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

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

use crate::{Error, Result};

cuda_export!(DecodeKernel = "libmir_cuda_clamped_routed_paged_attention_bf16"(
    query: &DeviceBuffer<bf16>, key_pages: &DeviceBuffer<u8>,
    value_pages: &DeviceBuffer<u8>, block_table: &DeviceBuffer<u32>,
    sinks: &DeviceBuffer<bf16>, output: &mut DeviceBuffer<bf16>,
    token_count: u32, block_count: u32, block_size: u32, query_heads: u32,
    kv_heads: u32, head_dim: u32, window: u32, scale: f32,
));
cuda_export!(PrefillKernel = "libmir_cuda_clamped_routed_paged_prefill_attention_bf16"(
    query: &DeviceBuffer<bf16>, key_pages: &DeviceBuffer<u8>,
    value_pages: &DeviceBuffer<u8>, block_table: &DeviceBuffer<u32>,
    sinks: &DeviceBuffer<bf16>, output: &mut DeviceBuffer<bf16>,
    query_tokens: u32, start_position: u32, block_count: u32, block_size: u32,
    query_heads: u32, kv_heads: u32, head_dim: u32, window: u32, scale: f32,
));

#[derive(Clone, Debug)]
pub struct ClampedRoutedAttention {
    decode: TypedKernel<DecodeKernel>,
    prefill: TypedKernel<PrefillKernel>,
    block_size: usize,
    query_heads: usize,
    kv_heads: usize,
    head_dim: usize,
}

impl ClampedRoutedAttention {
    pub(crate) fn compile(
        compiler: &Compiler,
        block_size: usize,
        query_heads: usize,
        kv_heads: usize,
        head_dim: usize,
        dtype: KvCacheDType,
    ) -> Result<Self> {
        if block_size == 0
            || query_heads == 0
            || kv_heads == 0
            || !query_heads.is_multiple_of(kv_heads)
            || head_dim == 0
            || head_dim > 256
        {
            return Err(Error::InvalidPagedKv("invalid clamped-routed attention geometry"));
        }
        let options = match dtype {
            KvCacheDType::Auto | KvCacheDType::BFloat16 => CompileOptions::default(),
            KvCacheDType::Fp8 | KvCacheDType::Fp8E4M3 => CompileOptions {
                extra_options: vec!["--define-macro=LIBMIR_KV_FP8=1".into()],
                ..CompileOptions::default()
            },
            _ => return Err(Error::InvalidPagedKv("unsupported clamped-routed KV dtype")),
        };
        let module = compiler.compile(
            cuda_kernel_file!("../../../kernels/clamped_routed_attention_bf16.cu"),
            &options,
        )?;
        Ok(Self {
            decode: module.kernel()?,
            prefill: module.kernel()?,
            block_size,
            query_heads,
            kv_heads,
            head_dim,
        })
    }

    #[allow(clippy::too_many_arguments)]
    pub(crate) fn execute(
        &self,
        stream: &Stream,
        query: &DeviceBuffer<bf16>,
        key_pages: &DeviceBuffer<u8>,
        value_pages: &DeviceBuffer<u8>,
        table: &DeviceBuffer<u32>,
        sinks: &DeviceBuffer<bf16>,
        output: &mut DeviceBuffer<bf16>,
        tokens: usize,
        blocks: usize,
        window: Option<usize>,
        scale: f32,
    ) -> Result<()> {
        Ok(self.decode.launch(
            stream,
            Self::launch(self.query_heads)?,
            (
                query,
                key_pages,
                value_pages,
                table,
                sinks,
                output,
                narrow(tokens)?,
                narrow(blocks)?,
                narrow(self.block_size)?,
                narrow(self.query_heads)?,
                narrow(self.kv_heads)?,
                narrow(self.head_dim)?,
                narrow(window.unwrap_or(0))?,
                scale,
            ),
        )?)
    }

    #[allow(clippy::too_many_arguments)]
    pub(crate) fn execute_prefill(
        &self,
        stream: &Stream,
        query: &DeviceBuffer<bf16>,
        key_pages: &DeviceBuffer<u8>,
        value_pages: &DeviceBuffer<u8>,
        table: &DeviceBuffer<u32>,
        sinks: &DeviceBuffer<bf16>,
        output: &mut DeviceBuffer<bf16>,
        query_tokens: usize,
        start: usize,
        blocks: usize,
        window: Option<usize>,
        scale: f32,
    ) -> Result<()> {
        Ok(self.prefill.launch(
            stream,
            Self::launch(query_tokens * self.query_heads)?,
            (
                query,
                key_pages,
                value_pages,
                table,
                sinks,
                output,
                narrow(query_tokens)?,
                narrow(start)?,
                narrow(blocks)?,
                narrow(self.block_size)?,
                narrow(self.query_heads)?,
                narrow(self.kv_heads)?,
                narrow(self.head_dim)?,
                narrow(window.unwrap_or(0))?,
                scale,
            ),
        )?)
    }

    fn launch(blocks: usize) -> Result<LaunchConfig> {
        Ok(LaunchConfig {
            grid: (narrow(blocks)?, 1, 1),
            block: (256, 1, 1),
            shared_memory_bytes: 0,
        })
    }
}

fn narrow(value: usize) -> Result<u32> {
    Ok(u32::try_from(value)?)
}