Advanced Encryption Standard (AES) Benchmark

The Advanced Encryption Standard (AES) Benchmark denotes the AES encryption (AES enc) and AES decryption (AES dec) workloads used as benchmarks in the evaluation of the FISH (Fast Instruction SyntHesis) custom-processor synthesis tool. In the FISH paper, AES is grouped with other cryptography and multimedia benchmarks (DES, SHA, IDEA, djpeg, g721 encode/decode, rawcaudio, rawdaudio) used to measure run-time performance of the subgraph enumeration and instruction identification algorithms [22f4cc90-8240-484c-bc69-091dfdbee814, 04914463-16db-46e3-ac08-64d2c3e3f00f].

AES Implementation Used

The FISH evaluation uses an optimized 32-bit implementation of the Advanced Encryption Standard (AES) for both encryption and decryption, drawn from the cryptography benchmark suite referenced in the paper ^[1]. The AES benchmarks are listed alongside a FIPS-compliant fully unrolled Data Encryption Standard (DES) implementation and other multimedia and cryptography kernels.

Role in the FISH Benchmark Suite

Within the FISH benchmark suite, the AES benchmark is part of the broader set of multimedia and cryptography benchmarks used to stress subgraph-extraction and instruction-synthesis algorithms. The benchmarks are processed under varying algorithmic configurations (Basic Preprocessing, Partial Clustering, Enhanced Clustering, and combinations with prior optimization techniques) to measure the run-time required to compute the subgraph that maximizes MM (or equivalent coverage metric) within the largest basic block.

Reported Run-Time Behavior (from FISH, Table IV)

Table IV of the FISH paper reports run-time (in seconds) for the subgraph-maximization step across the benchmark suite. For the AES benchmarks the reported values are:

• AES enc: 21.550 (Atasu et al. [19]), 25.209 (Li et al. [28]), 3.6332 (Bron–Kerbosch), 0.05642 (ILP / CPLEX), 0.4600 (ILP / lpsolve), 2.8031 and 1.1987 (Enum-CMB configurations) ^[2].
• AES dec: 12057 (Atasu et al. [19]), 134.05 (Li et al. [28]), 3.0706 (Bron–Kerbosch), 0.23837 (ILP / CPLEX), 1.6080 (ILP / lpsolve), 15879 and 0.67775 (Enum-CMB configurations) ^[2].

The paper notes that AES encryption, AES decryption, DES, and djpeg are the benchmarks where search time dominates, while for smaller benchmarks such as g721 encode/decode, rawcaudio, and rawdaudio, the setup time dominates the search time ^[2].

Notes on the Evidence

• The FISH paper's tables report run-time only; no quality-of-result (QoR) figures, area, or performance metrics for AES are included in the provided evidence.
• Memory access, branch, and division instructions were marked as forbidden instructions in the experiments ^[1].
• Some table fragments in the provided evidence are partially garbled (line-wrapping artifacts), but the AES enc/AES dec rows are reconstructible from the surrounding context.