1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * If compiled without -DRF_INLINE_MACROS then needs -lm at link time 28 * If compiled with -DRF_INLINE_MACROS then needs conv.il at compile time 29 * (i.e. cc <compiler_flags> -DRF_INLINE_MACROS conv.il mont_mulf.c ) 30 */ 31 32 #include <sys/types.h> 33 #include <math.h> 34 35 static const double TwoTo16 = 65536.0; 36 static const double TwoToMinus16 = 1.0/65536.0; 37 static const double Zero = 0.0; 38 static const double TwoTo32 = 65536.0 * 65536.0; 39 static const double TwoToMinus32 = 1.0 / (65536.0 * 65536.0); 40 41 #ifdef RF_INLINE_MACROS 42 43 double upper32(double); 44 double lower32(double, double); 45 double mod(double, double, double); 46 47 #else 48 49 static double 50 upper32(double x) 51 { 52 return (floor(x * TwoToMinus32)); 53 } 54 55 56 static double 57 lower32(double x, double y) 58 { 59 return (x - TwoTo32 * floor(x * TwoToMinus32)); 60 } 61 62 static double 63 mod(double x, double oneoverm, double m) 64 { 65 return (x - m * floor(x * oneoverm)); 66 } 67 68 #endif 69 70 71 static void 72 cleanup(double *dt, int from, int tlen) 73 { 74 int i; 75 double tmp, tmp1, x, x1; 76 77 tmp = tmp1 = Zero; 78 79 for (i = 2 * from; i < 2 * tlen; i += 2) { 80 x = dt[i]; 81 x1 = dt[i + 1]; 82 dt[i] = lower32(x, Zero) + tmp; 83 dt[i + 1] = lower32(x1, Zero) + tmp1; 84 tmp = upper32(x); 85 tmp1 = upper32(x1); 86 } 87 } 88 89 90 void 91 conv_d16_to_i32(uint32_t *i32, double *d16, int64_t *tmp, int ilen) 92 { 93 int i; 94 int64_t t, t1, /* Using int64_t and not uint64_t */ 95 a, b, c, d; /* because more efficient code is */ 96 /* generated this way, and there */ 97 /* is no overflow. */ 98 t1 = 0; 99 a = (int64_t)d16[0]; 100 b = (int64_t)d16[1]; 101 for (i = 0; i < ilen - 1; i++) { 102 c = (int64_t)d16[2 * i + 2]; 103 t1 += a & 0xffffffff; 104 t = (a >> 32); 105 d = (int64_t)d16[2 * i + 3]; 106 t1 += (b & 0xffff) << 16; 107 t += (b >> 16) + (t1 >> 32); 108 i32[i] = t1 & 0xffffffff; 109 t1 = t; 110 a = c; 111 b = d; 112 } 113 t1 += a & 0xffffffff; 114 t = (a >> 32); 115 t1 += (b & 0xffff) << 16; 116 i32[i] = t1 & 0xffffffff; 117 } 118 119 void 120 conv_i32_to_d32(double *d32, uint32_t *i32, int len) 121 { 122 int i; 123 124 #pragma pipeloop(0) 125 for (i = 0; i < len; i++) 126 d32[i] = (double)(i32[i]); 127 } 128 129 130 void 131 conv_i32_to_d16(double *d16, uint32_t *i32, int len) 132 { 133 int i; 134 uint32_t a; 135 136 #pragma pipeloop(0) 137 for (i = 0; i < len; i++) { 138 a = i32[i]; 139 d16[2 * i] = (double)(a & 0xffff); 140 d16[2 * i + 1] = (double)(a >> 16); 141 } 142 } 143 144 #ifdef RF_INLINE_MACROS 145 146 void 147 i16_to_d16_and_d32x4(const double *, /* 1/(2^16) */ 148 const double *, /* 2^16 */ 149 const double *, /* 0 */ 150 double *, /* result16 */ 151 double *, /* result32 */ 152 float *); /* source - should be unsigned int* */ 153 /* converted to float* */ 154 155 #else 156 157 158 static void 159 i16_to_d16_and_d32x4(const double *dummy1, /* 1/(2^16) */ 160 const double *dummy2, /* 2^16 */ 161 const double *dummy3, /* 0 */ 162 double *result16, 163 double *result32, 164 float *src) /* source - should be unsigned int* */ 165 /* converted to float* */ 166 { 167 uint32_t *i32; 168 uint32_t a, b, c, d; 169 170 i32 = (uint32_t *)src; 171 a = i32[0]; 172 b = i32[1]; 173 c = i32[2]; 174 d = i32[3]; 175 result16[0] = (double)(a & 0xffff); 176 result16[1] = (double)(a >> 16); 177 result32[0] = (double)a; 178 result16[2] = (double)(b & 0xffff); 179 result16[3] = (double)(b >> 16); 180 result32[1] = (double)b; 181 result16[4] = (double)(c & 0xffff); 182 result16[5] = (double)(c >> 16); 183 result32[2] = (double)c; 184 result16[6] = (double)(d & 0xffff); 185 result16[7] = (double)(d >> 16); 186 result32[3] = (double)d; 187 } 188 189 #endif 190 191 192 void 193 conv_i32_to_d32_and_d16(double *d32, double *d16, uint32_t *i32, int len) 194 { 195 int i; 196 uint32_t a; 197 198 #pragma pipeloop(0) 199 for (i = 0; i < len - 3; i += 4) { 200 i16_to_d16_and_d32x4(&TwoToMinus16, &TwoTo16, &Zero, 201 &(d16[2*i]), &(d32[i]), (float *)(&(i32[i]))); 202 } 203 for (; i < len; i++) { 204 a = i32[i]; 205 d32[i] = (double)(i32[i]); 206 d16[2 * i] = (double)(a & 0xffff); 207 d16[2 * i + 1] = (double)(a >> 16); 208 } 209 } 210 211 212 static void 213 adjust_montf_result(uint32_t *i32, uint32_t *nint, int len) 214 { 215 int64_t acc; 216 int i; 217 218 if (i32[len] > 0) 219 i = -1; 220 else { 221 for (i = len - 1; i >= 0; i--) { 222 if (i32[i] != nint[i]) break; 223 } 224 } 225 if ((i < 0) || (i32[i] > nint[i])) { 226 acc = 0; 227 for (i = 0; i < len; i++) { 228 acc = acc + (uint64_t)(i32[i]) - (uint64_t)(nint[i]); 229 i32[i] = acc & 0xffffffff; 230 acc = acc >> 32; 231 } 232 } 233 } 234 235 236 /* 237 * the lengths of the input arrays should be at least the following: 238 * result[nlen+1], dm1[nlen], dm2[2*nlen+1], dt[4*nlen+2], dn[nlen], nint[nlen] 239 * all of them should be different from one another 240 */ 241 void mont_mulf_noconv(uint32_t *result, 242 double *dm1, double *dm2, double *dt, 243 double *dn, uint32_t *nint, 244 int nlen, double dn0) 245 { 246 int i, j, jj; 247 double digit, m2j, a, b; 248 double *pdm1, *pdm2, *pdn, *pdtj, pdn_0, pdm1_0; 249 250 pdm1 = &(dm1[0]); 251 pdm2 = &(dm2[0]); 252 pdn = &(dn[0]); 253 pdm2[2 * nlen] = Zero; 254 255 if (nlen != 16) { 256 for (i = 0; i < 4 * nlen + 2; i++) 257 dt[i] = Zero; 258 a = dt[0] = pdm1[0] * pdm2[0]; 259 digit = mod(lower32(a, Zero) * dn0, TwoToMinus16, TwoTo16); 260 261 pdtj = &(dt[0]); 262 for (j = jj = 0; j < 2 * nlen; j++, jj++, pdtj++) { 263 m2j = pdm2[j]; 264 a = pdtj[0] + pdn[0] * digit; 265 b = pdtj[1] + pdm1[0] * pdm2[j + 1] + a * TwoToMinus16; 266 pdtj[1] = b; 267 268 #pragma pipeloop(0) 269 for (i = 1; i < nlen; i++) { 270 pdtj[2 * i] += pdm1[i] * m2j + pdn[i] * digit; 271 } 272 if (jj == 30) { 273 cleanup(dt, j / 2 + 1, 2 * nlen + 1); 274 jj = 0; 275 } 276 277 digit = mod(lower32(b, Zero) * dn0, 278 TwoToMinus16, TwoTo16); 279 } 280 } else { 281 a = dt[0] = pdm1[0] * pdm2[0]; 282 283 dt[65] = dt[64] = dt[63] = dt[62] = dt[61] = dt[60] = 284 dt[59] = dt[58] = dt[57] = dt[56] = dt[55] = 285 dt[54] = dt[53] = dt[52] = dt[51] = dt[50] = 286 dt[49] = dt[48] = dt[47] = dt[46] = dt[45] = 287 dt[44] = dt[43] = dt[42] = dt[41] = dt[40] = 288 dt[39] = dt[38] = dt[37] = dt[36] = dt[35] = 289 dt[34] = dt[33] = dt[32] = dt[31] = dt[30] = 290 dt[29] = dt[28] = dt[27] = dt[26] = dt[25] = 291 dt[24] = dt[23] = dt[22] = dt[21] = dt[20] = 292 dt[19] = dt[18] = dt[17] = dt[16] = dt[15] = 293 dt[14] = dt[13] = dt[12] = dt[11] = dt[10] = 294 dt[9] = dt[8] = dt[7] = dt[6] = dt[5] = dt[4] = 295 dt[3] = dt[2] = dt[1] = Zero; 296 297 pdn_0 = pdn[0]; 298 pdm1_0 = pdm1[0]; 299 300 digit = mod(lower32(a, Zero) * dn0, TwoToMinus16, TwoTo16); 301 pdtj = &(dt[0]); 302 303 for (j = 0; j < 32; j++, pdtj++) { 304 305 m2j = pdm2[j]; 306 a = pdtj[0] + pdn_0 * digit; 307 b = pdtj[1] + pdm1_0 * pdm2[j + 1] + a * TwoToMinus16; 308 pdtj[1] = b; 309 310 pdtj[2] += pdm1[1] *m2j + pdn[1] * digit; 311 pdtj[4] += pdm1[2] *m2j + pdn[2] * digit; 312 pdtj[6] += pdm1[3] *m2j + pdn[3] * digit; 313 pdtj[8] += pdm1[4] *m2j + pdn[4] * digit; 314 pdtj[10] += pdm1[5] *m2j + pdn[5] * digit; 315 pdtj[12] += pdm1[6] *m2j + pdn[6] * digit; 316 pdtj[14] += pdm1[7] *m2j + pdn[7] * digit; 317 pdtj[16] += pdm1[8] *m2j + pdn[8] * digit; 318 pdtj[18] += pdm1[9] *m2j + pdn[9] * digit; 319 pdtj[20] += pdm1[10] *m2j + pdn[10] * digit; 320 pdtj[22] += pdm1[11] *m2j + pdn[11] * digit; 321 pdtj[24] += pdm1[12] *m2j + pdn[12] * digit; 322 pdtj[26] += pdm1[13] *m2j + pdn[13] * digit; 323 pdtj[28] += pdm1[14] *m2j + pdn[14] * digit; 324 pdtj[30] += pdm1[15] *m2j + pdn[15] * digit; 325 /* no need for cleanup, cannot overflow */ 326 digit = mod(lower32(b, Zero) * dn0, 327 TwoToMinus16, TwoTo16); 328 } 329 } 330 331 conv_d16_to_i32(result, dt + 2 * nlen, (int64_t *)dt, nlen + 1); 332 adjust_montf_result(result, nint, nlen); 333 } 334