Horton

Thread

Commits

  1. Optimize hex_encode() and hex_decode() using SIMD.

  1. remove switch statement in vector8_shift_{left,right}

    Nathan Bossart <nathandbossart@gmail.com> — 2026-07-02T00:32:16Z

    In commit ec8719ccbf, I added a switch statement with all expected shift
counts for vector8_shift_{left,right} because AArch64's vshlq_n_u32 and
vshrq_n_u32 require an integer literal.  I discovered that we can use
vshlq_u32 instead for both cases, which avoids the need for the switch
statement and compiles to the same machine code on recent gcc/clang.

-- 
nathan

  2. Re: remove switch statement in vector8_shift_{left,right}

    John Naylor <johncnaylorls@gmail.com> — 2026-07-02T05:28:36Z

    On Thu, Jul 2, 2026 at 7:32 AM Nathan Bossart <nathandbossart@gmail.com> wrote:
> In commit ec8719ccbf, I added a switch statement with all expected shift
> counts for vector8_shift_{left,right} because AArch64's vshlq_n_u32 and
> vshrq_n_u32 require an integer literal.  I discovered that we can use
> vshlq_u32 instead for both cases, which avoids the need for the switch
> statement and compiles to the same machine code on recent gcc/clang.

Hmm! +1

--
John Naylor
Amazon Web Services

  3. Re: remove switch statement in vector8_shift_{left,right}

    Nathan Bossart <nathandbossart@gmail.com> — 2026-07-06T16:40:56Z

    Committed.

-- 
nathan