[mem]: Document mutex, rollback stack and tracking allocators

This commit is contained in:
flysand7
2024-09-14 10:46:35 +11:00
parent 3ed2ab6e2c
commit 016d1a84d4
4 changed files with 197 additions and 113 deletions

View File

@@ -3,16 +3,31 @@ package mem
import "core:sync"
/*
The data for mutex allocator.
*/
Mutex_Allocator :: struct {
backing: Allocator,
mutex: sync.Mutex,
}
/*
Initialize the mutex allocator.
This procedure initializes the mutex allocator using `backin_allocator` as the
allocator that will be used to pass all allocation requests through.
*/
mutex_allocator_init :: proc(m: ^Mutex_Allocator, backing_allocator: Allocator) {
m.backing = backing_allocator
m.mutex = {}
}
/*
Mutex allocator.
The mutex allocator is a wrapper for allocators that is used to serialize all
allocator requests across multiple threads.
*/
@(require_results)
mutex_allocator :: proc(m: ^Mutex_Allocator) -> Allocator {
return Allocator{

View File

@@ -4,68 +4,82 @@ import "base:builtin"
import "base:runtime"
/*
Mamory layout of the `any` type.
Memory layout of the `any` type.
*/
Raw_Any :: runtime.Raw_Any
/*
Mamory layout of the `string` type.
Memory layout of the `string` type.
*/
Raw_String :: runtime.Raw_String
/*
Mamory layout of the `cstring` type.
Memory layout of the `cstring` type.
*/
Raw_Cstring :: runtime.Raw_Cstring
/*
Mamory layout of `[]T` types.
Memory layout of `[]T` types.
*/
Raw_Slice :: runtime.Raw_Slice
/*
Mamory layout of `[dynamic]T` types.
Memory layout of `[dynamic]T` types.
*/
Raw_Dynamic_Array :: runtime.Raw_Dynamic_Array
/*
Mamory layout of `map[K]V` types.
Memory layout of `map[K]V` types.
*/
Raw_Map :: runtime.Raw_Map
/*
Mamory layout of `#soa []T` types.
Memory layout of `#soa []T` types.
*/
Raw_Soa_Pointer :: runtime.Raw_Soa_Pointer
/*
Mamory layout of the `complex32` type.
Memory layout of the `complex32` type.
*/
Raw_Complex32 :: runtime.Raw_Complex32
/*
Mamory layout of the `complex64` type.
Memory layout of the `complex64` type.
*/
Raw_Complex64 :: runtime.Raw_Complex64
/*
Mamory layout of the `complex128` type.
Memory layout of the `complex128` type.
*/
Raw_Complex128 :: runtime.Raw_Complex128
/*
Mamory layout of the `quaternion64` type.
Memory layout of the `quaternion64` type.
*/
Raw_Quaternion64 :: runtime.Raw_Quaternion64
/*
Mamory layout of the `quaternion128` type.
Memory layout of the `quaternion128` type.
*/
Raw_Quaternion128 :: runtime.Raw_Quaternion128
/*
Mamory layout of the `quaternion256` type.
Memory layout of the `quaternion256` type.
*/
Raw_Quaternion256 :: runtime.Raw_Quaternion256
/*
Mamory layout of the `quaternion64` type.
Memory layout of the `quaternion64` type.
*/
Raw_Quaternion64_Vector_Scalar :: runtime.Raw_Quaternion64_Vector_Scalar
/*
Mamory layout of the `quaternion128` type.
Memory layout of the `quaternion128` type.
*/
Raw_Quaternion128_Vector_Scalar :: runtime.Raw_Quaternion128_Vector_Scalar
/*
Mamory layout of the `quaternion256` type.
Memory layout of the `quaternion256` type.
*/
Raw_Quaternion256_Vector_Scalar :: runtime.Raw_Quaternion256_Vector_Scalar

View File

@@ -1,39 +1,15 @@
package mem
/*
The Rollback Stack Allocator was designed for the test runner to be fast,
able to grow, and respect the Tracking Allocator's requirement for
individual frees. It is not overly concerned with fragmentation, however.
It has support for expansion when configured with a block allocator and
limited support for out-of-order frees.
Allocation has constant-time best and usual case performance.
At worst, it is linear according to the number of memory blocks.
Allocation follows a first-fit strategy when there are multiple memory
blocks.
Freeing has constant-time best and usual case performance.
At worst, it is linear according to the number of memory blocks and number
of freed items preceding the last item in a block.
Resizing has constant-time performance, if it's the last item in a block, or
the new size is smaller. Naturally, this becomes linear-time if there are
multiple blocks to search for the pointer's owning block. Otherwise, the
allocator defaults to a combined alloc & free operation internally.
Out-of-order freeing is accomplished by collapsing a run of freed items
from the last allocation backwards.
Each allocation has an overhead of 8 bytes and any extra bytes to satisfy
the requested alignment.
*/
import "base:runtime"
/*
Rollback stack default block size.
*/
ROLLBACK_STACK_DEFAULT_BLOCK_SIZE :: 4 * Megabyte
/*
Rollback stack max head block size.
This limitation is due to the size of `prev_ptr`, but it is only for the
head block; any allocation in excess of the allocator's `block_size` is
valid, so long as the block allocator can handle it.
@@ -43,12 +19,18 @@ within is freed; they are immediately returned to the block allocator.
*/
ROLLBACK_STACK_MAX_HEAD_BLOCK_SIZE :: 2 * Gigabyte
/*
Allocation header of the rollback stack allocator.
*/
Rollback_Stack_Header :: bit_field u64 {
prev_offset: uintptr | 32,
is_free: bool | 1,
prev_ptr: uintptr | 31,
}
/*
Block header of the rollback stack allocator.
*/
Rollback_Stack_Block :: struct {
next_block: ^Rollback_Stack_Block,
last_alloc: rawptr,
@@ -56,6 +38,9 @@ Rollback_Stack_Block :: struct {
buffer: []byte,
}
/*
Rollback stack allocator data.
*/
Rollback_Stack :: struct {
head: ^Rollback_Stack_Block,
block_size: int,
@@ -111,6 +96,9 @@ rb_rollback_block :: proc(block: ^Rollback_Stack_Block, header: ^Rollback_Stack_
}
}
/*
Free memory to a rollback stack allocator.
*/
@(private="file", require_results)
rb_free :: proc(stack: ^Rollback_Stack, ptr: rawptr) -> Allocator_Error {
parent, block, header := rb_find_ptr(stack, ptr) or_return
@@ -129,6 +117,9 @@ rb_free :: proc(stack: ^Rollback_Stack, ptr: rawptr) -> Allocator_Error {
return nil
}
/*
Free all memory owned by the rollback stack allocator.
*/
@(private="file")
rb_free_all :: proc(stack: ^Rollback_Stack) {
for block := stack.head.next_block; block != nil; /**/ {
@@ -142,14 +133,16 @@ rb_free_all :: proc(stack: ^Rollback_Stack) {
stack.head.offset = 0
}
/*
Resize an allocation made on a rollback stack allocator.
*/
@(private="file", require_results)
rb_resize :: proc(stack: ^Rollback_Stack, ptr: rawptr, old_size, size, alignment: int) -> (result: []byte, err: Allocator_Error) {
rb_resize_non_zeroed :: proc(stack: ^Rollback_Stack, ptr: rawptr, old_size, size, alignment: int) -> (result: []byte, err: Allocator_Error) {
if ptr != nil {
if block, _, ok := rb_find_last_alloc(stack, ptr); ok {
// `block.offset` should never underflow because it is contingent
// on `old_size` in the first place, assuming sane arguments.
assert(block.offset >= cast(uintptr)old_size, "Rollback Stack Allocator received invalid `old_size`.")
if block.offset + cast(uintptr)size - cast(uintptr)old_size < cast(uintptr)len(block.buffer) {
// Prevent singleton allocations from fragmenting by forbidding
// them to shrink, removing the possibility of overflow bugs.
@@ -160,27 +153,26 @@ rb_resize :: proc(stack: ^Rollback_Stack, ptr: rawptr, old_size, size, alignment
}
}
}
result = rb_alloc(stack, size, alignment) or_return
result = rb_alloc_non_zeroed(stack, size, alignment) or_return
runtime.mem_copy_non_overlapping(raw_data(result), ptr, old_size)
err = rb_free(stack, ptr)
return
}
/*
Allocate memory using the rollback stack allocator.
*/
@(private="file", require_results)
rb_alloc :: proc(stack: ^Rollback_Stack, size, alignment: int) -> (result: []byte, err: Allocator_Error) {
rb_alloc_non_zeroed :: proc(stack: ^Rollback_Stack, size, alignment: int) -> (result: []byte, err: Allocator_Error) {
parent: ^Rollback_Stack_Block
for block := stack.head; /**/; block = block.next_block {
when !ODIN_DISABLE_ASSERT {
allocated_new_block: bool
}
if block == nil {
if stack.block_allocator.procedure == nil {
return nil, .Out_Of_Memory
}
minimum_size_required := size_of(Rollback_Stack_Header) + size + alignment - 1
new_block_size := max(minimum_size_required, stack.block_size)
block = rb_make_block(new_block_size, stack.block_allocator) or_return
@@ -189,10 +181,8 @@ rb_alloc :: proc(stack: ^Rollback_Stack, size, alignment: int) -> (result: []byt
allocated_new_block = true
}
}
start := raw_data(block.buffer)[block.offset:]
padding := cast(uintptr)calc_padding_with_header(cast(uintptr)start, cast(uintptr)alignment, size_of(Rollback_Stack_Header))
if block.offset + padding + cast(uintptr)size > cast(uintptr)len(block.buffer) {
when !ODIN_DISABLE_ASSERT {
if allocated_new_block {
@@ -202,54 +192,50 @@ rb_alloc :: proc(stack: ^Rollback_Stack, size, alignment: int) -> (result: []byt
parent = block
continue
}
header := cast(^Rollback_Stack_Header)(start[padding - size_of(Rollback_Stack_Header):])
ptr := start[padding:]
header^ = {
prev_offset = block.offset,
prev_ptr = uintptr(0) if block.last_alloc == nil else cast(uintptr)block.last_alloc - cast(uintptr)raw_data(block.buffer),
is_free = false,
}
block.last_alloc = ptr
block.offset += padding + cast(uintptr)size
if len(block.buffer) > stack.block_size {
// This block exceeds the allocator's standard block size and is considered a singleton.
// Prevent any further allocations on it.
block.offset = cast(uintptr)len(block.buffer)
}
#no_bounds_check return ptr[:size], nil
}
return nil, .Out_Of_Memory
}
@(private="file", require_results)
rb_make_block :: proc(size: int, allocator: Allocator) -> (block: ^Rollback_Stack_Block, err: Allocator_Error) {
buffer := runtime.mem_alloc(size_of(Rollback_Stack_Block) + size, align_of(Rollback_Stack_Block), allocator) or_return
block = cast(^Rollback_Stack_Block)raw_data(buffer)
#no_bounds_check block.buffer = buffer[size_of(Rollback_Stack_Block):]
return
}
/*
Initialize the rollback stack allocator using a fixed backing buffer.
*/
rollback_stack_init_buffered :: proc(stack: ^Rollback_Stack, buffer: []byte, location := #caller_location) {
MIN_SIZE :: size_of(Rollback_Stack_Block) + size_of(Rollback_Stack_Header) + size_of(rawptr)
assert(len(buffer) >= MIN_SIZE, "User-provided buffer to Rollback Stack Allocator is too small.", location)
block := cast(^Rollback_Stack_Block)raw_data(buffer)
block^ = {}
#no_bounds_check block.buffer = buffer[size_of(Rollback_Stack_Block):]
stack^ = {}
stack.head = block
stack.block_size = len(block.buffer)
}
/*
Initialize the rollback stack alocator using a backing block allocator.
*/
rollback_stack_init_dynamic :: proc(
stack: ^Rollback_Stack,
block_size : int = ROLLBACK_STACK_DEFAULT_BLOCK_SIZE,
@@ -262,22 +248,25 @@ rollback_stack_init_dynamic :: proc(
// size is insufficient; check only on platforms with big enough ints.
assert(block_size <= ROLLBACK_STACK_MAX_HEAD_BLOCK_SIZE, "Rollback Stack Allocators cannot support head blocks larger than 2 gigabytes.", location)
}
block := rb_make_block(block_size, block_allocator) or_return
stack^ = {}
stack.head = block
stack.block_size = block_size
stack.block_allocator = block_allocator
return nil
}
/*
Initialize the rollback stack.
*/
rollback_stack_init :: proc {
rollback_stack_init_buffered,
rollback_stack_init_dynamic,
}
/*
Destroy a rollback stack.
*/
rollback_stack_destroy :: proc(stack: ^Rollback_Stack) {
if stack.block_allocator.procedure != nil {
rb_free_all(stack)
@@ -286,6 +275,37 @@ rollback_stack_destroy :: proc(stack: ^Rollback_Stack) {
stack^ = {}
}
/*
Rollback stack allocator.
The Rollback Stack Allocator was designed for the test runner to be fast,
able to grow, and respect the Tracking Allocator's requirement for
individual frees. It is not overly concerned with fragmentation, however.
It has support for expansion when configured with a block allocator and
limited support for out-of-order frees.
Allocation has constant-time best and usual case performance.
At worst, it is linear according to the number of memory blocks.
Allocation follows a first-fit strategy when there are multiple memory
blocks.
Freeing has constant-time best and usual case performance.
At worst, it is linear according to the number of memory blocks and number
of freed items preceding the last item in a block.
Resizing has constant-time performance, if it's the last item in a block, or
the new size is smaller. Naturally, this becomes linear-time if there are
multiple blocks to search for the pointer's owning block. Otherwise, the
allocator defaults to a combined alloc & free operation internally.
Out-of-order freeing is accomplished by collapsing a run of freed items
from the last allocation backwards.
Each allocation has an overhead of 8 bytes and any extra bytes to satisfy
the requested alignment.
*/
@(require_results)
rollback_stack_allocator :: proc(stack: ^Rollback_Stack) -> Allocator {
return Allocator {
@@ -309,38 +329,31 @@ rollback_stack_allocator_proc :: proc(
case .Alloc, .Alloc_Non_Zeroed:
assert(size >= 0, "Size must be positive or zero.", location)
assert(is_power_of_two(cast(uintptr)alignment), "Alignment must be a power of two.", location)
result = rb_alloc(stack, size, alignment) or_return
result = rb_alloc_non_zeroed(stack, size, alignment) or_return
if mode == .Alloc {
zero_slice(result)
}
case .Free:
err = rb_free(stack, old_memory)
case .Free_All:
rb_free_all(stack)
case .Resize, .Resize_Non_Zeroed:
assert(size >= 0, "Size must be positive or zero.", location)
assert(old_size >= 0, "Old size must be positive or zero.", location)
assert(is_power_of_two(cast(uintptr)alignment), "Alignment must be a power of two.", location)
result = rb_resize(stack, old_memory, old_size, size, alignment) or_return
result = rb_resize_non_zeroed(stack, old_memory, old_size, size, alignment) or_return
#no_bounds_check if mode == .Resize && size > old_size {
zero_slice(result[old_size:])
}
case .Query_Features:
set := (^Allocator_Mode_Set)(old_memory)
if set != nil {
set^ = {.Alloc, .Alloc_Non_Zeroed, .Free, .Free_All, .Resize, .Resize_Non_Zeroed}
}
return nil, nil
case .Query_Info:
return nil, .Mode_Not_Implemented
}
return
}

View File

@@ -4,50 +4,38 @@ package mem
import "base:runtime"
import "core:sync"
/*
Allocation entry for the tracking allocator.
This structure stores the data related to an allocation.
*/
Tracking_Allocator_Entry :: struct {
memory: rawptr,
size: int,
// Pointer to an allocated region.
memory: rawptr,
// Size of the allocated memory region.
size: int,
// Requested alignment.
alignment: int,
mode: Allocator_Mode,
err: Allocator_Error,
// Mode of the operation.
mode: Allocator_Mode,
// Error.
err: Allocator_Error,
// Location of the allocation.
location: runtime.Source_Code_Location,
}
/*
Bad free entry for a tracking allocator.
*/
Tracking_Allocator_Bad_Free_Entry :: struct {
memory: rawptr,
// Pointer, on which free operation was called.
memory: rawptr,
// The source location of where the operation was called.
location: runtime.Source_Code_Location,
}
/*
An example of how to use the `Tracking_Allocator` to track subsequent allocations
in your program and report leaks and bad frees:
Example:
package foo
import "core:mem"
import "core:fmt"
_main :: proc() {
// do stuff
}
main :: proc() {
track: mem.Tracking_Allocator
mem.tracking_allocator_init(&track, context.allocator)
defer mem.tracking_allocator_destroy(&track)
context.allocator = mem.tracking_allocator(&track)
_main()
for _, leak in track.allocation_map {
fmt.printf("%v leaked %m\n", leak.location, leak.size)
}
for bad_free in track.bad_free_array {
fmt.printf("%v allocation %p was freed badly\n", bad_free.location, bad_free.memory)
}
}
Tracking allocator data.
*/
Tracking_Allocator :: struct {
backing: Allocator,
@@ -63,6 +51,13 @@ Tracking_Allocator :: struct {
current_memory_allocated: i64,
}
/*
Initialize the tracking allocator.
This procedure initializes the tracking allocator `t` with a backing allocator
specified with `backing_allocator`. The `internals_allocator` will used to
allocate the tracked data.
*/
tracking_allocator_init :: proc(t: ^Tracking_Allocator, backing_allocator: Allocator, internals_allocator := context.allocator) {
t.backing = backing_allocator
t.allocation_map.allocator = internals_allocator
@@ -72,12 +67,22 @@ tracking_allocator_init :: proc(t: ^Tracking_Allocator, backing_allocator: Alloc
}
}
/*
Destroy the tracking allocator.
*/
tracking_allocator_destroy :: proc(t: ^Tracking_Allocator) {
delete(t.allocation_map)
delete(t.bad_free_array)
}
// Clear only the current allocation data while keeping the totals intact.
/*
Clear the tracking allocator.
This procedure clears the tracked data from a tracking allocator.
**Note**: This procedure clears only the current allocation data while keeping
the totals intact.
*/
tracking_allocator_clear :: proc(t: ^Tracking_Allocator) {
sync.mutex_lock(&t.mutex)
clear(&t.allocation_map)
@@ -86,7 +91,11 @@ tracking_allocator_clear :: proc(t: ^Tracking_Allocator) {
sync.mutex_unlock(&t.mutex)
}
// Reset all of a Tracking Allocator's allocation data back to zero.
/*
Reset the tracking allocator.
Reset all of a Tracking Allocator's allocation data back to zero.
*/
tracking_allocator_reset :: proc(t: ^Tracking_Allocator) {
sync.mutex_lock(&t.mutex)
clear(&t.allocation_map)
@@ -100,6 +109,39 @@ tracking_allocator_reset :: proc(t: ^Tracking_Allocator) {
sync.mutex_unlock(&t.mutex)
}
/*
Tracking allocator.
The tracking allocator is an allocator wrapper that tracks memory allocations.
This allocator stores all the allocations in a map. Whenever a pointer that's
not inside of the map is freed, the `bad_free_array` entry is added.
An example of how to use the `Tracking_Allocator` to track subsequent allocations
in your program and report leaks and bad frees:
Example:
package foo
import "core:mem"
import "core:fmt"
main :: proc() {
track: mem.Tracking_Allocator
mem.tracking_allocator_init(&track, context.allocator)
defer mem.tracking_allocator_destroy(&track)
context.allocator = mem.tracking_allocator(&track)
do_stuff()
for _, leak in track.allocation_map {
fmt.printf("%v leaked %m\n", leak.location, leak.size)
}
for bad_free in track.bad_free_array {
fmt.printf("%v allocation %p was freed badly\n", bad_free.location, bad_free.memory)
}
}
*/
@(require_results)
tracking_allocator :: proc(data: ^Tracking_Allocator) -> Allocator {
return Allocator{