1 1.1 mrg dnl SPARC v9 32-bit mpn_mul_1 -- Multiply a limb vector with a limb and store 2 1.1 mrg dnl the result in a second limb vector. 3 1.1 mrg 4 1.1 mrg dnl Copyright 1998, 2000, 2001, 2003 Free Software Foundation, Inc. 5 1.1 mrg 6 1.1 mrg dnl This file is part of the GNU MP Library. 7 1.1.1.2 mrg dnl 8 1.1 mrg dnl The GNU MP Library is free software; you can redistribute it and/or modify 9 1.1.1.2 mrg dnl it under the terms of either: 10 1.1.1.2 mrg dnl 11 1.1.1.2 mrg dnl * the GNU Lesser General Public License as published by the Free 12 1.1.1.2 mrg dnl Software Foundation; either version 3 of the License, or (at your 13 1.1.1.2 mrg dnl option) any later version. 14 1.1.1.2 mrg dnl 15 1.1.1.2 mrg dnl or 16 1.1.1.2 mrg dnl 17 1.1.1.2 mrg dnl * the GNU General Public License as published by the Free Software 18 1.1.1.2 mrg dnl Foundation; either version 2 of the License, or (at your option) any 19 1.1.1.2 mrg dnl later version. 20 1.1.1.2 mrg dnl 21 1.1.1.2 mrg dnl or both in parallel, as here. 22 1.1.1.2 mrg dnl 23 1.1 mrg dnl The GNU MP Library is distributed in the hope that it will be useful, but 24 1.1 mrg dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 25 1.1.1.2 mrg dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 26 1.1.1.2 mrg dnl for more details. 27 1.1.1.2 mrg dnl 28 1.1.1.2 mrg dnl You should have received copies of the GNU General Public License and the 29 1.1.1.2 mrg dnl GNU Lesser General Public License along with the GNU MP Library. If not, 30 1.1.1.2 mrg dnl see https://www.gnu.org/licenses/. 31 1.1 mrg 32 1.1 mrg include(`../config.m4') 33 1.1 mrg 34 1.1 mrg C Algorithm: We use two floating-point multiplies per limb product, with the 35 1.1 mrg C invariant v operand split into two 16-bit pieces, and the u operand split 36 1.1 mrg C into 32-bit pieces. We convert the two 48-bit products and transfer them to 37 1.1 mrg C the integer unit. 38 1.1 mrg 39 1.1 mrg C cycles/limb 40 1.1 mrg C UltraSPARC 1&2: 6.5 41 1.1 mrg C UltraSPARC 3: ? 42 1.1 mrg 43 1.1 mrg C Possible optimizations: 44 1.1 mrg C 1. Combine 32-bit memory operations into 64-bit operations. Since we're 45 1.1 mrg C memory bandwidth limited, this could save 1.5 cycles/limb. 46 1.1 mrg C 2. Unroll the inner loop. Since we already use alternate temporary areas, 47 1.1 mrg C it is very straightforward to unroll, using an exit branch midways. 48 1.1 mrg C Unrolling would allow deeper scheduling which could improve speed for L2 49 1.1 mrg C cache case. 50 1.1 mrg C 3. For mpn_mul_1: Use more alternating temp areas. The std'es and ldx'es 51 1.1 mrg C aren't sufficiently apart-scheduled with just two temp areas. 52 1.1 mrg C 4. Specialize for particular v values. If its upper 16 bits are zero, we 53 1.1 mrg C could save many operations. 54 1.1 mrg 55 1.1 mrg C INPUT PARAMETERS 56 1.1 mrg C rp i0 57 1.1 mrg C up i1 58 1.1 mrg C n i2 59 1.1 mrg C v i3 60 1.1 mrg 61 1.1 mrg define(`FSIZE',224) 62 1.1 mrg 63 1.1 mrg ASM_START() 64 1.1 mrg PROLOGUE(mpn_mul_1) 65 1.1 mrg add %sp, -FSIZE, %sp 66 1.1 mrg sethi %hi(0xffff), %g1 67 1.1 mrg srl %o3, 16, %g2 68 1.1 mrg or %g1, %lo(0xffff), %g1 69 1.1 mrg and %o3, %g1, %g1 70 1.1 mrg stx %g1, [%sp+104] 71 1.1 mrg stx %g2, [%sp+112] 72 1.1 mrg ldd [%sp+104], %f6 73 1.1 mrg ldd [%sp+112], %f8 74 1.1 mrg fxtod %f6, %f6 75 1.1 mrg fxtod %f8, %f8 76 1.1 mrg ld [%sp+104], %f10 C zero f10 77 1.1 mrg 78 1.1 mrg mov 0, %g3 C cy = 0 79 1.1 mrg 80 1.1 mrg define(`fanop', `fitod %f18, %f0') C A quasi nop running in the FA pipe 81 1.1 mrg 82 1.1 mrg add %sp, 160, %o5 C point in scratch area 83 1.1 mrg and %o5, -32, %o5 C align at 0 (mod 32) in scratch area 84 1.1 mrg 85 1.1 mrg subcc %o2, 1, %o2 86 1.1 mrg ld [%o1], %f11 C read up[i] 87 1.1 mrg add %o1, 4, %o1 C up++ 88 1.1 mrg bne,pt %icc, .L_two_or_more 89 1.1 mrg fxtod %f10, %f2 90 1.1 mrg 91 1.1 mrg fmuld %f2, %f8, %f16 92 1.1 mrg fmuld %f2, %f6, %f4 93 1.1 mrg fdtox %f16, %f14 94 1.1 mrg fdtox %f4, %f12 95 1.1 mrg std %f14, [%o5+16] 96 1.1 mrg std %f12, [%o5+24] 97 1.1 mrg ldx [%o5+16], %g2 C p16 98 1.1 mrg ldx [%o5+24], %g1 C p0 99 1.1 mrg b .L1 100 1.1 mrg add %o0, -16, %o0 101 1.1 mrg 102 1.1 mrg .align 16 103 1.1 mrg .L_two_or_more: 104 1.1 mrg subcc %o2, 1, %o2 105 1.1 mrg ld [%o1], %f11 C read up[i] 106 1.1 mrg fmuld %f2, %f8, %f16 107 1.1 mrg fmuld %f2, %f6, %f4 108 1.1 mrg add %o1, 4, %o1 C up++ 109 1.1 mrg bne,pt %icc, .L_three_or_more 110 1.1 mrg fxtod %f10, %f2 111 1.1 mrg 112 1.1 mrg fdtox %f16, %f14 113 1.1 mrg fdtox %f4, %f12 114 1.1 mrg std %f14, [%o5+16] 115 1.1 mrg fmuld %f2, %f8, %f16 116 1.1 mrg std %f12, [%o5+24] 117 1.1 mrg fmuld %f2, %f6, %f4 118 1.1 mrg fdtox %f16, %f14 119 1.1 mrg fdtox %f4, %f12 120 1.1 mrg std %f14, [%o5+0] 121 1.1 mrg std %f12, [%o5+8] 122 1.1 mrg ldx [%o5+16], %g2 C p16 123 1.1 mrg ldx [%o5+24], %g1 C p0 124 1.1 mrg b .L2 125 1.1 mrg add %o0, -12, %o0 126 1.1 mrg 127 1.1 mrg .align 16 128 1.1 mrg .L_three_or_more: 129 1.1 mrg subcc %o2, 1, %o2 130 1.1 mrg ld [%o1], %f11 C read up[i] 131 1.1 mrg fdtox %f16, %f14 132 1.1 mrg fdtox %f4, %f12 133 1.1 mrg std %f14, [%o5+16] 134 1.1 mrg fmuld %f2, %f8, %f16 135 1.1 mrg std %f12, [%o5+24] 136 1.1 mrg fmuld %f2, %f6, %f4 137 1.1 mrg add %o1, 4, %o1 C up++ 138 1.1 mrg bne,pt %icc, .L_four_or_more 139 1.1 mrg fxtod %f10, %f2 140 1.1 mrg 141 1.1 mrg fdtox %f16, %f14 142 1.1 mrg fdtox %f4, %f12 143 1.1 mrg std %f14, [%o5+0] 144 1.1 mrg fmuld %f2, %f8, %f16 145 1.1 mrg std %f12, [%o5+8] 146 1.1 mrg fmuld %f2, %f6, %f4 147 1.1 mrg fdtox %f16, %f14 148 1.1 mrg ldx [%o5+16], %g2 C p16 149 1.1 mrg fdtox %f4, %f12 150 1.1 mrg ldx [%o5+24], %g1 C p0 151 1.1 mrg std %f14, [%o5+16] 152 1.1 mrg std %f12, [%o5+24] 153 1.1 mrg b .L3 154 1.1 mrg add %o0, -8, %o0 155 1.1 mrg 156 1.1 mrg .align 16 157 1.1 mrg .L_four_or_more: 158 1.1 mrg subcc %o2, 1, %o2 159 1.1 mrg ld [%o1], %f11 C read up[i] 160 1.1 mrg fdtox %f16, %f14 161 1.1 mrg fdtox %f4, %f12 162 1.1 mrg std %f14, [%o5+0] 163 1.1 mrg fmuld %f2, %f8, %f16 164 1.1 mrg std %f12, [%o5+8] 165 1.1 mrg fmuld %f2, %f6, %f4 166 1.1 mrg add %o1, 4, %o1 C up++ 167 1.1 mrg bne,pt %icc, .L_five_or_more 168 1.1 mrg fxtod %f10, %f2 169 1.1 mrg 170 1.1 mrg fdtox %f16, %f14 171 1.1 mrg ldx [%o5+16], %g2 C p16 172 1.1 mrg fdtox %f4, %f12 173 1.1 mrg ldx [%o5+24], %g1 C p0 174 1.1 mrg std %f14, [%o5+16] 175 1.1 mrg fmuld %f2, %f8, %f16 176 1.1 mrg std %f12, [%o5+24] 177 1.1 mrg fmuld %f2, %f6, %f4 178 1.1 mrg add %o1, 4, %o1 C up++ 179 1.1 mrg b .L4 180 1.1 mrg add %o0, -4, %o0 181 1.1 mrg 182 1.1 mrg .align 16 183 1.1 mrg .L_five_or_more: 184 1.1 mrg subcc %o2, 1, %o2 185 1.1 mrg ld [%o1], %f11 C read up[i] 186 1.1 mrg fdtox %f16, %f14 187 1.1 mrg ldx [%o5+16], %g2 C p16 188 1.1 mrg fdtox %f4, %f12 189 1.1 mrg ldx [%o5+24], %g1 C p0 190 1.1 mrg std %f14, [%o5+16] 191 1.1 mrg fmuld %f2, %f8, %f16 192 1.1 mrg std %f12, [%o5+24] 193 1.1 mrg fmuld %f2, %f6, %f4 194 1.1 mrg add %o1, 4, %o1 C up++ 195 1.1 mrg bne,pt %icc, .Loop 196 1.1 mrg fxtod %f10, %f2 197 1.1 mrg b,a .L5 198 1.1 mrg 199 1.1 mrg C BEGIN MAIN LOOP 200 1.1 mrg .align 16 201 1.1 mrg C -- 0 202 1.1 mrg .Loop: nop 203 1.1 mrg subcc %o2, 1, %o2 204 1.1 mrg ld [%o1], %f11 C read up[i] 205 1.1 mrg fdtox %f16, %f14 206 1.1 mrg C -- 1 207 1.1 mrg sllx %g2, 16, %g4 C (p16 << 16) 208 1.1 mrg add %o0, 4, %o0 C rp++ 209 1.1 mrg ldx [%o5+0], %g2 C p16 210 1.1 mrg fdtox %f4, %f12 211 1.1 mrg C -- 2 212 1.1 mrg nop 213 1.1 mrg add %g1, %g4, %g4 C p = p0 + (p16 << 16) 214 1.1 mrg ldx [%o5+8], %g1 C p0 215 1.1 mrg fanop 216 1.1 mrg C -- 3 217 1.1 mrg nop 218 1.1 mrg add %g3, %g4, %g4 C p += cy 219 1.1 mrg std %f14, [%o5+0] 220 1.1 mrg fmuld %f2, %f8, %f16 221 1.1 mrg C -- 4 222 1.1 mrg srlx %g4, 32, %g3 C new cy 223 1.1 mrg add %o1, 4, %o1 C up++ 224 1.1 mrg std %f12, [%o5+8] 225 1.1 mrg fmuld %f2, %f6, %f4 226 1.1 mrg C -- 5 227 1.1 mrg xor %o5, 16, %o5 C alternate scratch variables 228 1.1 mrg stw %g4, [%o0-4] 229 1.1 mrg bne,pt %icc, .Loop 230 1.1 mrg fxtod %f10, %f2 231 1.1 mrg C END MAIN LOOP 232 1.1 mrg 233 1.1 mrg .L5: fdtox %f16, %f14 234 1.1 mrg sllx %g2, 16, %g4 C (p16 << 16) 235 1.1 mrg ldx [%o5+0], %g2 C p16 236 1.1 mrg fdtox %f4, %f12 237 1.1 mrg add %g1, %g4, %g4 C p = p0 + (p16 << 16) 238 1.1 mrg ldx [%o5+8], %g1 C p0 239 1.1 mrg add %g4, %g3, %g4 C p += cy 240 1.1 mrg std %f14, [%o5+0] 241 1.1 mrg fmuld %f2, %f8, %f16 242 1.1 mrg std %f12, [%o5+8] 243 1.1 mrg fmuld %f2, %f6, %f4 244 1.1 mrg xor %o5, 16, %o5 245 1.1 mrg stw %g4, [%o0+0] 246 1.1 mrg srlx %g4, 32, %g3 C new cy 247 1.1 mrg 248 1.1 mrg .L4: fdtox %f16, %f14 249 1.1 mrg sllx %g2, 16, %g4 C (p16 << 16) 250 1.1 mrg ldx [%o5+0], %g2 C p16 251 1.1 mrg fdtox %f4, %f12 252 1.1 mrg add %g1, %g4, %g4 C p = p0 + (p16 << 16) 253 1.1 mrg ldx [%o5+8], %g1 C p0 254 1.1 mrg add %g3, %g4, %g4 C p += cy 255 1.1 mrg std %f14, [%o5+0] 256 1.1 mrg std %f12, [%o5+8] 257 1.1 mrg xor %o5, 16, %o5 258 1.1 mrg stw %g4, [%o0+4] 259 1.1 mrg srlx %g4, 32, %g3 C new cy 260 1.1 mrg 261 1.1 mrg .L3: sllx %g2, 16, %g4 C (p16 << 16) 262 1.1 mrg ldx [%o5+0], %g2 C p16 263 1.1 mrg add %g1, %g4, %g4 C p = p0 + (p16 << 16) 264 1.1 mrg ldx [%o5+8], %g1 C p0 265 1.1 mrg add %g3, %g4, %g4 C p += cy 266 1.1 mrg xor %o5, 16, %o5 267 1.1 mrg stw %g4, [%o0+8] 268 1.1 mrg srlx %g4, 32, %g3 C new cy 269 1.1 mrg 270 1.1 mrg .L2: sllx %g2, 16, %g4 C (p16 << 16) 271 1.1 mrg ldx [%o5+0], %g2 C p16 272 1.1 mrg add %g1, %g4, %g4 C p = p0 + (p16 << 16) 273 1.1 mrg ldx [%o5+8], %g1 C p0 274 1.1 mrg add %g3, %g4, %g4 C p += cy 275 1.1 mrg stw %g4, [%o0+12] 276 1.1 mrg srlx %g4, 32, %g3 C new cy 277 1.1 mrg 278 1.1 mrg .L1: sllx %g2, 16, %g4 C (p16 << 16) 279 1.1 mrg add %g1, %g4, %g4 C p = p0 + (p16 << 16) 280 1.1 mrg add %g3, %g4, %g4 C p += cy 281 1.1 mrg stw %g4, [%o0+16] 282 1.1 mrg srlx %g4, 32, %g3 C new cy 283 1.1 mrg 284 1.1 mrg mov %g3, %o0 285 1.1 mrg retl 286 1.1 mrg sub %sp, -FSIZE, %sp 287 1.1 mrg EPILOGUE(mpn_mul_1) 288
Indexes created Fri Apr 24 00:22:58 UTC 2026