libmir-cuda 0.2.0

CUDA inference backend for libmir
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

use mircuda::{
    CompileOptions, Compiler, DeviceBuffer, LaunchConfig, Stream, TypedKernel, bf16, cuda_export,
    cuda_kernel_file,
};

use super::super::geometry::{narrow, product, require};
use crate::{Error, Result};

cuda_export!(
    AttentionKernel = "libmir_cuda_vision_attention_bf16"(
        query: &DeviceBuffer<bf16>, key: &DeviceBuffer<bf16>,
        value: &DeviceBuffer<bf16>, output: &mut DeviceBuffer<bf16>,
        tokens: u32, query_heads: u32, kv_heads: u32, head_dim: u32, scale: f32,
    )
);
cuda_export!(
    SpatialRopeKernel = "libmir_cuda_vision_spatial_rope_bf16"(
        input: &DeviceBuffer<bf16>, positions: &DeviceBuffer<u32>,
        output: &mut DeviceBuffer<bf16>, tokens: u32, heads: u32,
        head_dim: u32, theta: f32,
    )
);

#[derive(Clone, Copy, Debug, PartialEq)]
pub struct VisionAttentionSpec {
    pub tokens: usize,
    pub query_heads: usize,
    pub kv_heads: usize,
    pub head_dim: usize,
    pub scale: f32,
}

#[derive(Clone, Debug)]
pub struct VisionAttention {
    kernel: TypedKernel<AttentionKernel>,
    spec: VisionAttentionSpec,
}

#[derive(Clone, Debug)]
pub struct VisionSpatialRope {
    kernel: TypedKernel<SpatialRopeKernel>,
    tokens: usize,
    heads: usize,
    head_dim: usize,
    theta: f32,
}

impl VisionAttention {
    pub fn compile(compiler: &Compiler, spec: VisionAttentionSpec) -> Result<Self> {
        validate(spec)?;
        let source = cuda_kernel_file!("../../../kernels/vision_attention_bf16.cu");
        let module = compiler.compile(source, &CompileOptions::default())?;
        Ok(Self { kernel: module.kernel()?, spec })
    }

    pub fn execute(
        &self,
        stream: &Stream,
        query: &DeviceBuffer<bf16>,
        key: &DeviceBuffer<bf16>,
        value: &DeviceBuffer<bf16>,
        output: &mut DeviceBuffer<bf16>,
    ) -> Result<()> {
        let query_elements =
            product(product(self.spec.tokens, self.spec.query_heads)?, self.spec.head_dim)?;
        let kv_elements =
            product(product(self.spec.tokens, self.spec.kv_heads)?, self.spec.head_dim)?;
        require("vision attention query", query_elements, query.len())?;
        require("vision attention key", kv_elements, key.len())?;
        require("vision attention value", kv_elements, value.len())?;
        require("vision attention output", query_elements, output.len())?;
        Ok(self.kernel.launch(
            stream,
            LaunchConfig {
                grid: (narrow(self.spec.tokens)?, narrow(self.spec.query_heads)?, 1),
                block: (256, 1, 1),
                shared_memory_bytes: 0,
            },
            (
                query,
                key,
                value,
                output,
                narrow(self.spec.tokens)?,
                narrow(self.spec.query_heads)?,
                narrow(self.spec.kv_heads)?,
                narrow(self.spec.head_dim)?,
                self.spec.scale,
            ),
        )?)
    }
}

impl VisionSpatialRope {
    pub fn compile(
        compiler: &Compiler,
        tokens: usize,
        heads: usize,
        head_dim: usize,
        theta: f32,
    ) -> Result<Self> {
        if tokens == 0
            || heads == 0
            || !head_dim.is_multiple_of(4)
            || head_dim > 256
            || !theta.is_finite()
            || theta <= 0.0
        {
            return Err(Error::InvalidVisionKernel("invalid spatial RoPE geometry"));
        }
        let source = cuda_kernel_file!("../../../kernels/vision_attention_bf16.cu");
        let module = compiler.compile(source, &CompileOptions::default())?;
        Ok(Self {
            kernel: module.kernel()?,
            tokens,
            heads,
            head_dim,
            theta,
        })
    }

    pub fn execute(
        &self,
        stream: &Stream,
        input: &DeviceBuffer<bf16>,
        positions: &DeviceBuffer<u32>,
        output: &mut DeviceBuffer<bf16>,
    ) -> Result<()> {
        let elements = product(product(self.tokens, self.heads)?, self.head_dim)?;
        require("vision RoPE input", elements, input.len())?;
        require("vision RoPE positions", self.tokens * 2, positions.len())?;
        require("vision RoPE output", elements, output.len())?;
        Ok(self.kernel.launch(
            stream,
            LaunchConfig {
                grid: (narrow(elements.div_ceil(256))?, 1, 1),
                block: (256, 1, 1),
                shared_memory_bytes: 0,
            },
            (
                input,
                positions,
                output,
                narrow(self.tokens)?,
                narrow(self.heads)?,
                narrow(self.head_dim)?,
                self.theta,
            ),
        )?)
    }
}

fn validate(spec: VisionAttentionSpec) -> Result<()> {
    if spec.tokens == 0
        || spec.query_heads == 0
        || spec.kv_heads == 0
        || !spec.query_heads.is_multiple_of(spec.kv_heads)
        || spec.head_dim == 0
        || spec.head_dim > 256
        || !spec.scale.is_finite()
        || spec.scale <= 0.0
    {
        Err(Error::InvalidVisionKernel("invalid attention geometry"))
    } else {
        Ok(())
    }
}