hpt 0.1.3

High Performance Tensor (HPT) - A fast, efficient, and user-friendly tensor computation library for Rust
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

use std::borrow::{Borrow, BorrowMut};

use crate::{
    backend::Cuda, backends::cuda::tensor_internal::matmul::gemm_with_out, tensor::Tensor,
};
use cudarc::{
    cublas::{CudaBlas, Gemm},
    driver::DeviceRepr,
};
use hpt_allocator::traits::{Allocator, AllocatorOutputRetrive};
use hpt_common::error::base::TensorError;
use hpt_traits::{ops::binary::Matmul, tensor::CommonBounds};
use hpt_types::dtype::CudaType;
impl<T, const CUDA_DEVICE: usize, Al> Matmul<Tensor<T, Cuda, CUDA_DEVICE, Al>>
    for Tensor<T, Cuda, CUDA_DEVICE, Al>
where
    T: CommonBounds + DeviceRepr + CudaType,
    CudaBlas: Gemm<T>,
    Al: Allocator,
    Al::Output: AllocatorOutputRetrive,
{
    type Output = Tensor<T, Cuda, CUDA_DEVICE, Al>;

    type OutputMeta = T;

    type InplaceOutput = Tensor<T, Cuda, CUDA_DEVICE, Al>;

    fn matmul(
        &self,
        rhs: Tensor<T, Cuda, CUDA_DEVICE, Al>,
    ) -> std::result::Result<Self::Output, TensorError> {
        Ok(gemm_with_out(
            self.inner.as_ref(),
            rhs.inner.as_ref(),
            T::ONE,
            T::ZERO,
            None::<Self::Output>,
        )?
        .into())
    }
    fn matmul_<U>(
        &self,
        rhs: Tensor<T, Cuda, CUDA_DEVICE, Al>,
        out: U,
    ) -> std::result::Result<Self::Output, TensorError>
    where
        U: Borrow<Self::InplaceOutput> + BorrowMut<Self::InplaceOutput>,
    {
        let out = out.borrow().inner.as_ref().clone();
        Ok(gemm_with_out(
            self.inner.as_ref(),
            rhs.inner.as_ref(),
            T::ONE,
            T::ZERO,
            Some(out),
        )?
        .into())
    }
}

impl<T, const CUDA_DEVICE: usize, Al> Matmul<&Tensor<T, Cuda, CUDA_DEVICE, Al>>
    for Tensor<T, Cuda, CUDA_DEVICE, Al>
where
    T: CommonBounds + DeviceRepr + CudaType,
    CudaBlas: Gemm<T>,
    Al: Allocator,
    Al::Output: AllocatorOutputRetrive,
{
    type Output = Tensor<T, Cuda, CUDA_DEVICE, Al>;

    type OutputMeta = T;

    type InplaceOutput = Tensor<T, Cuda, CUDA_DEVICE, Al>;

    fn matmul(
        &self,
        rhs: &Tensor<T, Cuda, CUDA_DEVICE, Al>,
    ) -> std::result::Result<Self::Output, TensorError> {
        Ok(gemm_with_out(
            self.inner.as_ref(),
            rhs.inner.as_ref(),
            T::ONE,
            T::ZERO,
            None::<Self::Output>,
        )?
        .into())
    }

    fn matmul_<U>(
        &self,
        rhs: &Tensor<T, Cuda, CUDA_DEVICE, Al>,
        out: U,
    ) -> std::result::Result<Self::Output, TensorError>
    where
        U: Borrow<Self::InplaceOutput> + BorrowMut<Self::InplaceOutput>,
    {
        let out = out.borrow().inner.as_ref().clone();
        Ok(gemm_with_out(
            self.inner.as_ref(),
            rhs.inner.as_ref(),
            T::ONE,
            T::ZERO,
            Some(out),
        )?
        .into())
    }
}