numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
Documentation 
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
305
306

//! AVX2 SIMD implementations for cumulative operations
//!
//! Vectorizes over the inner_size dimension - each SIMD lane maintains
//! its own independent accumulator through the scan dimension.

#[cfg(target_arch = "x86_64")]
use std::arch::x86_64::*;

// ============================================================================
// Cumsum Strided - AVX2
// ============================================================================

/// AVX2 strided cumsum for f32.
///
/// Processes 8 inner positions simultaneously, each with its own accumulator.
#[target_feature(enable = "avx2")]
pub unsafe fn cumsum_strided_f32(
    a: *const f32,
    out: *mut f32,
    scan_size: usize,
    outer_size: usize,
    inner_size: usize,
) {
    const LANES: usize = 8;
    let chunks = inner_size / LANES;

    for o in 0..outer_size {
        let outer_base = o * scan_size * inner_size;

        // Process SIMD chunks
        for chunk in 0..chunks {
            let i_base = chunk * LANES;

            // Initialize 8 accumulators to zero
            let mut acc = _mm256_setzero_ps();

            // Scan through the scan dimension
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i_base;

                // Load 8 consecutive inner elements
                let val = _mm256_loadu_ps(a.add(idx));

                // Add to running sum
                acc = _mm256_add_ps(acc, val);

                // Store cumulative sum
                _mm256_storeu_ps(out.add(idx), acc);
            }
        }

        // Handle remainder with scalar
        let i_start = chunks * LANES;
        for i in i_start..inner_size {
            let mut acc = 0.0f32;
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i;
                acc += *a.add(idx);
                *out.add(idx) = acc;
            }
        }
    }
}

/// AVX2 strided cumsum for f64.
#[target_feature(enable = "avx2")]
pub unsafe fn cumsum_strided_f64(
    a: *const f64,
    out: *mut f64,
    scan_size: usize,
    outer_size: usize,
    inner_size: usize,
) {
    const LANES: usize = 4;
    let chunks = inner_size / LANES;

    for o in 0..outer_size {
        let outer_base = o * scan_size * inner_size;

        // Process SIMD chunks
        for chunk in 0..chunks {
            let i_base = chunk * LANES;

            // Initialize 4 accumulators to zero
            let mut acc = _mm256_setzero_pd();

            // Scan through the scan dimension
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i_base;

                // Load 4 consecutive inner elements
                let val = _mm256_loadu_pd(a.add(idx));

                // Add to running sum
                acc = _mm256_add_pd(acc, val);

                // Store cumulative sum
                _mm256_storeu_pd(out.add(idx), acc);
            }
        }

        // Handle remainder with scalar
        let i_start = chunks * LANES;
        for i in i_start..inner_size {
            let mut acc = 0.0f64;
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i;
                acc += *a.add(idx);
                *out.add(idx) = acc;
            }
        }
    }
}

// ============================================================================
// Cumprod Strided - AVX2
// ============================================================================

/// AVX2 strided cumprod for f32.
#[target_feature(enable = "avx2")]
pub unsafe fn cumprod_strided_f32(
    a: *const f32,
    out: *mut f32,
    scan_size: usize,
    outer_size: usize,
    inner_size: usize,
) {
    const LANES: usize = 8;
    let chunks = inner_size / LANES;

    for o in 0..outer_size {
        let outer_base = o * scan_size * inner_size;

        // Process SIMD chunks
        for chunk in 0..chunks {
            let i_base = chunk * LANES;

            // Initialize 8 accumulators to one
            let mut acc = _mm256_set1_ps(1.0);

            // Scan through the scan dimension
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i_base;

                // Load 8 consecutive inner elements
                let val = _mm256_loadu_ps(a.add(idx));

                // Multiply with running product
                acc = _mm256_mul_ps(acc, val);

                // Store cumulative product
                _mm256_storeu_ps(out.add(idx), acc);
            }
        }

        // Handle remainder with scalar
        let i_start = chunks * LANES;
        for i in i_start..inner_size {
            let mut acc = 1.0f32;
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i;
                acc *= *a.add(idx);
                *out.add(idx) = acc;
            }
        }
    }
}

/// AVX2 strided cumprod for f64.
#[target_feature(enable = "avx2")]
pub unsafe fn cumprod_strided_f64(
    a: *const f64,
    out: *mut f64,
    scan_size: usize,
    outer_size: usize,
    inner_size: usize,
) {
    const LANES: usize = 4;
    let chunks = inner_size / LANES;

    for o in 0..outer_size {
        let outer_base = o * scan_size * inner_size;

        // Process SIMD chunks
        for chunk in 0..chunks {
            let i_base = chunk * LANES;

            // Initialize 4 accumulators to one
            let mut acc = _mm256_set1_pd(1.0);

            // Scan through the scan dimension
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i_base;

                // Load 4 consecutive inner elements
                let val = _mm256_loadu_pd(a.add(idx));

                // Multiply with running product
                acc = _mm256_mul_pd(acc, val);

                // Store cumulative product
                _mm256_storeu_pd(out.add(idx), acc);
            }
        }

        // Handle remainder with scalar
        let i_start = chunks * LANES;
        for i in i_start..inner_size {
            let mut acc = 1.0f64;
            for s in 0..scan_size {
                let idx = outer_base + s * inner_size + i;
                acc *= *a.add(idx);
                *out.add(idx) = acc;
            }
        }
    }
}

// ============================================================================
// Tests
// ============================================================================

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

    fn has_avx2() -> bool {
        is_x86_feature_detected!("avx2")
    }

    #[test]
    fn test_cumsum_strided_f32_avx2() {
        if !has_avx2() {
            return;
        }

        // 1 outer segment, scan_size=4, inner_size=16 (will use SIMD)
        let input: Vec<f32> = (0..64).map(|x| x as f32).collect();
        let mut output = vec![0.0f32; 64];

        unsafe {
            cumsum_strided_f32(input.as_ptr(), output.as_mut_ptr(), 4, 1, 16);
        }

        // Verify i=0: values at 0, 16, 32, 48
        // cumsum: 0, 0+16=16, 16+32=48, 48+48=96
        assert_eq!(output[0], 0.0);
        assert_eq!(output[16], 16.0);
        assert_eq!(output[32], 48.0);
        assert_eq!(output[48], 96.0);

        // Verify i=1: values at 1, 17, 33, 49
        // cumsum: 1, 1+17=18, 18+33=51, 51+49=100
        assert_eq!(output[1], 1.0);
        assert_eq!(output[17], 18.0);
        assert_eq!(output[33], 51.0);
        assert_eq!(output[49], 100.0);
    }

    #[test]
    fn test_cumprod_strided_f32_avx2() {
        if !has_avx2() {
            return;
        }

        // 1 outer segment, scan_size=3, inner_size=8
        let input = vec![
            2.0f32, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, // s=0
            3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, // s=1
            4.0, 4.0, 4.0, 4.0, 4.0, 4.0, 4.0, 4.0, // s=2
        ];
        let mut output = vec![0.0f32; 24];

        unsafe {
            cumprod_strided_f32(input.as_ptr(), output.as_mut_ptr(), 3, 1, 8);
        }

        // Each column should be: 2, 2*3=6, 6*4=24
        for i in 0..8 {
            assert_eq!(output[i], 2.0, "s=0, i={}", i);
            assert_eq!(output[8 + i], 6.0, "s=1, i={}", i);
            assert_eq!(output[16 + i], 24.0, "s=2, i={}", i);
        }
    }

    #[test]
    fn test_cumsum_strided_f64_avx2() {
        if !has_avx2() {
            return;
        }

        // 1 outer segment, scan_size=3, inner_size=8
        let input: Vec<f64> = (0..24).map(|x| x as f64).collect();
        let mut output = vec![0.0f64; 24];

        unsafe {
            cumsum_strided_f64(input.as_ptr(), output.as_mut_ptr(), 3, 1, 8);
        }

        // Verify i=0: values at 0, 8, 16
        // cumsum: 0, 0+8=8, 8+16=24
        assert_eq!(output[0], 0.0);
        assert_eq!(output[8], 8.0);
        assert_eq!(output[16], 24.0);
    }
}