vkml 0.0.2

High-level Vulkan-based machine learning library
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

use crate::{ComputeManager, instruction::Instruction, tensor::DeviceId, tensor_graph::TensorId};
use std::fmt::{Debug, Formatter, Result as FmtResult};

pub struct TransferToDeviceInstruction {
    pub src: TensorId,
    pub dst: TensorId,
    pub source_device: DeviceId,
    pub target_device: DeviceId,
}

impl Debug for TransferToDeviceInstruction {
    fn fmt(&self, f: &mut Formatter<'_>) -> FmtResult {
        write!(
            f,
            "TransferToDevice(src={}, dst={}, from={:?}, to={:?})",
            self.src, self.dst, self.source_device, self.target_device
        )
    }
}

impl Instruction for TransferToDeviceInstruction {
    fn get_input_tensor_ids(&self) -> Vec<TensorId> {
        vec![self.src]
    }

    fn get_output_tensor_ids(&self) -> Vec<TensorId> {
        vec![self.dst]
    }

    fn remap_tensor_ids(&mut self, new_inputs: &[TensorId], new_outputs: &[TensorId]) {
        if !new_inputs.is_empty() {
            self.src = new_inputs[0];
        }

        if !new_outputs.is_empty() {
            self.dst = new_outputs[0];
        }
    }

    fn must_execute_on_cpu(&self) -> bool {
        true
    }

    fn execute_cpu(&self, cm: &ComputeManager) {
        let data = cm.tensor_read(self.src).read();
        cm.tensor_write(self.dst).write(&data);
    }
}