-
Notifications
You must be signed in to change notification settings - Fork 18
/
Copy pathhp_stat.jl
214 lines (206 loc) · 7.62 KB
/
hp_stat.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
function hp_maximum(f, x::AbstractVector{T}) where {T}
n = length(x)
nt = Threads.nthreads()
cz = div(n, nt)
cz == 0 && return stat_maximum(f, x)
CT = Core.Compiler.return_type(f, Tuple{our_nonmissingtype(eltype(x))})
if T >: Missing
CT = Union{Missing,CT}
end
res = Vector{CT}(undef, nt)
Threads.@threads for i in 1:nt
i == nt ? hi = n : hi = i * cz
res[i] = stat_maximum(f, x, lo=(i - 1) * cz + 1, hi=hi)
end
stat_maximum(identity, res)
end
hp_maximum(x::AbstractVector{T}) where {T} = hp_maximum(identity, x)
function hp_minimum(f, x::AbstractVector{T}) where {T}
n = length(x)
nt = Threads.nthreads()
cz = div(n, nt)
cz == 0 && return stat_minimum(f, x)
CT = Core.Compiler.return_type(f, Tuple{our_nonmissingtype(eltype(x))})
if T >: Missing
CT = Union{Missing,CT}
end
res = Vector{CT}(undef, nt)
Threads.@threads for i in 1:nt
i == nt ? hi = n : hi = i * cz
res[i] = stat_minimum(f, x, lo=(i - 1) * cz + 1, hi=hi)
end
stat_minimum(identity, res)
end
hp_minimum(x::AbstractVector{T}) where {T} = hp_minimum(identity, x)
function hp_sum(f, x::AbstractVector{T}) where {T}
n = length(x)
nt = Threads.nthreads()
cz = div(n, nt)
cz == 0 && return stat_sum(f, x)
CT = Core.Compiler.return_type(f, Tuple{our_nonmissingtype(eltype(x))})
CT <: Base.SmallSigned ? CT = Int : nothing
CT <: Base.SmallUnsigned ? CT = UInt : nothing
CT <: Bool ? CT = Int : nothing
if T >: Missing
CT = Union{Missing,CT}
end
res = Vector{CT}(undef, nt)
Threads.@threads for i in 1:nt
i == nt ? hi = n : hi = i * cz
res[i] = stat_sum(f, x, lo=(i - 1) * cz + 1, hi=hi)
end
stat_sum(identity, res)
end
hp_sum(x::AbstractVector{T}) where {T} = hp_sum(identity, x)
Base.@propagate_inbounds function hp_topk_vals(x::AbstractVector{T}, k::Int, lt_fun::F, by) where {T} where {F}
k < 1 && throw(ArgumentError("k must be greater than 1"))
all(ismissing, x) && return Union{Missing,T}[missing]
nt = Threads.nthreads()
res = Vector{T}(undef, k * nt)
res_out = allowmissing(res)
fill!(res_out, missing)
cz = div(length(x), nt)
Threads.@threads for i in 1:nt
lo = (i - 1) * cz + 1
i == nt ? hi = length(x) : hi = i * cz
th_res = view(res, (i-1)*k+1:i*k)
th_x = view(x, lo:hi)
th_res_out = view(res_out, (i-1)*k+1:i*k)
idx, cnt = initiate_topk_res!(th_res, th_x, by)
topk_sort!(th_res, 1, cnt, lt_fun)
for i in idx+1:length(th_x)
if !ismissing(by(th_x[i]))
insert_fixed_sorted!(th_res, th_x[i], lt_fun)
cnt += 1
end
end
if cnt < k
view(th_res_out, 1:cnt) .= view(th_res, 1:cnt)
else
th_res_out .= th_res
end
end
topk_vals(res_out, k, lt_fun, by)
end
Base.@propagate_inbounds function hp_topk_vals(x::Union{Vector{T}, SubArray{T, N, Vector{T}, Tuple{I}, L}}, k::Int, lt_fun::F, by) where {T<:Union{Missing, FLOATS, INTEGERS}} where {F} where N where I <: UnitRange{Int} where L
k < 1 && throw(ArgumentError("k must be greater than 1"))
all(ismissing, x) && return Union{Missing,T}[missing]
nt = Threads.nthreads()
res = Vector{T}(undef, k * nt)
res_out = allowmissing(res)
fill!(res_out, missing)
cz = div(length(x), nt)
Threads.@threads for i in 1:nt
lo = (i - 1) * cz + 1
i == nt ? hi = length(x) : hi = i * cz
th_res = view(res, (i-1)*k+1:i*k)
th_x = view(x, lo:hi)
th_res_out = view(res_out, (i-1)*k+1:i*k)
idx, cnt = initiate_topk_res!(th_res, th_x, by)
topk_sort!(th_res, 1, cnt, lt_fun)
if k < 21
for i in idx+1:length(th_x)
if !ismissing(by(th_x[i]))
insert_fixed_sorted!(th_res, th_x[i], lt_fun)
cnt += 1
end
end
else
for i in idx+1:length(th_x)
if !ismissing(by(th_x[i]))
insert_fixed_sorted_binary!(th_res, th_x[i], lt_fun)
cnt += 1
end
end
end
if cnt < k
view(th_res_out, 1:cnt) .= view(th_res, 1:cnt)
else
th_res_out .= th_res
end
end
topk_vals(res_out, k, lt_fun, by)
end
Base.@propagate_inbounds function hp_topk_perm(x::AbstractVector{T}, k::Int, lt_fun::F, by) where {T} where {F}
k < 1 && throw(ArgumentError("k must be greater than 1"))
all(ismissing, x) && return Union{Missing,Int}[missing]
nt = Threads.nthreads()
res = Vector{T}(undef, k * nt)
res_out = allowmissing(res)
fill!(res_out, missing)
perm = zeros(Int, k * nt)
perm_out = allowmissing(perm)
fill!(perm_out, missing)
cz = div(length(x), nt)
Threads.@threads for i in 1:nt
lo = (i - 1) * cz + 1
i == nt ? hi = length(x) : hi = i * cz
th_res = view(res, (i-1)*k+1:i*k)
th_perm = view(perm, (i-1)*k+1:i*k)
th_x = view(x, lo:hi)
th_res_out = view(res_out, (i-1)*k+1:i*k)
th_perm_out = view(perm_out, (i-1)*k+1:i*k)
idx, cnt = initiate_topk_res_perm!(th_perm, th_res, th_x, by, offset = lo - 1)
topk_sort_permute!(th_res, th_perm, 1, cnt, lt_fun)
for i in idx+1:length(th_x)
if !ismissing(by(th_x[i]))
insert_fixed_sorted_perm!(th_perm, th_res, i + lo - 1, th_x[i], lt_fun)
cnt += 1
end
end
if cnt < k
view(th_res_out, 1:cnt) .= view(th_res, 1:cnt)
view(th_perm_out, 1:cnt) .= view(th_perm, 1:cnt)
else
th_res_out .= th_res
th_perm_out .= th_perm
end
end
perm_out[topk_perm(res_out, k, lt_fun, by)]
end
Base.@propagate_inbounds function hp_topk_perm(x::Union{Vector{T}, SubArray{T, N, Vector{T}, Tuple{I}, L}}, k::Int, lt_fun::F, by) where {T<:Union{Missing, FLOATS, INTEGERS}} where {F} where N where I <: UnitRange{Int} where L
k < 1 && throw(ArgumentError("k must be greater than 1"))
all(ismissing, x) && return Union{Missing,Int}[missing]
nt = Threads.nthreads()
res = Vector{T}(undef, k * nt)
res_out = allowmissing(res)
fill!(res_out, missing)
perm = zeros(Int, k * nt)
perm_out = allowmissing(perm)
fill!(perm_out, missing)
cz = div(length(x), nt)
Threads.@threads for i in 1:nt
lo = (i - 1) * cz + 1
i == nt ? hi = length(x) : hi = i * cz
th_res = view(res, (i-1)*k+1:i*k)
th_perm = view(perm, (i-1)*k+1:i*k)
th_x = view(x, lo:hi)
th_res_out = view(res_out, (i-1)*k+1:i*k)
th_perm_out = view(perm_out, (i-1)*k+1:i*k)
idx, cnt = initiate_topk_res_perm!(th_perm, th_res, th_x, by, offset = lo - 1)
topk_sort_permute!(th_res, th_perm, 1, cnt, lt_fun)
if k < 16
for i in idx+1:length(th_x)
if !ismissing(by(th_x[i]))
insert_fixed_sorted_perm!(th_perm, th_res, i + lo - 1, th_x[i], lt_fun)
cnt += 1
end
end
else
for i in idx+1:length(th_x)
if !ismissing(by(th_x[i]))
insert_fixed_sorted_perm_binary!(th_perm, th_res, i + lo - 1, th_x[i], lt_fun)
cnt += 1
end
end
end
if cnt < k
view(th_res_out, 1:cnt) .= view(th_res, 1:cnt)
view(th_perm_out, 1:cnt) .= view(th_perm, 1:cnt)
else
th_res_out .= th_res
th_perm_out .= th_perm
end
end
perm_out[topk_perm(res_out, k, lt_fun, by)]
end