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
|
.. meta::
:description: rocSOLVER library precision support overview
:keywords: rocSOLVER, ROCm, API, Linear Algebra, documentation, precision support, data types
.. _rocsolver-precision-support:
********************************************************************
rocSOLVER precision support
********************************************************************
This section provides an overview of the numerical precision types supported by the rocSOLVER library.
Supported precision types
=========================
rocSOLVER supports four primary precision types across its functions:
.. list-table::
:header-rows: 1
*
- Type prefix
- C++ type
- Description
*
- ``s``
- ``float``
- Single-precision real (32-bit)
*
- ``d``
- ``double``
- Double-precision real (64-bit)
*
- ``c``
- ``rocblas_float_complex``
- Single-precision complex (32-bit real, 32-bit imaginary)
*
- ``z``
- ``rocblas_double_complex``
- Double-precision complex (64-bit real, 64-bit imaginary)
Function naming convention
--------------------------
rocSOLVER follows the standard LAPACK naming convention where the first letter of the function name
indicates the precision type:
* Functions beginning with ``rocsolver_s`` operate on single-precision real data.
* Functions beginning with ``rocsolver_d`` operate on double-precision real data.
* Functions beginning with ``rocsolver_c`` operate on single-precision complex data.
* Functions beginning with ``rocsolver_z`` operate on double-precision complex data.
For example, the LU factorization function ``getrf`` is implemented as:
* ``rocsolver_sgetrf`` - For single-precision real matrices
* ``rocsolver_dgetrf`` - For double-precision real matrices
* ``rocsolver_cgetrf`` - For single-precision complex matrices
* ``rocsolver_zgetrf`` - For double-precision complex matrices
In the documentation, these are often represented generically as ``rocsolver_<type>getrf()``, where ``<type>``
is a placeholder for the precision type prefix.
Understanding precision in function signatures
----------------------------------------------
In the function signatures throughout the documentation, precision information is indicated directly in the
parameter types. For example:
.. code-block:: c
rocblas_status rocsolver_slarfb(rocblas_handle handle, /* ... */
float *v, /* ... */
float *t, /* ... */
float *a, /* ... */)
The parameter types (``float``, ``double``, ``rocblas_float_complex``, or ``rocblas_double_complex``) correspond
to the function prefix and indicate the precision used by that specific function variant.
Real versus complex precision
-----------------------------
Some LAPACK functions have different behaviors or names when operating on real versus complex data:
* Functions for symmetric matrices (prefix ``sy``) use the same name for both real precision types.
* Functions for Hermitian matrices (prefix ``he``) are used for complex precision types.
* Some auxiliary routines might be specific to real or complex precision types.
For example, ``rocsolver_ssytrd`` and ``rocsolver_dsytrd`` handle real symmetric matrices, while ``rocsolver_chetrd``
and ``rocsolver_zhetrd`` handle complex Hermitian matrices.
|
