feat: Add back support for big-endian targets

[clabby]

Overview

Adds back big-endian support after #17, cc @shekhirin

These changes should have no impact on the performance of this crate when targeting little-endian - All methods that require read-only access to the underlying U256 can use as_le_bytes, and methods that relied on as_le_slice_mut retain the existing implementation when cfg(target_endian = "little").

Also adds a new CI job to build/test the crate on big-endian 32-bit and 64-bit MIPS targets.

Meta

closes #31

[shekhirin]

LGTM, thanks! @DaniPopes please take a look too

[codspeed-hq]

CodSpeed Performance Report

Merging #32 will degrade performances by 25.69%

_{Comparing clabby:cl/big-endian-supp (647b4cd) with main (a2853a6)}

Summary

⚡ 18 improvements
❌ 9 regressions
✅ 84 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	common_prefix_length[16]	242.8 ns	213.6 ns	+13.65%
⚡	common_prefix_length[32]	395.6 ns	337.2 ns	+17.3%
⚡	common_prefix_length[64]	395.6 ns	337.2 ns	+17.3%
⚡	common_prefix_length[8]	242.8 ns	213.6 ns	+13.65%
❌	ends_with[16]	342.2 ns	400.6 ns	-14.56%
❌	ends_with[32]	342.2 ns	403.3 ns	-15.15%
❌	ends_with[64]	342.2 ns	460.6 ns	-25.69%
⚡	extend[16]	471.4 ns	413.1 ns	+14.12%
⚡	extend[32]	467.2 ns	408.9 ns	+14.27%
⚡	extend[8]	474.2 ns	415.8 ns	+14.03%
⚡	get_byte_unchecked[16]	279.4 ns	250.3 ns	+11.65%
⚡	get_byte_unchecked[32]	419.7 ns	361.4 ns	+16.14%
⚡	get_byte_unchecked[8]	223.9 ns	194.7 ns	+14.98%
❌	last[32]	123.1 ns	152.2 ns	-19.16%
❌	pack[32]	463.1 ns	521.4 ns	-11.19%
❌	set_at[8]	213.6 ns	242.8 ns	-12.01%
⚡	to_end[16]	290 ns	260.8 ns	+11.18%
⚡	to_end[32]	283.1 ns	253.9 ns	+11.49%
⚡	to_end[64]	283.1 ns	253.9 ns	+11.49%
⚡	to_end[8]	290 ns	260.8 ns	+11.18%
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

[DaniPopes]

lgtm; to_le_bytes works but it kills performance because of the heap copy, you can make a separate function because we know it's 32 bytes (u256) that makes a stack copy and reverses instead of one on the heap

[DaniPopes]

cant we add an include to the test matrix? or is it because of build-std

[clabby]

Yeah, build-std (always needs nightly) + dtolnay/rust-toolchain will error when installing a tier-3 target due to rust-std not being distributed for them. We could add some more conditionals to the original job, but figured splitting them up was a bit more readable.

[clabby]

lgtm; to_le_bytes works but it kills performance because of the heap copy, you can make a separate function because we know it's 32 bytes (u256) that makes a stack copy and reverses instead of one on the heap

Updated the impl to have a container that prevents touching the heap in the big-endian case.

e: whoops, fixing CI...

[DaniPopes]

Thanks! Performance is unaffected

[DaniPopes]

Released in 0.4.1