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
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

//! Binary operations for CUDA runtime
use crate::error::{Error, Result};
use crate::ops::BinaryOps;
use crate::runtime::cuda::kernels::{launch_fused_add_mul, launch_fused_mul_add};
use crate::runtime::cuda::ops::helpers::native_binary_op;
use crate::runtime::cuda::{CudaClient, CudaRuntime};
use crate::runtime::ensure_contiguous;
use crate::tensor::Tensor;

impl BinaryOps<CudaRuntime> for CudaClient {
    fn add(&self, a: &Tensor<CudaRuntime>, b: &Tensor<CudaRuntime>) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, a, b, "add")
    }

    fn sub(&self, a: &Tensor<CudaRuntime>, b: &Tensor<CudaRuntime>) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, a, b, "sub")
    }

    fn mul(&self, a: &Tensor<CudaRuntime>, b: &Tensor<CudaRuntime>) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, a, b, "mul")
    }

    fn div(&self, a: &Tensor<CudaRuntime>, b: &Tensor<CudaRuntime>) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, a, b, "div")
    }

    fn pow(&self, a: &Tensor<CudaRuntime>, b: &Tensor<CudaRuntime>) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, a, b, "pow")
    }

    fn maximum(
        &self,
        a: &Tensor<CudaRuntime>,
        b: &Tensor<CudaRuntime>,
    ) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, a, b, "max")
    }

    fn minimum(
        &self,
        a: &Tensor<CudaRuntime>,
        b: &Tensor<CudaRuntime>,
    ) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, a, b, "min")
    }

    fn atan2(
        &self,
        y: &Tensor<CudaRuntime>,
        x: &Tensor<CudaRuntime>,
    ) -> Result<Tensor<CudaRuntime>> {
        native_binary_op(self, y, x, "atan2")
    }

    fn fused_mul_add(
        &self,
        a: &Tensor<CudaRuntime>,
        b: &Tensor<CudaRuntime>,
        c: &Tensor<CudaRuntime>,
    ) -> Result<Tensor<CudaRuntime>> {
        let dtype = a.dtype();
        if b.dtype() != dtype || c.dtype() != dtype {
            return Err(Error::DTypeMismatch {
                lhs: dtype,
                rhs: if b.dtype() != dtype {
                    b.dtype()
                } else {
                    c.dtype()
                },
            });
        }
        let a_contig = ensure_contiguous(a);
        let b_contig = ensure_contiguous(b);
        let c_contig = ensure_contiguous(c);
        let out = Tensor::<CudaRuntime>::empty(a.shape(), dtype, &self.device);

        unsafe {
            launch_fused_mul_add(
                &self.context,
                &self.stream,
                self.device.index,
                dtype,
                a_contig.ptr(),
                b_contig.ptr(),
                c_contig.ptr(),
                out.ptr(),
                out.numel(),
            )?;
        }

        Ok(out)
    }

    fn fused_add_mul(
        &self,
        a: &Tensor<CudaRuntime>,
        b: &Tensor<CudaRuntime>,
        c: &Tensor<CudaRuntime>,
    ) -> Result<Tensor<CudaRuntime>> {
        let dtype = a.dtype();
        if b.dtype() != dtype || c.dtype() != dtype {
            return Err(Error::DTypeMismatch {
                lhs: dtype,
                rhs: if b.dtype() != dtype {
                    b.dtype()
                } else {
                    c.dtype()
                },
            });
        }
        let a_contig = ensure_contiguous(a);
        let b_contig = ensure_contiguous(b);
        let c_contig = ensure_contiguous(c);
        let out = Tensor::<CudaRuntime>::empty(a.shape(), dtype, &self.device);

        unsafe {
            launch_fused_add_mul(
                &self.context,
                &self.stream,
                self.device.index,
                dtype,
                a_contig.ptr(),
                b_contig.ptr(),
                c_contig.ptr(),
                out.ptr(),
                out.numel(),
            )?;
        }

        Ok(out)
    }
}