Improve deku performance

[wcampbell0x2a]

Still want to do benchmarks and more testing, but it's done-ish

See #344

[github-actions]

Benchmark for d1b38cf

Click to view benchmark

Test	Base	PR	%
deku_read_bits	608.6±0.50ns	608.7±0.69ns	+0.02%
deku_read_byte	17.6±0.09ns	8.4±0.04ns	-52.27%
deku_read_enum	30.1±0.54ns	16.7±0.12ns	-44.52%
deku_read_vec	1402.0±11.22ns	934.5±8.60ns	-33.35%
deku_read_vec_perf	1404.7±10.10ns	947.9±5.13ns	-32.52%
deku_write_bits	108.0±0.20ns	111.4±0.61ns	+3.15%
deku_write_byte	66.7±0.25ns	66.7±0.25ns	0.00%
deku_write_enum	111.8±0.37ns	115.0±0.63ns	+2.86%
deku_write_vec	5.2±0.02µs	5.2±0.01µs	0.00%
deku_write_vec_perf	5.3±0.01µs	5.2±0.01µs	-1.89%

[github-actions]

Benchmark for a443c4b

Click to view benchmark

Test	Base	PR	%
deku_read_bits	608.9±0.42ns	608.7±0.84ns	-0.03%
deku_read_byte	17.6±0.09ns	8.4±0.00ns	-52.27%
deku_read_enum	29.7±0.15ns	16.8±0.13ns	-43.43%
deku_read_vec	1399.1±4.71ns	934.4±8.82ns	-33.21%
deku_read_vec_perf	1398.2±5.90ns	949.8±5.92ns	-32.07%
deku_write_bits	108.2±0.24ns	110.9±0.41ns	+2.50%
deku_write_byte	66.7±0.34ns	66.4±0.26ns	-0.45%
deku_write_enum	111.7±0.38ns	114.9±0.44ns	+2.86%
deku_write_vec	5.2±0.01µs	5.2±0.01µs	0.00%
deku_write_vec_perf	5.3±0.01µs	5.2±0.01µs	-1.89%

[github-actions]

Benchmark for 1b73cc4

Click to view benchmark

Test	Base	PR	%
deku_read_bits	608.8±0.45ns	604.6±0.44ns	-0.69%
deku_read_byte	17.7±0.13ns	9.7±0.16ns	-45.20%
deku_read_enum	29.8±0.09ns	17.9±0.16ns	-39.93%
deku_read_vec	1404.0±6.45ns	940.2±4.42ns	-33.03%
deku_read_vec_perf	1403.3±8.17ns	909.6±4.37ns	-35.18%
deku_write_bits	112.2±0.55ns	110.3±0.56ns	-1.69%
deku_write_byte	68.5±6.98ns	64.7±0.16ns	-5.55%
deku_write_enum	111.7±0.58ns	110.2±0.17ns	-1.34%
deku_write_vec	5.2±0.01µs	5.1±0.01µs	-1.92%
deku_write_vec_perf	5.3±0.02µs	5.1±0.01µs	-3.77%

[v1gnesh]

Hey @wcampbell0x2a I have a dumb question -
Since in many cases, parsing is for static content (either sizes known fully at compile time or reasonable upper bounds may be known for dynamic content), would it help if memory for these were pre-allocated before getting into read?

[wcampbell0x2a]

Hey @wcampbell0x2a I have a dumb question - Since in many cases, parsing is for static content (either sizes known fully at compile time or reasonable upper bounds may be known for dynamic content), would it help if memory for these were pre-allocated before getting into read?

You can see some tests for allocation counts here: /~https://github.com/sharksforarms/deku/blob/master/tests/test_alloc.rs. Many of these allocations are quite dynamic based upon bit alignment when reading.

[wcampbell0x2a]

Benchmark for 1b73cc4

Click to view benchmark

This commit showed a %5 performance increase for deku deku_read_byte on my machine.

[sharksforarms]

Yeah not sure how much I trust the CI benchmarking, I'm not sure it's fully accurate. I'll take a look at this when I have more time, thanks for working on this!

[wcampbell0x2a]

This breaks the logging feature. I'll need to fix that.

[wcampbell0x2a]

I also need to test the old + new with both having LTO enabled. The performance might be worse with this patch.

not true