aesni-benchmarks

This project is simply a framework for benchmarking the performance of AES-NI accelerated brute-force efforts.

Hardware requirements

The framework requires an Intel platform with a micro architecture ≥ Westmere (although I've only tested it on Sandy Bridge and Haswell).

Output metrics

The framework reports on three different metrics: execution time (in seconds), throughput (in cycles per byte) and keys per second. I would recommending using throughput as the only meaningful metric but the others can be useful.

Parameters

The benchmarks.cpp contains the preprocessor definitions NUMER_OF_KEYS and NUMBER_OF_REPEATS that control the conditions for testing. These should be read in as command line arguments! ¯\_(ツ)_/¯

Adding tests

Implementations for testing should have the following prototype definition:

void testname(uint8_t *keys, uint8_t *data, uint8_t *dataOut);

and are included in the test suite by adding this call to benchmarks.cpp:

runBenchmarks(&testname, keys, data, cips, comp, STRIDE, "testname_label");

where testname is your new implementation, STRIDE is an integer that sets the number of keys tested per invocation of testname and testname_label is the label that will be printed when producing the benchmark stats.

Implementations included

Source	Description
intel_impl.c	Naïve Intel reference implementation ^[1]
bogdanov_impl.c	Implementation as described by Bogdanov et al.^[2]
luke_impl.c	Implementation as per an internal reference point
c_aestest.c	My own implementations written using Intrinsics
asm_aestest.c	An implementation written directly in assembly

[1] - https://software.intel.com/en-us/articles/intel-advanced-encryption-standard-aes-instructions-set
[2] - https://eprint.iacr.org/2015/795

Building

The project is written in a mix of C and C++ and includes a Makefile that should compile the code across any platform as long as GCC or clang is present.

Notes

The assembly implementation is omitted when compiling on a Windows platform as I suspect it's not compatible with MASM.
There are plenty of improvements to be made but this framework has served it's purpose and is not intended to be perfect.

jakelongo/aesni-benchmarks