runmat-accelerate 0.4.4

Pluggable GPU acceleration layer for RunMat (CUDA, ROCm, Metal, Vulkan/Spir-V)
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

pub const CIRCSHIFT_SHADER_F64: &str = r#"
const MAX_RANK: u32 = 128u;

struct PackedValue {
    value: u32,
    _pad0: u32,
    _pad1: u32,
    _pad2: u32,
};

alias PackedArray = array<PackedValue, MAX_RANK>;

struct Tensor {
    data: array<f64>,
};

struct Params {
    len: u32,
    offset: u32,
    rank: u32,
    _pad: u32,
    shape: PackedArray,
    strides: PackedArray,
    shifts: PackedArray,
};

@group(0) @binding(0) var<storage, read> Input: Tensor;
@group(0) @binding(1) var<storage, read_write> Output: Tensor;
@group(0) @binding(2) var<uniform> params: Params;

@compute @workgroup_size(@WG@)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
    let local_index = gid.x;
    if local_index >= params.len {
        return;
    }
    let global_index = params.offset + local_index;

    var coords: array<u32, MAX_RANK>;
    var tmp = global_index;
    var dim: u32 = 0u;
    loop {
        if dim >= params.rank {
            break;
        }
        let size = params.shape[dim].value;
        if size == 0u {
            coords[dim] = 0u;
        } else {
            coords[dim] = tmp % size;
            tmp = tmp / size;
        }
        dim = dim + 1u;
    }

    var src_index: u32 = 0u;
    dim = 0u;
    loop {
        if dim >= params.rank {
            break;
        }
        let size = params.shape[dim].value;
        let stride = params.strides[dim].value;
        var coord = coords[dim];
        if size > 1u {
            let shift = params.shifts[dim].value;
            if shift != 0u {
                let wrap = shift % size;
                coord = (coord + size - wrap) % size;
            }
        }
        src_index = src_index + coord * stride;
        dim = dim + 1u;
    }

    Output.data[global_index] = Input.data[src_index];
}
"#;

pub const CIRCSHIFT_SHADER_F32: &str = r#"
const MAX_RANK: u32 = 128u;

struct PackedValue {
    value: u32,
    _pad0: u32,
    _pad1: u32,
    _pad2: u32,
};

alias PackedArray = array<PackedValue, MAX_RANK>;

struct Tensor {
    data: array<f32>,
};

struct Params {
    len: u32,
    offset: u32,
    rank: u32,
    _pad: u32,
    shape: PackedArray,
    strides: PackedArray,
    shifts: PackedArray,
};

@group(0) @binding(0) var<storage, read> Input: Tensor;
@group(0) @binding(1) var<storage, read_write> Output: Tensor;
@group(0) @binding(2) var<uniform> params: Params;

@compute @workgroup_size(@WG@)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
    let local_index = gid.x;
    if local_index >= params.len {
        return;
    }
    let global_index = params.offset + local_index;

    var coords: array<u32, MAX_RANK>;
    var tmp = global_index;
    var dim: u32 = 0u;
    loop {
        if dim >= params.rank {
            break;
        }
        let size = params.shape[dim].value;
        if size == 0u {
            coords[dim] = 0u;
        } else {
            coords[dim] = tmp % size;
            tmp = tmp / size;
        }
        dim = dim + 1u;
    }

    var src_index: u32 = 0u;
    dim = 0u;
    loop {
        if dim >= params.rank {
            break;
        }
        let size = params.shape[dim].value;
        let stride = params.strides[dim].value;
        var coord = coords[dim];
        if size > 1u {
            let shift = params.shifts[dim].value;
            if shift != 0u {
                let wrap = shift % size;
                coord = (coord + size - wrap) % size;
            }
        }
        src_index = src_index + coord * stride;
        dim = dim + 1u;
    }

    Output.data[global_index] = Input.data[src_index];
}
"#;