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
// RLX — versatile ML compiler + runtime.
// Copyright (C) 2026 Eugene Hauptmann, Nataliya Kosmyna.
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 3.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <https://www.gnu.org/licenses/>.
//! Fused multi-scale deformable attention on Metal (host delegate, unified
//! memory — reads the GPU buffers in place and runs the shared `rlx_cpu` math).
use crate::op_registry::{MetalKernel, register_metal_kernel};
use rlx_ir::Shape;
use std::sync::Arc;
pub const OP_NAME: &str = "gdino.ms_deform_attn";
#[derive(Debug)]
struct MsDeformAttnMetal;
impl MetalKernel for MsDeformAttnMetal {
fn name(&self) -> &str {
OP_NAME
}
fn execute(
&self,
inputs: &[(&[u8], &Shape)],
output: (&mut [u8], &Shape),
attrs: &[u8],
) -> Result<(), String> {
let ins: Vec<Vec<f32>> = inputs
.iter()
.map(|(bytes, _)| {
if !bytes.len().is_multiple_of(4) {
return Err("ms_deform_attn: non-f32-aligned input".to_string());
}
Ok(bytemuck::cast_slice::<u8, f32>(bytes).to_vec())
})
.collect::<Result<_, String>>()?;
let in_refs: Vec<&[f32]> = ins.iter().map(|v| v.as_slice()).collect();
let out = output.0;
if !out.len().is_multiple_of(4) {
return Err("ms_deform_attn: non-f32-aligned output".into());
}
let out_f32 = bytemuck::cast_slice_mut::<u8, f32>(out);
rlx_cpu::ms_deform_attn::execute(&in_refs, attrs, out_f32)
}
}
pub fn register() {
register_metal_kernel(Arc::new(MsDeformAttnMetal));
}