_size,
initial_buddy_dedicated_size: config.initial_buddy_dedicated_size,
buddy_allocators: props.memory_types.as_ref().iter().map(|_| None).collect(),
freelist_allocators: props.memory_types.as_ref().iter().map(|_| None).collect(),
}
}
/// Allocates memory block from specified `device` according to the `request`.
///
/// # Safety
///
/// * `device` must be one with `DeviceProperties` that were provided to create this `GpuAllocator` instance.
/// * Same `device` instance must be used for all interactions with one `GpuAllocator` instance
/// and memory blocks allocated from it.
#[cfg_attr(feature = "tracing", tracing::instrument(skip(self, device)))]
pub unsafe fn alloc(
&mut self,
device: &impl MemoryDevice<M>,
request: Request,
) -> Result<MemoryBlock<M>, AllocationError> {
self.alloc_internal(device, request, None)
}
/// Allocates memory block from specified `device` according to the `request`.
/// This function allows user to force specific allocation strategy.
/// Improper use can lead to suboptimal performance or too large overhead.
/// Prefer `GpuAllocator::alloc` if doubt.
///
/// # Safety
///
/// * `device` must be one with `DeviceProperties` that were provided to create this `GpuAllocator` instance.
/// * Same `device` instance must be used for all interactions with one `GpuAllocator` instance
/// and memory blocks allocated from it.
#[cfg_attr(feature = "tracing", tracing::instrument(skip(self, device)))]
pub unsafe fn alloc_with_dedicated(
&mut self,
device: &impl MemoryDevice<M>,
request: Request,
dedicated: Dedicated,
) -> Result<MemoryBlock<M>, AllocationError> {
self.alloc_internal(device, request, Some(dedicated))
}
unsafe fn alloc_internal(
&mut self,
device: &impl MemoryDevice<M>,
mut request: Request,
dedicated: Option<Dedicated>,
) -> Result<MemoryBlock<M>, AllocationError> {
enum Strategy {
Buddy,
Dedicated,
FreeList,
}
request.usage = with_implicit_usage_flags(request.usage);
if request.usage.contains(UsageFlags::DEVICE_ADDRESS) {
assert!(self.buffer_device_address, "`DEVICE_ADDRESS` cannot be requested when `DeviceProperties::buffer_device_address` is false");
}
if request.size > self.max_memory_allocation_size {
return Err(AllocationError::OutOfDeviceMemory);
}
if let Some(Dedicated::Required) = dedicated {
if self.allocations_remains == 0 {
return Err(AllocationError::TooManyObjects);
}
}
if 0 == self.memory_for_usage.mask(request.usage) & request.memory_types {
#[cfg(feature = "tracing")]
tracing::error!(
"Cannot serve request {:?}, no memory among bitset `{}` support usage {:?}",
request,
request.memory_types,
request.usage
);
return Err(AllocationError::NoCompatibleMemoryTypes);
}
let transient = request.usage.contains(UsageFlags::TRANSIENT);
for &index in self.memory_for_usage.types(request.usage) {
if 0 == request.memory_types & (1 << index) {
// Skip memory type incompatible with the request.
continue;
}
let memory_type = &self.memory_types[index as usize];
let heap = memory_type.heap;
let heap = &mut self.memory_heaps[heap as usize];
if request.size > heap.size() {
// Impossible to use memory type from this heap.
continue;
}
let atom_mask = if host_visible_non_coherent(memory_type.props) {
self.non_coherent_atom_mask
} else {
0
};
let flags = if self.buffer_device_address {
AllocationFlags::DEVICE_ADDRESS
} else {
AllocationFlags::empty()
};
let strategy = match (dedicated, transient) {
(Some(Dedicated::Required), _) => Strategy::Dedicated,
(Some(Dedicated::Preferred), _)
if request.size >= self.preferred_dedicated_threshold =>
{
Strategy::Dedicated
}
(_, true) => {
let threshold = self.transient_dedicated_threshold.min(heap.size() / 32);
if request.size < threshold {
Strategy::FreeList
} else {
Strategy::Dedicated
}
}
(_, false) => {
let threshold = self.dedicated_threshold.min(heap.size() / 32);
if request.size < threshold {
Strategy::Buddy
} else {
Strategy::Dedicated
}
}
};
match strategy {
Strategy::Dedicated => {
#[cfg(feature = "tracing")]
tracing::debug!(
"Allocating memory object `{}@{:?}`",
request.size,
memory_type
);
match device.allocate_memory(request.size, index, flags) {
Ok(memory) => {
self.allocations_remains -= 1;
heap.alloc(request.size);
return Ok(MemoryBlock::new(
index,
memory_type.props,
0,
request.size,
atom_mask,
MemoryBlockFlavor::Dedicated { memory },
));
}
Err(OutOfMemory::OutOfDeviceMemory) => continue,
Err(OutOfMemory::OutOfHostMemory) => {
return Err(AllocationError::OutOfHostMemory)
}
}
}
Strategy::FreeList => {
let allocator = match &mut self.freelist_allocators[index as usize] {
Some(allocator) => allocator,
slot => {
let starting_free_list_chunk = match align_down(
self.starting_free_list_chunk.min(heap.size() / 32),
atom_mask,
) {
0 => atom_mask,
other => other,
};
let final_free_list_chunk = match align_down(
self.final_free_list_chunk
.max(self.starting_free_list_chunk)
.max(self.transient_dedicated_threshold)
.min(heap.size() / 32),
atom_mask,
) {
0 => atom_mask,
other => other,
};
slot.get_or_insert(FreeListAllocator::new(
starting_free_list_chunk,
final_free_list_chunk,
index,
memory_type.props,
if host_visible_non_coherent(memory_type.props) {
self.non_coherent_atom_mask
} else {
0
},
))
}
};
let result = allocator.alloc(
device,
request.size,
request.align_mask,
flags,
heap,
&mut self.allocations_remains,
);
match result {
Ok(block) => {
return Ok(MemoryBlock::new(
index,
memory_type.props,
block.offset,
block.size,
atom_mask,
MemoryBlockFlavor::FreeList {
chunk: block.chunk,
ptr: block.ptr,
memory: block.memory,
},
))
}
Err(AllocationError::OutOfDeviceMemory) => continue,
Err(err) => return Err(err),
}
}
Strategy::Buddy => {
let allocator = match &mut self.buddy_allocators[index as usize] {
Some(allocator) => allocator,
slot => {
let minimal_buddy_size = self
.minimal_buddy_size
.min(heap.size() / 1024)
.next_power_of_two();
let initial_buddy_dedicated_size = self
.initial_buddy_dedicated_size
.min(heap.size() / 32)
.next_power_of_two();
slot.get_or_insert(BuddyAllocator::new(
minimal_buddy_size,
initial_buddy_dedicated_size,
index,
memory_type.props,
if host_visible_non_coherent(memory_type.props) {
self.non_coherent_atom_mask
} else {
0
},
))
}
};
let result = allocator.alloc(
device,
request.size,
request.align_mask,
flags,
heap,
&mut self.allocations_remains,
);
match result {
Ok(block) => {
return Ok(MemoryBlock::new(
index,
memory_type.props,
block.offset,
block.size,
atom_mask,
MemoryBlockFlavor::Buddy {
chunk: block.chunk,
ptr: block.ptr,
index: block.index,
memory: block.memory,
},
))
}
Err(AllocationError::OutOfDeviceMemory) => continue,
Err(err) => return Err(err),
}
}
}
}
Err(AllocationError::OutOfDeviceMemory)
}
/// Creates a memory block from an existing memory allocation, transferring ownership to the allocator.
///
/// This function allows the [`GpuAllocator`] to manage memory allocated outside of the typical
/// [`GpuAllocator::alloc`] family of functions.
///
/// # Usage
///
/// If you need to import external memory, such as a Win32 `HANDLE` or a Linux `dmabuf`, import the device
/// memory using the graphics api and platform dependent functions. Once that is done, call this function
/// to make the [`GpuAllocator`] take ownership of the imported memory.
///
/// When calling this function, you **must** ensure there are [enough remaining allocations](GpuAllocator::remaining_allocations).
///
/// # Safety
///
/// - The `memory` must be allocated with the same device that was provided to create this [`GpuAllocator`]
/// instance.
/// - The `memory` must be valid.
/// - The `props`, `offset` and `size` must match the properties, offset and size of the memory allocation.
/// - The memory must have been allocated with the specified `memory_type`.
/// - There must be enough remaining allocations.
/// - The memory allocation must not come from an existing memory block created by this allocator.
/// - The underlying memory object must be deallocated using the returned [`MemoryBlock`] with
/// [`GpuAllocator::dealloc`].
pub unsafe fn import_memory(
&mut self,
memory: M,
memory_type: u32,
props: MemoryPropertyFlags,
offset: u64,
size: u64,
) -> MemoryBlock<M> {
// Get the heap which the imported memory is from.
let heap = self
.memory_types
.get(memory_type as usize)
.expect("Invalid memory type specified when importing memory")
.heap;
let heap = &mut self.memory_heaps[heap as usize];
#[cfg(feature = "tracing")]
tracing::debug!(
"Importing memory object {:?} `{}@{:?}`",
memory,
size,
memory_type
);
assert_ne!(
self.allocations_remains, 0,
"Out of allocations when importing a memory block. Ensure you check GpuAllocator::remaining_allocations before import."
);
self.allocations_remains -= 1;
let atom_mask = if host_visible_non_coherent(props) {
self.non_coherent_atom_mask
} else {
0
};
heap.alloc(size);
MemoryBlock::new(
memory_type,
props,
offset,
size,
atom_mask,
MemoryBlockFlavor::Dedicated { memory },
)
}
/// Deallocates memory block previously allocated from this `GpuAllocator` instance.
///
/// # Safety
///
/// * Memory block must have been allocated by this `GpuAllocator` instance
/// * `device` must be one with `DeviceProperties` that were provided to create this `GpuAllocator` instance
/// * Same `device` instance must be used for all interactions with one `GpuAllocator` instance
/// and memory blocks allocated from it
#[cfg_attr(feature = "tracing", tracing::instrument(skip(self, device)))]
pub unsafe fn dealloc(&mut self, device: &impl MemoryDevice<M>, block: MemoryBlock<M>) {
let memory_type = block.memory_type();
let offset = block.offset();
let size = block.size();
let flavor = block.deallocate();
match flavor {
MemoryBlockFlavor::Dedicated { memory } => {
let heap = self.memory_types[memory_type as usize].heap;
device.deallocate_memory(memory);
self.allocations_remains += 1;
self.memory_heaps[heap as usize].dealloc(size);
}
MemoryBlockFlavor::Buddy {
chunk,
ptr,
index,
memory,
} => {
let heap = self.memory_types[memory_type as usize].heap;
let heap = &mut self.memory_heaps[heap as usize];
let allocator = self.buddy_allocators[memory_type as usize]
.as_mut()
.expect("Allocator should exist");
allocator.dealloc(
device,
BuddyBlock {
memory,
ptr,
offset,
size,
chunk,
index,
},
heap,
&mut self.allocations_remains,
);
}
MemoryBlockFlavor::FreeList { chunk, ptr, memory } => {
let heap = self.memory_types[memory_type as usize].heap;
let heap = &mut self.memory_heaps[heap as usize];
let allocator = self.freelist_allocators[memory_type as usize]
.as_mut()
.expect("Allocator should exist");
allocator.dealloc(
device,
FreeListBlock {
memory,
ptr,
chunk,
offset,
size,
},
heap,
&mut self.allocations_remains,
);
}
}
}
/// Returns the maximum allocation size supported.
pub fn max_allocation_size(&self) -> u64 {
self.max_memory_allocation_size
}
/// Returns the number of remaining available allocations.
///
/// This may be useful if you need know if the allocator can allocate a number of allocations ahead of
/// time. This function is also useful for ensuring you do not allocate too much memory outside allocator
/// (such as external memory).
pub fn remaining_allocations(&self) -> u32 {
self.allocations_remains
}
/// Sets the number of remaining available allocations.
///
/// # Safety
///
/// The caller is responsible for ensuring the number of remaining allocations does not exceed how many
/// remaining allocations there actually are on the memory device.
pub unsafe fn set_remaining_allocations(&mut self, remaining: u32) {
self.allocations_remains = remaining;
}
/// Deallocates leftover memory objects.
/// Should be used before dropping.
///
/// # Safety
///
/// * `device` must be one with `DeviceProperties` that were provided to create this `GpuAllocator` instance
/// * Same `device` instance must be used for all interactions with one `GpuAllocator` instance
/// and memory blocks allocated from it
#[cfg_attr(feature = "tracing", tracing::instrument(skip(self, device)))]
pub unsafe fn cleanup(&mut self, device: &impl MemoryDevice<M>) {
for (index, allocator) in self
.freelist_allocators
.iter_mut()
.enumerate()
.filter_map(|(index, allocator)| Some((index, allocator.as_mut()?)))
{
let memory_type = &self.memory_types[index];
let heap = memory_type.heap;
let heap = &mut self.memory_heaps[heap as usize];
allocator.cleanup(device, heap, &mut self.allocations_remains);
}
}
}
fn host_visible_non_coherent(props: MemoryPropertyFlags) -> bool {
(props & (MemoryPropertyFlags::HOST_COHERENT | MemoryPropertyFlags::HOST_VISIBLE))
== MemoryPropertyFlags::HOST_VISIBLE
}
fn with_implicit_usage_flags(usage: UsageFlags) -> UsageFlags {
if usage.is_empty() {
UsageFlags::FAST_DEVICE_ACCESS
} else if usage.intersects(UsageFlags::DOWNLOAD | UsageFlags::UPLOAD) {
usage | UsageFlags::HOST_ACCESS
} else {
usage
}
}