trueno 0.17.1

High-performance SIMD compute library with GPU support for matrix operations
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

//! Reduction and softmax WGSL compute shaders.

/// Max reduction compute shader (WGSL)
///
/// Computes max(input) using parallel reduction
/// Used as first pass in softmax to ensure numerical stability
pub(crate) const MAX_REDUCTION_SHADER: &str = r#"
@group(0) @binding(0) var<storage, read> input: array<f32>;
@group(0) @binding(1) var<storage, read_write> result: array<f32>;

var<workgroup> partial_max: array<f32, 256>;

@compute @workgroup_size(256)
fn main(
    @builtin(global_invocation_id) global_id: vec3<u32>,
    @builtin(local_invocation_id) local_id: vec3<u32>,
) {
    let idx = global_id.x;
    let local_idx = local_id.x;
    let len = arrayLength(&input);

    // Load value or negative infinity
    var max_val: f32 = -3.402823466e+38; // -FLT_MAX
    if (idx < len) {
        max_val = input[idx];
    }
    partial_max[local_idx] = max_val;

    workgroupBarrier();

    // Parallel reduction within workgroup (find max)
    var stride: u32 = 128u;
    while (stride > 0u) {
        if (local_idx < stride) {
            partial_max[local_idx] = max(partial_max[local_idx], partial_max[local_idx + stride]);
        }
        stride = stride / 2u;
        workgroupBarrier();
    }

    // First thread writes workgroup result
    if (local_idx == 0u) {
        result[global_id.x / 256u] = partial_max[0];
    }
}
"#;

/// Sum reduction compute shader (WGSL)
///
/// Computes sum(input) using parallel reduction
/// Used in softmax to sum exp values
pub(crate) const SUM_REDUCTION_SHADER: &str = r#"
@group(0) @binding(0) var<storage, read> input: array<f32>;
@group(0) @binding(1) var<storage, read_write> result: array<f32>;

var<workgroup> partial_sums: array<f32, 256>;

@compute @workgroup_size(256)
fn main(
    @builtin(global_invocation_id) global_id: vec3<u32>,
    @builtin(local_invocation_id) local_id: vec3<u32>,
) {
    let idx = global_id.x;
    let local_idx = local_id.x;
    let len = arrayLength(&input);

    // Load value
    var sum: f32 = 0.0;
    if (idx < len) {
        sum = input[idx];
    }
    partial_sums[local_idx] = sum;

    workgroupBarrier();

    // Parallel reduction within workgroup
    var stride: u32 = 128u;
    while (stride > 0u) {
        if (local_idx < stride) {
            partial_sums[local_idx] = partial_sums[local_idx] + partial_sums[local_idx + stride];
        }
        stride = stride / 2u;
        workgroupBarrier();
    }

    // First thread writes workgroup result
    if (local_idx == 0u) {
        result[global_id.x / 256u] = partial_sums[0];
    }
}
"#;

/// Softmax exp-subtract compute shader (WGSL)
///
/// Computes exp(input[i] - max_val) for each element
/// Second pass in softmax: numerically stable exp computation
pub(crate) const SOFTMAX_EXP_SHADER: &str = r#"
@group(0) @binding(0) var<storage, read> input: array<f32>;
@group(0) @binding(1) var<storage, read_write> output: array<f32>;

struct MaxValue {
    max_val: f32,
}

@group(0) @binding(2) var<uniform> params: MaxValue;

@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
    let idx = global_id.x;
    let len = arrayLength(&input);

    if (idx < len) {
        // exp(x - max) for numerical stability
        output[idx] = exp(input[idx] - params.max_val);
    }
}
"#;

/// Softmax normalize compute shader (WGSL)
///
/// Computes output[i] = input[i] / sum_val for each element
/// Fourth pass in softmax: normalize by sum of exp values
pub(crate) const SOFTMAX_NORMALIZE_SHADER: &str = r#"
@group(0) @binding(0) var<storage, read> input: array<f32>;
@group(0) @binding(1) var<storage, read_write> output: array<f32>;

struct SumValue {
    sum_val: f32,
}

@group(0) @binding(2) var<uniform> params: SumValue;

@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
    let idx = global_id.x;
    let len = arrayLength(&input);

    if (idx < len) {
        // Normalize by sum
        output[idx] = input[idx] / params.sum_val;
    }
}
"#;

/// Log-softmax compute shader (WGSL)
///
/// Computes log_softmax[i] = input[i] - max_val - log(sum_val) for each element
/// Numerically stable log-softmax in single pass after reductions
pub(crate) const LOG_SOFTMAX_SHADER: &str = r#"
@group(0) @binding(0) var<storage, read> input: array<f32>;
@group(0) @binding(1) var<storage, read_write> output: array<f32>;

struct LogSoftmaxParams {
    max_val: f32,
    log_sum_exp: f32,
}

@group(0) @binding(2) var<uniform> params: LogSoftmaxParams;

@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
    let idx = global_id.x;
    let len = arrayLength(&input);

    if (idx < len) {
        // log_softmax(x)[i] = x[i] - max - log(sum(exp(x - max)))
        output[idx] = input[idx] - params.max_val - params.log_sum_exp;
    }
}
"#;