AMD AOCC 4.0 vs. GCC vs. LLVM Clang Compiler Benchmarks On Zen 4

Simdjson is a high performance JSON parser that can make use of AVX-512 on capable CPUs, such as with Zen 4. For some of the simdjson benchmarks the AOCC 4.0 compiler was yielding a measurable increase in performance but in other test scenarios the performance was comparable to LLVM Clang 14 upstream.

Of course, it generally pays off to provide the workloads of most importance to you or your company, as the compiler benefits can vary.

JPEG XL image encoding was showing some nice speed-ups using LLVM Clang 15 upstream. Hopefully the next AOCC (4.1) release will re-based from LLVM 14 to LLVM 15.

When it came to JPEG-XL image decoding, AOCC 4.0 did shoot out in front of the other compilers.

With WebP image encoding it was a much tighter race.