Fixing perf issue with getitem by Boolean mask

[kozlov-alexey]

No description provided.

[AlexanderKalistratov]

any performance data?

[AlexanderKalistratov]

Cool

[kozlov-alexey]

any performance data?

@AlexanderKalistratov TC builds are failing now, so I re-measured manually on nnlmlp01 (the data size was increased to 5 * 10 ** 7 to increase exec times):

With fix:

				median	min	max	compile	boxing
name	nthreads	type	size
DataFrame.getitem_idx_bool_array	1	Python	50000000	0.729936	0.728974	0.734385	NaN	NaN
		SDC	50000000	0.783687	0.780611	0.78992	0.454833	2.437608
	2	SDC	50000000	0.420206	0.418938	0.42046	0.466937	2.445182
	4	SDC	50000000	0.2299	0.228938	0.245967	0.471535	2.501849
	8	SDC	50000000	0.140959	0.140459	0.14146	0.569357	2.520889
	16	SDC	50000000	0.114258	0.109348	0.155166	0.488567	2.548175
	28	SDC	50000000	0.093318	0.092599	0.094481	0.527987	2.517458
	56	SDC	50000000	0.084853	0.082968	0.086843	0.469488	2.520272
DataFrame.getitem_idx_bool_series	1	Python	50000000	0.727004	0.726642	0.73025	NaN	NaN
		SDC	50000000	0.667441	0.660132	0.678571	0.464186	2.483425
	2	SDC	50000000	0.34452	0.343004	0.34593	0.481691	2.452096
	4	SDC	50000000	0.185001	0.183356	0.196492	0.606666	2.49135
	8	SDC	50000000	0.111713	0.105505	0.126037	0.421492	2.495467
	16	SDC	50000000	0.075595	0.065445	0.083412	0.494176	2.525917
	28	SDC	50000000	0.065051	0.064746	0.073442	0.446053	2.46728
	56	SDC	50000000	0.06671	0.065066	0.067768	0.430038	2.518629

On master (without fix):

				median	min	max	compile	boxing
name	nthreads	type	size
DataFrame.getitem_idx_bool_array	1	Python	50000000	0.709037	0.705619	0.712697	NaN	NaN
		SDC	50000000	1.295295	1.249202	1.332165	0.573321	2.476892
	2	SDC	50000000	0.734991	0.730225	0.774579	0.533662	2.522519
	4	SDC	50000000	0.430704	0.404743	0.436254	0.520294	2.499454
	8	SDC	50000000	0.27165	0.268377	0.279392	0.593541	2.522613
	16	SDC	50000000	0.193998	0.19062	0.19909	0.607101	2.523254
	28	SDC	50000000	0.183823	0.183236	0.184575	0.449276	2.489915
	56	SDC	50000000	0.182439	0.173317	0.190032	0.655567	2.55842
DataFrame.getitem_idx_bool_series	1	Python	50000000	0.722556	0.719484	0.724653	NaN	NaN
		SDC	50000000	0.887583	0.883327	0.891517	0.519036	2.51013
	2	SDC	50000000	0.515719	0.51462	0.519068	0.477245	2.470925
	4	SDC	50000000	0.281323	0.280721	0.289484	0.478673	2.472751
	8	SDC	50000000	0.179992	0.178943	0.183958	0.452823	2.4679
	16	SDC	50000000	0.130896	0.127235	0.139556	0.462505	2.51652
	28	SDC	50000000	0.098136	0.096969	0.108245	0.496178	2.503406
	56	SDC	50000000	0.096119	0.094569	0.100459	0.442157	2.560549