trueno 0.18.0

High-performance SIMD compute library with GPU support, LLM inference engine, and GGUF model loading
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

//! AVX2 reduction operations (dot, sum, max, min, argmax, argmin).

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

use crate::backends::VectorBackend;

/// AVX2 dot product with 4-accumulator unrolling for ILP.
#[inline]
#[target_feature(enable = "avx2,fma")]
// SAFETY: caller ensures preconditions are met for this unsafe function
pub(crate) unsafe fn dot(a: &[f32], b: &[f32]) -> f32 {
    unsafe {
        let len = a.len();
        let mut i = 0;

        let mut acc0 = _mm256_setzero_ps();
        let mut acc1 = _mm256_setzero_ps();
        let mut acc2 = _mm256_setzero_ps();
        let mut acc3 = _mm256_setzero_ps();

        while i + 32 <= len {
            let va0 = _mm256_loadu_ps(a.as_ptr().add(i));
            let vb0 = _mm256_loadu_ps(b.as_ptr().add(i));
            let va1 = _mm256_loadu_ps(a.as_ptr().add(i + 8));
            let vb1 = _mm256_loadu_ps(b.as_ptr().add(i + 8));
            let va2 = _mm256_loadu_ps(a.as_ptr().add(i + 16));
            let vb2 = _mm256_loadu_ps(b.as_ptr().add(i + 16));
            let va3 = _mm256_loadu_ps(a.as_ptr().add(i + 24));
            let vb3 = _mm256_loadu_ps(b.as_ptr().add(i + 24));

            acc0 = _mm256_fmadd_ps(va0, vb0, acc0);
            acc1 = _mm256_fmadd_ps(va1, vb1, acc1);
            acc2 = _mm256_fmadd_ps(va2, vb2, acc2);
            acc3 = _mm256_fmadd_ps(va3, vb3, acc3);

            i += 32;
        }

        while i + 8 <= len {
            let va = _mm256_loadu_ps(a.as_ptr().add(i));
            let vb = _mm256_loadu_ps(b.as_ptr().add(i));
            acc0 = _mm256_fmadd_ps(va, vb, acc0);
            i += 8;
        }

        let acc01 = _mm256_add_ps(acc0, acc1);
        let acc23 = _mm256_add_ps(acc2, acc3);
        let acc = _mm256_add_ps(acc01, acc23);

        let mut result = {
            let sum_halves = _mm_add_ps(_mm256_castps256_ps128(acc), _mm256_extractf128_ps(acc, 1));
            let temp = _mm_add_ps(sum_halves, _mm_movehl_ps(sum_halves, sum_halves));
            let temp = _mm_add_ss(temp, _mm_shuffle_ps(temp, temp, 1));
            _mm_cvtss_f32(temp)
        };

        result += a[i..].iter().zip(&b[i..]).map(|(x, y)| x * y).sum::<f32>();
        result
    }
}

/// AVX2 vector sum.
#[inline]
#[target_feature(enable = "avx2")]
// SAFETY: caller ensures preconditions are met for this unsafe function
pub(crate) unsafe fn sum(a: &[f32]) -> f32 {
    unsafe {
        let len = a.len();
        let mut i = 0;
        let mut acc = _mm256_setzero_ps();

        while i + 8 <= len {
            let va = _mm256_loadu_ps(a.as_ptr().add(i));
            acc = _mm256_add_ps(acc, va);
            i += 8;
        }

        let mut result = {
            let sum_halves = _mm_add_ps(_mm256_castps256_ps128(acc), _mm256_extractf128_ps(acc, 1));
            let temp = _mm_add_ps(sum_halves, _mm_movehl_ps(sum_halves, sum_halves));
            let temp = _mm_add_ss(temp, _mm_shuffle_ps(temp, temp, 1));
            _mm_cvtss_f32(temp)
        };

        result += a[i..].iter().sum::<f32>();
        result
    }
}

/// AVX2 vector max.
#[inline]
#[target_feature(enable = "avx2")]
// SAFETY: caller ensures preconditions are met for this unsafe function
pub(crate) unsafe fn max(a: &[f32]) -> f32 {
    unsafe {
        let len = a.len();
        let mut i = 0;
        let mut vmax = _mm256_set1_ps(a[0]);

        while i + 8 <= len {
            let va = _mm256_loadu_ps(a.as_ptr().add(i));
            vmax = _mm256_max_ps(vmax, va);
            i += 8;
        }

        let mut result = {
            let max_halves =
                _mm_max_ps(_mm256_castps256_ps128(vmax), _mm256_extractf128_ps(vmax, 1));
            let temp = _mm_max_ps(max_halves, _mm_movehl_ps(max_halves, max_halves));
            let temp = _mm_max_ss(temp, _mm_shuffle_ps(temp, temp, 1));
            _mm_cvtss_f32(temp)
        };

        for &val in &a[i..] {
            if val > result {
                result = val;
            }
        }
        result
    }
}

/// AVX2 vector min.
#[inline]
#[target_feature(enable = "avx2")]
// SAFETY: caller ensures preconditions are met for this unsafe function
pub(crate) unsafe fn min(a: &[f32]) -> f32 {
    unsafe {
        let len = a.len();
        let mut i = 0;
        let mut vmin = _mm256_set1_ps(a[0]);

        while i + 8 <= len {
            let va = _mm256_loadu_ps(a.as_ptr().add(i));
            vmin = _mm256_min_ps(vmin, va);
            i += 8;
        }

        let mut result = {
            let min_halves =
                _mm_min_ps(_mm256_castps256_ps128(vmin), _mm256_extractf128_ps(vmin, 1));
            let temp = _mm_min_ps(min_halves, _mm_movehl_ps(min_halves, min_halves));
            let temp = _mm_min_ss(temp, _mm_shuffle_ps(temp, temp, 1));
            _mm_cvtss_f32(temp)
        };

        for &val in &a[i..] {
            if val < result {
                result = val;
            }
        }
        result
    }
}

/// AVX2 argmax.
#[inline]
#[target_feature(enable = "avx2")]
// SAFETY: caller ensures preconditions are met for this unsafe function
pub(crate) unsafe fn argmax(a: &[f32]) -> usize {
    unsafe {
        let len = a.len();
        let mut max_idx: usize = 0;
        let mut max_val = a[0];
        let mut i = 0;

        let mut vmax = _mm256_set1_ps(a[0]);
        let mut vidx_max = _mm256_setzero_ps();
        let vidx_inc = _mm256_set1_ps(8.0);
        let mut vcurrent_idx = _mm256_set_ps(7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0, 0.0);

        while i + 8 <= len {
            let va = _mm256_loadu_ps(a.as_ptr().add(i));
            let mask = _mm256_cmp_ps(va, vmax, _CMP_GT_OQ);
            vmax = _mm256_blendv_ps(vmax, va, mask);
            vidx_max = _mm256_blendv_ps(vidx_max, vcurrent_idx, mask);
            vcurrent_idx = _mm256_add_ps(vcurrent_idx, vidx_inc);
            i += 8;
        }

        // Extract max from vector
        let mut vals = [0.0f32; 8];
        let mut idxs = [0.0f32; 8];
        _mm256_storeu_ps(vals.as_mut_ptr(), vmax);
        _mm256_storeu_ps(idxs.as_mut_ptr(), vidx_max);

        for j in 0..8 {
            if vals[j] > max_val {
                max_val = vals[j];
                max_idx = idxs[j] as usize;
            }
        }

        // Check remaining elements
        for (j, &val) in a[i..].iter().enumerate() {
            if val > max_val {
                max_val = val;
                max_idx = i + j;
            }
        }

        max_idx
    }
}

/// AVX2 argmin.
#[inline]
#[target_feature(enable = "avx2")]
// SAFETY: caller ensures preconditions are met for this unsafe function
pub(crate) unsafe fn argmin(a: &[f32]) -> usize {
    unsafe {
        let len = a.len();
        let mut min_idx: usize = 0;
        let mut min_val = a[0];
        let mut i = 0;

        let mut vmin = _mm256_set1_ps(a[0]);
        let mut vidx_min = _mm256_setzero_ps();
        let vidx_inc = _mm256_set1_ps(8.0);
        let mut vcurrent_idx = _mm256_set_ps(7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0, 0.0);

        while i + 8 <= len {
            let va = _mm256_loadu_ps(a.as_ptr().add(i));
            let mask = _mm256_cmp_ps(va, vmin, _CMP_LT_OQ);
            vmin = _mm256_blendv_ps(vmin, va, mask);
            vidx_min = _mm256_blendv_ps(vidx_min, vcurrent_idx, mask);
            vcurrent_idx = _mm256_add_ps(vcurrent_idx, vidx_inc);
            i += 8;
        }

        let mut vals = [0.0f32; 8];
        let mut idxs = [0.0f32; 8];
        _mm256_storeu_ps(vals.as_mut_ptr(), vmin);
        _mm256_storeu_ps(idxs.as_mut_ptr(), vidx_min);

        for j in 0..8 {
            if vals[j] < min_val {
                min_val = vals[j];
                min_idx = idxs[j] as usize;
            }
        }

        for (j, &val) in a[i..].iter().enumerate() {
            if val < min_val {
                min_val = val;
                min_idx = i + j;
            }
        }

        min_idx
    }
}

/// Kahan sum for numerical stability (delegates to scalar).
#[inline]
// SAFETY: caller ensures preconditions are met for this unsafe function
pub(crate) unsafe fn sum_kahan(a: &[f32]) -> f32 {
    unsafe { crate::backends::scalar::ScalarBackend::sum_kahan(a) }
}