1 /* { dg-do compile } */
2 /* { dg-options "-Os -Wstack-usage=2500" } */
3
4 /* This is a modified algorithm with bit-not "~" at the Sigma-blocks.
5 It improves the test coverage of one_cmpldi2 and subdi3 patterns.
6 Unlike the original test case these insns can reach the reload pass,
7 which may result in large stack usage. */
8
9 #define SHA_LONG64 unsigned long long
10 #define U64(C) C##ULL
11
12 #define SHA_LBLOCK 16
13 #define SHA512_CBLOCK (SHA_LBLOCK*8)
14
15 typedef struct SHA512state_st {
16 SHA_LONG64 h[8];
17 SHA_LONG64 Nl, Nh;
18 union {
19 SHA_LONG64 d[SHA_LBLOCK];
20 unsigned char p[SHA512_CBLOCK];
21 } u;
22 unsigned int num, md_len;
23 } SHA512_CTX;
24
25 static const SHA_LONG64 K512[80] = {
26 U64(0x428a2f98d728ae22), U64(0x7137449123ef65cd),
27 U64(0xb5c0fbcfec4d3b2f), U64(0xe9b5dba58189dbbc),
28 U64(0x3956c25bf348b538), U64(0x59f111f1b605d019),
29 U64(0x923f82a4af194f9b), U64(0xab1c5ed5da6d8118),
30 U64(0xd807aa98a3030242), U64(0x12835b0145706fbe),
31 U64(0x243185be4ee4b28c), U64(0x550c7dc3d5ffb4e2),
32 U64(0x72be5d74f27b896f), U64(0x80deb1fe3b1696b1),
33 U64(0x9bdc06a725c71235), U64(0xc19bf174cf692694),
34 U64(0xe49b69c19ef14ad2), U64(0xefbe4786384f25e3),
35 U64(0x0fc19dc68b8cd5b5), U64(0x240ca1cc77ac9c65),
36 U64(0x2de92c6f592b0275), U64(0x4a7484aa6ea6e483),
37 U64(0x5cb0a9dcbd41fbd4), U64(0x76f988da831153b5),
38 U64(0x983e5152ee66dfab), U64(0xa831c66d2db43210),
39 U64(0xb00327c898fb213f), U64(0xbf597fc7beef0ee4),
40 U64(0xc6e00bf33da88fc2), U64(0xd5a79147930aa725),
41 U64(0x06ca6351e003826f), U64(0x142929670a0e6e70),
42 U64(0x27b70a8546d22ffc), U64(0x2e1b21385c26c926),
43 U64(0x4d2c6dfc5ac42aed), U64(0x53380d139d95b3df),
44 U64(0x650a73548baf63de), U64(0x766a0abb3c77b2a8),
45 U64(0x81c2c92e47edaee6), U64(0x92722c851482353b),
46 U64(0xa2bfe8a14cf10364), U64(0xa81a664bbc423001),
47 U64(0xc24b8b70d0f89791), U64(0xc76c51a30654be30),
48 U64(0xd192e819d6ef5218), U64(0xd69906245565a910),
49 U64(0xf40e35855771202a), U64(0x106aa07032bbd1b8),
50 U64(0x19a4c116b8d2d0c8), U64(0x1e376c085141ab53),
51 U64(0x2748774cdf8eeb99), U64(0x34b0bcb5e19b48a8),
52 U64(0x391c0cb3c5c95a63), U64(0x4ed8aa4ae3418acb),
53 U64(0x5b9cca4f7763e373), U64(0x682e6ff3d6b2b8a3),
54 U64(0x748f82ee5defb2fc), U64(0x78a5636f43172f60),
55 U64(0x84c87814a1f0ab72), U64(0x8cc702081a6439ec),
56 U64(0x90befffa23631e28), U64(0xa4506cebde82bde9),
57 U64(0xbef9a3f7b2c67915), U64(0xc67178f2e372532b),
58 U64(0xca273eceea26619c), U64(0xd186b8c721c0c207),
59 U64(0xeada7dd6cde0eb1e), U64(0xf57d4f7fee6ed178),
60 U64(0x06f067aa72176fba), U64(0x0a637dc5a2c898a6),
61 U64(0x113f9804bef90dae), U64(0x1b710b35131c471b),
62 U64(0x28db77f523047d84), U64(0x32caab7b40c72493),
63 U64(0x3c9ebe0a15c9bebc), U64(0x431d67c49c100d4c),
64 U64(0x4cc5d4becb3e42b6), U64(0x597f299cfc657e2a),
65 U64(0x5fcb6fab3ad6faec), U64(0x6c44198c4a475817)
66 };
67
68 #define B(x,j) (((SHA_LONG64)(*(((const unsigned char *)(&x))+j)))<<((7-j)*8))
69 #define PULL64(x) (B(x,0)|B(x,1)|B(x,2)|B(x,3)|B(x,4)|B(x,5)|B(x,6)|B(x,7))
70 #define ROTR(x,s) (((x)>>s) | (x)<<(64-s))
71 #define Sigma0(x) ~(ROTR((x),28) ^ ROTR((x),34) ^ ROTR((x),39))
72 #define Sigma1(x) ~(ROTR((x),14) ^ ROTR((x),18) ^ ROTR((x),41))
73 #define sigma0(x) ~(ROTR((x),1) ^ ROTR((x),8) ^ ((x)>>7))
74 #define sigma1(x) ~(ROTR((x),19) ^ ROTR((x),61) ^ ((x)>>6))
75 #define Ch(x,y,z) (((x) & (y)) ^ ((~(x)) & (z)))
76 #define Maj(x,y,z) (((x) & (y)) ^ ((x) & (z)) ^ ((y) & (z)))
77
78 #define ROUND_00_15(i,a,b,c,d,e,f,g,h) do { \
79 T1 += h + Sigma1(e) + Ch(e,f,g) + K512[i]; \
80 h = Sigma0(a) + Maj(a,b,c); \
81 d += T1; h += T1; } while (0)
82 #define ROUND_16_80(i,j,a,b,c,d,e,f,g,h,X) do { \
83 s0 = X[(j+1)&0x0f]; s0 = sigma0(s0); \
84 s1 = X[(j+14)&0x0f]; s1 = sigma1(s1); \
85 T1 = X[(j)&0x0f] += s0 + s1 + X[(j+9)&0x0f]; \
86 ROUND_00_15(i+j,a,b,c,d,e,f,g,h); } while (0)
87 void sha512_block_data_order(SHA512_CTX *ctx, const void *in,
88 unsigned int num)
89 {
90 const SHA_LONG64 *W = in;
91 SHA_LONG64 a, b, c, d, e, f, g, h, s0, s1, T1;
92 SHA_LONG64 X[16];
93 int i;
94
95 while (num--) {
96
97 a = ctx->h[0];
98 b = ctx->h[1];
99 c = ctx->h[2];
100 d = ctx->h[3];
101 e = ctx->h[4];
102 f = ctx->h[5];
103 g = ctx->h[6];
104 h = ctx->h[7];
105
106 T1 = X[0] = PULL64(W[0]);
107 ROUND_00_15(0, a, b, c, d, e, f, g, h);
108 T1 = X[1] = PULL64(W[1]);
109 ROUND_00_15(1, h, a, b, c, d, e, f, g);
110 T1 = X[2] = PULL64(W[2]);
111 ROUND_00_15(2, g, h, a, b, c, d, e, f);
112 T1 = X[3] = PULL64(W[3]);
113 ROUND_00_15(3, f, g, h, a, b, c, d, e);
114 T1 = X[4] = PULL64(W[4]);
115 ROUND_00_15(4, e, f, g, h, a, b, c, d);
116 T1 = X[5] = PULL64(W[5]);
117 ROUND_00_15(5, d, e, f, g, h, a, b, c);
118 T1 = X[6] = PULL64(W[6]);
119 ROUND_00_15(6, c, d, e, f, g, h, a, b);
120 T1 = X[7] = PULL64(W[7]);
121 ROUND_00_15(7, b, c, d, e, f, g, h, a);
122 T1 = X[8] = PULL64(W[8]);
123 ROUND_00_15(8, a, b, c, d, e, f, g, h);
124 T1 = X[9] = PULL64(W[9]);
125 ROUND_00_15(9, h, a, b, c, d, e, f, g);
126 T1 = X[10] = PULL64(W[10]);
127 ROUND_00_15(10, g, h, a, b, c, d, e, f);
128 T1 = X[11] = PULL64(W[11]);
129 ROUND_00_15(11, f, g, h, a, b, c, d, e);
130 T1 = X[12] = PULL64(W[12]);
131 ROUND_00_15(12, e, f, g, h, a, b, c, d);
132 T1 = X[13] = PULL64(W[13]);
133 ROUND_00_15(13, d, e, f, g, h, a, b, c);
134 T1 = X[14] = PULL64(W[14]);
135 ROUND_00_15(14, c, d, e, f, g, h, a, b);
136 T1 = X[15] = PULL64(W[15]);
137 ROUND_00_15(15, b, c, d, e, f, g, h, a);
138
139 for (i = 16; i < 80; i += 16) {
140 ROUND_16_80(i, 0, a, b, c, d, e, f, g, h, X);
141 ROUND_16_80(i, 1, h, a, b, c, d, e, f, g, X);
142 ROUND_16_80(i, 2, g, h, a, b, c, d, e, f, X);
143 ROUND_16_80(i, 3, f, g, h, a, b, c, d, e, X);
144 ROUND_16_80(i, 4, e, f, g, h, a, b, c, d, X);
145 ROUND_16_80(i, 5, d, e, f, g, h, a, b, c, X);
146 ROUND_16_80(i, 6, c, d, e, f, g, h, a, b, X);
147 ROUND_16_80(i, 7, b, c, d, e, f, g, h, a, X);
148 ROUND_16_80(i, 8, a, b, c, d, e, f, g, h, X);
149 ROUND_16_80(i, 9, h, a, b, c, d, e, f, g, X);
150 ROUND_16_80(i, 10, g, h, a, b, c, d, e, f, X);
151 ROUND_16_80(i, 11, f, g, h, a, b, c, d, e, X);
152 ROUND_16_80(i, 12, e, f, g, h, a, b, c, d, X);
153 ROUND_16_80(i, 13, d, e, f, g, h, a, b, c, X);
154 ROUND_16_80(i, 14, c, d, e, f, g, h, a, b, X);
155 ROUND_16_80(i, 15, b, c, d, e, f, g, h, a, X);
156 }
157
158 ctx->h[0] += a;
159 ctx->h[1] += b;
160 ctx->h[2] += c;
161 ctx->h[3] += d;
162 ctx->h[4] += e;
163 ctx->h[5] += f;
164 ctx->h[6] += g;
165 ctx->h[7] += h;
166
167 W += SHA_LBLOCK;
168 }
169 }