5072963 Need an optimized AES implementation for amd64

*** 24,45 ****
   */
  
  #ifndef _AES_IMPL_H
  #define _AES_IMPL_H
  
! #pragma ident   "@(#)aes_impl.h 1.3     08/02/26 SMI"
  
  /*
   * Common definitions used by AES.
   */
  
  #ifdef  __cplusplus
  extern "C" {
  #endif
  
! #define AES_BLOCK_LEN 16
  
  #define AES_COPY_BLOCK(src, dst) \
          (dst)[0] = (src)[0]; \
          (dst)[1] = (src)[1]; \
          (dst)[2] = (src)[2]; \
          (dst)[3] = (src)[3]; \
--- 24,53 ----
   */
  
  #ifndef _AES_IMPL_H
  #define _AES_IMPL_H
  
! #pragma ident   "@(#)aes_impl.h 1.4     08/05/19 SMI"
  
  /*
   * Common definitions used by AES.
   */
  
  #ifdef  __cplusplus
  extern "C" {
  #endif
  
! #include <sys/types.h>
  
+ /* Similar to sysmacros.h IS_P2ALIGNED, but checks two pointers: */
+ #define IS_P2ALIGNED2(v, w, a) \
+         ((((uintptr_t)(v) | (uintptr_t)(w)) & ((uintptr_t)(a) - 1)) == 0)
+ 
+ #define AES_BLOCK_LEN   16      /* bytes */
+ /* Round constant length, in number of 32-bit elements: */
+ #define RC_LENGTH       (5 * ((AES_BLOCK_LEN) / 4 - 2))
+ 
  #define AES_COPY_BLOCK(src, dst) \
          (dst)[0] = (src)[0]; \
          (dst)[1] = (src)[1]; \
          (dst)[2] = (src)[2]; \
          (dst)[3] = (src)[3]; \
*** 72,99 ****
          (dst)[12] ^= (src)[12]; \
          (dst)[13] ^= (src)[13]; \
          (dst)[14] ^= (src)[14]; \
          (dst)[15] ^= (src)[15]
  
  #define AES_MINBITS             128
! #define AES_MINBYTES            (AES_MINBITS >> 3)
  #define AES_MAXBITS             256
! #define AES_MAXBYTES            (AES_MAXBITS >> 3)
  
! #define AES_MIN_KEY_BYTES       (AES_MINBITS >> 3)
! #define AES_MAX_KEY_BYTES       (AES_MAXBITS >> 3)
  #define AES_192_KEY_BYTES       24
  #define AES_IV_LEN              16
  
  #define AES_32BIT_KS            32
  #define AES_64BIT_KS            64
  
! #define MAX_AES_NR              14
  
  typedef union {
!         uint64_t        ks64[(MAX_AES_NR + 1) * 4];
!         uint32_t        ks32[(MAX_AES_NR + 1) * 4];
  } aes_ks_t;
  
  typedef struct aes_key aes_key_t;
  struct aes_key {
          int             nr;
--- 80,112 ----
          (dst)[12] ^= (src)[12]; \
          (dst)[13] ^= (src)[13]; \
          (dst)[14] ^= (src)[14]; \
          (dst)[15] ^= (src)[15]
  
+ /* AES key size definitions */
  #define AES_MINBITS             128
! #define AES_MINBYTES            ((AES_MINBITS) >> 3)
  #define AES_MAXBITS             256
! #define AES_MAXBYTES            ((AES_MAXBITS) >> 3)
  
! #define AES_MIN_KEY_BYTES       ((AES_MINBITS) >> 3)
! #define AES_MAX_KEY_BYTES       ((AES_MAXBITS) >> 3)
  #define AES_192_KEY_BYTES       24
  #define AES_IV_LEN              16
  
+ /* AES key schedule may be implemented with 32- or 64-bit elements: */
  #define AES_32BIT_KS            32
  #define AES_64BIT_KS            64
  
! #define MAX_AES_NR              14 /* Maximum number of rounds */
! #define MAX_AES_NB              4  /* Number of columns comprising a state */
  
  typedef union {
! #ifdef  sun4u
!         uint64_t        ks64[((MAX_AES_NR) + 1) * (MAX_AES_NB)];
! #endif
!         uint32_t        ks32[((MAX_AES_NR) + 1) * (MAX_AES_NB)];
  } aes_ks_t;
  
  typedef struct aes_key aes_key_t;
  struct aes_key {
          int             nr;
*** 100,117 ****
          int             type;
          aes_ks_t        encr_ks;
          aes_ks_t        decr_ks;
  };
  
! extern void aes_encrypt_block(void *, uint8_t *, uint8_t *);
! extern void aes_decrypt_block(void *, uint8_t *, uint8_t *);
! extern void aes_init_keysched(uint8_t *, uint_t, void *);
! extern void *aes_alloc_keysched(size_t *, int);
! extern void aes_encrypt_impl(const aes_ks_t *ks, int Nr, const uint32_t pt[4],
!     uint32_t ct[4]);
! extern void aes_decrypt_impl(const aes_ks_t *ks, int Nr, const uint32_t ct[4],
!     uint32_t pt[4]);
  
  #ifdef  __cplusplus
  }
  #endif
  
--- 113,127 ----
          int             type;
          aes_ks_t        encr_ks;
          aes_ks_t        decr_ks;
  };
  
! extern void aes_encrypt_block(const void *ks, const uint8_t *pt, uint8_t *ct);
! extern void aes_decrypt_block(const void *ks, const uint8_t *ct, uint8_t *pt);
! extern void aes_init_keysched(const uint8_t *cipherKey, uint_t keyBits,
!         void *keysched);
! extern void *aes_alloc_keysched(size_t *size, int kmflag);
  
  #ifdef  __cplusplus
  }
  #endif