src/port/pg_crc32c_sse42.c

   1 /*-------------------------------------------------------------------------
   2  *
   3  * pg_crc32c_sse42.c
   4  *        Compute CRC-32C checksum using Intel SSE 4.2 instructions.
   5  *
   6  * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
   7  * Portions Copyright (c) 1994, Regents of the University of California
   8  *
   9  *
  10  * IDENTIFICATION
  11  *        src/port/pg_crc32c_sse42.c
  12  *
  13  *-------------------------------------------------------------------------
  14  */
  15 #include "c.h"
  16
  17 #include <nmmintrin.h>
  18
  19 #include "port/pg_crc32c.h"
  20
  21 pg_attribute_no_sanitize_alignment()
  22 pg_attribute_target("sse4.2")
  23 pg_crc32c
  24 pg_comp_crc32c_sse42(pg_crc32c crc, const void *data, size_t len)
  25 {
  26         const unsigned char *p = data;
  27         const unsigned char *pend = p + len;
  28
  29         /*
  30          * Process eight bytes of data at a time.
  31          *
  32          * NB: We do unaligned accesses here. The Intel architecture allows that,
  33          * and performance testing didn't show any performance gain from aligning
  34          * the begin address.
  35          */
  36 #ifdef __x86_64__
  37         while (p + 8 <= pend)
  38         {
  39                 crc = (uint32) _mm_crc32_u64(crc, *((const uint64 *) p));
  40                 p += 8;
  41         }
  42
  43         /* Process remaining full four bytes if any */
  44         if (p + 4 <= pend)
  45         {
  46                 crc = _mm_crc32_u32(crc, *((const unsigned int *) p));
  47                 p += 4;
  48         }
  49 #else
  50
  51         /*
  52          * Process four bytes at a time. (The eight byte instruction is not
  53          * available on the 32-bit x86 architecture).
  54          */
  55         while (p + 4 <= pend)
  56         {
  57                 crc = _mm_crc32_u32(crc, *((const unsigned int *) p));
  58                 p += 4;
  59         }
  60 #endif                                                  /* __x86_64__ */
  61
  62         /* Process any remaining bytes one at a time. */
  63         while (p < pend)
  64         {
  65                 crc = _mm_crc32_u8(crc, *p);
  66                 p++;
  67         }
  68
  69         return crc;
  70 }