代码

#include <stdint.h>
#include <sys/auxv.h>
#include <unistd.h>
#include <immintrin.h>
struct DuckInfo {
  uint64_t abi_version;
  const char *stdin_ptr; uint64_t stdin_size;
  char *stdout_ptr; uint64_t stdout_limit; uint64_t stdout_size;
  char *stderr_ptr; uint64_t stderr_limit; uint64_t stderr_size;
  const char *IB_ptr; uint64_t IB_limit;
  char *OB_ptr; uint64_t OB_limit;
  uint64_t tsc_frequency;
} __attribute__((packed));

__attribute__((target("avx2"))) static unsigned test_simd(const char *p){
    __m256i a = _mm256_set1_epi8('0');
    __m256i b = _mm256_loadu_si256((const __m256i*)p);
    __m256i c = _mm256_sub_epi8(b, a);
    return (unsigned)_mm256_movemask_epi8(_mm256_cmpgt_epi8(c, _mm256_set1_epi8(9)));
}

int main(){
    struct DuckInfo *di = (struct DuckInfo*)getauxval(0x6b637564);
    const char *p = di->stdin_ptr;
    uint64_t sz = di->stdin_size;
    volatile uint64_t acc = 0;
    for (int i = 0; i < 64; i++) acc += (uint64_t)(unsigned char)p[sz + i];
    acc += test_simd(p);
    char *o = di->stdout_ptr;
    for (int i = 0; i < 16; i++){ int nib = (acc >> (4*(15-i))) & 15; o[i] = nib < 10 ? '0'+nib : 'a'+nib-10; }
    di->stdout_size = 16;
    asm volatile("mov $60, %%eax; xor %%edi, %%edi; syscall" ::: "rax","rdi","memory");
    __builtin_unreachable();
}

评测结果