mirror of
https://github.com/ruby/ruby.git
synced 2025-09-17 09:33:59 +02:00
a3ec53bbb0
(https://github.com/ruby/json/pull/743) See the pull request for the long development history: https://github.com/ruby/json/pull/743 ``` == Encoding activitypub.json (52595 bytes) ruby 3.4.2 (2025-02-15 revisiond2930f8e7a) +YJIT +PRISM [arm64-darwin24] Warming up -------------------------------------- after 2.913k i/100ms Calculating ------------------------------------- after 29.377k (± 2.0%) i/s (34.04 μs/i) - 148.563k in 5.059169s Comparison: before: 23314.1 i/s after: 29377.3 i/s - 1.26x faster == Encoding citm_catalog.json (500298 bytes) ruby 3.4.2 (2025-02-15 revisiond2930f8e7a) +YJIT +PRISM [arm64-darwin24] Warming up -------------------------------------- after 152.000 i/100ms Calculating ------------------------------------- after 1.569k (± 0.8%) i/s (637.49 μs/i) - 7.904k in 5.039001s Comparison: before: 1485.6 i/s after: 1568.7 i/s - 1.06x faster == Encoding twitter.json (466906 bytes) ruby 3.4.2 (2025-02-15 revisiond2930f8e7a) +YJIT +PRISM [arm64-darwin24] Warming up -------------------------------------- after 309.000 i/100ms Calculating ------------------------------------- after 3.115k (± 3.1%) i/s (321.01 μs/i) - 15.759k in 5.063776s Comparison: before: 2508.3 i/s after: 3115.2 i/s - 1.24x faster ```49003523da
112 lines
2.4 KiB
C
112 lines
2.4 KiB
C
#include "extconf.h"
|
|
|
|
typedef enum {
|
|
SIMD_NONE,
|
|
SIMD_NEON,
|
|
SIMD_SSE2
|
|
} SIMD_Implementation;
|
|
|
|
#ifdef ENABLE_SIMD
|
|
|
|
#ifdef __clang__
|
|
#if __has_builtin(__builtin_ctzll)
|
|
#define HAVE_BUILTIN_CTZLL 1
|
|
#else
|
|
#define HAVE_BUILTIN_CTZLL 0
|
|
#endif
|
|
#elif defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))
|
|
#define HAVE_BUILTIN_CTZLL 1
|
|
#else
|
|
#define HAVE_BUILTIN_CTZLL 0
|
|
#endif
|
|
|
|
static inline uint32_t trailing_zeros64(uint64_t input) {
|
|
#if HAVE_BUILTIN_CTZLL
|
|
return __builtin_ctzll(input);
|
|
#else
|
|
uint32_t trailing_zeros = 0;
|
|
uint64_t temp = input;
|
|
while ((temp & 1) == 0 && temp > 0) {
|
|
trailing_zeros++;
|
|
temp >>= 1;
|
|
}
|
|
return trailing_zeros;
|
|
#endif
|
|
}
|
|
|
|
static inline int trailing_zeros(int input) {
|
|
#if HAVE_BUILTIN_CTZLL
|
|
return __builtin_ctz(input);
|
|
#else
|
|
int trailing_zeros = 0;
|
|
int temp = input;
|
|
while ((temp & 1) == 0 && temp > 0) {
|
|
trailing_zeros++;
|
|
temp >>= 1;
|
|
}
|
|
return trailing_zeros;
|
|
#endif
|
|
}
|
|
|
|
#define SIMD_MINIMUM_THRESHOLD 6
|
|
|
|
#if defined(__ARM_NEON) || defined(__ARM_NEON__) || defined(__aarch64__) || defined(_M_ARM64)
|
|
#include <arm_neon.h>
|
|
|
|
#define FIND_SIMD_IMPLEMENTATION_DEFINED 1
|
|
static SIMD_Implementation find_simd_implementation(void) {
|
|
return SIMD_NEON;
|
|
}
|
|
|
|
#define HAVE_SIMD_NEON 1
|
|
|
|
uint8x16x4_t load_uint8x16_4(const unsigned char *table) {
|
|
uint8x16x4_t tab;
|
|
tab.val[0] = vld1q_u8(table);
|
|
tab.val[1] = vld1q_u8(table+16);
|
|
tab.val[2] = vld1q_u8(table+32);
|
|
tab.val[3] = vld1q_u8(table+48);
|
|
return tab;
|
|
}
|
|
|
|
#endif /* ARM Neon Support.*/
|
|
|
|
#if defined(__amd64__) || defined(__amd64) || defined(__x86_64__) || defined(__x86_64) || defined(_M_X64) || defined(_M_AMD64)
|
|
|
|
#ifdef HAVE_X86INTRIN_H
|
|
#include <x86intrin.h>
|
|
|
|
#define HAVE_SIMD_SSE2 1
|
|
|
|
#ifdef HAVE_CPUID_H
|
|
#define FIND_SIMD_IMPLEMENTATION_DEFINED 1
|
|
|
|
#include <cpuid.h>
|
|
#endif /* HAVE_CPUID_H */
|
|
|
|
static SIMD_Implementation find_simd_implementation(void) {
|
|
|
|
#if defined(__GNUC__ ) || defined(__clang__)
|
|
#ifdef __GNUC__
|
|
__builtin_cpu_init();
|
|
#endif /* __GNUC__ */
|
|
|
|
// TODO Revisit. I think the SSE version now only uses SSE2 instructions.
|
|
if (__builtin_cpu_supports("sse2")) {
|
|
return SIMD_SSE2;
|
|
}
|
|
#endif /* __GNUC__ || __clang__*/
|
|
|
|
return SIMD_NONE;
|
|
}
|
|
|
|
#endif /* HAVE_X86INTRIN_H */
|
|
#endif /* X86_64 Support */
|
|
|
|
#endif /* ENABLE_SIMD */
|
|
|
|
#ifndef FIND_SIMD_IMPLEMENTATION_DEFINED
|
|
static SIMD_Implementation find_simd_implementation(void) {
|
|
return SIMD_NONE;
|
|
}
|
|
#endif
|