lmutils 0.2.92

A set of utilities for working with linear regression models
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
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304

#![allow(clippy::needless_range_loop, clippy::missing_safety_doc)]

#[inline(always)]
pub fn mean(data: &[f64]) -> f64 {
    if is_x86_feature_detected!("avx512f") {
        unsafe { mean_avx512(data).0 }
    } else if is_x86_feature_detected!("avx2") {
        unsafe { mean_avx2(data).0 }
    } else if is_x86_feature_detected!("sse4.1") {
        unsafe { mean_sse4(data).0 }
    } else {
        mean_naive(data).0
    }
}

#[inline(always)]
pub fn mean_naive(data: &[f64]) -> (f64, u64) {
    let mut sum = 0.0;
    let mut count = 0;
    for i in 0..data.len() {
        let d = data[i];
        if !d.is_nan() {
            count += 1;
            sum += d;
        }
    }
    (if count == 0 { 0.0 } else { sum / count as f64 }, count)
}

#[inline(always)]
pub unsafe fn mean_sse4(data: &[f64]) -> (f64, u64) {
    let mut sum: f64;
    let mut count: u64;
    core::arch::asm! {
        // xmm1 = sum
        "xorpd xmm1, xmm1",
        // xmm2 = count
        "pxor xmm2, xmm2",
        // xmm3 = 1
        "movddup xmm3, xmm3",
        // xmm4 = 0
        "xorpd xmm4, xmm4",
        "test rax, rax",
        "jz 3f",
            "2:",
            // load from memory
            "movupd xmm0, [rsi]",
            "movupd xmm5, xmm0",
            // check for NaNs
            "cmppd xmm0, xmm0, 3",
            // replace NaNs with 0.0
            "blendvpd xmm5, xmm4",
            // replace NaNs with 1 for counting
            "pand xmm0, xmm3",
            // count NaNs
            "paddq xmm2, xmm0",
            // sum
            "addpd xmm1, xmm5",

            "add rsi, 16",
            "dec rax",
            "jnz 2b",
        "3:",
        "haddpd xmm1, xmm1",
        // sum xmm2
        "movupd xmm3, xmm2",
        "psrldq xmm2, 8",
        "paddq xmm2, xmm3",

        out("xmm0") _,
        out("xmm1") sum,
        out("xmm2") count,
        inout("xmm3") 1 => _,
        out("xmm4") _,
        out("xmm5") _,
        inout("rsi") data.as_ptr() => _,
        inout("rax") data.len() / 2 => _,
        options(readonly, nostack),
    }
    if data.len() % 2 != 0 {
        let d = data[data.len() - 1];
        if d.is_nan() {
            count += 1;
        } else {
            sum += d;
        }
    }
    count = data.len() as u64 - count;
    (if count == 0 { 0.0 } else { sum / count as f64 }, count)
}

#[inline(always)]
pub unsafe fn mean_avx2(data: &[f64]) -> (f64, u64) {
    let mut sum: f64;
    let mut count: u64;
    core::arch::asm! {
        // ymm0 = sum
        "vxorpd ymm0, ymm0, ymm0",
        // ymm1 = count
        "vpxor ymm1, ymm1, ymm1",
        // ymm2 = 1
        "vbroadcastsd ymm2, xmm2",
        // ymm3 = 0
        "vxorpd ymm3, ymm3, ymm3",
        "test rax, rax",
        "jz 3f",
            "2:",
            // load from memory
            "vmovupd ymm5, [rsi]",
            // check for NaNs
            "vcmppd ymm4, ymm5, ymm5, 3",
            // replace NaNs with 0.0
            "vblendvpd ymm5, ymm5, ymm3, ymm4",
            // replace NaNs with 1 for counting
            "vpand ymm4, ymm4, ymm2",
            // count NaNs
            "vpaddq ymm1, ymm1, ymm4",
            // sum
            "vaddpd ymm0, ymm0, ymm5",

            "add rsi, 32",
            "dec rax",
            "jnz 2b",
        "3:",
        // extract sum
        "vextractf128 xmm2, ymm0, 1",
        "vaddpd xmm0, xmm0, xmm2",
        "vhaddpd xmm0, xmm0, xmm0",
        // extract count
        "vextracti128 xmm2, ymm1, 1",
        "vpaddq xmm1, xmm1, xmm2",
        "movupd xmm2, xmm1",
        "psrldq xmm1, 8",
        "paddq xmm1, xmm2",
        "vzeroupper",

        out("xmm0") sum,
        out("xmm1") count,
        inout("xmm2") 1 => _,
        out("ymm3") _,
        out("ymm4") _,
        out("ymm5") _,
        inout("rax") data.len() / 4 => _,
        inout("rsi") data.as_ptr() => _,
        options(readonly, nostack),
    }
    if data.len() % 4 != 0 {
        for i in (data.len() - data.len() % 4)..data.len() {
            let d = data[i];
            if d.is_nan() {
                count += 1;
            } else {
                sum += d;
            }
        }
    }
    count = data.len() as u64 - count;
    (if count == 0 { 0.0 } else { sum / count as f64 }, count)
}

#[inline(always)]
pub unsafe fn mean_avx512(data: &[f64]) -> (f64, u64) {
    let mut sum: f64;
    let mut non_nan: u64;
    core::arch::asm! {
        // zmm0 = sum
        "vxorpd zmm0, zmm0, zmm0",
        // zmm1 = count
        "vpxorq zmm1, zmm1, zmm1",
        // zmm2 = 1
        "vbroadcastsd zmm2, xmm2",
        // zmm3 = 0
        "vxorpd zmm3, zmm3, zmm3",
        "test rax, rax",
        "jz 3f",
            "2:",
            // load from memory
            "vmovupd zmm4, [rsi]",
            // check for NaNs
            "vcmppd k1, zmm4, zmm4, 0",
            // count non-NaNs
            "vpaddq zmm1{{k1}}, zmm1, zmm2",
            // sum
            "vaddpd zmm0{{k1}}, zmm0, zmm4",

            "add rsi, 64",
            "dec rax",
            "jnz 2b",
        "3:",
        // extract sum
        "vextractf64x4 ymm2, zmm0, 1",
        "vaddpd ymm0, ymm0, ymm2",
        "vextractf64x2 xmm2, ymm0, 1",
        "vaddpd xmm0, xmm0, xmm2",
        "vhaddpd xmm0, xmm0, xmm0",
        // extract count
        "vextracti64x4 ymm2, zmm1, 1",
        "vpaddq ymm1, ymm1, ymm2",
        "vextracti64x2 xmm2, ymm1, 1",
        "vpaddq xmm1, xmm1, xmm2",
        "movupd xmm2, xmm1",
        "psrldq xmm1, 8",
        "paddq xmm1, xmm2",
        "vzeroupper",

        out("xmm0") sum,
        out("xmm1") non_nan,
        inout("xmm2") 1 => _,
        out("zmm3") _,
        out("zmm4") _,
        inout("rax") data.len() / 8 => _,
        inout("rsi") data.as_ptr() => _,
        options(readonly, nostack),
    }
    if data.len() % 8 != 0 {
        for i in (data.len() - data.len() % 8)..data.len() {
            let d = data[i];
            if !d.is_nan() {
                non_nan += 1;
                sum += d;
            }
        }
    }
    (
        if non_nan == 0 {
            0.0
        } else {
            sum / non_nan as f64
        },
        non_nan,
    )
}

#[cfg(test)]
mod tests {
    use super::*;

    fn data() -> Vec<f64> {
        [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0]
            .iter()
            .cycle()
            .take(8 * 1000 - 1)
            .copied()
            .collect::<Vec<f64>>()
    }

    fn data_nan() -> Vec<f64> {
        [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, f64::NAN, f64::NAN]
            .iter()
            .cycle()
            .take(8 * 1000 - 1)
            .copied()
            .collect::<Vec<f64>>()
    }

    const MEAN: (f64, u64) = ((36000.0 - 8.0) / 7999.0, 7999);
    const MEAN_NAN: (f64, u64) = (21000.0 / 6000.0, 6000);

    #[test]
    fn test_mean_naive() {
        assert_eq!(mean_naive(&data()), MEAN);
    }

    #[test]
    fn test_mean_naive_nan() {
        assert_eq!(mean_naive(&data_nan()), MEAN_NAN);
    }

    #[test]
    #[cfg_attr(not(target_feature = "sse4.1"), ignore)]
    fn test_mean_sse4() {
        assert_eq!(unsafe { mean_sse4(&data()) }, MEAN);
    }

    #[test]
    #[cfg_attr(not(target_feature = "sse4.1"), ignore)]
    fn test_mean_sse4_nan() {
        assert_eq!(unsafe { mean_sse4(&data_nan()) }, MEAN_NAN);
    }

    #[test]
    #[cfg_attr(not(target_feature = "avx2"), ignore)]
    fn test_mean_avx2() {
        assert_eq!(unsafe { mean_avx2(&data()) }, MEAN);
    }

    #[test]
    #[cfg_attr(not(target_feature = "avx2"), ignore)]
    fn test_mean_avx2_nan() {
        assert_eq!(unsafe { mean_avx2(&data_nan()) }, MEAN_NAN);
    }

    #[test]
    #[cfg_attr(not(target_feature = "avx512f"), ignore)]
    fn test_mean_avx512() {
        assert_eq!(unsafe { mean_avx512(&data()) }, MEAN);
    }

    #[test]
    #[cfg_attr(not(target_feature = "avx512f"), ignore)]
    fn test_mean_avx512_nan() {
        assert_eq!(unsafe { mean_avx512(&data_nan()) }, MEAN_NAN);
    }
}