There are three build options for crypto operations when using wolfSSL on Xilinx/AMD Ultrascale+ devices. The lightweight wolfSSL library can use a software only implementation, make use of the ARMv8 crypto extensions along with custom ARM assembly, or offload the operation to the CSU. Each has its trade offs. Recently wolfSSL has made improvements to the ARMv8 optimizations for use with AES-GCM operations.
- Offloading to the CSU (labeled here as the hardened option) free’s up the CPU for other operations and it leverages the hardening available which provides enhancements like additional protections against DPA (differential power analysis)
- ARMv8 crypto extensions is very performant for smaller block sizes and can be taken through a FIPS OE with use of PAA (Processor Algorithm Acceleration)
The following tables are raw numbers of the throughput collected. Collecting the performance on even larger block sizes would show that Xilinx/AMD hardened crypto accelerators continue on linearly until reaching their maximum. The hardened numbers were collected previously using FreeRTOS, the software and ARMv8 were collected while running on Petalinux with the latest wolfSSL version 5.8.2. A VMK180 Versal board was used.
| Algorithm | Hardened – MB/s | Block Size |
| AES-256-GCM-enc-no_AAD | 0.19188 | 16 |
| AES-256-GCM-enc-no_AAD | 6.324667 | 528 |
| AES-256-GCM-enc-no_AAD | 12.254902 | 1024 |
| AES-256-GCM-enc-no_AAD | 49.01886 | 4112 |
| AES-256-GCM-enc-no_AAD | 89.60888 | 7696 |
| AES-256-GCM-enc-no_AAD | 181.00591 | 15888 |
| AES-256-GCM-enc-no_AAD | 350.444225 | 32768 |
| AES-256-GCM-enc-no_AAD | 633.100698 | 65535 |
| Algorithm | Software – MB/s | Block Size |
| AES-256-GCM-enc-no_AAD | 15.0984 | 16 |
| AES-256-GCM-enc-no_AAD | 31.0764 | 528 |
| AES-256-GCM-enc-no_AAD | 31.5839 | 1024 |
| AES-256-GCM-enc-no_AAD | 32.0214 | 4112 |
| AES-256-GCM-enc-no_AAD | 32.0883 | 7696 |
| AES-256-GCM-enc-no_AAD | 32.1052 | 15888 |
| AES-256-GCM-enc-no_AAD | 32.1038 | 32768 |
| AES-256-GCM-enc-no_AAD | 32.1293 | 65535 |
| Algorithm | ARMv8 – MB/s | Block Size |
| AES-256-GCM-enc-no_AAD | 120.862503 | 16 |
| AES-256-GCM-enc-no_AAD | 633.607939 | 528 |
| AES-256-GCM-enc-no_AAD | 715.517677 | 1024 |
| AES-256-GCM-enc-no_AAD | 776.28316 | 4112 |
| AES-256-GCM-enc-no_AAD | 783.198307 | 7696 |
| AES-256-GCM-enc-no_AAD | 793.405041 | 15888 |
| AES-256-GCM-enc-no_AAD | 793.122663 | 32768 |
| AES-256-GCM-enc-no_AAD | 797.332681 | 65535 |
For RSA operations the following chart shows performance differences using a 4096 bit key for private key operations. SP stands for Single Precision.
If you have questions about any of the above, please contact us at facts@wolfssl.com or call +1 425 245 8247.
Download wolfSSL Now