3 ##############################################################################
5 # Copyright 2014 Intel Corporation #
7 # Licensed under the Apache License, Version 2.0 (the "License"); #
8 # you may not use this file except in compliance with the License. #
9 # You may obtain a copy of the License at #
11 # http://www.apache.org/licenses/LICENSE-2.0 #
13 # Unless required by applicable law or agreed to in writing, software #
14 # distributed under the License is distributed on an "AS IS" BASIS, #
15 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #
16 # See the License for the specific language governing permissions and #
17 # limitations under the License. #
19 ##############################################################################
21 # Developers and authors: #
22 # Shay Gueron (1, 2), and Vlad Krasnov (1) #
23 # (1) Intel Corporation, Israel Development Center #
24 # (2) University of Haifa #
26 # S.Gueron and V.Krasnov, "Fast Prime Field Elliptic Curve Cryptography with#
29 ##############################################################################
31 # Further optimization by <appro@openssl.org>:
44 # Ranges denote minimum and maximum improvement coefficients depending
49 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
51 $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
53 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
54 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
55 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
56 die "can't locate x86_64-xlate.pl";
58 open OUT,"| \"$^X\" $xlate $flavour $output";
61 if (`$ENV{CC} -Wa,-v -c -o /dev/null -x assembler /dev/null 2>&1`
62 =~ /GNU assembler version ([2-9]\.[0-9]+)/) {
63 $avx = ($1>=2.19) + ($1>=2.22);
67 if (!$addx && $win64 && ($flavour =~ /nasm/ || $ENV{ASM} =~ /nasm/) &&
68 `nasm -v 2>&1` =~ /NASM version ([2-9]\.[0-9]+)/) {
69 $avx = ($1>=2.09) + ($1>=2.10);
73 if (!$addx && $win64 && ($flavour =~ /masm/ || $ENV{ASM} =~ /ml64/) &&
74 `ml64 2>&1` =~ /Version ([0-9]+)\./) {
75 $avx = ($1>=10) + ($1>=11);
79 if (!$addx && `$ENV{CC} -v 2>&1` =~ /(^clang version|based on LLVM) ([3-9])\.([0-9]+)/) {
80 my $ver = $2 + $3/100.0; # 3.1->3.01, 3.10->3.10
81 $avx = ($ver>=3.0) + ($ver>=3.01);
87 .extern OPENSSL_ia32cap_P
92 .quad 0xffffffffffffffff, 0x00000000ffffffff, 0x0000000000000000, 0xffffffff00000001
94 # 2^512 mod P precomputed for NIST P256 polynomial
96 .quad 0x0000000000000003, 0xfffffffbffffffff, 0xfffffffffffffffe, 0x00000004fffffffd
101 .long 2,2,2,2,2,2,2,2
103 .long 3,3,3,3,3,3,3,3
105 .quad 0x0000000000000001, 0xffffffff00000000, 0xffffffffffffffff, 0x00000000fffffffe
109 ################################################################################
110 # void ecp_nistz256_mul_by_2(uint64_t res[4], uint64_t a[4]);
112 my ($a0,$a1,$a2,$a3)=map("%r$_",(8..11));
113 my ($t0,$t1,$t2,$t3,$t4)=("%rax","%rdx","%rcx","%r12","%r13");
114 my ($r_ptr,$a_ptr,$b_ptr)=("%rdi","%rsi","%rdx");
118 .globl ecp_nistz256_mul_by_2
119 .type ecp_nistz256_mul_by_2,\@function,2
121 ecp_nistz256_mul_by_2:
127 add $a0, $a0 # a0:a3+a0:a3
131 lea .Lpoly(%rip), $a_ptr
158 .size ecp_nistz256_mul_by_2,.-ecp_nistz256_mul_by_2
160 ################################################################################
161 # void ecp_nistz256_div_by_2(uint64_t res[4], uint64_t a[4]);
162 .globl ecp_nistz256_div_by_2
163 .type ecp_nistz256_div_by_2,\@function,2
165 ecp_nistz256_div_by_2:
174 lea .Lpoly(%rip), $a_ptr
185 xor $a_ptr, $a_ptr # borrow $a_ptr
194 mov $a1, $t0 # a0:a3>>1
218 .size ecp_nistz256_div_by_2,.-ecp_nistz256_div_by_2
220 ################################################################################
221 # void ecp_nistz256_mul_by_3(uint64_t res[4], uint64_t a[4]);
222 .globl ecp_nistz256_mul_by_3
223 .type ecp_nistz256_mul_by_3,\@function,2
225 ecp_nistz256_mul_by_3:
232 add $a0, $a0 # a0:a3+a0:a3
244 sbb .Lpoly+8*1(%rip), $a1
247 sbb .Lpoly+8*3(%rip), $a3
256 add 8*0($a_ptr), $a0 # a0:a3+=a_ptr[0:3]
266 sbb .Lpoly+8*1(%rip), $a1
269 sbb .Lpoly+8*3(%rip), $a3
284 .size ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
286 ################################################################################
287 # void ecp_nistz256_add(uint64_t res[4], uint64_t a[4], uint64_t b[4]);
288 .globl ecp_nistz256_add
289 .type ecp_nistz256_add,\@function,3
300 lea .Lpoly(%rip), $a_ptr
330 .size ecp_nistz256_add,.-ecp_nistz256_add
332 ################################################################################
333 # void ecp_nistz256_sub(uint64_t res[4], uint64_t a[4], uint64_t b[4]);
334 .globl ecp_nistz256_sub
335 .type ecp_nistz256_sub,\@function,3
346 lea .Lpoly(%rip), $a_ptr
376 .size ecp_nistz256_sub,.-ecp_nistz256_sub
378 ################################################################################
379 # void ecp_nistz256_neg(uint64_t res[4], uint64_t a[4]);
380 .globl ecp_nistz256_neg
381 .type ecp_nistz256_neg,\@function,2
398 lea .Lpoly(%rip), $a_ptr
422 .size ecp_nistz256_neg,.-ecp_nistz256_neg
426 my ($r_ptr,$a_ptr,$b_org,$b_ptr)=("%rdi","%rsi","%rdx","%rbx");
427 my ($acc0,$acc1,$acc2,$acc3,$acc4,$acc5,$acc6,$acc7)=map("%r$_",(8..15));
428 my ($t0,$t1,$t2,$t3,$t4)=("%rcx","%rbp","%rbx","%rdx","%rax");
429 my ($poly1,$poly3)=($acc6,$acc7);
432 ################################################################################
433 # void ecp_nistz256_to_mont(
436 .globl ecp_nistz256_to_mont
437 .type ecp_nistz256_to_mont,\@function,2
439 ecp_nistz256_to_mont:
441 $code.=<<___ if ($addx);
443 and OPENSSL_ia32cap_P+8(%rip), %ecx
446 lea .LRR(%rip), $b_org
448 .size ecp_nistz256_to_mont,.-ecp_nistz256_to_mont
450 ################################################################################
451 # void ecp_nistz256_mul_mont(
456 .globl ecp_nistz256_mul_mont
457 .type ecp_nistz256_mul_mont,\@function,3
459 ecp_nistz256_mul_mont:
461 $code.=<<___ if ($addx);
463 and OPENSSL_ia32cap_P+8(%rip), %ecx
474 $code.=<<___ if ($addx);
480 mov 8*0($b_org), %rax
481 mov 8*0($a_ptr), $acc1
482 mov 8*1($a_ptr), $acc2
483 mov 8*2($a_ptr), $acc3
484 mov 8*3($a_ptr), $acc4
486 call __ecp_nistz256_mul_montq
488 $code.=<<___ if ($addx);
494 mov 8*0($b_org), %rdx
495 mov 8*0($a_ptr), $acc1
496 mov 8*1($a_ptr), $acc2
497 mov 8*2($a_ptr), $acc3
498 mov 8*3($a_ptr), $acc4
499 lea -128($a_ptr), $a_ptr # control u-op density
501 call __ecp_nistz256_mul_montx
512 .size ecp_nistz256_mul_mont,.-ecp_nistz256_mul_mont
514 .type __ecp_nistz256_mul_montq,\@abi-omnipotent
516 __ecp_nistz256_mul_montq:
517 ########################################################################
521 mov .Lpoly+8*1(%rip),$poly1
527 mov .Lpoly+8*3(%rip),$poly3
546 ########################################################################
547 # First reduction step
548 # Basically now we want to multiply acc[0] by p256,
549 # and add the result to the acc.
550 # Due to the special form of p256 we do some optimizations
552 # acc[0] x p256[0] = acc[0] x 2^64 - acc[0]
553 # then we add acc[0] and get acc[0] x 2^64
557 add $acc0, $acc1 # +=acc[0]*2^64
562 # acc[0] x p256[2] = 0
571 mov 8*1($b_ptr), %rax
575 ########################################################################
608 ########################################################################
609 # Second reduction step
624 mov 8*2($b_ptr), %rax
628 ########################################################################
661 ########################################################################
662 # Third reduction step
677 mov 8*3($b_ptr), %rax
681 ########################################################################
714 ########################################################################
715 # Final reduction step
723 #adc \$0, $t0 # doesn't overflow
734 ########################################################################
735 # Branch-less conditional subtraction of P
736 sub \$-1, $acc4 # .Lpoly[0]
738 sbb $poly1, $acc5 # .Lpoly[1]
739 sbb \$0, $acc0 # .Lpoly[2]
741 sbb $poly3, $acc1 # .Lpoly[3]
746 mov $acc4, 8*0($r_ptr)
748 mov $acc5, 8*1($r_ptr)
750 mov $acc0, 8*2($r_ptr)
751 mov $acc1, 8*3($r_ptr)
754 .size __ecp_nistz256_mul_montq,.-__ecp_nistz256_mul_montq
756 ################################################################################
757 # void ecp_nistz256_sqr_mont(
761 # we optimize the square according to S.Gueron and V.Krasnov,
762 # "Speeding up Big-Number Squaring"
763 .globl ecp_nistz256_sqr_mont
764 .type ecp_nistz256_sqr_mont,\@function,2
766 ecp_nistz256_sqr_mont:
768 $code.=<<___ if ($addx);
770 and OPENSSL_ia32cap_P+8(%rip), %ecx
780 $code.=<<___ if ($addx);
785 mov 8*0($a_ptr), %rax
786 mov 8*1($a_ptr), $acc6
787 mov 8*2($a_ptr), $acc7
788 mov 8*3($a_ptr), $acc0
790 call __ecp_nistz256_sqr_montq
792 $code.=<<___ if ($addx);
797 mov 8*0($a_ptr), %rdx
798 mov 8*1($a_ptr), $acc6
799 mov 8*2($a_ptr), $acc7
800 mov 8*3($a_ptr), $acc0
801 lea -128($a_ptr), $a_ptr # control u-op density
803 call __ecp_nistz256_sqr_montx
814 .size ecp_nistz256_sqr_mont,.-ecp_nistz256_sqr_mont
816 .type __ecp_nistz256_sqr_montq,\@abi-omnipotent
818 __ecp_nistz256_sqr_montq:
820 mulq $acc6 # a[1]*a[0]
825 mulq $acc5 # a[0]*a[2]
831 mulq $acc5 # a[0]*a[3]
837 #################################
838 mulq $acc6 # a[1]*a[2]
844 mulq $acc6 # a[1]*a[3]
852 #################################
853 mulq $acc7 # a[2]*a[3]
856 mov 8*0($a_ptr), %rax
860 add $acc1, $acc1 # acc1:6<<1
870 mov 8*1($a_ptr), %rax
876 mov 8*2($a_ptr), %rax
883 mov 8*3($a_ptr), %rax
893 mov .Lpoly+8*1(%rip), $a_ptr
894 mov .Lpoly+8*3(%rip), $t1
896 ##########################################
905 adc %rdx, $acc2 # doesn't overflow
917 ##########################################
925 adc %rdx, $acc3 # doesn't overflow
937 ##########################################
945 adc %rdx, $acc4 # doesn't overflow
957 ###########################################
965 adc %rdx, $acc0 # doesn't overflow
976 ############################################
977 # Add the rest of the acc
985 sub \$-1, $acc4 # .Lpoly[0]
987 sbb $a_ptr, $acc5 # .Lpoly[1]
988 sbb \$0, $acc6 # .Lpoly[2]
990 sbb $t1, $acc7 # .Lpoly[3]
995 mov $acc4, 8*0($r_ptr)
997 mov $acc5, 8*1($r_ptr)
999 mov $acc6, 8*2($r_ptr)
1000 mov $acc7, 8*3($r_ptr)
1003 .size __ecp_nistz256_sqr_montq,.-__ecp_nistz256_sqr_montq
1008 .type __ecp_nistz256_mul_montx,\@abi-omnipotent
1010 __ecp_nistz256_mul_montx:
1011 ########################################################################
1013 mulx $acc1, $acc0, $acc1
1014 mulx $acc2, $t0, $acc2
1016 xor $acc5, $acc5 # cf=0
1017 mulx $acc3, $t1, $acc3
1018 mov .Lpoly+8*3(%rip), $poly3
1020 mulx $acc4, $t0, $acc4
1023 shlx $poly1,$acc0,$t1
1025 shrx $poly1,$acc0,$t0
1028 ########################################################################
1029 # First reduction step
1030 xor $acc0, $acc0 # $acc0=0,cf=0,of=0
1034 mulx $poly3, $t0, $t1
1035 mov 8*1($b_ptr), %rdx
1040 adcx $acc0, $acc5 # cf=0
1041 adox $acc0, $acc5 # of=0
1043 ########################################################################
1045 mulx 8*0+128($a_ptr), $t0, $t1
1049 mulx 8*1+128($a_ptr), $t0, $t1
1053 mulx 8*2+128($a_ptr), $t0, $t1
1057 mulx 8*3+128($a_ptr), $t0, $t1
1060 shlx $poly1, $acc1, $t0
1062 shrx $poly1, $acc1, $t1
1068 ########################################################################
1069 # Second reduction step
1070 xor $acc1 ,$acc1 # $acc1=0,cf=0,of=0
1074 mulx $poly3, $t0, $t1
1075 mov 8*2($b_ptr), %rdx
1080 adcx $acc1, $acc0 # cf=0
1081 adox $acc1, $acc0 # of=0
1083 ########################################################################
1085 mulx 8*0+128($a_ptr), $t0, $t1
1089 mulx 8*1+128($a_ptr), $t0, $t1
1093 mulx 8*2+128($a_ptr), $t0, $t1
1097 mulx 8*3+128($a_ptr), $t0, $t1
1100 shlx $poly1, $acc2, $t0
1102 shrx $poly1, $acc2, $t1
1108 ########################################################################
1109 # Third reduction step
1110 xor $acc2, $acc2 # $acc2=0,cf=0,of=0
1114 mulx $poly3, $t0, $t1
1115 mov 8*3($b_ptr), %rdx
1120 adcx $acc2, $acc1 # cf=0
1121 adox $acc2, $acc1 # of=0
1123 ########################################################################
1125 mulx 8*0+128($a_ptr), $t0, $t1
1129 mulx 8*1+128($a_ptr), $t0, $t1
1133 mulx 8*2+128($a_ptr), $t0, $t1
1137 mulx 8*3+128($a_ptr), $t0, $t1
1140 shlx $poly1, $acc3, $t0
1142 shrx $poly1, $acc3, $t1
1148 ########################################################################
1149 # Fourth reduction step
1150 xor $acc3, $acc3 # $acc3=0,cf=0,of=0
1154 mulx $poly3, $t0, $t1
1156 mov .Lpoly+8*1(%rip), $poly1
1166 ########################################################################
1167 # Branch-less conditional subtraction of P
1169 sbb \$-1, $acc4 # .Lpoly[0]
1170 sbb $poly1, $acc5 # .Lpoly[1]
1171 sbb \$0, $acc0 # .Lpoly[2]
1173 sbb $poly3, $acc1 # .Lpoly[3]
1178 mov $acc4, 8*0($r_ptr)
1180 mov $acc5, 8*1($r_ptr)
1182 mov $acc0, 8*2($r_ptr)
1183 mov $acc1, 8*3($r_ptr)
1186 .size __ecp_nistz256_mul_montx,.-__ecp_nistz256_mul_montx
1188 .type __ecp_nistz256_sqr_montx,\@abi-omnipotent
1190 __ecp_nistz256_sqr_montx:
1191 mulx $acc6, $acc1, $acc2 # a[0]*a[1]
1192 mulx $acc7, $t0, $acc3 # a[0]*a[2]
1195 mulx $acc0, $t1, $acc4 # a[0]*a[3]
1199 xor $acc5, $acc5 # $acc5=0,cf=0,of=0
1201 #################################
1202 mulx $acc7, $t0, $t1 # a[1]*a[2]
1206 mulx $acc0, $t0, $t1 # a[1]*a[3]
1212 #################################
1213 mulx $acc0, $t0, $acc6 # a[2]*a[3]
1214 mov 8*0+128($a_ptr), %rdx
1215 xor $acc7, $acc7 # $acc7=0,cf=0,of=0
1216 adcx $acc1, $acc1 # acc1:6<<1
1219 adox $acc7, $acc6 # of=0
1221 mulx %rdx, $acc0, $t1
1222 mov 8*1+128($a_ptr), %rdx
1227 mov 8*2+128($a_ptr), %rdx
1233 mov 8*3+128($a_ptr), %rdx
1243 shlx $a_ptr, $acc0, $t0
1245 shrx $a_ptr, $acc0, $t4
1246 mov .Lpoly+8*3(%rip), $t1
1256 shlx $a_ptr, $acc1, $t0
1258 shrx $a_ptr, $acc1, $t4
1269 shlx $a_ptr, $acc2, $t0
1271 shrx $a_ptr, $acc2, $t4
1282 shlx $a_ptr, $acc3, $t0
1284 shrx $a_ptr, $acc3, $t4
1298 adc $acc0, $acc4 # accumulate upper half
1299 mov .Lpoly+8*1(%rip), $a_ptr
1307 xor %eax, %eax # cf=0
1308 sbb \$-1, $acc4 # .Lpoly[0]
1310 sbb $a_ptr, $acc5 # .Lpoly[1]
1311 sbb \$0, $acc6 # .Lpoly[2]
1313 sbb $t1, $acc7 # .Lpoly[3]
1318 mov $acc4, 8*0($r_ptr)
1320 mov $acc5, 8*1($r_ptr)
1322 mov $acc6, 8*2($r_ptr)
1323 mov $acc7, 8*3($r_ptr)
1326 .size __ecp_nistz256_sqr_montx,.-__ecp_nistz256_sqr_montx
1331 my ($r_ptr,$in_ptr)=("%rdi","%rsi");
1332 my ($acc0,$acc1,$acc2,$acc3,$acc4)=map("%r$_",(8..12));
1333 my ($t0,$t1)=("%rcx","%rsi");
1336 ################################################################################
1337 # void ecp_nistz256_from_mont(
1340 # This one performs Montgomery multiplication by 1, so we only need the reduction
1342 .globl ecp_nistz256_from_mont
1343 .type ecp_nistz256_from_mont,\@function,2
1345 ecp_nistz256_from_mont:
1349 mov 8*0($in_ptr), %rax
1350 mov 8*1($in_ptr), $acc1
1351 mov 8*2($in_ptr), $acc2
1352 mov 8*3($in_ptr), $acc3
1353 lea .Lpoly(%rip), $in_ptr
1357 #########################################
1377 #########################################
1397 ##########################################
1417 ###########################################
1435 mov 0*8($in_ptr), %rax
1436 mov 1*8($in_ptr), %rdx
1437 mov 2*8($in_ptr), $t0
1438 mov 3*8($in_ptr), $t1
1447 mov $acc4, 8*0($r_ptr)
1449 mov $acc0, 8*1($r_ptr)
1451 mov $acc1, 8*2($r_ptr)
1452 mov $acc2, 8*3($r_ptr)
1457 .size ecp_nistz256_from_mont,.-ecp_nistz256_from_mont
1461 my ($val,$in_t,$index)=$win64?("%rcx","%rdx","%r8d"):("%rdi","%rsi","%edx");
1462 my ($ONE,$INDEX,$Ra,$Rb,$Rc,$Rd,$Re,$Rf)=map("%xmm$_",(0..7));
1463 my ($M0,$T0a,$T0b,$T0c,$T0d,$T0e,$T0f,$TMP0)=map("%xmm$_",(8..15));
1464 my ($M1,$T2a,$T2b,$TMP2,$M2,$T2a,$T2b,$TMP2)=map("%xmm$_",(8..15));
1467 ################################################################################
1468 # void ecp_nistz256_select_w5(uint64_t *val, uint64_t *in_t, int index);
1469 .globl ecp_nistz256_select_w5
1470 .type ecp_nistz256_select_w5,\@abi-omnipotent
1472 ecp_nistz256_select_w5:
1474 $code.=<<___ if ($avx>1);
1475 mov OPENSSL_ia32cap_P+8(%rip), %eax
1477 jnz .Lavx2_select_w5
1479 $code.=<<___ if ($win64);
1480 lea -0x88(%rsp), %rax
1481 .LSEH_begin_ecp_nistz256_select_w5:
1482 .byte 0x48,0x8d,0x60,0xe0 #lea -0x20(%rax), %rsp
1483 .byte 0x0f,0x29,0x70,0xe0 #movaps %xmm6, -0x20(%rax)
1484 .byte 0x0f,0x29,0x78,0xf0 #movaps %xmm7, -0x10(%rax)
1485 .byte 0x44,0x0f,0x29,0x00 #movaps %xmm8, 0(%rax)
1486 .byte 0x44,0x0f,0x29,0x48,0x10 #movaps %xmm9, 0x10(%rax)
1487 .byte 0x44,0x0f,0x29,0x50,0x20 #movaps %xmm10, 0x20(%rax)
1488 .byte 0x44,0x0f,0x29,0x58,0x30 #movaps %xmm11, 0x30(%rax)
1489 .byte 0x44,0x0f,0x29,0x60,0x40 #movaps %xmm12, 0x40(%rax)
1490 .byte 0x44,0x0f,0x29,0x68,0x50 #movaps %xmm13, 0x50(%rax)
1491 .byte 0x44,0x0f,0x29,0x70,0x60 #movaps %xmm14, 0x60(%rax)
1492 .byte 0x44,0x0f,0x29,0x78,0x70 #movaps %xmm15, 0x70(%rax)
1495 movdqa .LOne(%rip), $ONE
1506 pshufd \$0, $INDEX, $INDEX
1509 .Lselect_loop_sse_w5:
1513 pcmpeqd $INDEX, $TMP0
1515 movdqa 16*0($in_t), $T0a
1516 movdqa 16*1($in_t), $T0b
1517 movdqa 16*2($in_t), $T0c
1518 movdqa 16*3($in_t), $T0d
1519 movdqa 16*4($in_t), $T0e
1520 movdqa 16*5($in_t), $T0f
1521 lea 16*6($in_t), $in_t
1537 jnz .Lselect_loop_sse_w5
1539 movdqu $Ra, 16*0($val)
1540 movdqu $Rb, 16*1($val)
1541 movdqu $Rc, 16*2($val)
1542 movdqu $Rd, 16*3($val)
1543 movdqu $Re, 16*4($val)
1544 movdqu $Rf, 16*5($val)
1546 $code.=<<___ if ($win64);
1547 movaps (%rsp), %xmm6
1548 movaps 0x10(%rsp), %xmm7
1549 movaps 0x20(%rsp), %xmm8
1550 movaps 0x30(%rsp), %xmm9
1551 movaps 0x40(%rsp), %xmm10
1552 movaps 0x50(%rsp), %xmm11
1553 movaps 0x60(%rsp), %xmm12
1554 movaps 0x70(%rsp), %xmm13
1555 movaps 0x80(%rsp), %xmm14
1556 movaps 0x90(%rsp), %xmm15
1557 lea 0xa8(%rsp), %rsp
1558 .LSEH_end_ecp_nistz256_select_w5:
1562 .size ecp_nistz256_select_w5,.-ecp_nistz256_select_w5
1564 ################################################################################
1565 # void ecp_nistz256_select_w7(uint64_t *val, uint64_t *in_t, int index);
1566 .globl ecp_nistz256_select_w7
1567 .type ecp_nistz256_select_w7,\@abi-omnipotent
1569 ecp_nistz256_select_w7:
1571 $code.=<<___ if ($avx>1);
1572 mov OPENSSL_ia32cap_P+8(%rip), %eax
1574 jnz .Lavx2_select_w7
1576 $code.=<<___ if ($win64);
1577 lea -0x88(%rsp), %rax
1578 .LSEH_begin_ecp_nistz256_select_w7:
1579 .byte 0x48,0x8d,0x60,0xe0 #lea -0x20(%rax), %rsp
1580 .byte 0x0f,0x29,0x70,0xe0 #movaps %xmm6, -0x20(%rax)
1581 .byte 0x0f,0x29,0x78,0xf0 #movaps %xmm7, -0x10(%rax)
1582 .byte 0x44,0x0f,0x29,0x00 #movaps %xmm8, 0(%rax)
1583 .byte 0x44,0x0f,0x29,0x48,0x10 #movaps %xmm9, 0x10(%rax)
1584 .byte 0x44,0x0f,0x29,0x50,0x20 #movaps %xmm10, 0x20(%rax)
1585 .byte 0x44,0x0f,0x29,0x58,0x30 #movaps %xmm11, 0x30(%rax)
1586 .byte 0x44,0x0f,0x29,0x60,0x40 #movaps %xmm12, 0x40(%rax)
1587 .byte 0x44,0x0f,0x29,0x68,0x50 #movaps %xmm13, 0x50(%rax)
1588 .byte 0x44,0x0f,0x29,0x70,0x60 #movaps %xmm14, 0x60(%rax)
1589 .byte 0x44,0x0f,0x29,0x78,0x70 #movaps %xmm15, 0x70(%rax)
1592 movdqa .LOne(%rip), $M0
1601 pshufd \$0, $INDEX, $INDEX
1604 .Lselect_loop_sse_w7:
1607 movdqa 16*0($in_t), $T0a
1608 movdqa 16*1($in_t), $T0b
1609 pcmpeqd $INDEX, $TMP0
1610 movdqa 16*2($in_t), $T0c
1611 movdqa 16*3($in_t), $T0d
1612 lea 16*4($in_t), $in_t
1621 prefetcht0 255($in_t)
1625 jnz .Lselect_loop_sse_w7
1627 movdqu $Ra, 16*0($val)
1628 movdqu $Rb, 16*1($val)
1629 movdqu $Rc, 16*2($val)
1630 movdqu $Rd, 16*3($val)
1632 $code.=<<___ if ($win64);
1633 movaps (%rsp), %xmm6
1634 movaps 0x10(%rsp), %xmm7
1635 movaps 0x20(%rsp), %xmm8
1636 movaps 0x30(%rsp), %xmm9
1637 movaps 0x40(%rsp), %xmm10
1638 movaps 0x50(%rsp), %xmm11
1639 movaps 0x60(%rsp), %xmm12
1640 movaps 0x70(%rsp), %xmm13
1641 movaps 0x80(%rsp), %xmm14
1642 movaps 0x90(%rsp), %xmm15
1643 lea 0xa8(%rsp), %rsp
1644 .LSEH_end_ecp_nistz256_select_w7:
1648 .size ecp_nistz256_select_w7,.-ecp_nistz256_select_w7
1652 my ($val,$in_t,$index)=$win64?("%rcx","%rdx","%r8d"):("%rdi","%rsi","%edx");
1653 my ($TWO,$INDEX,$Ra,$Rb,$Rc)=map("%ymm$_",(0..4));
1654 my ($M0,$T0a,$T0b,$T0c,$TMP0)=map("%ymm$_",(5..9));
1655 my ($M1,$T1a,$T1b,$T1c,$TMP1)=map("%ymm$_",(10..14));
1658 ################################################################################
1659 # void ecp_nistz256_avx2_select_w5(uint64_t *val, uint64_t *in_t, int index);
1660 .type ecp_nistz256_avx2_select_w5,\@abi-omnipotent
1662 ecp_nistz256_avx2_select_w5:
1666 $code.=<<___ if ($win64);
1667 lea -0x88(%rsp), %rax
1668 .LSEH_begin_ecp_nistz256_avx2_select_w5:
1669 .byte 0x48,0x8d,0x60,0xe0 #lea -0x20(%rax), %rsp
1670 .byte 0xc5,0xf8,0x29,0x70,0xe0 #vmovaps %xmm6, -0x20(%rax)
1671 .byte 0xc5,0xf8,0x29,0x78,0xf0 #vmovaps %xmm7, -0x10(%rax)
1672 .byte 0xc5,0x78,0x29,0x40,0x00 #vmovaps %xmm8, 8(%rax)
1673 .byte 0xc5,0x78,0x29,0x48,0x10 #vmovaps %xmm9, 0x10(%rax)
1674 .byte 0xc5,0x78,0x29,0x50,0x20 #vmovaps %xmm10, 0x20(%rax)
1675 .byte 0xc5,0x78,0x29,0x58,0x30 #vmovaps %xmm11, 0x30(%rax)
1676 .byte 0xc5,0x78,0x29,0x60,0x40 #vmovaps %xmm12, 0x40(%rax)
1677 .byte 0xc5,0x78,0x29,0x68,0x50 #vmovaps %xmm13, 0x50(%rax)
1678 .byte 0xc5,0x78,0x29,0x70,0x60 #vmovaps %xmm14, 0x60(%rax)
1679 .byte 0xc5,0x78,0x29,0x78,0x70 #vmovaps %xmm15, 0x70(%rax)
1682 vmovdqa .LTwo(%rip), $TWO
1688 vmovdqa .LOne(%rip), $M0
1689 vmovdqa .LTwo(%rip), $M1
1692 vpermd $INDEX, $Ra, $INDEX
1695 .Lselect_loop_avx2_w5:
1697 vmovdqa 32*0($in_t), $T0a
1698 vmovdqa 32*1($in_t), $T0b
1699 vmovdqa 32*2($in_t), $T0c
1701 vmovdqa 32*3($in_t), $T1a
1702 vmovdqa 32*4($in_t), $T1b
1703 vmovdqa 32*5($in_t), $T1c
1705 vpcmpeqd $INDEX, $M0, $TMP0
1706 vpcmpeqd $INDEX, $M1, $TMP1
1708 vpaddd $TWO, $M0, $M0
1709 vpaddd $TWO, $M1, $M1
1710 lea 32*6($in_t), $in_t
1712 vpand $TMP0, $T0a, $T0a
1713 vpand $TMP0, $T0b, $T0b
1714 vpand $TMP0, $T0c, $T0c
1715 vpand $TMP1, $T1a, $T1a
1716 vpand $TMP1, $T1b, $T1b
1717 vpand $TMP1, $T1c, $T1c
1719 vpxor $T0a, $Ra, $Ra
1720 vpxor $T0b, $Rb, $Rb
1721 vpxor $T0c, $Rc, $Rc
1722 vpxor $T1a, $Ra, $Ra
1723 vpxor $T1b, $Rb, $Rb
1724 vpxor $T1c, $Rc, $Rc
1727 jnz .Lselect_loop_avx2_w5
1729 vmovdqu $Ra, 32*0($val)
1730 vmovdqu $Rb, 32*1($val)
1731 vmovdqu $Rc, 32*2($val)
1734 $code.=<<___ if ($win64);
1735 movaps (%rsp), %xmm6
1736 movaps 0x10(%rsp), %xmm7
1737 movaps 0x20(%rsp), %xmm8
1738 movaps 0x30(%rsp), %xmm9
1739 movaps 0x40(%rsp), %xmm10
1740 movaps 0x50(%rsp), %xmm11
1741 movaps 0x60(%rsp), %xmm12
1742 movaps 0x70(%rsp), %xmm13
1743 movaps 0x80(%rsp), %xmm14
1744 movaps 0x90(%rsp), %xmm15
1745 lea 0xa8(%rsp), %rsp
1746 .LSEH_end_ecp_nistz256_avx2_select_w5:
1750 .size ecp_nistz256_avx2_select_w5,.-ecp_nistz256_avx2_select_w5
1754 my ($val,$in_t,$index)=$win64?("%rcx","%rdx","%r8d"):("%rdi","%rsi","%edx");
1755 my ($THREE,$INDEX,$Ra,$Rb)=map("%ymm$_",(0..3));
1756 my ($M0,$T0a,$T0b,$TMP0)=map("%ymm$_",(4..7));
1757 my ($M1,$T1a,$T1b,$TMP1)=map("%ymm$_",(8..11));
1758 my ($M2,$T2a,$T2b,$TMP2)=map("%ymm$_",(12..15));
1762 ################################################################################
1763 # void ecp_nistz256_avx2_select_w7(uint64_t *val, uint64_t *in_t, int index);
1764 .globl ecp_nistz256_avx2_select_w7
1765 .type ecp_nistz256_avx2_select_w7,\@abi-omnipotent
1767 ecp_nistz256_avx2_select_w7:
1771 $code.=<<___ if ($win64);
1772 lea -0x88(%rsp), %rax
1773 .LSEH_begin_ecp_nistz256_avx2_select_w7:
1774 .byte 0x48,0x8d,0x60,0xe0 #lea -0x20(%rax), %rsp
1775 .byte 0xc5,0xf8,0x29,0x70,0xe0 #vmovaps %xmm6, -0x20(%rax)
1776 .byte 0xc5,0xf8,0x29,0x78,0xf0 #vmovaps %xmm7, -0x10(%rax)
1777 .byte 0xc5,0x78,0x29,0x40,0x00 #vmovaps %xmm8, 8(%rax)
1778 .byte 0xc5,0x78,0x29,0x48,0x10 #vmovaps %xmm9, 0x10(%rax)
1779 .byte 0xc5,0x78,0x29,0x50,0x20 #vmovaps %xmm10, 0x20(%rax)
1780 .byte 0xc5,0x78,0x29,0x58,0x30 #vmovaps %xmm11, 0x30(%rax)
1781 .byte 0xc5,0x78,0x29,0x60,0x40 #vmovaps %xmm12, 0x40(%rax)
1782 .byte 0xc5,0x78,0x29,0x68,0x50 #vmovaps %xmm13, 0x50(%rax)
1783 .byte 0xc5,0x78,0x29,0x70,0x60 #vmovaps %xmm14, 0x60(%rax)
1784 .byte 0xc5,0x78,0x29,0x78,0x70 #vmovaps %xmm15, 0x70(%rax)
1787 vmovdqa .LThree(%rip), $THREE
1792 vmovdqa .LOne(%rip), $M0
1793 vmovdqa .LTwo(%rip), $M1
1794 vmovdqa .LThree(%rip), $M2
1797 vpermd $INDEX, $Ra, $INDEX
1798 # Skip index = 0, because it is implicitly the point at infinity
1801 .Lselect_loop_avx2_w7:
1803 vmovdqa 32*0($in_t), $T0a
1804 vmovdqa 32*1($in_t), $T0b
1806 vmovdqa 32*2($in_t), $T1a
1807 vmovdqa 32*3($in_t), $T1b
1809 vmovdqa 32*4($in_t), $T2a
1810 vmovdqa 32*5($in_t), $T2b
1812 vpcmpeqd $INDEX, $M0, $TMP0
1813 vpcmpeqd $INDEX, $M1, $TMP1
1814 vpcmpeqd $INDEX, $M2, $TMP2
1816 vpaddd $THREE, $M0, $M0
1817 vpaddd $THREE, $M1, $M1
1818 vpaddd $THREE, $M2, $M2
1819 lea 32*6($in_t), $in_t
1821 vpand $TMP0, $T0a, $T0a
1822 vpand $TMP0, $T0b, $T0b
1823 vpand $TMP1, $T1a, $T1a
1824 vpand $TMP1, $T1b, $T1b
1825 vpand $TMP2, $T2a, $T2a
1826 vpand $TMP2, $T2b, $T2b
1828 vpxor $T0a, $Ra, $Ra
1829 vpxor $T0b, $Rb, $Rb
1830 vpxor $T1a, $Ra, $Ra
1831 vpxor $T1b, $Rb, $Rb
1832 vpxor $T2a, $Ra, $Ra
1833 vpxor $T2b, $Rb, $Rb
1836 jnz .Lselect_loop_avx2_w7
1839 vmovdqa 32*0($in_t), $T0a
1840 vmovdqa 32*1($in_t), $T0b
1842 vpcmpeqd $INDEX, $M0, $TMP0
1844 vpand $TMP0, $T0a, $T0a
1845 vpand $TMP0, $T0b, $T0b
1847 vpxor $T0a, $Ra, $Ra
1848 vpxor $T0b, $Rb, $Rb
1850 vmovdqu $Ra, 32*0($val)
1851 vmovdqu $Rb, 32*1($val)
1854 $code.=<<___ if ($win64);
1855 movaps (%rsp), %xmm6
1856 movaps 0x10(%rsp), %xmm7
1857 movaps 0x20(%rsp), %xmm8
1858 movaps 0x30(%rsp), %xmm9
1859 movaps 0x40(%rsp), %xmm10
1860 movaps 0x50(%rsp), %xmm11
1861 movaps 0x60(%rsp), %xmm12
1862 movaps 0x70(%rsp), %xmm13
1863 movaps 0x80(%rsp), %xmm14
1864 movaps 0x90(%rsp), %xmm15
1865 lea 0xa8(%rsp), %rsp
1866 .LSEH_end_ecp_nistz256_avx2_select_w7:
1870 .size ecp_nistz256_avx2_select_w7,.-ecp_nistz256_avx2_select_w7
1874 .globl ecp_nistz256_avx2_select_w7
1875 .type ecp_nistz256_avx2_select_w7,\@function,3
1877 ecp_nistz256_avx2_select_w7:
1878 .byte 0x0f,0x0b # ud2
1880 .size ecp_nistz256_avx2_select_w7,.-ecp_nistz256_avx2_select_w7
1884 ########################################################################
1885 # This block implements higher level point_double, point_add and
1886 # point_add_affine. The key to performance in this case is to allow
1887 # out-of-order execution logic to overlap computations from next step
1888 # with tail processing from current step. By using tailored calling
1889 # sequence we minimize inter-step overhead to give processor better
1890 # shot at overlapping operations...
1892 # You will notice that input data is copied to stack. Trouble is that
1893 # there are no registers to spare for holding original pointers and
1894 # reloading them, pointers, would create undesired dependencies on
1895 # effective addresses calculation paths. In other words it's too done
1896 # to favour out-of-order execution logic.
1897 # <appro@openssl.org>
1899 my ($r_ptr,$a_ptr,$b_org,$b_ptr)=("%rdi","%rsi","%rdx","%rbx");
1900 my ($acc0,$acc1,$acc2,$acc3,$acc4,$acc5,$acc6,$acc7)=map("%r$_",(8..15));
1901 my ($t0,$t1,$t2,$t3,$t4)=("%rax","%rbp","%rcx",$acc4,$acc4);
1902 my ($poly1,$poly3)=($acc6,$acc7);
1904 sub load_for_mul () {
1905 my ($a,$b,$src0) = @_;
1906 my $bias = $src0 eq "%rax" ? 0 : -128;
1912 lea $bias+$a, $a_ptr
1917 sub load_for_sqr () {
1919 my $bias = $src0 eq "%rax" ? 0 : -128;
1923 lea $bias+$a, $a_ptr
1929 ########################################################################
1930 # operate in 4-5-0-1 "name space" that matches multiplication output
1932 my ($a0,$a1,$a2,$a3,$t3,$t4)=($acc4,$acc5,$acc0,$acc1,$acc2,$acc3);
1935 .type __ecp_nistz256_add_toq,\@abi-omnipotent
1937 __ecp_nistz256_add_toq:
1938 add 8*0($b_ptr), $a0
1939 adc 8*1($b_ptr), $a1
1941 adc 8*2($b_ptr), $a2
1942 adc 8*3($b_ptr), $a3
1956 mov $a0, 8*0($r_ptr)
1958 mov $a1, 8*1($r_ptr)
1960 mov $a2, 8*2($r_ptr)
1961 mov $a3, 8*3($r_ptr)
1964 .size __ecp_nistz256_add_toq,.-__ecp_nistz256_add_toq
1966 .type __ecp_nistz256_sub_fromq,\@abi-omnipotent
1968 __ecp_nistz256_sub_fromq:
1969 sub 8*0($b_ptr), $a0
1970 sbb 8*1($b_ptr), $a1
1972 sbb 8*2($b_ptr), $a2
1973 sbb 8*3($b_ptr), $a3
1987 mov $a0, 8*0($r_ptr)
1989 mov $a1, 8*1($r_ptr)
1991 mov $a2, 8*2($r_ptr)
1992 mov $a3, 8*3($r_ptr)
1995 .size __ecp_nistz256_sub_fromq,.-__ecp_nistz256_sub_fromq
1997 .type __ecp_nistz256_subq,\@abi-omnipotent
1999 __ecp_nistz256_subq:
2022 .size __ecp_nistz256_subq,.-__ecp_nistz256_subq
2024 .type __ecp_nistz256_mul_by_2q,\@abi-omnipotent
2026 __ecp_nistz256_mul_by_2q:
2027 add $a0, $a0 # a0:a3+a0:a3
2045 mov $a0, 8*0($r_ptr)
2047 mov $a1, 8*1($r_ptr)
2049 mov $a2, 8*2($r_ptr)
2050 mov $a3, 8*3($r_ptr)
2053 .size __ecp_nistz256_mul_by_2q,.-__ecp_nistz256_mul_by_2q
2058 my ($src0,$sfx,$bias);
2059 my ($S,$M,$Zsqr,$in_x,$tmp0)=map(32*$_,(0..4));
2067 .globl ecp_nistz256_point_double
2068 .type ecp_nistz256_point_double,\@function,2
2070 ecp_nistz256_point_double:
2072 $code.=<<___ if ($addx);
2074 and OPENSSL_ia32cap_P+8(%rip), %ecx
2084 .type ecp_nistz256_point_doublex,\@function,2
2086 ecp_nistz256_point_doublex:
2099 movdqu 0x00($a_ptr), %xmm0 # copy *(P256_POINT *)$a_ptr.x
2100 mov $a_ptr, $b_ptr # backup copy
2101 movdqu 0x10($a_ptr), %xmm1
2102 mov 0x20+8*0($a_ptr), $acc4 # load in_y in "5-4-0-1" order
2103 mov 0x20+8*1($a_ptr), $acc5
2104 mov 0x20+8*2($a_ptr), $acc0
2105 mov 0x20+8*3($a_ptr), $acc1
2106 mov .Lpoly+8*1(%rip), $poly1
2107 mov .Lpoly+8*3(%rip), $poly3
2108 movdqa %xmm0, $in_x(%rsp)
2109 movdqa %xmm1, $in_x+0x10(%rsp)
2110 lea 0x20($r_ptr), $acc2
2111 lea 0x40($r_ptr), $acc3
2116 lea $S(%rsp), $r_ptr
2117 call __ecp_nistz256_mul_by_2$x # p256_mul_by_2(S, in_y);
2119 mov 0x40+8*0($a_ptr), $src0
2120 mov 0x40+8*1($a_ptr), $acc6
2121 mov 0x40+8*2($a_ptr), $acc7
2122 mov 0x40+8*3($a_ptr), $acc0
2123 lea 0x40-$bias($a_ptr), $a_ptr
2124 lea $Zsqr(%rsp), $r_ptr
2125 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Zsqr, in_z);
2127 `&load_for_sqr("$S(%rsp)", "$src0")`
2128 lea $S(%rsp), $r_ptr
2129 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(S, S);
2131 mov 0x20($b_ptr), $src0 # $b_ptr is still valid
2132 mov 0x40+8*0($b_ptr), $acc1
2133 mov 0x40+8*1($b_ptr), $acc2
2134 mov 0x40+8*2($b_ptr), $acc3
2135 mov 0x40+8*3($b_ptr), $acc4
2136 lea 0x40-$bias($b_ptr), $a_ptr
2137 lea 0x20($b_ptr), $b_ptr
2139 call __ecp_nistz256_mul_mont$x # p256_mul_mont(res_z, in_z, in_y);
2140 call __ecp_nistz256_mul_by_2$x # p256_mul_by_2(res_z, res_z);
2142 mov $in_x+8*0(%rsp), $acc4 # "5-4-0-1" order
2143 mov $in_x+8*1(%rsp), $acc5
2144 lea $Zsqr(%rsp), $b_ptr
2145 mov $in_x+8*2(%rsp), $acc0
2146 mov $in_x+8*3(%rsp), $acc1
2147 lea $M(%rsp), $r_ptr
2148 call __ecp_nistz256_add_to$x # p256_add(M, in_x, Zsqr);
2150 mov $in_x+8*0(%rsp), $acc4 # "5-4-0-1" order
2151 mov $in_x+8*1(%rsp), $acc5
2152 lea $Zsqr(%rsp), $b_ptr
2153 mov $in_x+8*2(%rsp), $acc0
2154 mov $in_x+8*3(%rsp), $acc1
2155 lea $Zsqr(%rsp), $r_ptr
2156 call __ecp_nistz256_sub_from$x # p256_sub(Zsqr, in_x, Zsqr);
2158 `&load_for_sqr("$S(%rsp)", "$src0")`
2160 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(res_y, S);
2163 ######## ecp_nistz256_div_by_2(res_y, res_y); ##########################
2164 # operate in 4-5-6-7 "name space" that matches squaring output
2166 my ($poly1,$poly3)=($a_ptr,$t1);
2167 my ($a0,$a1,$a2,$a3,$t3,$t4,$t1)=($acc4,$acc5,$acc6,$acc7,$acc0,$acc1,$acc2);
2180 xor $a_ptr, $a_ptr # borrow $a_ptr
2189 mov $a1, $t0 # a0:a3>>1
2200 mov $a0, 8*0($r_ptr)
2202 mov $a1, 8*1($r_ptr)
2206 mov $a2, 8*2($r_ptr)
2207 mov $a3, 8*3($r_ptr)
2211 `&load_for_mul("$M(%rsp)", "$Zsqr(%rsp)", "$src0")`
2212 lea $M(%rsp), $r_ptr
2213 call __ecp_nistz256_mul_mont$x # p256_mul_mont(M, M, Zsqr);
2215 lea $tmp0(%rsp), $r_ptr
2216 call __ecp_nistz256_mul_by_2$x
2218 lea $M(%rsp), $b_ptr
2219 lea $M(%rsp), $r_ptr
2220 call __ecp_nistz256_add_to$x # p256_mul_by_3(M, M);
2222 `&load_for_mul("$S(%rsp)", "$in_x(%rsp)", "$src0")`
2223 lea $S(%rsp), $r_ptr
2224 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S, S, in_x);
2226 lea $tmp0(%rsp), $r_ptr
2227 call __ecp_nistz256_mul_by_2$x # p256_mul_by_2(tmp0, S);
2229 `&load_for_sqr("$M(%rsp)", "$src0")`
2231 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(res_x, M);
2233 lea $tmp0(%rsp), $b_ptr
2234 mov $acc6, $acc0 # harmonize sqr output and sub input
2238 call __ecp_nistz256_sub_from$x # p256_sub(res_x, res_x, tmp0);
2240 mov $S+8*0(%rsp), $t0
2241 mov $S+8*1(%rsp), $t1
2242 mov $S+8*2(%rsp), $t2
2243 mov $S+8*3(%rsp), $acc2 # "4-5-0-1" order
2244 lea $S(%rsp), $r_ptr
2245 call __ecp_nistz256_sub$x # p256_sub(S, S, res_x);
2248 lea $M(%rsp), $b_ptr
2249 mov $acc4, $acc6 # harmonize sub output and mul input
2251 mov $acc4, $S+8*0(%rsp) # have to save:-(
2253 mov $acc5, $S+8*1(%rsp)
2255 mov $acc0, $S+8*2(%rsp)
2256 lea $S-$bias(%rsp), $a_ptr
2258 mov $acc1, $S+8*3(%rsp)
2260 lea $S(%rsp), $r_ptr
2261 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S, S, M);
2265 call __ecp_nistz256_sub_from$x # p256_sub(res_y, S, res_y);
2275 .size ecp_nistz256_point_double$sfx,.-ecp_nistz256_point_double$sfx
2282 my ($src0,$sfx,$bias);
2283 my ($H,$Hsqr,$R,$Rsqr,$Hcub,
2285 $res_x,$res_y,$res_z,
2286 $in1_x,$in1_y,$in1_z,
2287 $in2_x,$in2_y,$in2_z)=map(32*$_,(0..17));
2288 my ($Z1sqr, $Z2sqr) = ($Hsqr, $Rsqr);
2296 .globl ecp_nistz256_point_add
2297 .type ecp_nistz256_point_add,\@function,3
2299 ecp_nistz256_point_add:
2301 $code.=<<___ if ($addx);
2303 and OPENSSL_ia32cap_P+8(%rip), %ecx
2313 .type ecp_nistz256_point_addx,\@function,3
2315 ecp_nistz256_point_addx:
2328 movdqu 0x00($a_ptr), %xmm0 # copy *(P256_POINT *)$a_ptr
2329 movdqu 0x10($a_ptr), %xmm1
2330 movdqu 0x20($a_ptr), %xmm2
2331 movdqu 0x30($a_ptr), %xmm3
2332 movdqu 0x40($a_ptr), %xmm4
2333 movdqu 0x50($a_ptr), %xmm5
2334 mov $a_ptr, $b_ptr # reassign
2335 mov $b_org, $a_ptr # reassign
2336 movdqa %xmm0, $in1_x(%rsp)
2337 movdqa %xmm1, $in1_x+0x10(%rsp)
2339 movdqa %xmm2, $in1_y(%rsp)
2340 movdqa %xmm3, $in1_y+0x10(%rsp)
2342 movdqa %xmm4, $in1_z(%rsp)
2343 movdqa %xmm5, $in1_z+0x10(%rsp)
2346 movdqu 0x00($a_ptr), %xmm0 # copy *(P256_POINT *)$b_ptr
2347 pshufd \$0xb1, %xmm3, %xmm5
2348 movdqu 0x10($a_ptr), %xmm1
2349 movdqu 0x20($a_ptr), %xmm2
2351 movdqu 0x30($a_ptr), %xmm3
2352 mov 0x40+8*0($a_ptr), $src0 # load original in2_z
2353 mov 0x40+8*1($a_ptr), $acc6
2354 mov 0x40+8*2($a_ptr), $acc7
2355 mov 0x40+8*3($a_ptr), $acc0
2356 movdqa %xmm0, $in2_x(%rsp)
2357 pshufd \$0x1e, %xmm5, %xmm4
2358 movdqa %xmm1, $in2_x+0x10(%rsp)
2360 movq $r_ptr, %xmm0 # save $r_ptr
2361 movdqa %xmm2, $in2_y(%rsp)
2362 movdqa %xmm3, $in2_y+0x10(%rsp)
2368 lea 0x40-$bias($a_ptr), $a_ptr # $a_ptr is still valid
2369 mov $src0, $in2_z+8*0(%rsp) # make in2_z copy
2370 mov $acc6, $in2_z+8*1(%rsp)
2371 mov $acc7, $in2_z+8*2(%rsp)
2372 mov $acc0, $in2_z+8*3(%rsp)
2373 lea $Z2sqr(%rsp), $r_ptr # Z2^2
2374 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Z2sqr, in2_z);
2376 pcmpeqd %xmm4, %xmm5
2377 pshufd \$0xb1, %xmm3, %xmm4
2379 pshufd \$0, %xmm5, %xmm5 # in1infty
2380 pshufd \$0x1e, %xmm4, %xmm3
2383 pcmpeqd %xmm3, %xmm4
2384 pshufd \$0, %xmm4, %xmm4 # in2infty
2385 mov 0x40+8*0($b_ptr), $src0 # load original in1_z
2386 mov 0x40+8*1($b_ptr), $acc6
2387 mov 0x40+8*2($b_ptr), $acc7
2388 mov 0x40+8*3($b_ptr), $acc0
2390 lea 0x40-$bias($b_ptr), $a_ptr
2391 lea $Z1sqr(%rsp), $r_ptr # Z1^2
2392 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Z1sqr, in1_z);
2394 `&load_for_mul("$Z2sqr(%rsp)", "$in2_z(%rsp)", "$src0")`
2395 lea $S1(%rsp), $r_ptr # S1 = Z2^3
2396 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S1, Z2sqr, in2_z);
2398 `&load_for_mul("$Z1sqr(%rsp)", "$in1_z(%rsp)", "$src0")`
2399 lea $S2(%rsp), $r_ptr # S2 = Z1^3
2400 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S2, Z1sqr, in1_z);
2402 `&load_for_mul("$S1(%rsp)", "$in1_y(%rsp)", "$src0")`
2403 lea $S1(%rsp), $r_ptr # S1 = Y1*Z2^3
2404 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S1, S1, in1_y);
2406 `&load_for_mul("$S2(%rsp)", "$in2_y(%rsp)", "$src0")`
2407 lea $S2(%rsp), $r_ptr # S2 = Y2*Z1^3
2408 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S2, S2, in2_y);
2410 lea $S1(%rsp), $b_ptr
2411 lea $R(%rsp), $r_ptr # R = S2 - S1
2412 call __ecp_nistz256_sub_from$x # p256_sub(R, S2, S1);
2414 or $acc5, $acc4 # see if result is zero
2418 por %xmm5, %xmm2 # in1infty || in2infty
2421 `&load_for_mul("$Z2sqr(%rsp)", "$in1_x(%rsp)", "$src0")`
2422 lea $U1(%rsp), $r_ptr # U1 = X1*Z2^2
2423 call __ecp_nistz256_mul_mont$x # p256_mul_mont(U1, in1_x, Z2sqr);
2425 `&load_for_mul("$Z1sqr(%rsp)", "$in2_x(%rsp)", "$src0")`
2426 lea $U2(%rsp), $r_ptr # U2 = X2*Z1^2
2427 call __ecp_nistz256_mul_mont$x # p256_mul_mont(U2, in2_x, Z1sqr);
2429 lea $U1(%rsp), $b_ptr
2430 lea $H(%rsp), $r_ptr # H = U2 - U1
2431 call __ecp_nistz256_sub_from$x # p256_sub(H, U2, U1);
2433 or $acc5, $acc4 # see if result is zero
2437 .byte 0x3e # predict taken
2438 jnz .Ladd_proceed$x # is_equal(U1,U2)?
2442 jnz .Ladd_proceed$x # (in1infty || in2infty)?
2444 jz .Ladd_proceed$x # is_equal(S1,S2)?
2446 movq %xmm0, $r_ptr # restore $r_ptr
2448 movdqu %xmm0, 0x00($r_ptr)
2449 movdqu %xmm0, 0x10($r_ptr)
2450 movdqu %xmm0, 0x20($r_ptr)
2451 movdqu %xmm0, 0x30($r_ptr)
2452 movdqu %xmm0, 0x40($r_ptr)
2453 movdqu %xmm0, 0x50($r_ptr)
2458 `&load_for_sqr("$R(%rsp)", "$src0")`
2459 lea $Rsqr(%rsp), $r_ptr # R^2
2460 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Rsqr, R);
2462 `&load_for_mul("$H(%rsp)", "$in1_z(%rsp)", "$src0")`
2463 lea $res_z(%rsp), $r_ptr # Z3 = H*Z1*Z2
2464 call __ecp_nistz256_mul_mont$x # p256_mul_mont(res_z, H, in1_z);
2466 `&load_for_sqr("$H(%rsp)", "$src0")`
2467 lea $Hsqr(%rsp), $r_ptr # H^2
2468 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Hsqr, H);
2470 `&load_for_mul("$res_z(%rsp)", "$in2_z(%rsp)", "$src0")`
2471 lea $res_z(%rsp), $r_ptr # Z3 = H*Z1*Z2
2472 call __ecp_nistz256_mul_mont$x # p256_mul_mont(res_z, res_z, in2_z);
2474 `&load_for_mul("$Hsqr(%rsp)", "$H(%rsp)", "$src0")`
2475 lea $Hcub(%rsp), $r_ptr # H^3
2476 call __ecp_nistz256_mul_mont$x # p256_mul_mont(Hcub, Hsqr, H);
2478 `&load_for_mul("$Hsqr(%rsp)", "$U1(%rsp)", "$src0")`
2479 lea $U2(%rsp), $r_ptr # U1*H^2
2480 call __ecp_nistz256_mul_mont$x # p256_mul_mont(U2, U1, Hsqr);
2483 #######################################################################
2484 # operate in 4-5-0-1 "name space" that matches multiplication output
2486 my ($acc0,$acc1,$acc2,$acc3,$t3,$t4)=($acc4,$acc5,$acc0,$acc1,$acc2,$acc3);
2487 my ($poly1, $poly3)=($acc6,$acc7);
2490 #lea $U2(%rsp), $a_ptr
2491 #lea $Hsqr(%rsp), $r_ptr # 2*U1*H^2
2492 #call __ecp_nistz256_mul_by_2 # ecp_nistz256_mul_by_2(Hsqr, U2);
2494 add $acc0, $acc0 # a0:a3+a0:a3
2495 lea $Rsqr(%rsp), $a_ptr
2512 mov 8*0($a_ptr), $t0
2514 mov 8*1($a_ptr), $t1
2516 mov 8*2($a_ptr), $t2
2518 mov 8*3($a_ptr), $t3
2520 call __ecp_nistz256_sub$x # p256_sub(res_x, Rsqr, Hsqr);
2522 lea $Hcub(%rsp), $b_ptr
2523 lea $res_x(%rsp), $r_ptr
2524 call __ecp_nistz256_sub_from$x # p256_sub(res_x, res_x, Hcub);
2526 mov $U2+8*0(%rsp), $t0
2527 mov $U2+8*1(%rsp), $t1
2528 mov $U2+8*2(%rsp), $t2
2529 mov $U2+8*3(%rsp), $t3
2530 lea $res_y(%rsp), $r_ptr
2532 call __ecp_nistz256_sub$x # p256_sub(res_y, U2, res_x);
2534 mov $acc0, 8*0($r_ptr) # save the result, as
2535 mov $acc1, 8*1($r_ptr) # __ecp_nistz256_sub doesn't
2536 mov $acc2, 8*2($r_ptr)
2537 mov $acc3, 8*3($r_ptr)
2541 `&load_for_mul("$S1(%rsp)", "$Hcub(%rsp)", "$src0")`
2542 lea $S2(%rsp), $r_ptr
2543 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S2, S1, Hcub);
2545 `&load_for_mul("$R(%rsp)", "$res_y(%rsp)", "$src0")`
2546 lea $res_y(%rsp), $r_ptr
2547 call __ecp_nistz256_mul_mont$x # p256_mul_mont(res_y, R, res_y);
2549 lea $S2(%rsp), $b_ptr
2550 lea $res_y(%rsp), $r_ptr
2551 call __ecp_nistz256_sub_from$x # p256_sub(res_y, res_y, S2);
2553 movq %xmm0, $r_ptr # restore $r_ptr
2555 movdqa %xmm5, %xmm0 # copy_conditional(res_z, in2_z, in1infty);
2557 pandn $res_z(%rsp), %xmm0
2559 pandn $res_z+0x10(%rsp), %xmm1
2561 pand $in2_z(%rsp), %xmm2
2562 pand $in2_z+0x10(%rsp), %xmm3
2566 movdqa %xmm4, %xmm0 # copy_conditional(res_z, in1_z, in2infty);
2572 pand $in1_z(%rsp), %xmm2
2573 pand $in1_z+0x10(%rsp), %xmm3
2576 movdqu %xmm2, 0x40($r_ptr)
2577 movdqu %xmm3, 0x50($r_ptr)
2579 movdqa %xmm5, %xmm0 # copy_conditional(res_x, in2_x, in1infty);
2581 pandn $res_x(%rsp), %xmm0
2583 pandn $res_x+0x10(%rsp), %xmm1
2585 pand $in2_x(%rsp), %xmm2
2586 pand $in2_x+0x10(%rsp), %xmm3
2590 movdqa %xmm4, %xmm0 # copy_conditional(res_x, in1_x, in2infty);
2596 pand $in1_x(%rsp), %xmm2
2597 pand $in1_x+0x10(%rsp), %xmm3
2600 movdqu %xmm2, 0x00($r_ptr)
2601 movdqu %xmm3, 0x10($r_ptr)
2603 movdqa %xmm5, %xmm0 # copy_conditional(res_y, in2_y, in1infty);
2605 pandn $res_y(%rsp), %xmm0
2607 pandn $res_y+0x10(%rsp), %xmm1
2609 pand $in2_y(%rsp), %xmm2
2610 pand $in2_y+0x10(%rsp), %xmm3
2614 movdqa %xmm4, %xmm0 # copy_conditional(res_y, in1_y, in2infty);
2620 pand $in1_y(%rsp), %xmm2
2621 pand $in1_y+0x10(%rsp), %xmm3
2624 movdqu %xmm2, 0x20($r_ptr)
2625 movdqu %xmm3, 0x30($r_ptr)
2636 .size ecp_nistz256_point_add$sfx,.-ecp_nistz256_point_add$sfx
2641 sub gen_add_affine () {
2643 my ($src0,$sfx,$bias);
2644 my ($U2,$S2,$H,$R,$Hsqr,$Hcub,$Rsqr,
2645 $res_x,$res_y,$res_z,
2646 $in1_x,$in1_y,$in1_z,
2647 $in2_x,$in2_y)=map(32*$_,(0..14));
2656 .globl ecp_nistz256_point_add_affine
2657 .type ecp_nistz256_point_add_affine,\@function,3
2659 ecp_nistz256_point_add_affine:
2661 $code.=<<___ if ($addx);
2663 and OPENSSL_ia32cap_P+8(%rip), %ecx
2665 je .Lpoint_add_affinex
2673 .type ecp_nistz256_point_add_affinex,\@function,3
2675 ecp_nistz256_point_add_affinex:
2676 .Lpoint_add_affinex:
2688 movdqu 0x00($a_ptr), %xmm0 # copy *(P256_POINT *)$a_ptr
2689 mov $b_org, $b_ptr # reassign
2690 movdqu 0x10($a_ptr), %xmm1
2691 movdqu 0x20($a_ptr), %xmm2
2692 movdqu 0x30($a_ptr), %xmm3
2693 movdqu 0x40($a_ptr), %xmm4
2694 movdqu 0x50($a_ptr), %xmm5
2695 mov 0x40+8*0($a_ptr), $src0 # load original in1_z
2696 mov 0x40+8*1($a_ptr), $acc6
2697 mov 0x40+8*2($a_ptr), $acc7
2698 mov 0x40+8*3($a_ptr), $acc0
2699 movdqa %xmm0, $in1_x(%rsp)
2700 movdqa %xmm1, $in1_x+0x10(%rsp)
2702 movdqa %xmm2, $in1_y(%rsp)
2703 movdqa %xmm3, $in1_y+0x10(%rsp)
2705 movdqa %xmm4, $in1_z(%rsp)
2706 movdqa %xmm5, $in1_z+0x10(%rsp)
2709 movdqu 0x00($b_ptr), %xmm0 # copy *(P256_POINT_AFFINE *)$b_ptr
2710 pshufd \$0xb1, %xmm3, %xmm5
2711 movdqu 0x10($b_ptr), %xmm1
2712 movdqu 0x20($b_ptr), %xmm2
2714 movdqu 0x30($b_ptr), %xmm3
2715 movdqa %xmm0, $in2_x(%rsp)
2716 pshufd \$0x1e, %xmm5, %xmm4
2717 movdqa %xmm1, $in2_x+0x10(%rsp)
2719 movq $r_ptr, %xmm0 # save $r_ptr
2720 movdqa %xmm2, $in2_y(%rsp)
2721 movdqa %xmm3, $in2_y+0x10(%rsp)
2727 lea 0x40-$bias($a_ptr), $a_ptr # $a_ptr is still valid
2728 lea $Z1sqr(%rsp), $r_ptr # Z1^2
2729 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Z1sqr, in1_z);
2731 pcmpeqd %xmm4, %xmm5
2732 pshufd \$0xb1, %xmm3, %xmm4
2733 mov 0x00($b_ptr), $src0 # $b_ptr is still valid
2734 #lea 0x00($b_ptr), $b_ptr
2735 mov $acc4, $acc1 # harmonize sqr output and mul input
2737 pshufd \$0, %xmm5, %xmm5 # in1infty
2738 pshufd \$0x1e, %xmm4, %xmm3
2743 pcmpeqd %xmm3, %xmm4
2744 pshufd \$0, %xmm4, %xmm4 # in2infty
2746 lea $Z1sqr-$bias(%rsp), $a_ptr
2748 lea $U2(%rsp), $r_ptr # U2 = X2*Z1^2
2749 call __ecp_nistz256_mul_mont$x # p256_mul_mont(U2, Z1sqr, in2_x);
2751 lea $in1_x(%rsp), $b_ptr
2752 lea $H(%rsp), $r_ptr # H = U2 - U1
2753 call __ecp_nistz256_sub_from$x # p256_sub(H, U2, in1_x);
2755 `&load_for_mul("$Z1sqr(%rsp)", "$in1_z(%rsp)", "$src0")`
2756 lea $S2(%rsp), $r_ptr # S2 = Z1^3
2757 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S2, Z1sqr, in1_z);
2759 `&load_for_mul("$H(%rsp)", "$in1_z(%rsp)", "$src0")`
2760 lea $res_z(%rsp), $r_ptr # Z3 = H*Z1*Z2
2761 call __ecp_nistz256_mul_mont$x # p256_mul_mont(res_z, H, in1_z);
2763 `&load_for_mul("$S2(%rsp)", "$in2_y(%rsp)", "$src0")`
2764 lea $S2(%rsp), $r_ptr # S2 = Y2*Z1^3
2765 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S2, S2, in2_y);
2767 lea $in1_y(%rsp), $b_ptr
2768 lea $R(%rsp), $r_ptr # R = S2 - S1
2769 call __ecp_nistz256_sub_from$x # p256_sub(R, S2, in1_y);
2771 `&load_for_sqr("$H(%rsp)", "$src0")`
2772 lea $Hsqr(%rsp), $r_ptr # H^2
2773 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Hsqr, H);
2775 `&load_for_sqr("$R(%rsp)", "$src0")`
2776 lea $Rsqr(%rsp), $r_ptr # R^2
2777 call __ecp_nistz256_sqr_mont$x # p256_sqr_mont(Rsqr, R);
2779 `&load_for_mul("$H(%rsp)", "$Hsqr(%rsp)", "$src0")`
2780 lea $Hcub(%rsp), $r_ptr # H^3
2781 call __ecp_nistz256_mul_mont$x # p256_mul_mont(Hcub, Hsqr, H);
2783 `&load_for_mul("$Hsqr(%rsp)", "$in1_x(%rsp)", "$src0")`
2784 lea $U2(%rsp), $r_ptr # U1*H^2
2785 call __ecp_nistz256_mul_mont$x # p256_mul_mont(U2, in1_x, Hsqr);
2788 #######################################################################
2789 # operate in 4-5-0-1 "name space" that matches multiplication output
2791 my ($acc0,$acc1,$acc2,$acc3,$t3,$t4)=($acc4,$acc5,$acc0,$acc1,$acc2,$acc3);
2792 my ($poly1, $poly3)=($acc6,$acc7);
2795 #lea $U2(%rsp), $a_ptr
2796 #lea $Hsqr(%rsp), $r_ptr # 2*U1*H^2
2797 #call __ecp_nistz256_mul_by_2 # ecp_nistz256_mul_by_2(Hsqr, U2);
2799 add $acc0, $acc0 # a0:a3+a0:a3
2800 lea $Rsqr(%rsp), $a_ptr
2817 mov 8*0($a_ptr), $t0
2819 mov 8*1($a_ptr), $t1
2821 mov 8*2($a_ptr), $t2
2823 mov 8*3($a_ptr), $t3
2825 call __ecp_nistz256_sub$x # p256_sub(res_x, Rsqr, Hsqr);
2827 lea $Hcub(%rsp), $b_ptr
2828 lea $res_x(%rsp), $r_ptr
2829 call __ecp_nistz256_sub_from$x # p256_sub(res_x, res_x, Hcub);
2831 mov $U2+8*0(%rsp), $t0
2832 mov $U2+8*1(%rsp), $t1
2833 mov $U2+8*2(%rsp), $t2
2834 mov $U2+8*3(%rsp), $t3
2835 lea $H(%rsp), $r_ptr
2837 call __ecp_nistz256_sub$x # p256_sub(H, U2, res_x);
2839 mov $acc0, 8*0($r_ptr) # save the result, as
2840 mov $acc1, 8*1($r_ptr) # __ecp_nistz256_sub doesn't
2841 mov $acc2, 8*2($r_ptr)
2842 mov $acc3, 8*3($r_ptr)
2846 `&load_for_mul("$Hcub(%rsp)", "$in1_y(%rsp)", "$src0")`
2847 lea $S2(%rsp), $r_ptr
2848 call __ecp_nistz256_mul_mont$x # p256_mul_mont(S2, Hcub, in1_y);
2850 `&load_for_mul("$H(%rsp)", "$R(%rsp)", "$src0")`
2851 lea $H(%rsp), $r_ptr
2852 call __ecp_nistz256_mul_mont$x # p256_mul_mont(H, H, R);
2854 lea $S2(%rsp), $b_ptr
2855 lea $res_y(%rsp), $r_ptr
2856 call __ecp_nistz256_sub_from$x # p256_sub(res_y, H, S2);
2858 movq %xmm0, $r_ptr # restore $r_ptr
2860 movdqa %xmm5, %xmm0 # copy_conditional(res_z, ONE, in1infty);
2862 pandn $res_z(%rsp), %xmm0
2864 pandn $res_z+0x10(%rsp), %xmm1
2866 pand .LONE_mont(%rip), %xmm2
2867 pand .LONE_mont+0x10(%rip), %xmm3
2871 movdqa %xmm4, %xmm0 # copy_conditional(res_z, in1_z, in2infty);
2877 pand $in1_z(%rsp), %xmm2
2878 pand $in1_z+0x10(%rsp), %xmm3
2881 movdqu %xmm2, 0x40($r_ptr)
2882 movdqu %xmm3, 0x50($r_ptr)
2884 movdqa %xmm5, %xmm0 # copy_conditional(res_x, in2_x, in1infty);
2886 pandn $res_x(%rsp), %xmm0
2888 pandn $res_x+0x10(%rsp), %xmm1
2890 pand $in2_x(%rsp), %xmm2
2891 pand $in2_x+0x10(%rsp), %xmm3
2895 movdqa %xmm4, %xmm0 # copy_conditional(res_x, in1_x, in2infty);
2901 pand $in1_x(%rsp), %xmm2
2902 pand $in1_x+0x10(%rsp), %xmm3
2905 movdqu %xmm2, 0x00($r_ptr)
2906 movdqu %xmm3, 0x10($r_ptr)
2908 movdqa %xmm5, %xmm0 # copy_conditional(res_y, in2_y, in1infty);
2910 pandn $res_y(%rsp), %xmm0
2912 pandn $res_y+0x10(%rsp), %xmm1
2914 pand $in2_y(%rsp), %xmm2
2915 pand $in2_y+0x10(%rsp), %xmm3
2919 movdqa %xmm4, %xmm0 # copy_conditional(res_y, in1_y, in2infty);
2925 pand $in1_y(%rsp), %xmm2
2926 pand $in1_y+0x10(%rsp), %xmm3
2929 movdqu %xmm2, 0x20($r_ptr)
2930 movdqu %xmm3, 0x30($r_ptr)
2940 .size ecp_nistz256_point_add_affine$sfx,.-ecp_nistz256_point_add_affine$sfx
2943 &gen_add_affine("q");
2945 ########################################################################
2949 ########################################################################
2950 # operate in 4-5-0-1 "name space" that matches multiplication output
2952 my ($a0,$a1,$a2,$a3,$t3,$t4)=($acc4,$acc5,$acc0,$acc1,$acc2,$acc3);
2955 .type __ecp_nistz256_add_tox,\@abi-omnipotent
2957 __ecp_nistz256_add_tox:
2959 adc 8*0($b_ptr), $a0
2960 adc 8*1($b_ptr), $a1
2962 adc 8*2($b_ptr), $a2
2963 adc 8*3($b_ptr), $a3
2978 mov $a0, 8*0($r_ptr)
2980 mov $a1, 8*1($r_ptr)
2982 mov $a2, 8*2($r_ptr)
2983 mov $a3, 8*3($r_ptr)
2986 .size __ecp_nistz256_add_tox,.-__ecp_nistz256_add_tox
2988 .type __ecp_nistz256_sub_fromx,\@abi-omnipotent
2990 __ecp_nistz256_sub_fromx:
2992 sbb 8*0($b_ptr), $a0
2993 sbb 8*1($b_ptr), $a1
2995 sbb 8*2($b_ptr), $a2
2996 sbb 8*3($b_ptr), $a3
3011 mov $a0, 8*0($r_ptr)
3013 mov $a1, 8*1($r_ptr)
3015 mov $a2, 8*2($r_ptr)
3016 mov $a3, 8*3($r_ptr)
3019 .size __ecp_nistz256_sub_fromx,.-__ecp_nistz256_sub_fromx
3021 .type __ecp_nistz256_subx,\@abi-omnipotent
3023 __ecp_nistz256_subx:
3048 .size __ecp_nistz256_subx,.-__ecp_nistz256_subx
3050 .type __ecp_nistz256_mul_by_2x,\@abi-omnipotent
3052 __ecp_nistz256_mul_by_2x:
3054 adc $a0, $a0 # a0:a3+a0:a3
3073 mov $a0, 8*0($r_ptr)
3075 mov $a1, 8*1($r_ptr)
3077 mov $a2, 8*2($r_ptr)
3078 mov $a3, 8*3($r_ptr)
3081 .size __ecp_nistz256_mul_by_2x,.-__ecp_nistz256_mul_by_2x
3086 &gen_add_affine("x");
3090 $code =~ s/\`([^\`]*)\`/eval $1/gem;