hodu_cpu_kernels 0.2.4

hodu cpu kernels
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

use hodu_cpu_kernels::*;

fn approx(v: Vec<f32>, digits: i32) -> Vec<f32> {
    let b = 10f32.powi(digits);
    v.iter().map(|t| f32::round(t * b) / b).collect()
}

fn run_binary(lhs: &[f32], rhs: &[f32], kernel: Kernel) -> Vec<f32> {
    assert_eq!(lhs.len(), rhs.len());
    let mut output = vec![0.0f32; lhs.len()];
    let shape = vec![lhs.len()];
    let strides = vec![1];

    let num_els = lhs.len();
    let num_dims = 1;
    let mut metadata = Vec::new();
    metadata.push(num_els);
    metadata.push(num_dims);
    metadata.extend(&shape); // lhs_shape
    metadata.extend(&shape); // rhs_shape
    metadata.extend(&strides); // lhs_strides
    metadata.extend(&strides); // rhs_strides
    metadata.push(0); // lhs_offset
    metadata.push(0); // rhs_offset

    call_ops_binary(
        kernel,
        lhs.as_ptr() as *const core::ffi::c_void,
        rhs.as_ptr() as *const core::ffi::c_void,
        output.as_mut_ptr() as *mut core::ffi::c_void,
        &metadata,
    )
    .unwrap();
    output
}

fn run_binary_to_bool(lhs: &[f32], rhs: &[f32], kernel: Kernel) -> Vec<u8> {
    assert_eq!(lhs.len(), rhs.len());
    let mut output = vec![0u8; lhs.len()];
    let shape = vec![lhs.len()];
    let strides = vec![1];

    let num_els = lhs.len();
    let num_dims = 1;
    let mut metadata = Vec::new();
    metadata.push(num_els);
    metadata.push(num_dims);
    metadata.extend(&shape); // lhs_shape
    metadata.extend(&shape); // rhs_shape
    metadata.extend(&strides); // lhs_strides
    metadata.extend(&strides); // rhs_strides
    metadata.push(0); // lhs_offset
    metadata.push(0); // rhs_offset

    call_ops_binary(
        kernel,
        lhs.as_ptr() as *const core::ffi::c_void,
        rhs.as_ptr() as *const core::ffi::c_void,
        output.as_mut_ptr() as *mut core::ffi::c_void,
        &metadata,
    )
    .unwrap();
    output
}

// binary
#[test]
fn test_add_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![5.0f32, 6.0, 7.0, 8.0];
    let output = run_binary(&lhs, &rhs, add::F32);
    assert_eq!(approx(output, 4), vec![6.0, 8.0, 10.0, 12.0]);
}

#[test]
fn test_sub_f32() {
    let lhs = vec![10.0f32, 20.0, 30.0, 40.0];
    let rhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let output = run_binary(&lhs, &rhs, sub::F32);
    assert_eq!(approx(output, 4), vec![9.0, 18.0, 27.0, 36.0]);
}

#[test]
fn test_mul_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![2.0f32, 3.0, 4.0, 5.0];
    let output = run_binary(&lhs, &rhs, mul::F32);
    assert_eq!(approx(output, 4), vec![2.0, 6.0, 12.0, 20.0]);
}

#[test]
fn test_div_f32() {
    let lhs = vec![10.0f32, 20.0, 30.0, 40.0];
    let rhs = vec![2.0f32, 4.0, 5.0, 8.0];
    let output = run_binary(&lhs, &rhs, div::F32);
    assert_eq!(approx(output, 4), vec![5.0, 5.0, 6.0, 5.0]);
}

#[test]
fn test_pow_f32() {
    let lhs = vec![2.0f32, 3.0, 4.0, 5.0];
    let rhs = vec![2.0f32, 2.0, 2.0, 2.0];
    let output = run_binary(&lhs, &rhs, pow::F32);
    assert_eq!(approx(output, 4), vec![4.0, 9.0, 16.0, 25.0]);
}

#[test]
fn test_maximum_f32() {
    let lhs = vec![1.0f32, 5.0, 3.0, 8.0];
    let rhs = vec![3.0f32, 2.0, 7.0, 6.0];
    let output = run_binary(&lhs, &rhs, maximum::F32);
    assert_eq!(approx(output, 4), vec![3.0, 5.0, 7.0, 8.0]);
}

#[test]
fn test_minimum_f32() {
    let lhs = vec![1.0f32, 5.0, 3.0, 8.0];
    let rhs = vec![3.0f32, 2.0, 7.0, 6.0];
    let output = run_binary(&lhs, &rhs, minimum::F32);
    assert_eq!(approx(output, 4), vec![1.0, 2.0, 3.0, 6.0]);
}

// binary logical
#[test]
fn test_logical_and_f32() {
    let lhs = vec![0.0f32, 0.0, 1.0, 2.0];
    let rhs = vec![0.0f32, 3.0, 0.0, 4.0];
    let output = run_binary_to_bool(&lhs, &rhs, logical_and::F32);
    assert_eq!(output, vec![0, 0, 0, 1]);
}

#[test]
fn test_logical_or_f32() {
    let lhs = vec![0.0f32, 0.0, 1.0, 2.0];
    let rhs = vec![0.0f32, 3.0, 0.0, 4.0];
    let output = run_binary_to_bool(&lhs, &rhs, logical_or::F32);
    assert_eq!(output, vec![0, 1, 1, 1]);
}

#[test]
fn test_logical_xor_f32() {
    let lhs = vec![0.0f32, 0.0, 1.0, 2.0];
    let rhs = vec![0.0f32, 3.0, 0.0, 4.0];
    let output = run_binary_to_bool(&lhs, &rhs, logical_xor::F32);
    assert_eq!(output, vec![0, 1, 1, 0]);
}

// cmp
#[test]
fn test_eq_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![1.0f32, 3.0, 3.0, 5.0];
    let output = run_binary_to_bool(&lhs, &rhs, eq::F32);
    assert_eq!(output, vec![1, 0, 1, 0]);
}

#[test]
fn test_ne_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![1.0f32, 3.0, 3.0, 5.0];
    let output = run_binary_to_bool(&lhs, &rhs, ne::F32);
    assert_eq!(output, vec![0, 1, 0, 1]);
}

#[test]
fn test_lt_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![2.0f32, 2.0, 2.0, 3.0];
    let output = run_binary_to_bool(&lhs, &rhs, lt::F32);
    assert_eq!(output, vec![1, 0, 0, 0]);
}

#[test]
fn test_le_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![2.0f32, 2.0, 2.0, 3.0];
    let output = run_binary_to_bool(&lhs, &rhs, le::F32);
    assert_eq!(output, vec![1, 1, 0, 0]);
}

#[test]
fn test_gt_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![2.0f32, 2.0, 2.0, 3.0];
    let output = run_binary_to_bool(&lhs, &rhs, gt::F32);
    assert_eq!(output, vec![0, 0, 1, 1]);
}

#[test]
fn test_ge_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![2.0f32, 2.0, 2.0, 3.0];
    let output = run_binary_to_bool(&lhs, &rhs, ge::F32);
    assert_eq!(output, vec![0, 1, 1, 1]);
}

// Edge cases
#[test]
fn test_div_by_zero_f32() {
    let lhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let rhs = vec![1.0f32, 0.0, 3.0, 0.0];
    let output = run_binary(&lhs, &rhs, div::F32);
    // Division by zero should produce infinity
    // Use approximate comparison for floating point (SIMD may have slight precision differences)
    assert!((output[0] - 1.0).abs() < 1e-5, "Expected ~1.0, got {}", output[0]);
    assert!(output[1].is_infinite());
    assert!((output[2] - 1.0).abs() < 1e-5, "Expected ~1.0, got {}", output[2]);
    assert!(output[3].is_infinite());
}

#[test]
fn test_negative_values_f32() {
    let lhs = vec![-1.0f32, -2.0, -3.0, -4.0];
    let rhs = vec![1.0f32, 2.0, 3.0, 4.0];
    let output = run_binary(&lhs, &rhs, add::F32);
    assert_eq!(approx(output, 4), vec![0.0, 0.0, 0.0, 0.0]);
}