Re: [PATCH] Hex-coding optimizations using SVE on ARM.

Auto-vectorization is working now with the following addition on Graviton 3 (m7g.4xlarge) with GCC 11.4, and the results match yours. Previously, auto-vectorization had no effect because we missed the -march=native option.

      encode.o: CFLAGS += ${CFLAGS_VECTORIZE} -march=native

There is a 30% improvement using auto-vectorization.

 buf   | default | auto_vec | SVE

--------+-------+--------+-------

     16 |     16  |      12  |    8

     64 |     58  |      40  |    9

    256 |    223  |     152  |   18

   1024 |    934  |     613  |   54

   4096 |   3533  |    2430  |  202

  16384 |  14081  |    9831  |  800

  65536 |  56374  |   38702  | 3202

Auto-vectorization had no effect on hex_decode due to the presence of control flow.

-----

Here is a comment snippet from src/include/port/simd.h

Currently, it is assumed that all aarch64 machine support NEON, but for newer advanced SIMD like SVE (and AVX512 for x86) this assumption may not hold. We need a runtime check to be sure.. Using src/include/port/simd.h to abstract away these advanced SIMD implementations may be difficult.