summaryrefslogtreecommitdiff
path: root/ext/json/generator
diff options
context:
space:
mode:
authorJean Boussier <[email protected]>2025-04-28 15:23:48 +0200
committerJean Boussier <[email protected]>2025-04-30 08:12:41 +0200
commitb5beed45a22265f9537d23717637f6067458c834 ()
treed14e9522710958b153f4f2e0c47108c051f1c2fd /ext/json/generator
parentb7d1e33f26be7610df27d5c4538f443e81f986fa (diff)
[ruby/json] SIMD: Match control char and double quote in one pass
`c < 32 || c == 34` is equivalent to `c ^ 2 < 33`. Found in: https://lemire.me/blog/2025/04/13/detect-control-characters-quotes-and-backslashes-efficiently-using-swar/ The gain seem mostly present on micro-benchmark, and even there aren't very consistent, but it's never slower. ``` == Encoding long string (124001 bytes) ruby 3.4.2 (2025-02-15 revision https://.com/ruby/json/commit/d2930f8e7a) +YJIT +PRISM [arm64-darwin24] Warming up -------------------------------------- after 5.295k i/100ms Calculating ------------------------------------- after 55.796k (± 3.4%) i/s (17.92 μs/i) - 280.635k in 5.035690s Comparison: before: 49840.7 i/s after: 55795.8 i/s - 1.12x faster ``` https://.com/ruby/json/commit/034c5debd8
-rw-r--r--ext/json/generator/generator.c25
1 files changed, 10 insertions, 15 deletions
@@ -320,14 +320,12 @@ static inline FORCE_INLINE uint64_t neon_rules_update(const char *ptr)
{
uint8x16_t chunk = vld1q_u8((const unsigned char *)ptr);
- const uint8x16_t lower_bound = vdupq_n_u8(' ');
- const uint8x16_t backslash = vdupq_n_u8('\\');
- const uint8x16_t dblquote = vdupq_n_u8('\"');
- uint8x16_t too_low = vcltq_u8(chunk, lower_bound);
- uint8x16_t has_backslash = vceqq_u8(chunk, backslash);
- uint8x16_t has_dblquote = vceqq_u8(chunk, dblquote);
- uint8x16_t needs_escape = vorrq_u8(too_low, vorrq_u8(has_backslash, has_dblquote));
return neon_match_mask(needs_escape);
}
@@ -467,14 +465,11 @@ static inline TARGET_SSE2 FORCE_INLINE int sse2_update(const char *ptr)
{
__m128i chunk = _mm_loadu_si128((__m128i const*)ptr);
- const __m128i lower_bound = _mm_set1_epi8(' ');
- const __m128i backslash = _mm_set1_epi8('\\');
- const __m128i dblquote = _mm_set1_epi8('\"');
-
- __m128i too_low = _mm_cmplt_epu8(chunk, lower_bound);
- __m128i has_backslash = _mm_cmpeq_epi8(chunk, backslash);
- __m128i has_dblquote = _mm_cmpeq_epi8(chunk, dblquote);
- __m128i needs_escape = _mm_or_si128(too_low, _mm_or_si128(has_backslash, has_dblquote));
return _mm_movemask_epi8(needs_escape);
}