Phoronix article – AMD A0CC 1.2 (2018-05-20)
Phoronix posted an article benchmarking the newest AMD AOCC compiler.
The article compares performance with gcc 7, gcc 8 and llvm 6. This article goes through some of the benchmarks to understand what is measured. Overall, not sure the benchmarks chosen are particularly noteworthy. Some very small (tscp), non-cpu (redis) and some subtests strange (scimark2) but overall at least a few to give a generic comparison without big things standing out in the results.
| Benchmark | Phoronix observations | My observations | Analysis |
|---|---|---|---|
| fhourstones | AMD AOCC is fastest. | Single-threaded. High branch misses, also some backend stalls. | Analysis |
| scimark2 | Four workloads and AMD AOCC is in position #1 (matrix multiply), #2 (composite), #3 (lu factorization) and #4 (jacobi). | Strange choice of tests since composite is an average of 5 and two others are missing. Benchmark is single-threaded with backend stalls being primary limiter. | Analysis |
| tscp | LLVM better than GCC, AOCC slightly slower than llvm 6. | Single-threaded toy benchmark that runs in 1.2 seconds. | Analysis |
| graphics-magick | gcc is better overall than clang; AOCC very close to llvm 6. | On_Cpu is 30%; IPC is moderate with issues varying by workload | Analysis |
| himeno | llvm 6 is fastest and AOCC is slowest. | Single-threaded. Small grid size but still limited by backend memory stalls. | Analysis |
| build-php | AOCC is slowest. | Compiletime performance measurement. Much of the time close to 100% CPU and then slow at the end. | Analysis |
| stockfish | AOCC is fastest. | Frontend stalls with some branch misses as well. | Analysis |
| aobench | GCC is faster with AOCC close. | Single-threaded. Moderately high IPC with some backend stalls. | Analysis |
| encode-flac | AOCC similar to llvm 6 and faster than gcc | Single-threaded. Five short processes run in quick succession. High IPC with some backend stalls. | Analysis |
| encode-mp3 | GCC faster and AOCC similar to llvm 6 | Single-threaded with some backend stalls. | Analysis |
| redis | AOCC often slightly better | Not CPU bound, short intervals where context switches are most important. | Analysis |
| nginx | All compilers almost the same. | Single-threaded backend server. | Analysis |
Comments
Phoronix article – AMD A0CC 1.2 (2018-05-20) — No Comments
HTML tags allowed in your comment: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>