3 # Copyright (c) 2010-2011 Intel Corp.
4 # Author: Vinodh.Gopal@intel.com
6 # Erdinc.Ozturk@intel.com
7 # Maxim.Perminov@intel.com
9 # More information about algorithm used can be found at:
10 # http://www.cse.buffalo.edu/srds2009/escs2009_submission_Gopal.pdf
12 # ====================================================================
13 # Copyright (c) 2011 The OpenSSL Project. All rights reserved.
15 # Redistribution and use in source and binary forms, with or without
16 # modification, are permitted provided that the following conditions
19 # 1. Redistributions of source code must retain the above copyright
20 # notice, this list of conditions and the following disclaimer.
22 # 2. Redistributions in binary form must reproduce the above copyright
23 # notice, this list of conditions and the following disclaimer in
24 # the documentation and/or other materials provided with the
27 # 3. All advertising materials mentioning features or use of this
28 # software must display the following acknowledgment:
29 # "This product includes software developed by the OpenSSL Project
30 # for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
32 # 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
33 # endorse or promote products derived from this software without
34 # prior written permission. For written permission, please contact
35 # licensing@OpenSSL.org.
37 # 5. Products derived from this software may not be called "OpenSSL"
38 # nor may "OpenSSL" appear in their names without prior written
39 # permission of the OpenSSL Project.
41 # 6. Redistributions of any form whatsoever must retain the following
43 # "This product includes software developed by the OpenSSL Project
44 # for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
46 # THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
47 # EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
48 # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
49 # PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR
50 # ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
51 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
52 # NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
53 # LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
54 # HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
55 # STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
56 # ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
57 # OF THE POSSIBILITY OF SUCH DAMAGE.
58 # ====================================================================
62 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
64 my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
66 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
67 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
68 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
69 die "can't locate x86_64-xlate.pl";
71 open STDOUT,"| $^X $xlate $flavour $output";
81 #MULSTEP_512_ADD MACRO x7, x6, x5, x4, x3, x2, x1, x0, dst, src1, src2, add_src, tmp1, tmp2
83 # uses rax, rdx, and args
86 my ($x, $DST, $SRC2, $ASRC, $OP, $TMP)=@_;
87 my @X=@$x; # make a copy
89 mov (+8*0)($SRC2), %rax
90 mul $OP # rdx:rax = %OP * [0]
96 for(my $i=1;$i<8;$i++) {
100 mov (+8*$i)($SRC2), %rax
101 mul $OP # rdx:rax = %OP * [$i]
102 mov (+8*$i)($ASRC), $X[$i]
114 #MULSTEP_512 MACRO x7, x6, x5, x4, x3, x2, x1, x0, dst, src2, src1_val, tmp
116 # uses rax, rdx, and args
119 my ($x, $DST, $SRC2, $OP, $TMP)=@_;
120 my @X=@$x; # make a copy
122 mov (+8*0)($SRC2), %rax
123 mul $OP # rdx:rax = %OP * [0]
128 for(my $i=1;$i<8;$i++) {
132 mov (+8*$i)($SRC2), %rax
133 mul $OP # rdx:rax = %OP * [$i]
149 # macro to copy data from flat space to swizzled table
150 #MACRO swizzle pDst, pSrc, tmp1, tmp2
151 # pDst and pSrc are modified
154 my ($pDst, $pSrc, $cnt, $d0)=@_;
161 mov $d0#w, (+64*1)($pDst)
163 mov $d0#w, (+64*2)($pDst)
165 mov $d0#w, (+64*3)($pDst)
167 lea 64*4($pDst), $pDst
175 # macro to copy data from swizzled table to flat space
176 #MACRO unswizzle pDst, pSrc, tmp*3
179 my ($pDst, $pSrc, $cnt, $d0, $d1)=@_;
183 movzxw (+64*3+256*0)($pSrc), $d0
184 movzxw (+64*3+256*1)($pSrc), $d1
187 mov (+64*2+256*0)($pSrc), $d0#w
188 mov (+64*2+256*1)($pSrc), $d1#w
191 mov (+64*1+256*0)($pSrc), $d0#w
192 mov (+64*1+256*1)($pSrc), $d1#w
195 mov (+64*0+256*0)($pSrc), $d0#w
196 mov (+64*0+256*1)($pSrc), $d1#w
197 mov $d0, (+8*0)($pDst)
198 mov $d1, (+8*1)($pDst)
199 lea 256*2($pSrc), $pSrc
200 lea 8*2($pDst), $pDst
229 # 058 X1[11] P[9] Z[8]
230 # 050 X1[10] P[8] Z[7]
231 # 048 X1[9] P[7] Z[6]
232 # 040 X1[8] P[6] Z[5]
233 # 038 X1[7] P[5] Z[4]
234 # 030 X1[6] P[4] Z[3]
235 # 028 X1[5] P[3] Z[2]
236 # 020 X1[4] P[2] Z[1]
237 # 018 X1[3] P[1] Z[0]
238 # 010 X1[2] P[0] Y[2]
239 # 008 X1[1] Q[1] Y[1]
240 # 000 X1[0] Q[0] Y[0]
242 my $X1_offset = 0; # 13 qwords
243 my $X2_offset = $X1_offset + 13*8; # 11 qwords
244 my $Carries_offset = $X2_offset + 11*8; # 1 qword
245 my $Q_offset = 0; # 2 qwords
246 my $P_offset = $Q_offset + 2*8; # 11 qwords
247 my $Y_offset = 0; # 3 qwords
248 my $Z_offset = $Y_offset + 3*8; # 9 qwords
250 my $Red_Data_Size = $Carries_offset + 1*8; # (25 qwords)
257 # ... <old stack contents>
281 # 088 reduce result addr
294 # 008 pResult ; arg 1
295 # 000 rsp ; stack pointer before subtract
298 my $pResult_offset = 8*1 + $rsp_offset;
299 my $pG_offset = 8*1 + $pResult_offset;
300 my $pData_offset = 8*1 + $pG_offset;
301 my $i_offset = 8*1 + $pData_offset;
302 my $pg_offset = 8*1 + $i_offset;
303 my $loop_idx_offset = 8*1 + $pg_offset;
304 my $reserved1_offset = 8*1 + $loop_idx_offset;
305 my $exp_offset = 8*1 + $reserved1_offset;
306 my $red_result_addr_offset= 8*9 + $exp_offset;
307 my $reserved2_offset = 8*1 + $red_result_addr_offset;
308 my $Reduce_Data_offset = 8*5 + $reserved2_offset;
309 my $GT_offset = $Red_Data_Size + $Reduce_Data_offset;
310 my $tmp_offset = 8*8 + $GT_offset;
311 my $tmp16_offset = 8*8 + $tmp_offset;
312 my $garray_offset = 8*16 + $tmp16_offset;
313 my $mem_size = 8*8*32 + $garray_offset;
316 # Offsets within Reduce Data
319 # struct MODF_2FOLD_MONT_512_C1_DATA {
322 # UINT64 m1[8]; /* 2^768 % m */
323 # UINT64 m2[8]; /* 2^640 % m */
324 # UINT64 k1[2]; /* (- 1/m) % 2^128 */
328 my $M = 512; # = 8 * 8 * 8
329 my $M1 = 576; # = 8 * 8 * 9 /* += 8 * 8 */
330 my $M2 = 640; # = 8 * 8 * 10 /* += 8 * 8 */
331 my $K1 = 704; # = 8 * 8 * 11 /* += 8 * 8 */
339 # MULADD_128x512 : Function to multiply 128-bits (2 qwords) by 512-bits (8 qwords)
340 # and add 512-bits (8 qwords)
341 # to get 640 bits (10 qwords)
342 # Input: 128-bit mul source: [rdi+8*1], rbp
343 # 512-bit mul source: [rsi+8*n]
344 # 512-bit add source: r15, r14, ..., r9, r8
345 # Output: r9, r8, r15, r14, r13, r12, r11, r10, [rcx+8*1], [rcx+8*0]
346 # Clobbers all regs except: rcx, rsi, rdi
348 .type MULADD_128x512,\@abi-omnipotent
352 &MULSTEP_512([map("%r$_",(8..15))], "(+8*0)(%rcx)", "%rsi", "%rbp", "%rbx");
354 mov (+8*1)(%rdi), %rbp
356 &MULSTEP_512([map("%r$_",(9..15,8))], "(+8*1)(%rcx)", "%rsi", "%rbp", "%rbx");
359 .size MULADD_128x512,.-MULADD_128x512
364 #MULADD_256x512 MACRO pDst, pA, pB, OP, TMP, X7, X6, X5, X4, X3, X2, X1, X0
366 # Inputs: pDst: Destination (768 bits, 12 qwords)
367 # pA: Multiplicand (1024 bits, 16 qwords)
368 # pB: Multiplicand (512 bits, 8 qwords)
370 # where Ah is (in qwords) A[15:12] (256 bits) and Al is A[7:0] (512 bits)
371 # Results in X3 X2 X1 X0 X7 X6 X5 X4 Dst[3:0]
372 # Uses registers: arguments, RAX, RDX
375 my ($pDst, $pA, $pB, $OP, $TMP, $X)=@_;
377 mov (+8*12)($pA), $OP
379 &MULSTEP_512_ADD($X, "(+8*0)($pDst)", $pB, $pA, $OP, $TMP);
380 push(@$X,shift(@$X));
383 mov (+8*13)($pA), $OP
385 &MULSTEP_512($X, "(+8*1)($pDst)", $pB, $OP, $TMP);
386 push(@$X,shift(@$X));
389 mov (+8*14)($pA), $OP
391 &MULSTEP_512($X, "(+8*2)($pDst)", $pB, $OP, $TMP);
392 push(@$X,shift(@$X));
395 mov (+8*15)($pA), $OP
397 &MULSTEP_512($X, "(+8*3)($pDst)", $pB, $OP, $TMP);
398 push(@$X,shift(@$X));
402 # mont_reduce(UINT64 *x, /* 1024 bits, 16 qwords */
403 # UINT64 *m, /* 512 bits, 8 qwords */
404 # MODF_2FOLD_MONT_512_C1_DATA *data,
405 # UINT64 *r) /* 512 bits, 8 qwords */
406 # Input: x (number to be reduced): tmp16 (Implicit)
407 # m (modulus): [pM] (Implicit)
408 # data (reduce data): [pData] (Implicit)
409 # Output: r (result): Address in [red_res_addr]
410 # result also in: r9, r8, r15, r14, r13, r12, r11, r10
412 my @X=map("%r$_",(8..15));
415 .type mont_reduce,\@abi-omnipotent
424 lea (+$Reduce_Data_offset+$X1_offset+$STACK_DEPTH)(%rsp), %rdi # pX1 (Dst) 769 bits, 13 qwords
425 mov (+$pData_offset+$STACK_DEPTH)(%rsp), %rsi # pM1 (Bsrc) 512 bits, 8 qwords
427 lea (+$tmp16_offset+$STACK_DEPTH)(%rsp), %rcx # X (Asrc) 1024 bits, 16 qwords
431 &MULADD_256x512("%rdi", "%rcx", "%rsi", "%rbp", "%rbx", \@X); # rotates @X 4 times
432 # results in r11, r10, r9, r8, r15, r14, r13, r12, X1[3:0]
437 add (+8*8)(%rcx), $X[4]
438 adc (+8*9)(%rcx), $X[5]
439 adc (+8*10)(%rcx), $X[6]
440 adc (+8*11)(%rcx), $X[7]
442 # X1 is now rax, r11-r8, r15-r12, tmp16[3:0]
445 # check for carry ;; carry stored in rax
446 mov $X[4], (+8*8)(%rdi) # rdi points to X1
447 mov $X[5], (+8*9)(%rdi)
449 mov $X[7], (+8*11)(%rdi)
451 mov %rax, (+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp)
453 mov (+8*0)(%rdi), $X[4]
454 mov (+8*1)(%rdi), $X[5]
455 mov (+8*2)(%rdi), $X[6]
456 mov (+8*3)(%rdi), $X[7]
458 # X1 is now stored in: X1[11], rbp, X1[9:8], r15-r8
464 # do first part (X2 = Xh * M2)
465 add \$8*10, %rdi # rdi -> pXh ; 128 bits, 2 qwords
466 # Xh is actually { [rdi+8*1], rbp }
467 add \$($M2-$M1), %rsi # rsi -> M2
468 lea (+$Reduce_Data_offset+$X2_offset+$STACK_DEPTH)(%rsp), %rcx # rcx -> pX2 ; 641 bits, 11 qwords
470 unshift(@X,pop(@X)); unshift(@X,pop(@X));
473 call MULADD_128x512 # args in rcx, rdi / rbp, rsi, r15-r8
474 # result in r9, r8, r15, r14, r13, r12, r11, r10, X2[1:0]
475 mov (+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp), %rax
478 add (+8*8-8*10)(%rdi), $X[6] # (-8*10) is to adjust rdi -> Xh to Xl
479 adc (+8*9-8*10)(%rdi), $X[7]
480 mov $X[6], (+8*8)(%rcx)
481 mov $X[7], (+8*9)(%rcx)
484 mov %rax, (+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp)
486 lea (+$Reduce_Data_offset+$Q_offset+$STACK_DEPTH)(%rsp), %rdi # rdi -> pQ ; 128 bits, 2 qwords
487 add \$($K1-$M2), %rsi # rsi -> pK1 ; 128 bits, 2 qwords
489 # MUL_128x128t128 rdi, rcx, rsi ; Q = X2 * K1 (bottom half)
490 # B1:B0 = rsi[1:0] = K1[1:0]
491 # A1:A0 = rcx[1:0] = X2[1:0]
492 # Result = rdi[1],rbp = Q[1],rbp
494 mov (+8*1)(%rsi), %rbx # B1
496 mov (%rcx), %rax # A0
501 mov (+8*1)(%rcx), %rax # A1
505 mov (%rcx), %rax # A0
509 mov %r9, (+8*1)(%rdi)
510 # end MUL_128x128t128
515 mov (+8*1)(%rcx), $X[7] # r9:r8 = X2[1:0]
517 call MULADD_128x512 # args in rcx, rdi / rbp, rsi, r15-r8
518 # result in r9, r8, r15, r14, r13, r12, r11, r10, X2[1:0]
520 # load first half of m to rdx, rdi, rbx, rax
521 # moved this here for efficiency
522 mov (+8*0)(%rsi), %rax
523 mov (+8*1)(%rsi), %rbx
524 mov (+8*2)(%rsi), %rdi
525 mov (+8*3)(%rsi), %rdx
527 # continue with reduction
528 mov (+$Reduce_Data_offset+$Carries_offset+$STACK_DEPTH)(%rsp), %rbp
530 add (+8*8)(%rcx), $X[6]
531 adc (+8*9)(%rcx), $X[7]
533 #accumulate the final carry to rbp
536 # Add in overflow corrections: R = (X2>>128) += T[overflow]
537 # R = {r9, r8, r15, r14, ..., r10}
539 mov (+$pData_offset+$STACK_DEPTH)(%rsp), %rcx # rsi -> Data (and points to T)
540 add %rcx, %rbp # pT ; 512 bits, 8 qwords, spread out
542 # rsi will be used to generate a mask after the addition
545 add (+8*8*0)(%rbp), $X[0]
546 adc (+8*8*1)(%rbp), $X[1]
547 adc (+8*8*2)(%rbp), $X[2]
548 adc (+8*8*3)(%rbp), $X[3]
549 adc (+8*8*4)(%rbp), $X[4]
550 adc (+8*8*5)(%rbp), $X[5]
551 adc (+8*8*6)(%rbp), $X[6]
552 adc (+8*8*7)(%rbp), $X[7]
554 # if there is a carry: rsi = 0xFFFFFFFFFFFFFFFF
555 # if carry is clear: rsi = 0x0000000000000000
558 # if carry is clear, subtract 0. Otherwise, subtract 256 bits of m
570 # if there is a borrow: rbp = 0
571 # if there is no borrow: rbp = 1
572 # this is used to save the borrows in between the first half and the 2nd half of the subtraction of m
575 #load second half of m to rdx, rdi, rbx, rax
578 mov (+8*4)(%rcx), %rax
579 mov (+8*5)(%rcx), %rbx
580 mov (+8*6)(%rcx), %rdi
581 mov (+8*7)(%rcx), %rdx
583 # use the rsi mask as before
584 # if carry is clear, subtract 0. Otherwise, subtract 256 bits of m
590 # if rbp = 0, there was a borrow before, it is moved to the carry flag
591 # if rbp = 1, there was not a borrow before, carry flag is cleared
599 # write R back to memory
601 mov (+$red_result_addr_offset+$STACK_DEPTH)(%rsp), %rsi
602 mov $X[0], (+8*0)(%rsi)
603 mov $X[1], (+8*1)(%rsi)
604 mov $X[2], (+8*2)(%rsi)
605 mov $X[3], (+8*3)(%rsi)
606 mov $X[4], (+8*4)(%rsi)
607 mov $X[5], (+8*5)(%rsi)
608 mov $X[6], (+8*6)(%rsi)
609 mov $X[7], (+8*7)(%rsi)
612 .size mont_reduce,.-mont_reduce
617 #MUL_512x512 MACRO pDst, pA, pB, x7, x6, x5, x4, x3, x2, x1, x0, tmp*2
619 # Inputs: pDst: Destination (1024 bits, 16 qwords)
620 # pA: Multiplicand (512 bits, 8 qwords)
621 # pB: Multiplicand (512 bits, 8 qwords)
622 # Uses registers rax, rdx, args
623 # B operand in [pB] and also in x7...x0
626 my ($pDst, $pA, $pB, $x, $OP, $TMP, $pDst_o)=@_;
627 my ($pDst, $pDst_o) = ($pDst =~ m/([^+]*)\+?(.*)?/);
628 my @X=@$x; # make a copy
634 mul $OP # rdx:rax = %OP * [0]
635 mov %rax, (+$pDst_o+8*0)($pDst)
638 for(my $i=1;$i<8;$i++) {
641 mul $OP # rdx:rax = %OP * [$i]
648 for(my $i=1;$i<8;$i++) {
650 mov (+8*$i)($pA), $OP
653 &MULSTEP_512(\@X, "(+$pDst_o+8*$i)($pDst)", $pB, $OP, $TMP);
658 mov $X[0], (+$pDst_o+8*8)($pDst)
659 mov $X[1], (+$pDst_o+8*9)($pDst)
660 mov $X[2], (+$pDst_o+8*10)($pDst)
661 mov $X[3], (+$pDst_o+8*11)($pDst)
662 mov $X[4], (+$pDst_o+8*12)($pDst)
663 mov $X[5], (+$pDst_o+8*13)($pDst)
664 mov $X[6], (+$pDst_o+8*14)($pDst)
665 mov $X[7], (+$pDst_o+8*15)($pDst)
670 # mont_mul_a3b : subroutine to compute (Src1 * Src2) % M (all 512-bits)
671 # Input: src1: Address of source 1: rdi
672 # src2: Address of source 2: rsi
673 # Output: dst: Address of destination: [red_res_addr]
674 # src2 and result also in: r9, r8, r15, r14, r13, r12, r11, r10
675 # Temp: Clobbers [tmp16], all registers
677 .type mont_mul_a3b,\@abi-omnipotent
681 # multiply tmp = src1 * src2
682 # For multiply: dst = rcx, src1 = rdi, src2 = rsi
683 # stack depth is extra 8 from call
685 &MUL_512x512("%rsp+$tmp16_offset+8", "%rdi", "%rsi", [map("%r$_",(10..15,8..9))], "%rbp", "%rbx");
689 # Call reduce(tmp, m, data, dst)
691 # tail recursion optimization: jmp to mont_reduce and return from there
695 .size mont_mul_a3b,.-mont_mul_a3b
700 #SQR_512 MACRO pDest, pA, x7, x6, x5, x4, x3, x2, x1, x0, tmp*4
702 # Input in memory [pA] and also in x7...x0
703 # Uses all argument registers plus rax and rdx
705 # This version computes all of the off-diagonal terms into memory,
706 # and then it adds in the diagonal terms
710 my ($pDst, $pA, $x, $A, $tmp, $x7, $x6, $pDst_o)=@_;
711 my ($pDst, $pDst_o) = ($pDst =~ m/([^+]*)\+?(.*)?/);
712 my @X=@$x; # make a copy
721 mov %rax, (+$pDst_o+8*1)($pDst)
723 for(my $i=2;$i<8;$i++) {
735 mov $X[0], (+$pDst_o+8*2)($pDst)
738 # second pass 12...17
747 mov $X[1], (+$pDst_o+8*3)($pDst)
756 mov $X[2], (+$pDst_o+8*4)($pDst)
801 mov $X[3], (+$pDst_o+8*5)($pDst)
810 mov $X[4], (+$pDst_o+8*6)($pDst)
839 # fourth pass 34...37
848 mov $X[5], (+$pDst_o+8*7)($pDst)
857 mov $x7, (+$pDst_o+8*8)($pDst)
886 mov $X[1], (+$pDst_o+8*9)($pDst)
895 mov $X[2], (+$pDst_o+8*10)($pDst)
916 mov $X[5], (+$pDst_o+8*11)($pDst)
925 mov $X[1], (+$pDst_o+8*12)($pDst)
938 mov $X[2], (+$pDst_o+8*13)($pDst)
940 mov %rdx, (+$pDst_o+8*14)($pDst)
942 # start finalize (add in squares, and double off-terms)
943 mov (+$pDst_o+8*1)($pDst), $X[0]
944 mov (+$pDst_o+8*2)($pDst), $X[1]
945 mov (+$pDst_o+8*3)($pDst), $X[2]
946 mov (+$pDst_o+8*4)($pDst), $X[3]
947 mov (+$pDst_o+8*5)($pDst), $X[4]
948 mov (+$pDst_o+8*6)($pDst), $X[5]
950 mov (+8*3)($pA), %rax
963 mov (+8*0)($pA), %rax
965 mov %rax, (+$pDst_o+8*0)($pDst)
968 mov (+8*1)($pA), %rax
976 mov $X[0], (+$pDst_o+8*1)($pDst)
977 mov $X[1], (+$pDst_o+8*2)($pDst)
979 mov (+8*2)($pA), %rax
988 mov $X[2], (+$pDst_o+8*3)($pDst)
989 mov $X[3], (+$pDst_o+8*4)($pDst)
996 mov $X[4], (+$pDst_o+8*5)($pDst)
997 mov $X[5], (+$pDst_o+8*6)($pDst)
999 # %%tmp has 0/1 in column 7
1000 # %%A6 has a full value in column 7
1002 mov (+$pDst_o+8*7)($pDst), $X[0]
1003 mov (+$pDst_o+8*8)($pDst), $X[1]
1004 mov (+$pDst_o+8*9)($pDst), $X[2]
1005 mov (+$pDst_o+8*10)($pDst), $X[3]
1006 mov (+$pDst_o+8*11)($pDst), $X[4]
1007 mov (+$pDst_o+8*12)($pDst), $X[5]
1008 mov (+$pDst_o+8*13)($pDst), $x6
1009 mov (+$pDst_o+8*14)($pDst), $x7
1028 mov (+8*4)($pA), %rax
1037 mov $X[0], (+$pDst_o+8*7)($pDst)
1038 mov $X[1], (+$pDst_o+8*8)($pDst)
1040 mov (+8*5)($pA), %rax
1049 mov $X[2], (+$pDst_o+8*9)($pDst)
1050 mov $X[3], (+$pDst_o+8*10)($pDst)
1052 mov (+8*6)($pA), %rax
1059 mov $X[4], (+$pDst_o+8*11)($pDst)
1060 mov $X[5], (+$pDst_o+8*12)($pDst)
1066 mov $x6, (+$pDst_o+8*13)($pDst)
1067 mov $x7, (+$pDst_o+8*14)($pDst)
1068 mov $A, (+$pDst_o+8*15)($pDst)
1073 # sqr_reduce: subroutine to compute Result = reduce(Result * Result)
1075 # input and result also in: r9, r8, r15, r14, r13, r12, r11, r10
1078 .type sqr_reduce,\@abi-omnipotent
1081 mov (+$pResult_offset+8)(%rsp), %rcx
1083 &SQR_512("%rsp+$tmp16_offset+8", "%rcx", [map("%r$_",(10..15,8..9))], "%rbx", "%rbp", "%rsi", "%rdi");
1085 # tail recursion optimization: jmp to mont_reduce and return from there
1089 .size sqr_reduce,.-sqr_reduce
1097 #mod_exp_512(UINT64 *result, /* 512 bits, 8 qwords */
1098 # UINT64 *g, /* 512 bits, 8 qwords */
1099 # UINT64 *exp, /* 512 bits, 8 qwords */
1100 # struct mod_ctx_512 *data)
1103 # table size = 2^5 = 32
1104 #table_entries equ 32
1105 #table_size equ table_entries * 8
1108 .type mod_exp_512,\@function,4
1117 # adjust stack down and then align it with cache boundary
1119 sub \$($mem_size), %rsp
1122 # store previous stack pointer and arguments
1123 mov %r8, (+$rsp_offset)(%rsp)
1124 mov %rdi, (+$pResult_offset)(%rsp)
1125 mov %rsi, (+$pG_offset)(%rsp)
1126 mov %rcx, (+$pData_offset)(%rsp)
1128 # transform g into montgomery space
1129 # GT = reduce(g * C2) = reduce(g * (2^256))
1130 # reduce expects to have the input in [tmp16]
1132 movdqu (+16*0)(%rsi), %xmm0
1133 movdqu (+16*1)(%rsi), %xmm1
1134 movdqu (+16*2)(%rsi), %xmm2
1135 movdqu (+16*3)(%rsi), %xmm3
1136 movdqa %xmm4, (+$tmp16_offset+16*0)(%rsp)
1137 movdqa %xmm4, (+$tmp16_offset+16*1)(%rsp)
1138 movdqa %xmm4, (+$tmp16_offset+16*6)(%rsp)
1139 movdqa %xmm4, (+$tmp16_offset+16*7)(%rsp)
1140 movdqa %xmm0, (+$tmp16_offset+16*2)(%rsp)
1141 movdqa %xmm1, (+$tmp16_offset+16*3)(%rsp)
1142 movdqa %xmm2, (+$tmp16_offset+16*4)(%rsp)
1143 movdqa %xmm3, (+$tmp16_offset+16*5)(%rsp)
1145 # load pExp before rdx gets blown away
1146 movdqu (+16*0)(%rdx), %xmm0
1147 movdqu (+16*1)(%rdx), %xmm1
1148 movdqu (+16*2)(%rdx), %xmm2
1149 movdqu (+16*3)(%rdx), %xmm3
1151 lea (+$GT_offset)(%rsp), %rbx
1152 mov %rbx, (+$red_result_addr_offset)(%rsp)
1155 # Initialize tmp = C
1156 lea (+$tmp_offset)(%rsp), %rcx
1158 mov %rax, (+8*0)(%rcx)
1159 mov %rax, (+8*1)(%rcx)
1160 mov %rax, (+8*3)(%rcx)
1161 mov %rax, (+8*4)(%rcx)
1162 mov %rax, (+8*5)(%rcx)
1163 mov %rax, (+8*6)(%rcx)
1164 mov %rax, (+8*7)(%rcx)
1165 mov %rax, (+$exp_offset+8*8)(%rsp)
1166 movq \$1, (+8*2)(%rcx)
1168 lea (+$garray_offset)(%rsp), %rbp
1169 mov %rcx, %rsi # pTmp
1170 mov %rbp, %rdi # Garray[][0]
1173 &swizzle("%rdi", "%rcx", "%rax", "%rbx");
1175 # for (rax = 31; rax != 0; rax--) {
1176 # tmp = reduce(tmp * G)
1181 mov %rax, (+$i_offset)(%rsp)
1182 mov %rbp, (+$pg_offset)(%rsp)
1184 mov %rsi, (+$red_result_addr_offset)(%rsp)
1185 mov (+8*0)(%rsi), %r10
1186 mov (+8*1)(%rsi), %r11
1187 mov (+8*2)(%rsi), %r12
1188 mov (+8*3)(%rsi), %r13
1189 mov (+8*4)(%rsi), %r14
1190 mov (+8*5)(%rsi), %r15
1191 mov (+8*6)(%rsi), %r8
1192 mov (+8*7)(%rsi), %r9
1194 lea (+$GT_offset)(%rsp), %rdi
1196 lea (+$tmp_offset)(%rsp), %rsi
1197 mov (+$pg_offset)(%rsp), %rbp
1199 mov %rbp, (+$pg_offset)(%rsp)
1200 mov %rsi, %rcx # rcx = rsi = addr of tmp
1203 &swizzle("%rbp", "%rcx", "%rax", "%rbx");
1205 mov (+$i_offset)(%rsp), %rax
1207 mov %rax, (+$i_offset)(%rsp)
1211 # Copy exponent onto stack
1212 movdqa %xmm0, (+$exp_offset+16*0)(%rsp)
1213 movdqa %xmm1, (+$exp_offset+16*1)(%rsp)
1214 movdqa %xmm2, (+$exp_offset+16*2)(%rsp)
1215 movdqa %xmm3, (+$exp_offset+16*3)(%rsp)
1220 # Initialize result to G[exp{511:507}]
1221 mov (+$exp_offset+62)(%rsp), %eax
1225 mov %edx, (+$exp_offset+62)(%rsp)
1226 lea (+$garray_offset)(%rsp,%rax,2), %rsi
1227 mov (+$pResult_offset)(%rsp), %rdx
1230 &unswizzle("%rdx", "%rsi", "%rbp", "%rbx", "%rax");
1234 # rcx = [loop_idx] = index: 510-5 to 0 by 5
1236 movq \$505, (+$loop_idx_offset)(%rsp)
1238 mov (+$pResult_offset)(%rsp), %rcx
1239 mov %rcx, (+$red_result_addr_offset)(%rsp)
1240 mov (+8*0)(%rcx), %r10
1241 mov (+8*1)(%rcx), %r11
1242 mov (+8*2)(%rcx), %r12
1243 mov (+8*3)(%rcx), %r13
1244 mov (+8*4)(%rcx), %r14
1245 mov (+8*5)(%rcx), %r15
1246 mov (+8*6)(%rcx), %r8
1247 mov (+8*7)(%rcx), %r9
1259 # Do multiply, first look up proper value in Garray
1260 mov (+$loop_idx_offset)(%rsp), %rcx # bit index
1262 shr \$4, %rax # rax is word pointer
1263 mov (+$exp_offset)(%rsp,%rax,2), %edx
1268 lea (+$garray_offset)(%rsp,%rdx,2), %rsi
1269 lea (+$tmp_offset)(%rsp), %rdx
1273 &unswizzle("%rdx", "%rsi", "%rbp", "%rbx", "%rax");
1277 # Call mod_mul_a1(pDst, pSrc1, pSrc2, pM, pData)
1278 # result result pG M Data
1280 mov (+$pResult_offset)(%rsp), %rsi
1285 mov (+$loop_idx_offset)(%rsp), %rcx
1287 mov %rcx, (+$loop_idx_offset)(%rsp)
1293 # transform result out of Montgomery space
1294 # result = reduce(result)
1295 mov (+$pResult_offset)(%rsp), %rdx
1297 movdqu (+16*0)(%rdx), %xmm0
1298 movdqu (+16*1)(%rdx), %xmm1
1299 movdqu (+16*2)(%rdx), %xmm2
1300 movdqu (+16*3)(%rdx), %xmm3
1301 movdqa %xmm4, (+$tmp16_offset+16*4)(%rsp)
1302 movdqa %xmm4, (+$tmp16_offset+16*5)(%rsp)
1303 movdqa %xmm4, (+$tmp16_offset+16*6)(%rsp)
1304 movdqa %xmm4, (+$tmp16_offset+16*7)(%rsp)
1305 movdqa %xmm0, (+$tmp16_offset+16*0)(%rsp)
1306 movdqa %xmm1, (+$tmp16_offset+16*1)(%rsp)
1307 movdqa %xmm2, (+$tmp16_offset+16*2)(%rsp)
1308 movdqa %xmm3, (+$tmp16_offset+16*3)(%rsp)
1311 # If result > m, subract m
1312 # load result into r15:r8
1313 mov (+$pResult_offset)(%rsp), %rax
1314 mov (+8*0)(%rax), %r8
1315 mov (+8*1)(%rax), %r9
1316 mov (+8*2)(%rax), %r10
1317 mov (+8*3)(%rax), %r11
1318 mov (+8*4)(%rax), %r12
1319 mov (+8*5)(%rax), %r13
1320 mov (+8*6)(%rax), %r14
1321 mov (+8*7)(%rax), %r15
1324 mov (+$pData_offset)(%rsp), %rbx
1327 sub (+8*0)(%rbx), %r8
1328 sbb (+8*1)(%rbx), %r9
1329 sbb (+8*2)(%rbx), %r10
1330 sbb (+8*3)(%rbx), %r11
1331 sbb (+8*4)(%rbx), %r12
1332 sbb (+8*5)(%rbx), %r13
1333 sbb (+8*6)(%rbx), %r14
1334 sbb (+8*7)(%rbx), %r15
1336 # if Carry is clear, replace result with difference
1337 mov (+8*0)(%rax), %rsi
1338 mov (+8*1)(%rax), %rdi
1339 mov (+8*2)(%rax), %rcx
1340 mov (+8*3)(%rax), %rdx
1345 mov %rsi, (+8*0)(%rax)
1346 mov %rdi, (+8*1)(%rax)
1347 mov %rcx, (+8*2)(%rax)
1348 mov %rdx, (+8*3)(%rax)
1350 mov (+8*4)(%rax), %rsi
1351 mov (+8*5)(%rax), %rdi
1352 mov (+8*6)(%rax), %rcx
1353 mov (+8*7)(%rax), %rdx
1358 mov %rsi, (+8*4)(%rax)
1359 mov %rdi, (+8*5)(%rax)
1360 mov %rcx, (+8*6)(%rax)
1361 mov %rdx, (+8*7)(%rax)
1363 mov (+$rsp_offset)(%rsp), %rsi
1373 .size mod_exp_512, . - mod_exp_512
1377 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1378 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1385 .extern __imp_RtlVirtualUnwind
1386 .type mod_exp_512_se_handler,\@abi-omnipotent
1388 mod_exp_512_se_handler:
1400 mov 120($context),%rax # pull context->Rax
1401 mov 248($context),%rbx # pull context->Rip
1403 lea .Lbody(%rip),%r10
1404 cmp %r10,%rbx # context->Rip<prologue label
1407 mov 152($context),%rax # pull context->Rsp
1409 lea .Lepilogue(%rip),%r10
1410 cmp %r10,%rbx # context->Rip>=epilogue label
1413 mov $rsp_offset(%rax),%rax # pull saved Rsp
1422 mov %rbx,144($context) # restore context->Rbx
1423 mov %rbp,160($context) # restore context->Rbp
1424 mov %r12,216($context) # restore context->R12
1425 mov %r13,224($context) # restore context->R13
1426 mov %r14,232($context) # restore context->R14
1427 mov %r15,240($context) # restore context->R15
1432 mov %rax,152($context) # restore context->Rsp
1433 mov %rsi,168($context) # restore context->Rsi
1434 mov %rdi,176($context) # restore context->Rdi
1436 mov 40($disp),%rdi # disp->ContextRecord
1437 mov $context,%rsi # context
1438 mov \$154,%ecx # sizeof(CONTEXT)
1439 .long 0xa548f3fc # cld; rep movsq
1442 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
1443 mov 8(%rsi),%rdx # arg2, disp->ImageBase
1444 mov 0(%rsi),%r8 # arg3, disp->ControlPc
1445 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
1446 mov 40(%rsi),%r10 # disp->ContextRecord
1447 lea 56(%rsi),%r11 # &disp->HandlerData
1448 lea 24(%rsi),%r12 # &disp->EstablisherFrame
1449 mov %r10,32(%rsp) # arg5
1450 mov %r11,40(%rsp) # arg6
1451 mov %r12,48(%rsp) # arg7
1452 mov %rcx,56(%rsp) # arg8, (NULL)
1453 call *__imp_RtlVirtualUnwind(%rip)
1455 mov \$1,%eax # ExceptionContinueSearch
1467 .size mod_exp_512_se_handler,.-mod_exp_512_se_handler
1471 .rva .LSEH_begin_mod_exp_512
1472 .rva .LSEH_end_mod_exp_512
1473 .rva .LSEH_info_mod_exp_512
1477 .LSEH_info_mod_exp_512:
1479 .rva mod_exp_512_se_handler
1485 if ($reg =~ /%r[0-9]+/) { $reg .= $conv; }
1486 elsif ($conv eq "b") { $reg =~ s/%[er]([^x]+)x?/%$1l/; }
1487 elsif ($conv eq "w") { $reg =~ s/%[er](.+)/%$1/; }
1488 elsif ($conv eq "d") { $reg =~ s/%[er](.+)/%e$1/; }
1492 $code =~ s/(%[a-z0-9]+)#([bwd])/reg_part($1,$2)/gem;
1493 $code =~ s/\`([^\`]*)\`/eval $1/gem;
1494 $code =~ s/(\(\+[^)]+\))/eval $1/gem;