wasi: round up the size for aligned_alloc

[cuviper]

C11 aligned_alloc requires that the size be a multiple of the
alignment. This is enforced in the wasi-libc emmalloc implementation,
which always returns NULL if the size is not a multiple.
(The default MALLOC_IMPL=dlmalloc does not currently check this.)

[rustbot]

r? @thomcc

(rustbot has picked a reviewer for you, use r? to override)

[thomcc]

@bors r+

[bors]

📌 Commit 1c6d867 has been approved by thomcc

It is now in the queue for this repository.

[thomcc]

Note that this behavior has been proposed to change, so we should consider reverting this in the future if https://open-std.org/JTC1/SC22/WG14/www/docs/n2072.htm is standardized.

[cuviper]

we should consider reverting this in the future if n2072 is standardized.

We could... but I wouldn't bother unless there's some evidence that it's useful, and not just the allocator rounding up on your behalf or otherwise wasting that padding.

[thomcc]

It's definitely useful, to the point where it's surprising this restriction exists.

Consider allocating 1 byte at alignment 1024 from a bump nursery (this is true for other types of allocators as well but less easy to show). You'd do something like (curptr + 1023) & !1023 to get the allocation base, then add another byte to curptr to account for the size, this will waste between 0 and 1023 bytes. However, if you rounded the size up to 1024 even though you only need 1 byte, you'd be wasting between 1023 and 2047 bytes.

To put it another way, if I'm an allocator, I only need to allocate size + align bytes to satisfy a given layout, so in the previous example, at most 1025 bytes in order to be guaranteed to find 1 byte which is aligned to 1024. Rounding up size to alignment means that I now have to allocate up to 2048 bytes.

(Note: if I have an off-by-one anywhere in here, I blame the fact that I have the flu)

[bors]

⌛ Testing commit 1c6d867 with merge 1baf77a...

[bors]

☀️ Test successful - checks-actions
Approved by: thomcc
Pushing 1baf77a to master...

[rust-timer]

Finished benchmarking commit (1baf77a): comparison URL.

Overall result: no relevant changes - no action needed

@rustbot label: -perf-regression

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.1%	[1.2%, 3.9%]	4
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 630.803s -> 631.998s (0.19%)
Artifact size: 316.08 MiB -> 316.03 MiB (-0.02%)

[RalfJung]

C11 aligned_alloc requires that the size be a multiple of the
alignment.

FWIW, I can't find this requirement in C18 or C23 any more, so wasi may have to be updated here.

EDIT: #125003