numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
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

//! Scalar operation helpers for CPU tensors

use super::super::kernels;
use super::super::{CpuClient, CpuRuntime};
use crate::dispatch_dtype;
use crate::error::Result;
use crate::ops::BinaryOp;
use crate::runtime::ensure_contiguous;
use crate::tensor::Tensor;

/// Helper for scalar operations (scalar arithmetic with tensors)
pub fn scalar_op_impl(
    client: &CpuClient,
    op: BinaryOp,
    a: &Tensor<CpuRuntime>,
    scalar: f64,
    op_name: &'static str,
) -> Result<Tensor<CpuRuntime>> {
    let dtype = a.dtype();
    let a_contig = ensure_contiguous(a);
    let out = Tensor::<CpuRuntime>::empty(a.shape(), dtype, &client.device);

    let len = a.numel();
    let a_ptr = a_contig.ptr();
    let out_ptr = out.ptr();

    dispatch_dtype!(dtype, T => {
        unsafe {
            kernels::scalar_op_kernel::<T>(
                op,
                a_ptr as *const T,
                scalar,
                out_ptr as *mut T,
                len,
            );
        }
    }, op_name);

    Ok(out)
}

/// Reverse scalar subtract: scalar - a
pub fn rsub_scalar_op_impl(
    client: &CpuClient,
    a: &Tensor<CpuRuntime>,
    scalar: f64,
) -> Result<Tensor<CpuRuntime>> {
    let dtype = a.dtype();
    let a_contig = ensure_contiguous(a);
    let out = Tensor::<CpuRuntime>::empty(a.shape(), dtype, &client.device);

    let len = a.numel();
    let a_ptr = a_contig.ptr();
    let out_ptr = out.ptr();

    dispatch_dtype!(dtype, T => {
        unsafe {
            kernels::rsub_scalar_kernel::<T>(
                a_ptr as *const T,
                scalar,
                out_ptr as *mut T,
                len,
            );
        }
    }, "rsub_scalar");

    Ok(out)
}