The library provides software crypto acceleration primarily targeting packet processing applications. It can be used for application such as: IPsec, TLS, Wireless (RAN), Cable or MPEG DRM.
The library is hosted on GitHub and is used as software crypto provider in DPDK, Intel(R) QAT Engine and FD.io.
Using crypto interfaces from the above frameworks gives freedom to change providers without subsequent application modifications. The library can also be used directly through its native API.
Key differentiating features:
- operation chaining (encryption and authentication)
- advanced cryptographic pipelining
- job manager with scheduling and dispatching functions
- API hides underlying implementation details from an application
- multi-buffer and function stitching innovations
- low level implementations using latest instruction extensions
- Overview
- Processor Extensions
- Recommendations
- Package Content
- Documentation
- Compilation
- Security Considerations & Options for Increased Security
- Installation
- Backwards compatibility
- Disclaimer (ZUC, KASUMI, SNOW3G)
- Legal Disclaimer
- FIPS Compliance
Intel Multi-Buffer Crypto for IPsec Library is highly-optimized software implementations of the core cryptographic processing for IPsec, which provides industry-leading performance on a range of Intel(R) Processors.
For information on how the library works, see the Intel White Paper: "Fast Multi-buffer IPsec Implementations on Intel Architecture Processors". Jim Guilford, Sean Gulley, et. al.
The easiest way to find it is to search the Internet for the title and Intel White Paper.
Table 1. List of supported cipher algorithms and their implementations.
+----------------------------------------------------------------------+
| | Implementation |
| Encryption +-----------------------------------------------------|
| | x86_64 | SSE | AVX | AVX2 | AVX512 | VAES(5)|
|----------------+--------+--------+--------+--------+--------+--------|
| AES128-GCM | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| AES192-GCM | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| AES256-GCM | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| AES128-CCM | N | Y by8 | Y by8 | N | N | Y by16 |
| AES256-CCM | N | Y by8 | Y by8 | N | N | Y by16 |
| AES128-CBC | N | Y(1) | Y(3) | N | N | Y(6) |
| AES192-CBC | N | Y(1) | Y(3) | N | N | Y(6) |
| AES256-CBC | N | Y(1) | Y(3) | N | N | Y(6) |
| AES128-CTR | N | Y by8 | Y by8 | N | N | Y by16 |
| AES192-CTR | N | Y by8 | Y by8 | N | N | Y by16 |
| AES256-CTR | N | Y by8 | Y by8 | N | N | Y by16 |
| AES128-ECB | N | Y(1) | Y by8 | Y(10) | N | Y by16 |
| AES192-ECB | N | Y(1) | Y by8 | Y(10) | N | Y by16 |
| AES256-ECB | N | Y(1) | Y by8 | Y(10) | N | Y by16 |
| NULL | Y | N | N | N | N | N |
| AES128-DOCSIS | N | Y(2) | Y(4) | N | Y(7) | Y(8) |
| AES256-DOCSIS | N | Y(2) | Y(4) | N | Y(7) | Y(8) |
| DES-DOCSIS | Y | N | N | N | Y x16 | N |
| 3DES | Y | N | N | N | Y x16 | N |
| DES | Y | N | N | N | Y x16 | N |
| KASUMI-F8 | Y | N | N | N | N | N |
| ZUC-EEA3 | N | Y x4 | Y x4 | Y x8 | Y x16 | Y x16 |
| ZUC-EEA3-256 | N | Y x4 | Y x4 | Y x8 | Y x16 | Y x16 |
| SNOW3G-UEA2 | N | Y x4 | Y | Y | Y x16 | Y x16 |
| AES128-CBCS(9) | N | Y(1) | Y(3) | N | N | Y(6) |
| Chacha20 | N | Y | Y | Y | Y | N |
| Chacha20 AEAD | N | Y | Y | Y | Y | N |
| SNOW-V | N | Y | Y | N | N | N |
| SNOW-V AEAD | N | Y | Y | N | N | N |
| PON-CRC-BIP | N | Y by8 | Y by8 | N | N | Y |
+----------------------------------------------------------------------+
Notes:
(1,2) - By default, decryption is by4 and encryption is x4.
On CPU's supporting GFNI, decryption is by8 and encryption is x8.
(3,4) - decryption is by8 and encryption is x8
(5) - AVX512 plus VAES, VPCLMULQDQ and GFNI extensions
(6) - decryption is by16 and encryption is x16
(7) - same as AES128-CBC for AVX, combines cipher and CRC32
(8) - decryption is by16 and encryption is x16
(9) - currently 1:9 crypt:skip pattern supported
(10) - by default, decryption and encryption are AVX by8.
On CPUs supporting VAES, decryption and encryption are AVX2-VAES by16.
Legend:
byY - single buffer Y blocks at a time
xY - Y buffers at a time
As an example of how to read table 1 and 2, if one uses AVX512 interface to perform AES128-CBC encryption then there is no native AVX512 implementation for this cipher. In such case, the library uses best available implementation which is AVX for AES128-CBC.
Table 2. List of supported integrity algorithms and their implementations.
+-------------------------------------------------------------------------+
| | Implementation |
| Integrity +-----------------------------------------------------|
| | x86_64 | SSE | AVX | AVX2 | AVX512 | VAES(3)|
|-------------------+--------+--------+--------+--------+--------+--------|
| AES-XCBC-96 | N | Y x4 | Y x8 | N | N | Y x16 |
| HMAC-MD5-96 | Y(1) | Y x4x2 | Y x4x2 | Y x8x2 | N | N |
| HMAC-SHA1-96 | N | Y(2)x4 | Y x4 | Y x8 | Y x16 | N |
| HMAC-SHA2-224_112 | N | Y(2)x4 | Y x4 | Y x8 | Y x16 | N |
| HMAC-SHA2-256_128 | N | Y(2)x4 | Y x4 | Y x8 | Y x16 | N |
| HMAC-SHA2-384_192 | N | Y x2 | Y x2 | Y x4 | Y x8 | N |
| HMAC-SHA2-512_256 | N | Y x2 | Y x2 | Y x4 | Y x8 | N |
| SHA1 | N | Y(2)x4 | Y x4 | Y x8 | Y x16 | N |
| SHA2-224 | N | Y(2)x4 | Y x4 | Y x8 | Y x16 | N |
| SHA2-256 | N | Y(2)x4 | Y x4 | Y x8 | Y x16 | N |
| SHA2-384 | N | Y x2 | Y x2 | Y x4 | Y x8 | N |
| SHA2-512 | N | Y x2 | Y x2 | Y x4 | Y x8 | N |
| AES128-GMAC | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| AES192-GMAC | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| AES256-GMAC | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| NULL | Y | N | N | N | N | N |
| AES128-CCM | N | Y(5)x4 | Y x8 | N | N | Y x16 |
| AES256-CCM | N | Y(5)x4 | Y x8 | N | N | Y x16 |
| AES128-CMAC-96 | Y | Y(5)x4 | Y x8 | N | N | Y x16 |
| AES256-CMAC-96 | Y | Y(5)x4 | Y x8 | N | N | Y x16 |
| KASUMI-F9 | Y | N | N | N | N | N |
| ZUC-EIA3 | N | Y x4 | Y x4 | Y x8 | Y x16 | Y x16 |
| ZUC-EIA3-256 | N | Y x4 | Y x4 | Y x8 | Y x16 | Y x16 |
| SNOW3G-UIA2(8) | N | Y by4 | Y by4 | N | Y by32 | Y by32 |
| DOCSIS-CRC32(4) | N | Y | Y | N | Y | Y |
| HEC | N | Y | Y | N | N | N |
| POLY1305 | Y | N | N | N | Y | Y |
| POLY1305 AEAD | Y | N | N | N | Y | Y |
| SNOW-V AEAD | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| GHASH | N | Y by8 | Y by8 | Y by8 | Y by8 | Y by48 |
| CRC(6) | N | Y by8 | Y by8 | N | N | Y by16 |
| PON-CRC-BIP(7) | N | Y | Y | N | N | Y |
+-------------------------------------------------------------------------+
Notes:
(1) - MD5 over one block implemented in C
(2) - Implementation using SHANI extensions is x2
(3) - AVX512 plus VAES, VPCLMULQDQ, GFNI and IFMA extensions
(4) - used only with AES256-DOCSIS and AES128-DOCSIS ciphers
(5) - x8 on selected CPU's supporting GFNI
(6) - Supported CRC types:
- CRC32: Ethernet FCS, SCTP, WIMAX OFDMA
- CRC24: LTE A, LTE B
- CRC16: X25, FP data
- CRC11: FP header
- CRC10: IUUP data
- CRC8: WIMAX OFDMA HCS
- CRC7: FP header
- CRC6: IUUP header
(7) - used only with PON-AES128-CTR cipher
(8) - x4/x16 for init keystream generation, then by4/by32
Legend:
byY- single buffer Y blocks at a time
xY- Y buffers at a time
Table 3. Encryption and integrity algorithm combinations
+---------------------------------------------------------------------+
| Encryption | Allowed Integrity Algorithms |
|---------------+-----------------------------------------------------|
| AES128-GCM | AES128-GMAC |
|---------------+-----------------------------------------------------|
| AES192-GCM | AES192-GMAC |
|---------------+-----------------------------------------------------|
| AES256-GCM | AES256-GMAC |
|---------------+-----------------------------------------------------|
| AES128-CCM | AES128-CCM |
|---------------+-----------------------------------------------------|
| AES256-CCM | AES256-CCM |
|---------------+-----------------------------------------------------|
| AES128-CBC, | AES-XCBC-96, |
| AES192-CBC, | HMAC-SHA1-96, HMAC-SHA2-224_112, HMAC-SHA2-256_128, |
| AES256-CBC, | HMAC-SHA2-384_192, HMAC-SHA2-512_256, |
| AES128-CTR, | AES128-CMAC-96, |
| AES192-CTR, | NULL, |
| AES256-CTR, | KASUMI-F9, |
| AES128-ECB, | ZUC-EIA3, ZUC-EIA3-256, |
| AES192-ECB, | SNOW3G-UIA3, |
| AES256-ECB, | POLY1305, |
| NULL, | AES128-GMAC, AES192-GMAC, AES256-GMAC, GHASH |
| AES128-DOCSIS,| |
| AES256-DOCSIS,| |
| DES-DOCSIS, | |
| 3DES, | |
| DES, | |
| Chacha20, | |
| KASUMI-F8, | |
| ZUC-EEA3, | |
| ZUC-EEA3-256, | |
| SNOW3G-UEA3 | |
| SNOW-V | |
|---------------+-----------------------------------------------------|
| AES128-DOCSIS,| DOCSIS-CRC32 |
| AES256-DOCSIS | |
|---------------+-----------------------------------------------------|
| PON-AES128-CTR| PON-CRC-BIP |
|---------------+-----------------------------------------------------|
| CHACHA20 AEAD | POLY1305 AEAD |
+---------------+-----------------------------------------------------+
| SNOW-V AEAD | SNOW-V AEAD (GHASH) |
+---------------+-----------------------------------------------------+
Table 4. Processor extensions used in the library
+-------------------------------------------------------------------------+
| Algorithm | Interface | Extensions |
|-------------------+-----------+-----------------------------------------|
| HMAC-SHA1-96, | AVX512 | AVX512F, AVX512BW, AVX512VL |
| HMAC-SHA2-224_112,| | |
| HMAC-SHA2-256_128,| | |
| HMAC-SHA2-384_192,| | |
| HMAC-SHA2-512_256 | | |
|-------------------+-----------+-----------------------------------------|
| DES, 3DES, | AVX512 | AVX512F, AVX512BW |
| DOCSIS-DES | | |
|-------------------+-----------+-----------------------------------------|
| HMAC-SHA1-96, | SSE | SHANI |
| HMAC-SHA2-224_112,| | - presence is autodetected and library |
| HMAC-SHA2-256_128,| | falls back to SSE implementation |
| HMAC-SHA2-384_192,| | if not present |
| HMAC-SHA2-512_256 | | |
+-------------------+-----------+-----------------------------------------+
Legacy or to be avoided algorithms listed in the table below are implemented in the library in order to support legacy applications. Please use corresponding alternative algorithms instead.
+--------------------------------------------------------------+
| # | Algorithm | Recommendation | Alternative |
|---+---------------------+----------------+-------------------|
| 1 | DES encryption | Avoid | AES encryption |
|---+---------------------+----------------+-------------------|
| 2 | 3DES encryption | Avoid | AES encryption |
|---+---------------------+----------------+-------------------|
| 3 | HMAC-MD5 integrity | Legacy | HMAC-SHA256 |
|---+---------------------+----------------+-------------------|
| 4 | AES-ECB encryption | Avoid | AES-CBC, AES-CNTR |
|---+---------------------+----------------+-------------------|
| 3 | HMAC-SHA1 integrity | Avoid | HMAC-SHA256 |
+--------------------------------------------------------------+
Intel(R) Multi-Buffer Crypto for IPsec Library depends on C library and it is recommended to use its latest version.
Applications using the Intel(R) Multi-Buffer Crypto for IPsec Library rely on Operating System to provide process isolation. As the result, it is recommended to use latest Operating System patches and security updates.
- test - Library test applications
- perf - Library performance application
- lib - Library source files
- lib/sse - Intel(R) SSE optimized routines
- lib/avx - Intel(R) AVX optimized routines
- lib/avx2 - Intel(R) AVX2 optimized routines
- lib/avx512 - Intel(R) AVX512 optimized routines
- lib/no-aesni - Non-AESNI accelerated routines
Note:
There is just one branch used in the project. All development is done on the master branch.
Code taken from the tip of the master branch should not be considered fit for production.
Refer to the releases tab for stable code versions:
https://github.com/intel/intel-ipsec-mb/releases
Full documentation can be found at: https://intel.github.io/intel-ipsec-mb
To generate documentation locally, run:
> make doxy
Required tools:
- GNU make
- NASM version 2.14 (or newer)
- gcc (GCC) 4.8.3 (or newer)
Shared library:
> make
Static library:
> make SHARED=n
Clean the build:
> make clean
or
> make clean SHARED=n
Build with debugging information:
> make DEBUG=y
Build with AESNI emulation support (disabled by default):
> make AESNI_EMU=y
Note: Building with debugging information is not advised for production use.
For more build options and their explanation run:
> make help
Required tools:
- Microsoft (R) Visual Studio 2015:
- NMAKE: Microsoft (R) Program Maintenance Utility Version 14.00.24210.0
- CL: Microsoft (R) C/C++ Optimizing Compiler Version 19.00.24215.1 for x64
- LIB: Microsoft (R) Library Manager Version 14.00.24215.1
- LINK: Microsoft (R) Incremental Linker Version 14.00.24215.1
- Note: Building on later versions should work but is not verified
- NASM version 2.14 (or newer)
Shared library (DLL):
> nmake /f win_x64.mak
Static library:
> nmake /f win_x64.mak SHARED=n
Clean the build:
> nmake /f win_x64.mak clean
or
> nmake /f win_x64.mak clean SHARED=n
Build without safety features:
- SAFE_DATA clears sensitive information stored temporarily on stack, registers or internal data structures
- SAFE_PARAM adds extra checks on input parameters
- SAFE_LOOKUP uses constant-time lookups (enabled by default)
- SAFE_OPTIONS additional option to disable all safe options. Enabled by default.
Disable to turn off: SAFE_DATA, SAFE_PARAM and SAFE_LOOKUP.
> nmake /f win_x64.mak SAFE_DATA=n SAFE_PARAM=n
> nmake /f win_x64.mak SAFE_OPTIONS=n
Build with debugging information:
> nmake /f win_x64.mak DEBUG=y
Build with AESNI emulation support (disabled by default):
> nmake /f win_x64.mak AESNI_EMU=y
Note: Building with debugging information is not advised for production use.
For more build options and their explanation run:
> nmake /f win_x64.mak help
Required tools:
- GNU mingw32-make.exe
- NASM version 2.14 (or newer)
- gcc (GCC) 10.3.0 (or newer)
Shared library:
> mingw32-make.exe
Static library:
> mingw32-make.exe SHARED=n
Clean the build:
> mingw32-make.exe clean
or
> mingw32-make.exe clean SHARED=n
Build with debugging information:
> mingw32-make.exe DEBUG=y
Note: Building with debugging information is not advised for production use.
For more build options and their explanation run:
> mingw32-make.exe help
Required tools:
- GNU make
- NASM version 2.14 (or newer)
- gcc (GCC) 4.8.3 (or newer) / clang 5.0 (or newer)
Shared library:
> gmake
Static library:
> gmake SHARED=n
Clean the build:
> gmake clean
or
> gmake clean SHARED=n
Build with debugging information:
> gmake DEBUG=y
Note: Building with debugging information is not advised for production use.
For more build options and their explanation run:
> gmake help
The security of a system that uses cryptography depends on the strength of the cryptographic algorithms as well as the strength of the keys. Cryptographic key strength is dependent on several factors, with some of the most important factors including the length of the key, the entropy of the key bits, and maintaining the secrecy of the key.
The selection of an appropriate algorithm and mode of operation critically affects the security of a system. Appropriate selection criteria is beyond the scope of this document and should be determined based upon usage, appropriate standards and consultation with a cryptographic expert. This library includes some algorithms, which are considered cryptographically weak and are included only for legacy and interoperability reasons. See the "Recommendations" section for more details.
Secure creation of key material is not a part of this library. This library assumes that cryptographic keys have been created using approved methods with an appropriate and secure entropy source. Users of this library are referred to NIST SP800-133 Revision 1, Recommendation for Cryptographic Key Generation, found at https://nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.800-133r1.pdf
Even with the use of strong cryptographic algorithms and robustly generated keys, software implementations of cryptographic algorithms may be attacked at the implementation through cache-timing attacks, buffer-over-reads, and other software vulnerabilities. Counter-measures against these types of attacks are possible but require additional processing cycles. Whether a particular system should provide such counter-measures depends on the threats to that system, and cannot be determined by a general library such as this one. In order to provide the most flexible implementation, this library allows certain counter-measures to be enabled or disabled at compile time. These options are listed below as the "Options for Increased Security" and are enabled through various build flags.
There are three build options that are used to increase safety in the code and help protect external functions from incorrect input data. The SAFE_DATA, SAFE_PARAM and SAFE_LOOKUP options are enabled by default. Due to the potential performance impact associated to the extra code, these can be disabled by setting the parameter equal to "n" (e.g. make SAFE_LOOKUP=n).
No specific code has been added, and no specific validation or security tests have been performed to help protect against or check for side-channel attacks.
Stack and registers containing sensitive information, such as keys or IVs, are cleared upon completion of a function call.
Input parameters are checked, looking generally for NULL pointers or an incorrect input length.
Lookups which depend on sensitive information are implemented with constant time functions.
Algorithms where these constant time functions are used are the following:
- AESNI emulation
- DES: SSE, AVX and AVX2 implementations
- KASUMI: all architectures
- SNOW3G: all architectures
If SAFE_LOOKUP is not enabled in the build (e.g. make SAFE_LOOKUP=n) then the algorithms listed above may be susceptible to timing attacks which could expose the cryptographic key.
SAFE_OPTIONS is a parameter that can be used to disable/enable
all supported safe options (i.e. SAFE_DATA, SAFE_PARAM, SAFE_LOOKUP).
It is set to y by default and all safe options are enabled.
SAFE_OPTIONS=n disables all safe options.
Force clearing/zeroing of memory
IMB_DLL_EXPORT void imb_clear_mem(void *mem, const size_t size);To assist in clearing sensitive application data such as keys, plaintext etc.
the library provides the imb_clear_mem() API. This API zeros 'size' bytes
of memory pointed to by 'mem' followed by the sfence instruction to
ensure memory is cleared before the function returns.
The library GCM and GMAC implementation provides flexibility as to tag size selection. As explained in NIST Special Publication 800-38D section 5.2.1.2 and Appendix C, using tag sizes shorter than 96 bits can be insecure. Please refer to the aforementioned sections to understand the details, trade offs and mitigations of using shorter tag sizes.
First compile the library and then install:
> make
> sudo make install
To uninstall the library run:
> sudo make uninstall
If you want to change install location then define PREFIX:
> sudo make install PREFIX=<path>
If there is no need to run ldconfig at install stage please use NOLDCONFIG=y option:
> sudo make install NOLDCONFIG=y
If library was compiled as an archive (not a default option) then install it using SHARED=n option:
> sudo make install SHARED=n
First compile the library and then install from a command prompt in administrator mode:
> nmake /f win_x64.mak
> nmake /f win_x64.mak install
To uninstall the library run:
> nmake /f win_x64.mak uninstall
If you want to change install location then define PREFIX (default C:\Program Files):
> nmake /f win_x64.mak install PREFIX=<path>
If library was compiled as a static library (not a default option) then install it using SHARED=n option:
> nmake /f win_x64.mak install SHARED=n
First compile the library and then install:
> gmake
> sudo gmake install
To uninstall the library run:
> sudo gmake uninstall
If you want to change install location then define PREFIX:
> sudo gmake install PREFIX=<path>
If there is no need to run ldconfig at install stage please use NOLDCONFIG=y option:
> sudo gmake install NOLDCONFIG=y
If library was compiled as an archive (not a default option) then install it using SHARED=n option:
> sudo gmake install SHARED=n
In version 0.54, some symbols have been renamed to avoid too generic names (such as cipher modes or hash algorithms).
When building an application and linking it against the IPSec Multi Buffer library, by default the old symbols (up to v0.53) are exported, to maintain backwards compatibility.
Applications should move to the new API as soon as possible, as the old symbols are marked as deprecated and will be removed in a future release.
For applications which face symbol conflicts due to these old generic names, they should be compiled with the flag -DNO_COMPAT_IMB_API_053, which will not export the old symbols.
Please note that cryptographic material, such as ciphering algorithms, may be subject to national regulations. What is more, use of some algorithms in real networks and production equipment can be subject to agreement or licensing by the GSMA and/or the ETSI.
For more details please see:
- GSMA https://www.gsma.com/security/security-algorithms/
- ETSI https://www.etsi.org/security-algorithms-and-codes/cellular-algorithm-licences
THIS SOFTWARE IS PROVIDED BY INTEL"AS IS". NO LICENSE, EXPRESS OR
IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS
ARE GRANTED THROUGH USE. EXCEPT AS PROVIDED IN INTEL'S TERMS AND
CONDITIONS OF SALE, INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL
DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO SALE AND/OR
USE OF INTEL PRODUCTS INCLUDING LIABILITY OR WARRANTIES RELATING TO
FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT
OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT.
The library does not fulfill technical requirements to achieve Cryptographic Module (CMVP) certification as a standalone component. It is fit for Cryptographic Algorithm validation and certification (CAVP) and it can be part of CMVP as one of the components.
ACVP test application located in test directory is to support CAVP process. It implements validation of the following algorithms:
- AES-GCM
- AES-GMAC
- AES-CCM
- AES-CBC
- AES-CTR
- AES-CMAC
- SHA1
- SHA224
- SHA256
- SHA384
- SHA512
- HMAC-SHA1
- HMAC-SHA224
- HMAC-SHA256
- HMAC-SHA384
- HMAC-SHA512
In order to support CMVP, the library implements Self-Test functionality that is available with all compilation options. The test is always performed as part of library initialization (power-up). There is no conditional self-test functionality as none of such conditions occur (i.e. pair-wise consistency test, software/firmware load test, manual key entry test, continuous random number generator test, and bypass test). The self-test consists of Cryptographic algorithm test (known answer test) on following algorithms:
- AES-GCM
- AES-GMAC
- AES-CCM
- AES-CBC
- AES-CTR
- AES-CMAC
- SHA1
- SHA224
- SHA256
- SHA384
- SHA512
- HMAC-SHA1
- HMAC-SHA224
- HMAC-SHA256
- HMAC-SHA384
- HMAC-SHA512
Detecting library self-test failure in the application:
IMB_ARCH arch;
IMB_MGR *p_mgr = alloc_mb_mgr(0);
init_mb_mgr_auto(p_mgr, &arch); /* or init_mb_mgr_sse/avx/avx2/avx512 */
if (imb_get_errno(p_mgr) == IMB_ERR_SELFTEST) {
/* self-test error */
exit(EXIT_FAILURE);
}