skip_whitespace: use a non-ref callback

[hkBst]

This simplifies code ever so slightly, but seems to optimize better with bench_skip_ascii_whitespace going from 13k ns to 12k ns for a ~7% perf win. This function is called by the unescape functions, but not the check functions, but none of the benchmarks for the unescape functions include whitespace escapes, so they should be just noise. Full results anyway:

bench name	current	new	diff
bench_check_raw_byte_str_ascii	44329.05 ns/iter (+/- 84.62)	44563.79 ns/iter (+/- 73.15)	0.5%
bench_check_raw_c_str_ascii	41291.01 ns/iter (+/- 47.98)	41074.82 ns/iter (+/- 55.79)	-0.5%
bench_check_raw_c_str_non_ascii	52422.74 ns/iter (+/- 77.9)	52528.6 ns/iter (+/- 43.03)	0.2%
bench_check_raw_c_str_unicode	174775.2 ns/iter (+/- 967.73)	174510.83 ns/iter (+/- 754.32)	-0.2%
bench_check_raw_str_ascii	41029.81 ns/iter (+/- 69.24)	40801.5 ns/iter (+/- 148.06)	-0.6%
bench_check_raw_str_non_ascii	52347.3 ns/iter (+/- 68.18)	50534.07 ns/iter (+/- 77.04)	-3.5%
bench_check_raw_str_unicode	171870.68 ns/iter (+/- 468.3)	168191.82 ns/iter (+/- 487.05)	-2.1%
bench_skip_ascii_whitespace	13069.39 ns/iter (+/- 10.35)	12146.74 ns/iter (+/- 41.47)	-7.1%
bench_unescape_byte_str_ascii	69227.9 ns/iter (+/- 115.92)	68957.02 ns/iter (+/- 128.93)	-0.4%
bench_unescape_byte_str_ascii_escape	85243.79 ns/iter (+/- 116.16)	81618.08 ns/iter (+/- 154.13)	-4.3%
bench_unescape_byte_str_hex_escape	125803.1 ns/iter (+/- 163.28)	124808.68 ns/iter (+/- 184.41)	-0.8%
bench_unescape_byte_str_mixed_escape	362076.7 ns/iter (+/- 1303.98)	360177.25 ns/iter (+/- 1211.76)	-0.5%
bench_unescape_c_str_ascii	84739.67 ns/iter (+/- 170.68)	82777.93 ns/iter (+/- 130.72)	-2.3%
bench_unescape_c_str_ascii_escape	97622.13 ns/iter (+/- 171.56)	100090.55 ns/iter (+/- 762.39)	2.5%
bench_unescape_c_str_hex_escape_ascii	164471.32 ns/iter (+/- 342.05)	166104.42 ns/iter (+/- 431.08)	1.0%
bench_unescape_c_str_hex_escape_byte	166290.54 ns/iter (+/- 276.57)	165707.0 ns/iter (+/- 270.56)	-0.4%
bench_unescape_c_str_mixed_escape	1061116.4 ns/iter (+/- 3354.17)	1064596.3 ns/iter (+/- 3814.25)	0.3%
bench_unescape_c_str_non_ascii	110599.81 ns/iter (+/- 196.05)	110626.81 ns/iter (+/- 269.57)	0.0%
bench_unescape_c_str_unicode	395626.4 ns/iter (+/- 1532.71)	390032.55 ns/iter (+/- 1455.75)	-1.4%
bench_unescape_c_str_unicode_escape	312402.87 ns/iter (+/- 1000.7)	311308.72 ns/iter (+/- 1546.52)	-0.4%
bench_unescape_str_ascii	71954.11 ns/iter (+/- 78.88)	73341.17 ns/iter (+/- 149.8)	1.9%
bench_unescape_str_ascii_escape	105721.6 ns/iter (+/- 180.65)	105908.94 ns/iter (+/- 156.22)	0.2%
bench_unescape_str_hex_escape	164854.82 ns/iter (+/- 216.09)	164763.26 ns/iter (+/- 419.93)	-0.1%
bench_unescape_str_mixed_escape	905346.95 ns/iter (+/- 1517.71)	905614.95 ns/iter (+/- 3783.9)	0.0%
bench_unescape_str_non_ascii	98009.93 ns/iter (+/- 164.66)	98224.14 ns/iter (+/- 82.23)	0.2%
bench_unescape_str_unicode	339978.93 ns/iter (+/- 618.14)	343732.22 ns/iter (+/- 563.01)	1.1%
bench_unescape_str_unicode_escape	597426.3 ns/iter (+/- 6292.9)	597124.2 ns/iter (+/- 6211.08)	-0.1%

[GuillaumeGomez]

Pretty neat, thanks!

I'm gonna send a follow-up to increase the version for a new release, unless you plan to send more before next release?

In any case, cc @Urgau

[hkBst]

I'm gonna send a follow-up to increase the version for a new release, unless you plan to send more before next release?

NonZero got approved, so I should really revive that code. Will be a tiny API break.

[GuillaumeGomez]

Then I'll wait for this new PR to be merged before making a new release. Can't wait to see what you'll come up with.

[hkBst]

@GuillaumeGomez If you're really curious, then see rust-lang/rust#141001 (comment).

[hkBst]

I think now that the perf wins here are a result of the simpler code triggering more inlining on ARM.