[AST] [Evaluation] Use 'SmallArray' in 'Case'

[effectfully]

Another competitor. Previous in the series: #5818, #5813, #5816.

[effectfully]

/benchmark nofib

[github-actions]

Click here to check the status of your benchmark.

[github-actions]

Comparing benchmark results of 'nofib' on 'fe9876c4b' (base) and '512301cd0' (PR)

Results table

Script	fe9876c	512301c	Change
clausify/formula1	3.796 ms	2.886 ms	-24.0%
clausify/formula2	5.000 ms	3.844 ms	-23.1%
clausify/formula3	13.56 ms	10.43 ms	-23.1%
clausify/formula4	28.11 ms	23.33 ms	-17.0%
clausify/formula5	65.47 ms	50.51 ms	-22.9%
knights/4x4	17.97 ms	15.96 ms	-11.2%
knights/6x6	47.01 ms	42.09 ms	-10.5%
knights/8x8	82.80 ms	73.66 ms	-11.0%
primetest/05digits	8.965 ms	8.963 ms	-0.0%
primetest/08digits	14.36 ms	14.43 ms	+0.5%
primetest/10digits	17.24 ms	17.28 ms	+0.2%
primetest/20digits	34.42 ms	34.44 ms	+0.1%
primetest/30digits	52.38 ms	53.16 ms	+1.5%
primetest/40digits	71.17 ms	71.78 ms	+0.9%
primetest/50digits	86.68 ms	87.66 ms	+1.1%
queens4x4/bt	5.547 ms	5.009 ms	-9.7%
queens4x4/bm	7.302 ms	6.542 ms	-10.4%
queens4x4/bjbt1	6.879 ms	6.201 ms	-9.9%
queens4x4/bjbt2	6.355 ms	5.749 ms	-9.5%
queens4x4/fc	15.04 ms	13.36 ms	-11.2%
queens5x5/bt	74.50 ms	67.37 ms	-9.6%
queens5x5/bm	81.69 ms	73.70 ms	-9.8%
queens5x5/bjbt1	87.65 ms	79.59 ms	-9.2%
queens5x5/bjbt2	84.95 ms	76.90 ms	-9.5%
queens5x5/fc	189.6 ms	169.0 ms	-10.9%
TOTAL	1.108 s	1.014 s	-8.5%

[michaelpj]

Looks like Vector is faster? Could be measurement error, but the difference looks big and consistent enough to be real?

[effectfully]

/benchmark nofib

[github-actions]

Click here to check the status of your benchmark.

[effectfully]

Looks like Vector is faster? Could be measurement error, but the difference looks big and consistent enough to be real?

Yes, let me rerun and check the diff again, but otherwise Vector is clearly better, which I find really surprising. Maybe it's fusion that's kicking in somehow.

[michaelpj]

I'll rebase my branch and rerun the benchmarks in case something has changed

[github-actions]

Comparing benchmark results of 'nofib' on 'fe9876c4b' (base) and '512301cd0' (PR)

Results table

Script	fe9876c	512301c	Change
clausify/formula1	3.778 ms	2.870 ms	-24.0%
clausify/formula2	4.977 ms	3.845 ms	-22.7%
clausify/formula3	13.49 ms	10.46 ms	-22.5%
clausify/formula4	27.99 ms	23.38 ms	-16.5%
clausify/formula5	65.24 ms	50.62 ms	-22.4%
knights/4x4	17.84 ms	15.90 ms	-10.9%
knights/6x6	46.80 ms	42.19 ms	-9.9%
knights/8x8	82.05 ms	73.60 ms	-10.3%
primetest/05digits	8.834 ms	8.973 ms	+1.6%
primetest/08digits	14.19 ms	14.39 ms	+1.4%
primetest/10digits	17.05 ms	17.23 ms	+1.1%
primetest/20digits	34.19 ms	34.49 ms	+0.9%
primetest/30digits	52.10 ms	52.96 ms	+1.7%
primetest/40digits	70.34 ms	71.93 ms	+2.3%
primetest/50digits	85.98 ms	87.84 ms	+2.2%
queens4x4/bt	5.525 ms	4.998 ms	-9.5%
queens4x4/bm	7.263 ms	6.527 ms	-10.1%
queens4x4/bjbt1	6.815 ms	6.169 ms	-9.5%
queens4x4/bjbt2	6.317 ms	5.729 ms	-9.3%
queens4x4/fc	14.84 ms	13.28 ms	-10.5%
queens5x5/bt	73.88 ms	67.07 ms	-9.2%
queens5x5/bm	80.99 ms	73.33 ms	-9.5%
queens5x5/bjbt1	87.07 ms	79.23 ms	-9.0%
queens5x5/bjbt2	84.56 ms	76.43 ms	-9.6%
queens5x5/fc	188.8 ms	167.8 ms	-11.1%
TOTAL	1.101 s	1.011 s	-8.1%

[effectfully]

CPU performance isn't that big of a deal anymore and what we can get here should be fairly insignificant, so I added it to the epic and let's close it for now.