File: Windows_Install_Guide.rst

package info (click to toggle)
hipblas 6.4.3-3
  • links: PTS, VCS
  • area: main
  • in suites: sid
  • size: 26,548 kB
  • sloc: cpp: 209,856; f90: 49,800; python: 4,680; ansic: 1,141; sh: 799; makefile: 50; xml: 23
file content (143 lines) | stat: -rw-r--r-- 7,778 bytes parent folder | download
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
133
134
135
136
137
138
139
140
141
142
143
.. meta::
  :description: Installing and Building hipBLAS for Windows
  :keywords: hipBLAS, rocBLAS, BLAS, ROCm, API, Linear Algebra, documentation, Windows installation, build

.. _windows-install:

*********************************************
Installing and building for Microsoft Windows
*********************************************

This topic covers how to install hipBLAS on Microsoft Windows from a package and how to build and install it from the source code.
For a list of installation prerequisites, see :doc:`hipBLAS prerequisites <prerequisites>`.

Installing prebuilt packages
=============================

hipBLAS can be installed on Windows 10 or 11 using the :doc:`AMD HIP SDK installer <rocm-install-on-windows:index>`.

The simplest way to use hipBLAS in your code is to use CMake. To install hipBLAS on Windows, follow these steps:

#. Add the SDK installation location to ``CMAKE_PREFIX_PATH`` in the CMake configuration command:

   ::

      -DCMAKE_PREFIX_PATH="C:\hipSDK"


#. Add the following lines to ``CMakeLists.txt``:

   ::

      find_package(hipblas)
      target_link_libraries( your_exe PRIVATE roc::hipblas )

Building and installing hipBLAS
===============================

Most users do not need to build hipBLAS from source because it can be used after installing the prebuilt packages as described above.
If necessary, users can follow these instructions to build hipBLAS from source.

Download the hipBLAS source code
--------------------------------

The hipBLAS source code is available on the `hipBLAS GitHub <https://github.com/ROCm/hipBLAS>`_.
The ROCm HIP SDK version might be shown in the default installation path,
but you can run the HIP SDK compiler to display the version from the ``bin/`` folder:

::

    hipcc --version

The HIP version includes the major, minor, and patch fields, and possibly a build-specific identifier.
As an example, the HIP version might be ``5.4.22880-135e1ab4``.
This corresponds to major release = ``5``, minor release = ``4``, patch = ``22880``, and the build identifier ``135e1ab4``.
The hipBLAS GitHub contains branches with names like ``release/rocm-rel-major.minor``. Select the branch where the major and minor
fields correspond to the HIP version.
For example, use the following command format to download hipBLAS:

::

    git clone -b release/rocm-rel-x.y https://github.com/ROCm/hipBLAS.git

Replace ``x.y`` in this command with the version of the HIP SDK installed on your machine.
For example, if you have HIP 6.2 installed, then use ``-b release/rocm-rel-6.2``.
You can add the SDK tools to your path with an entry such as:

::

    %HIP_PATH%\bin

Building the library and library dependencies
---------------------------------------------

The root of the hipBLAS repository includes a helper Python script named ``rmake.py`` which lets you build and install hipBLAS
with a single command. It accepts several options but has a hard-coded configuration
that you can override by invoking ``cmake`` directly. However, it's a great way to get started quickly and serves
as an example of how to build and install hipBLAS.
A few commands in the script require administrator access, so you might be prompted for a password.

Common examples showing how to use ``rmake.py`` to build the library dependencies and library are listed
in this table:

.. csv-table::
   :header: "Command","Description"
   :widths: 30, 100

   "``python3 rmake.py -h``", "Help information."
   "``python3 rmake.py -d``", "Build the library dependencies and library in your local directory. The ``-d`` flag only has to be used once. For subsequent invocations of ``rmake.py`` it's not necessary to rebuild the dependencies."
   "``python3 rmake.py``", "Build the library in your local directory. It is assumed the dependencies have been built."
   "``python3 rmake.py -i``", "Build the library, then build and install the hipBLAS package in ``C:\\hipSDK``. You will be prompted for admin access. This installs it for all users. To restrict hipBLAS to your local directory, do not use the ``-i`` flag."
   "``python3 rmake.py -n``", "Build the library without the functionality provided by rocSOLVER. The rocSOLVER, rocSPARSE, and rocPRIM dependencies will not be required. This flag has no effect when building with a NVIDIA CUDA backend."


Building the library, client, and library and client dependencies
-------------------------------------------------------------------

The client contains the executables listed in the table below.

================= ====================================================
Executable name   Description
================= ====================================================
hipblas-test      Runs Google Tests to test the library
hipblas-bench     An executable to benchmark or test individual functions
hipblas-example-* Various examples showing how to use hipBLAS
================= ====================================================

Common ways to use ``rmake.py`` to build the dependencies, library, and client are
listed in this table.

.. csv-table::
   :header: "Command","Description"
   :widths: 33, 97

   "``python3 rmake.py -dc``", "Build the library dependencies, client dependencies, library, and client in your local directory. The ``-d`` flag only has to be used once. For subsequent invocations of ``rmake.py``, it is not necessary to rebuild the dependencies."
   "``python3 rmake.py -c``", "Build the library and client in your local directory. It is assumed the dependencies have been built."
   "``python3 rmake.py -idc``", "Build the library dependencies, client dependencies, library, and client, then build and install the hipBLAS package. You will be prompted for administrator access. To install hipBLAS for all users, use the ``-i`` flag. To restrict hipBLAS to your local directory, do not use the ``-i`` flag."
   "``python3 rmake.py -ic``", "Build and install the hipBLAS package and build the client. You will be prompted for administrator access. This installs it for all users. To restrict hipBLAS to your local directory, do not use the ``-i`` flag."

Dependencies for building the library
=====================================

Use ``rmake.py`` with the ``-d`` option to install the dependencies required to build the library.
This does not install the hipblas-common, rocBLAS, rocSOLVER, rocSPARSE, and rocPRIM dependencies.
When building hipBLAS, it is important to take note of the version dependencies for other libraries. The hipblas-common,
rocBLAS, and rocSOLVER versions required to build for an AMD backend are listed in the top-level ``CMakeLists.txt`` file.
rocSPARSE and rocPRIM are currently dependencies of rocSOLVER. To build these libraries from source,
see the :doc:`rocBLAS <rocblas:index>`,
:doc:`rocSOLVER <rocsolver:index>`, :doc:`rocSPARSE <rocsparse:index>`,
and :doc:`rocPRIM <rocprim:index>` documentation..

The minimum version of CMake is currently 3.16.8. See the ``--cmake_install`` flag in ``rmake.py`` to
upgrade automatically.

To use the test and benchmark clients' host reference functions, you must manually download and install
AMD's `ILP64 version of the AOCL libraries <https://www.amd.com/en/developer/aocl.html>`_ version 4.2.
If you download and run the full Windows AOCL installer to the default location (``C:\Program Files\AMD\AOCL-Windows\``),
then the AOCL reference BLAS (``amd-blis``) should be found by the clients' ``CMakeLists.txt`` file.

.. note::

   If you only use the ``rmake.py -d`` dependency script and change the default CMake option ``LINK_BLIS=ON``,
   you might experience ``hipblas-test`` stress test failures due to a 32-bit integer overflow
   on the host. To resolve this issue, exclude the stress tests using the command line argument ``--gtest_filter=-*stress*``.