Reorganize local header files
[openssl.git] / crypto / des / des_local.h
diff --git a/crypto/des/des_local.h b/crypto/des/des_local.h
new file mode 100644 (file)
index 0000000..230a30f
--- /dev/null
@@ -0,0 +1,226 @@
+/*
+ * Copyright 1995-2016 The OpenSSL Project Authors. All Rights Reserved.
+ *
+ * Licensed under the Apache License 2.0 (the "License").  You may not use
+ * this file except in compliance with the License.  You can obtain a copy
+ * in the file LICENSE in the source distribution or at
+ * https://www.openssl.org/source/license.html
+ */
+
+#ifndef HEADER_DES_LOCL_H
+# define HEADER_DES_LOCL_H
+
+# include <openssl/e_os2.h>
+
+# include <stdio.h>
+# include <stdlib.h>
+# include <string.h>
+
+# include <openssl/des.h>
+
+# ifdef OPENSSL_BUILD_SHLIBCRYPTO
+#  undef OPENSSL_EXTERN
+#  define OPENSSL_EXTERN OPENSSL_EXPORT
+# endif
+
+# define ITERATIONS 16
+# define HALF_ITERATIONS 8
+
+# define c2l(c,l)        (l =((DES_LONG)(*((c)++)))    , \
+                         l|=((DES_LONG)(*((c)++)))<< 8L, \
+                         l|=((DES_LONG)(*((c)++)))<<16L, \
+                         l|=((DES_LONG)(*((c)++)))<<24L)
+
+/* NOTE - c is not incremented as per c2l */
+# define c2ln(c,l1,l2,n) { \
+                        c+=n; \
+                        l1=l2=0; \
+                        switch (n) { \
+                        case 8: l2 =((DES_LONG)(*(--(c))))<<24L; \
+                        /* fall thru */                          \
+                        case 7: l2|=((DES_LONG)(*(--(c))))<<16L; \
+                        /* fall thru */                          \
+                        case 6: l2|=((DES_LONG)(*(--(c))))<< 8L; \
+                        /* fall thru */                          \
+                        case 5: l2|=((DES_LONG)(*(--(c))));      \
+                        /* fall thru */                          \
+                        case 4: l1 =((DES_LONG)(*(--(c))))<<24L; \
+                        /* fall thru */                          \
+                        case 3: l1|=((DES_LONG)(*(--(c))))<<16L; \
+                        /* fall thru */                          \
+                        case 2: l1|=((DES_LONG)(*(--(c))))<< 8L; \
+                        /* fall thru */                          \
+                        case 1: l1|=((DES_LONG)(*(--(c))));      \
+                                } \
+                        }
+
+# define l2c(l,c)        (*((c)++)=(unsigned char)(((l)     )&0xff), \
+                         *((c)++)=(unsigned char)(((l)>> 8L)&0xff), \
+                         *((c)++)=(unsigned char)(((l)>>16L)&0xff), \
+                         *((c)++)=(unsigned char)(((l)>>24L)&0xff))
+
+/*
+ * replacements for htonl and ntohl since I have no idea what to do when
+ * faced with machines with 8 byte longs.
+ */
+
+# define n2l(c,l)        (l =((DES_LONG)(*((c)++)))<<24L, \
+                         l|=((DES_LONG)(*((c)++)))<<16L, \
+                         l|=((DES_LONG)(*((c)++)))<< 8L, \
+                         l|=((DES_LONG)(*((c)++))))
+
+# define l2n(l,c)        (*((c)++)=(unsigned char)(((l)>>24L)&0xff), \
+                         *((c)++)=(unsigned char)(((l)>>16L)&0xff), \
+                         *((c)++)=(unsigned char)(((l)>> 8L)&0xff), \
+                         *((c)++)=(unsigned char)(((l)     )&0xff))
+
+/* NOTE - c is not incremented as per l2c */
+# define l2cn(l1,l2,c,n) { \
+                        c+=n; \
+                        switch (n) { \
+                        case 8: *(--(c))=(unsigned char)(((l2)>>24L)&0xff); \
+                        /* fall thru */                                     \
+                        case 7: *(--(c))=(unsigned char)(((l2)>>16L)&0xff); \
+                        /* fall thru */                                     \
+                        case 6: *(--(c))=(unsigned char)(((l2)>> 8L)&0xff); \
+                        /* fall thru */                                     \
+                        case 5: *(--(c))=(unsigned char)(((l2)     )&0xff); \
+                        /* fall thru */                                     \
+                        case 4: *(--(c))=(unsigned char)(((l1)>>24L)&0xff); \
+                        /* fall thru */                                     \
+                        case 3: *(--(c))=(unsigned char)(((l1)>>16L)&0xff); \
+                        /* fall thru */                                     \
+                        case 2: *(--(c))=(unsigned char)(((l1)>> 8L)&0xff); \
+                        /* fall thru */                                     \
+                        case 1: *(--(c))=(unsigned char)(((l1)     )&0xff); \
+                                } \
+                        }
+
+# if defined(_MSC_VER)
+#  define ROTATE(a,n)     (_lrotr(a,n))
+# elif defined(__ICC)
+#  define ROTATE(a,n)     (_rotr(a,n))
+# elif defined(__GNUC__) && __GNUC__>=2 && !defined(__STRICT_ANSI__) && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM) && !defined(PEDANTIC)
+#  if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
+#   define ROTATE(a,n)   ({ register unsigned int ret;   \
+                                asm ("rorl %1,%0"       \
+                                        : "=r"(ret)     \
+                                        : "I"(n),"0"(a) \
+                                        : "cc");        \
+                           ret;                         \
+                        })
+#  endif
+# endif
+# ifndef ROTATE
+#  define ROTATE(a,n)     (((a)>>(n))+((a)<<(32-(n))))
+# endif
+
+/*
+ * Don't worry about the LOAD_DATA() stuff, that is used by fcrypt() to add
+ * it's little bit to the front
+ */
+
+# ifdef DES_FCRYPT
+
+#  define LOAD_DATA_tmp(R,S,u,t,E0,E1) \
+        { DES_LONG tmp; LOAD_DATA(R,S,u,t,E0,E1,tmp); }
+
+#  define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
+        t=R^(R>>16L); \
+        u=t&E0; t&=E1; \
+        tmp=(u<<16); u^=R^s[S  ]; u^=tmp; \
+        tmp=(t<<16); t^=R^s[S+1]; t^=tmp
+# else
+#  define LOAD_DATA_tmp(a,b,c,d,e,f) LOAD_DATA(a,b,c,d,e,f,g)
+#  define LOAD_DATA(R,S,u,t,E0,E1,tmp) \
+        u=R^s[S  ]; \
+        t=R^s[S+1]
+# endif
+
+/*
+ * It recently occurred to me that 0^0^0^0^0^0^0 == 0, so there is no reason
+ * to not xor all the sub items together.  This potentially saves a register
+ * since things can be xored directly into L
+ */
+
+# define D_ENCRYPT(LL,R,S) { \
+        LOAD_DATA_tmp(R,S,u,t,E0,E1); \
+        t=ROTATE(t,4); \
+        LL^= \
+            DES_SPtrans[0][(u>> 2L)&0x3f]^ \
+            DES_SPtrans[2][(u>>10L)&0x3f]^ \
+            DES_SPtrans[4][(u>>18L)&0x3f]^ \
+            DES_SPtrans[6][(u>>26L)&0x3f]^ \
+            DES_SPtrans[1][(t>> 2L)&0x3f]^ \
+            DES_SPtrans[3][(t>>10L)&0x3f]^ \
+            DES_SPtrans[5][(t>>18L)&0x3f]^ \
+            DES_SPtrans[7][(t>>26L)&0x3f]; }
+
+        /*-
+         * IP and FP
+         * The problem is more of a geometric problem that random bit fiddling.
+         0  1  2  3  4  5  6  7      62 54 46 38 30 22 14  6
+         8  9 10 11 12 13 14 15      60 52 44 36 28 20 12  4
+        16 17 18 19 20 21 22 23      58 50 42 34 26 18 10  2
+        24 25 26 27 28 29 30 31  to  56 48 40 32 24 16  8  0
+
+        32 33 34 35 36 37 38 39      63 55 47 39 31 23 15  7
+        40 41 42 43 44 45 46 47      61 53 45 37 29 21 13  5
+        48 49 50 51 52 53 54 55      59 51 43 35 27 19 11  3
+        56 57 58 59 60 61 62 63      57 49 41 33 25 17  9  1
+
+        The output has been subject to swaps of the form
+        0 1 -> 3 1 but the odd and even bits have been put into
+        2 3    2 0
+        different words.  The main trick is to remember that
+        t=((l>>size)^r)&(mask);
+        r^=t;
+        l^=(t<<size);
+        can be used to swap and move bits between words.
+
+        So l =  0  1  2  3  r = 16 17 18 19
+                4  5  6  7      20 21 22 23
+                8  9 10 11      24 25 26 27
+               12 13 14 15      28 29 30 31
+        becomes (for size == 2 and mask == 0x3333)
+           t =   2^16  3^17 -- --   l =  0  1 16 17  r =  2  3 18 19
+                 6^20  7^21 -- --        4  5 20 21       6  7 22 23
+                10^24 11^25 -- --        8  9 24 25      10 11 24 25
+                14^28 15^29 -- --       12 13 28 29      14 15 28 29
+
+        Thanks for hints from Richard Outerbridge - he told me IP&FP
+        could be done in 15 xor, 10 shifts and 5 ands.
+        When I finally started to think of the problem in 2D
+        I first got ~42 operations without xors.  When I remembered
+        how to use xors :-) I got it to its final state.
+        */
+# define PERM_OP(a,b,t,n,m) ((t)=((((a)>>(n))^(b))&(m)),\
+        (b)^=(t),\
+        (a)^=((t)<<(n)))
+
+# define IP(l,r) \
+        { \
+        register DES_LONG tt; \
+        PERM_OP(r,l,tt, 4,0x0f0f0f0fL); \
+        PERM_OP(l,r,tt,16,0x0000ffffL); \
+        PERM_OP(r,l,tt, 2,0x33333333L); \
+        PERM_OP(l,r,tt, 8,0x00ff00ffL); \
+        PERM_OP(r,l,tt, 1,0x55555555L); \
+        }
+
+# define FP(l,r) \
+        { \
+        register DES_LONG tt; \
+        PERM_OP(l,r,tt, 1,0x55555555L); \
+        PERM_OP(r,l,tt, 8,0x00ff00ffL); \
+        PERM_OP(l,r,tt, 2,0x33333333L); \
+        PERM_OP(r,l,tt,16,0x0000ffffL); \
+        PERM_OP(l,r,tt, 4,0x0f0f0f0fL); \
+        }
+
+extern const DES_LONG DES_SPtrans[8][64];
+
+void fcrypt_body(DES_LONG *out, DES_key_schedule *ks,
+                 DES_LONG Eswap0, DES_LONG Eswap1);
+
+#endif