1;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; 2; Copyright(c) 2011-2015 Intel Corporation All rights reserved. 3; 4; Redistribution and use in source and binary forms, with or without 5; modification, are permitted provided that the following conditions 6; are met: 7; * Redistributions of source code must retain the above copyright 8; notice, this list of conditions and the following disclaimer. 9; * Redistributions in binary form must reproduce the above copyright 10; notice, this list of conditions and the following disclaimer in 11; the documentation and/or other materials provided with the 12; distribution. 13; * Neither the name of Intel Corporation nor the names of its 14; contributors may be used to endorse or promote products derived 15; from this software without specific prior written permission. 16; 17; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 18; "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 19; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 20; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 21; OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22; SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 23; LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24; DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25; THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; 29 30;;; 31;;; gf_6vect_mad_avx2(len, vec, vec_i, mul_array, src, dest); 32;;; 33 34%include "reg_sizes.asm" 35 36%define PS 8 37 38%ifidn __OUTPUT_FORMAT__, win64 39 %define arg0 rcx 40 %define arg0.w ecx 41 %define arg1 rdx 42 %define arg2 r8 43 %define arg3 r9 44 %define arg4 r12 45 %define arg5 r15 46 %define tmp r11 47 %define tmp.w r11d 48 %define tmp.b r11b 49 %define tmp2 r10 50 %define tmp3 r13 51 %define return rax 52 %define return.w eax 53 %define stack_size 16*10 + 3*8 54 %define arg(x) [rsp + stack_size + PS + PS*x] 55 %define func(x) proc_frame x 56 57%macro FUNC_SAVE 0 58 sub rsp, stack_size 59 movdqa [rsp+16*0],xmm6 60 movdqa [rsp+16*1],xmm7 61 movdqa [rsp+16*2],xmm8 62 movdqa [rsp+16*3],xmm9 63 movdqa [rsp+16*4],xmm10 64 movdqa [rsp+16*5],xmm11 65 movdqa [rsp+16*6],xmm12 66 movdqa [rsp+16*7],xmm13 67 movdqa [rsp+16*8],xmm14 68 movdqa [rsp+16*9],xmm15 69 save_reg r12, 10*16 + 0*8 70 save_reg r13, 10*16 + 1*8 71 save_reg r15, 10*16 + 2*8 72 end_prolog 73 mov arg4, arg(4) 74 mov arg5, arg(5) 75%endmacro 76 77%macro FUNC_RESTORE 0 78 movdqa xmm6, [rsp+16*0] 79 movdqa xmm7, [rsp+16*1] 80 movdqa xmm8, [rsp+16*2] 81 movdqa xmm9, [rsp+16*3] 82 movdqa xmm10, [rsp+16*4] 83 movdqa xmm11, [rsp+16*5] 84 movdqa xmm12, [rsp+16*6] 85 movdqa xmm13, [rsp+16*7] 86 movdqa xmm14, [rsp+16*8] 87 movdqa xmm15, [rsp+16*9] 88 mov r12, [rsp + 10*16 + 0*8] 89 mov r13, [rsp + 10*16 + 1*8] 90 mov r15, [rsp + 10*16 + 2*8] 91 add rsp, stack_size 92%endmacro 93 94%elifidn __OUTPUT_FORMAT__, elf64 95 %define arg0 rdi 96 %define arg0.w edi 97 %define arg1 rsi 98 %define arg2 rdx 99 %define arg3 rcx 100 %define arg4 r8 101 %define arg5 r9 102 %define tmp r11 103 %define tmp.w r11d 104 %define tmp.b r11b 105 %define tmp2 r10 106 %define tmp3 r12 107 %define return rax 108 %define return.w eax 109 110 %define func(x) x: endbranch 111 %macro FUNC_SAVE 0 112 push r12 113 %endmacro 114 %macro FUNC_RESTORE 0 115 pop r12 116 %endmacro 117%endif 118 119;;; gf_6vect_mad_avx2(len, vec, vec_i, mul_array, src, dest) 120%define len arg0 121%define len.w arg0.w 122%define vec arg1 123%define vec_i arg2 124%define mul_array arg3 125%define src arg4 126%define dest1 arg5 127%define pos return 128%define pos.w return.w 129 130%define dest2 tmp3 131%define dest3 tmp2 132%define dest4 mul_array 133%define dest5 vec 134%define dest6 vec_i 135 136%ifndef EC_ALIGNED_ADDR 137;;; Use Un-aligned load/store 138 %define XLDR vmovdqu 139 %define XSTR vmovdqu 140%else 141;;; Use Non-temporal load/stor 142 %ifdef NO_NT_LDST 143 %define XLDR vmovdqa 144 %define XSTR vmovdqa 145 %else 146 %define XLDR vmovntdqa 147 %define XSTR vmovntdq 148 %endif 149%endif 150 151 152default rel 153 154[bits 64] 155section .text 156 157%define xmask0f ymm15 158%define xmask0fx xmm15 159%define xgft1_lo ymm14 160%define xgft2_lo ymm13 161%define xgft3_lo ymm12 162%define xgft4_lo ymm11 163%define xgft5_lo ymm10 164%define xgft6_lo ymm9 165 166%define x0 ymm0 167%define xtmpa ymm1 168%define xtmpl ymm2 169%define xtmplx xmm2 170%define xtmph ymm3 171%define xtmphx xmm3 172%define xd1 ymm4 173%define xd2 ymm5 174%define xd3 ymm6 175%define xd4 ymm7 176%define xd5 ymm8 177%define xd6 xd1 178 179align 16 180mk_global gf_6vect_mad_avx2, function 181func(gf_6vect_mad_avx2) 182 FUNC_SAVE 183 sub len, 32 184 jl .return_fail 185 xor pos, pos 186 mov tmp.b, 0x0f 187 vpinsrb xmask0fx, xmask0fx, tmp.w, 0 188 vpbroadcastb xmask0f, xmask0fx ;Construct mask 0x0f0f0f... 189 190 sal vec_i, 5 ;Multiply by 32 191 sal vec, 5 ;Multiply by 32 192 lea tmp, [mul_array + vec_i] 193 mov vec_i, vec 194 mov mul_array, vec 195 sal vec_i, 1 196 sal mul_array, 1 197 add vec_i, vec ;vec_i=vec*96 198 add mul_array, vec_i ;vec_i=vec*160 199 200 vmovdqu xgft1_lo, [tmp] ;Load array Ax{00}, Ax{01}, ..., Ax{0f} 201 ; " Ax{00}, Ax{10}, ..., Ax{f0} 202 vmovdqu xgft2_lo, [tmp+vec] ;Load array Bx{00}, Bx{01}, ..., Bx{0f} 203 ; " Bx{00}, Bx{10}, ..., Bx{f0} 204 vmovdqu xgft3_lo, [tmp+2*vec] ;Load array Cx{00}, Cx{01}, ..., Cx{0f} 205 ; " Cx{00}, Cx{10}, ..., Cx{f0} 206 vmovdqu xgft4_lo, [tmp+vec_i] ;Load array Fx{00}, Fx{01}, ..., Fx{0f} 207 ; " Fx{00}, Fx{10}, ..., Fx{f0} 208 vmovdqu xgft5_lo, [tmp+4*vec] ;Load array Ex{00}, Ex{01}, ..., Ex{0f} 209 ; " Ex{00}, Ex{10}, ..., Ex{f0} 210 vmovdqu xgft6_lo, [tmp+mul_array] ;Load array Dx{00}, Dx{01}, ..., Dx{0f} 211 ; " Dx{00}, Dx{10}, ..., Dx{f0} 212 213 mov dest2, [dest1+PS] ; reuse tmp3 214 mov dest3, [dest1+2*PS] ; reuse tmp2 215 mov dest4, [dest1+3*PS] ; reuse mul_array 216 mov dest5, [dest1+4*PS] ; reuse vec 217 mov dest6, [dest1+5*PS] ; reuse vec_i 218 mov dest1, [dest1] 219 220.loop32: 221 XLDR x0, [src+pos] ;Get next source vector 222 XLDR xd1, [dest1+pos] ;Get next dest vector 223 XLDR xd2, [dest2+pos] ;Get next dest vector 224 XLDR xd3, [dest3+pos] ;Get next dest vector 225 XLDR xd4, [dest4+pos] ;Get next dest vector 226 XLDR xd5, [dest5+pos] ;Get next dest vector 227 228 vpand xtmpl, x0, xmask0f ;Mask low src nibble in bits 4-0 229 vpsraw x0, x0, 4 ;Shift to put high nibble into bits 4-0 230 vpand x0, x0, xmask0f ;Mask high src nibble in bits 4-0 231 vperm2i128 xtmpa, xtmpl, x0, 0x30 ;swap xtmpa from 1lo|2lo to 1lo|2hi 232 vperm2i128 x0, xtmpl, x0, 0x12 ;swap x0 from 1hi|2hi to 1hi|2lo 233 234 ;dest1 235 vperm2i128 xtmph, xgft1_lo, xgft1_lo, 0x01 ; swapped to hi | lo 236 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 237 vpshufb xtmpl, xgft1_lo, xtmpa ;Lookup mul table of low nibble 238 vpxor xtmph, xtmph, xtmpl ;GF add high and low partials 239 vpxor xd1, xd1, xtmph ;xd1 += partial 240 241 XSTR [dest1+pos], xd1 ;Store result into dest1 242 243 ;dest2 244 vperm2i128 xtmph, xgft2_lo, xgft2_lo, 0x01 ; swapped to hi | lo 245 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 246 vpshufb xtmpl, xgft2_lo, xtmpa ;Lookup mul table of low nibble 247 vpxor xtmph, xtmph, xtmpl ;GF add high and low partials 248 vpxor xd2, xd2, xtmph ;xd2 += partial 249 250 ;dest3 251 vperm2i128 xtmph, xgft3_lo, xgft3_lo, 0x01 ; swapped to hi | lo 252 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 253 vpshufb xtmpl, xgft3_lo, xtmpa ;Lookup mul table of low nibble 254 vpxor xtmph, xtmph, xtmpl ;GF add high and low partials 255 vpxor xd3, xd3, xtmph ;xd3 += partial 256 257 XLDR xd6, [dest6+pos] ;reuse xd1. Get next dest vector 258 259 ;dest4 260 vperm2i128 xtmph, xgft4_lo, xgft4_lo, 0x01 ; swapped to hi | lo 261 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 262 vpshufb xtmpl, xgft4_lo, xtmpa ;Lookup mul table of low nibble 263 vpxor xtmph, xtmph, xtmpl ;GF add high and low partials 264 vpxor xd4, xd4, xtmph ;xd4 += partial 265 266 ;dest5 267 vperm2i128 xtmph, xgft5_lo, xgft5_lo, 0x01 ; swapped to hi | lo 268 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 269 vpshufb xtmpl, xgft5_lo, xtmpa ;Lookup mul table of low nibble 270 vpxor xtmph, xtmph, xtmpl ;GF add high and low partials 271 vpxor xd5, xd5, xtmph ;xd5 += partial 272 273 ;dest6 274 vperm2i128 xtmph, xgft6_lo, xgft6_lo, 0x01 ; swapped to hi | lo 275 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 276 vpshufb xtmpl, xgft6_lo, xtmpa ;Lookup mul table of low nibble 277 vpxor xtmph, xtmph, xtmpl ;GF add high and low partials 278 vpxor xd6, xd6, xtmph ;xd6 += partial 279 280 XSTR [dest2+pos], xd2 ;Store result into dest2 281 XSTR [dest3+pos], xd3 ;Store result into dest3 282 XSTR [dest4+pos], xd4 ;Store result into dest4 283 XSTR [dest5+pos], xd5 ;Store result into dest5 284 XSTR [dest6+pos], xd6 ;Store result into dest6 285 286 add pos, 32 ;Loop on 32 bytes at a time 287 cmp pos, len 288 jle .loop32 289 290 lea tmp, [len + 32] 291 cmp pos, tmp 292 je .return_pass 293 294.lessthan32: 295 ;; Tail len 296 ;; Do one more overlap pass 297 mov tmp.b, 0x1f 298 vpinsrb xtmphx, xtmphx, tmp.w, 0 299 vpbroadcastb xtmph, xtmphx ;Construct mask 0x1f1f1f... 300 301 mov tmp, len ;Overlapped offset length-32 302 303 XLDR x0, [src+tmp] ;Get next source vector 304 XLDR xd1, [dest1+tmp] ;Get next dest vector 305 XLDR xd2, [dest2+tmp] ;Get next dest vector 306 XLDR xd3, [dest3+tmp] ;Get next dest vector 307 XLDR xd4, [dest4+tmp] ;Get next dest vector 308 XLDR xd5, [dest5+tmp] ;Get next dest vector 309 310 sub len, pos 311 312 vpinsrb xtmplx, xtmplx, len.w, 15 313 vinserti128 xtmpl, xtmpl, xtmplx, 1 ;swapped to xtmplx | xtmplx 314 vpshufb xtmpl, xtmpl, xtmph ;Broadcast len to all bytes. xtmph=0x1f1f1f... 315 vpcmpgtb xtmpl, xtmpl, [constip32] 316 317 vpand xtmph, x0, xmask0f ;Mask low src nibble in bits 4-0 318 vpsraw x0, x0, 4 ;Shift to put high nibble into bits 4-0 319 vpand x0, x0, xmask0f ;Mask high src nibble in bits 4-0 320 vperm2i128 xtmpa, xtmph, x0, 0x30 ;swap xtmpa from 1lo|2lo to 1lo|2hi 321 vperm2i128 x0, xtmph, x0, 0x12 ;swap x0 from 1hi|2hi to 1hi|2lo 322 323 ;dest1 324 vperm2i128 xtmph, xgft1_lo, xgft1_lo, 0x01 ; swapped to hi | lo 325 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 326 vpshufb xgft1_lo, xgft1_lo, xtmpa ;Lookup mul table of low nibble 327 vpxor xtmph, xtmph, xgft1_lo ;GF add high and low partials 328 vpand xtmph, xtmph, xtmpl 329 vpxor xd1, xd1, xtmph ;xd1 += partial 330 331 XSTR [dest1+tmp], xd1 ;Store result into dest1 332 333 ;dest2 334 vperm2i128 xtmph, xgft2_lo, xgft2_lo, 0x01 ; swapped to hi | lo 335 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 336 vpshufb xgft2_lo, xgft2_lo, xtmpa ;Lookup mul table of low nibble 337 vpxor xtmph, xtmph, xgft2_lo ;GF add high and low partials 338 vpand xtmph, xtmph, xtmpl 339 vpxor xd2, xd2, xtmph ;xd2 += partial 340 341 ;dest3 342 vperm2i128 xtmph, xgft3_lo, xgft3_lo, 0x01 ; swapped to hi | lo 343 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 344 vpshufb xgft3_lo, xgft3_lo, xtmpa ;Lookup mul table of low nibble 345 vpxor xtmph, xtmph, xgft3_lo ;GF add high and low partials 346 vpand xtmph, xtmph, xtmpl 347 vpxor xd3, xd3, xtmph ;xd3 += partial 348 349 XLDR xd6, [dest6+tmp] ;reuse xd1. Get next dest vector 350 351 ;dest4 352 vperm2i128 xtmph, xgft4_lo, xgft4_lo, 0x01 ; swapped to hi | lo 353 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 354 vpshufb xgft4_lo, xgft4_lo, xtmpa ;Lookup mul table of low nibble 355 vpxor xtmph, xtmph, xgft4_lo ;GF add high and low partials 356 vpand xtmph, xtmph, xtmpl 357 vpxor xd4, xd4, xtmph ;xd4 += partial 358 359 ;dest5 360 vperm2i128 xtmph, xgft5_lo, xgft5_lo, 0x01 ; swapped to hi | lo 361 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 362 vpshufb xgft5_lo, xgft5_lo, xtmpa ;Lookup mul table of low nibble 363 vpxor xtmph, xtmph, xgft5_lo ;GF add high and low partials 364 vpand xtmph, xtmph, xtmpl 365 vpxor xd5, xd5, xtmph ;xd5 += partial 366 367 ;dest6 368 vperm2i128 xtmph, xgft6_lo, xgft6_lo, 0x01 ; swapped to hi | lo 369 vpshufb xtmph, xtmph, x0 ;Lookup mul table of high nibble 370 vpshufb xgft6_lo, xgft6_lo, xtmpa ;Lookup mul table of low nibble 371 vpxor xtmph, xtmph, xgft6_lo ;GF add high and low partials 372 vpand xtmph, xtmph, xtmpl 373 vpxor xd6, xd6, xtmph ;xd6 += partial 374 375 XSTR [dest2+tmp], xd2 ;Store result into dest2 376 XSTR [dest3+tmp], xd3 ;Store result into dest3 377 XSTR [dest4+tmp], xd4 ;Store result into dest4 378 XSTR [dest5+tmp], xd5 ;Store result into dest5 379 XSTR [dest6+tmp], xd6 ;Store result into dest6 380 381.return_pass: 382 FUNC_RESTORE 383 mov return, 0 384 ret 385 386.return_fail: 387 FUNC_RESTORE 388 mov return, 1 389 ret 390 391endproc_frame 392 393section .data 394align 32 395constip32: 396 dq 0xf8f9fafbfcfdfeff, 0xf0f1f2f3f4f5f6f7 397 dq 0xe8e9eaebecedeeef, 0xe0e1e2e3e4e5e6e7 398 399;;; func core, ver, snum 400slversion gf_6vect_mad_avx2, 04, 01, 0211 401