block-sha1: improve code on large-register-set machines
[git/git.git] / block-sha1 / sha1.c
1 /*
2 * Based on the Mozilla SHA1 (see mozilla-sha1/sha1.c),
3 * optimized to do word accesses rather than byte accesses,
4 * and to avoid unnecessary copies into the context array.
5 */
6
7 #include <string.h>
8 #include <arpa/inet.h>
9
10 #include "sha1.h"
11
12 /* Hash one 64-byte block of data */
13 static void blk_SHA1Block(blk_SHA_CTX *ctx, const unsigned int *data);
14
15 void blk_SHA1_Init(blk_SHA_CTX *ctx)
16 {
17 ctx->size = 0;
18
19 /* Initialize H with the magic constants (see FIPS180 for constants)
20 */
21 ctx->H[0] = 0x67452301;
22 ctx->H[1] = 0xefcdab89;
23 ctx->H[2] = 0x98badcfe;
24 ctx->H[3] = 0x10325476;
25 ctx->H[4] = 0xc3d2e1f0;
26 }
27
28
29 void blk_SHA1_Update(blk_SHA_CTX *ctx, const void *data, unsigned long len)
30 {
31 int lenW = ctx->size & 63;
32
33 ctx->size += len;
34
35 /* Read the data into W and process blocks as they get full
36 */
37 if (lenW) {
38 int left = 64 - lenW;
39 if (len < left)
40 left = len;
41 memcpy(lenW + (char *)ctx->W, data, left);
42 lenW = (lenW + left) & 63;
43 len -= left;
44 data += left;
45 if (lenW)
46 return;
47 blk_SHA1Block(ctx, ctx->W);
48 }
49 while (len >= 64) {
50 blk_SHA1Block(ctx, data);
51 data += 64;
52 len -= 64;
53 }
54 if (len)
55 memcpy(ctx->W, data, len);
56 }
57
58
59 void blk_SHA1_Final(unsigned char hashout[20], blk_SHA_CTX *ctx)
60 {
61 static const unsigned char pad[64] = { 0x80 };
62 unsigned int padlen[2];
63 int i;
64
65 /* Pad with a binary 1 (ie 0x80), then zeroes, then length
66 */
67 padlen[0] = htonl(ctx->size >> 29);
68 padlen[1] = htonl(ctx->size << 3);
69
70 i = ctx->size & 63;
71 blk_SHA1_Update(ctx, pad, 1+ (63 & (55 - i)));
72 blk_SHA1_Update(ctx, padlen, 8);
73
74 /* Output hash
75 */
76 for (i = 0; i < 5; i++)
77 ((unsigned int *)hashout)[i] = htonl(ctx->H[i]);
78 }
79
80 #if defined(__i386__) || defined(__x86_64__)
81
82 #define SHA_ASM(op, x, n) ({ unsigned int __res; __asm__(op " %1,%0":"=r" (__res):"i" (n), "0" (x)); __res; })
83 #define SHA_ROL(x,n) SHA_ASM("rol", x, n)
84 #define SHA_ROR(x,n) SHA_ASM("ror", x, n)
85 #define SMALL_REGISTER_SET
86
87 #else
88
89 #define SHA_ROT(X,l,r) (((X) << (l)) | ((X) >> (r)))
90 #define SHA_ROL(X,n) SHA_ROT(X,n,32-(n))
91 #define SHA_ROR(X,n) SHA_ROT(X,32-(n),n)
92
93 #endif
94
95 /* This "rolls" over the 512-bit array */
96 #define W(x) (array[(x)&15])
97
98 /*
99 * If you have 32 registers or more, the compiler can (and should)
100 * try to change the array[] accesses into registers. However, on
101 * machines with less than ~25 registers, that won't really work,
102 * and at least gcc will make an unholy mess of it.
103 *
104 * So to avoid that mess which just slows things down, we force
105 * the stores to memory to actually happen (we might be better off
106 * with a 'W(t)=(val);asm("":"+m" (W(t))' there instead, as
107 * suggested by Artur Skawina - that will also make gcc unable to
108 * try to do the silly "optimize away loads" part because it won't
109 * see what the value will be).
110 *
111 * Ben Herrenschmidt reports that on PPC, the C version comes close
112 * to the optimized asm with this (ie on PPC you don't want that
113 * 'volatile', since there are lots of registers).
114 */
115 #ifdef SMALL_REGISTER_SET
116 #define setW(x, val) (*(volatile unsigned int *)&W(x) = (val))
117 #else
118 #define setW(x, val) (W(x) = (val))
119 #endif
120
121 /*
122 * Where do we get the source from? The first 16 iterations get it from
123 * the input data, the next mix it from the 512-bit array.
124 */
125 #define SHA_SRC(t) htonl(data[t])
126 #define SHA_MIX(t) SHA_ROL(W(t+13) ^ W(t+8) ^ W(t+2) ^ W(t), 1)
127
128 #define SHA_ROUND(t, input, fn, constant, A, B, C, D, E) do { \
129 unsigned int TEMP = input(t); setW(t, TEMP); \
130 E += TEMP + SHA_ROL(A,5) + (fn) + (constant); \
131 B = SHA_ROR(B, 2); } while (0)
132
133 #define T_0_15(t, A, B, C, D, E) SHA_ROUND(t, SHA_SRC, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
134 #define T_16_19(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (((C^D)&B)^D) , 0x5a827999, A, B, C, D, E )
135 #define T_20_39(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0x6ed9eba1, A, B, C, D, E )
136 #define T_40_59(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, ((B&C)+(D&(B^C))) , 0x8f1bbcdc, A, B, C, D, E )
137 #define T_60_79(t, A, B, C, D, E) SHA_ROUND(t, SHA_MIX, (B^C^D) , 0xca62c1d6, A, B, C, D, E )
138
139 static void blk_SHA1Block(blk_SHA_CTX *ctx, const unsigned int *data)
140 {
141 unsigned int A,B,C,D,E;
142 unsigned int array[16];
143
144 A = ctx->H[0];
145 B = ctx->H[1];
146 C = ctx->H[2];
147 D = ctx->H[3];
148 E = ctx->H[4];
149
150 /* Round 1 - iterations 0-16 take their input from 'data' */
151 T_0_15( 0, A, B, C, D, E);
152 T_0_15( 1, E, A, B, C, D);
153 T_0_15( 2, D, E, A, B, C);
154 T_0_15( 3, C, D, E, A, B);
155 T_0_15( 4, B, C, D, E, A);
156 T_0_15( 5, A, B, C, D, E);
157 T_0_15( 6, E, A, B, C, D);
158 T_0_15( 7, D, E, A, B, C);
159 T_0_15( 8, C, D, E, A, B);
160 T_0_15( 9, B, C, D, E, A);
161 T_0_15(10, A, B, C, D, E);
162 T_0_15(11, E, A, B, C, D);
163 T_0_15(12, D, E, A, B, C);
164 T_0_15(13, C, D, E, A, B);
165 T_0_15(14, B, C, D, E, A);
166 T_0_15(15, A, B, C, D, E);
167
168 /* Round 1 - tail. Input from 512-bit mixing array */
169 T_16_19(16, E, A, B, C, D);
170 T_16_19(17, D, E, A, B, C);
171 T_16_19(18, C, D, E, A, B);
172 T_16_19(19, B, C, D, E, A);
173
174 /* Round 2 */
175 T_20_39(20, A, B, C, D, E);
176 T_20_39(21, E, A, B, C, D);
177 T_20_39(22, D, E, A, B, C);
178 T_20_39(23, C, D, E, A, B);
179 T_20_39(24, B, C, D, E, A);
180 T_20_39(25, A, B, C, D, E);
181 T_20_39(26, E, A, B, C, D);
182 T_20_39(27, D, E, A, B, C);
183 T_20_39(28, C, D, E, A, B);
184 T_20_39(29, B, C, D, E, A);
185 T_20_39(30, A, B, C, D, E);
186 T_20_39(31, E, A, B, C, D);
187 T_20_39(32, D, E, A, B, C);
188 T_20_39(33, C, D, E, A, B);
189 T_20_39(34, B, C, D, E, A);
190 T_20_39(35, A, B, C, D, E);
191 T_20_39(36, E, A, B, C, D);
192 T_20_39(37, D, E, A, B, C);
193 T_20_39(38, C, D, E, A, B);
194 T_20_39(39, B, C, D, E, A);
195
196 /* Round 3 */
197 T_40_59(40, A, B, C, D, E);
198 T_40_59(41, E, A, B, C, D);
199 T_40_59(42, D, E, A, B, C);
200 T_40_59(43, C, D, E, A, B);
201 T_40_59(44, B, C, D, E, A);
202 T_40_59(45, A, B, C, D, E);
203 T_40_59(46, E, A, B, C, D);
204 T_40_59(47, D, E, A, B, C);
205 T_40_59(48, C, D, E, A, B);
206 T_40_59(49, B, C, D, E, A);
207 T_40_59(50, A, B, C, D, E);
208 T_40_59(51, E, A, B, C, D);
209 T_40_59(52, D, E, A, B, C);
210 T_40_59(53, C, D, E, A, B);
211 T_40_59(54, B, C, D, E, A);
212 T_40_59(55, A, B, C, D, E);
213 T_40_59(56, E, A, B, C, D);
214 T_40_59(57, D, E, A, B, C);
215 T_40_59(58, C, D, E, A, B);
216 T_40_59(59, B, C, D, E, A);
217
218 /* Round 4 */
219 T_60_79(60, A, B, C, D, E);
220 T_60_79(61, E, A, B, C, D);
221 T_60_79(62, D, E, A, B, C);
222 T_60_79(63, C, D, E, A, B);
223 T_60_79(64, B, C, D, E, A);
224 T_60_79(65, A, B, C, D, E);
225 T_60_79(66, E, A, B, C, D);
226 T_60_79(67, D, E, A, B, C);
227 T_60_79(68, C, D, E, A, B);
228 T_60_79(69, B, C, D, E, A);
229 T_60_79(70, A, B, C, D, E);
230 T_60_79(71, E, A, B, C, D);
231 T_60_79(72, D, E, A, B, C);
232 T_60_79(73, C, D, E, A, B);
233 T_60_79(74, B, C, D, E, A);
234 T_60_79(75, A, B, C, D, E);
235 T_60_79(76, E, A, B, C, D);
236 T_60_79(77, D, E, A, B, C);
237 T_60_79(78, C, D, E, A, B);
238 T_60_79(79, B, C, D, E, A);
239
240 ctx->H[0] += A;
241 ctx->H[1] += B;
242 ctx->H[2] += C;
243 ctx->H[3] += D;
244 ctx->H[4] += E;
245 }