3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
10 # This module implements Poly1305 hash for SPARCv9, vanilla, as well
11 # as VIS3 and FMA extensions.
15 # Numbers are cycles per processed byte with poly1305_blocks alone.
19 # UltraSPARC III 12.3(**)
21 # SPARC T4 1.70(***) 6.55
24 # (*) Comparison to compiler-generated code is really problematic,
25 # because latter's performance varies too much depending on too
26 # many variables. For example, one can measure from 5x to 15x
27 # improvement on T4 for gcc-4.6. Well, in T4 case it's a bit
28 # unfair comparison, because compiler doesn't use VIS3, but
29 # given same initial conditions coefficient varies from 3x to 9x.
30 # (**) Pre-III performance should be even worse; floating-point
31 # performance for UltraSPARC I-IV on the other hand is reported
32 # to be 4.25 for hand-coded assembly, but they are just too old
34 # (***) Multi-process benchmark saturates at ~12.5x single-process
35 # result on 8-core processor, or ~21GBps per 2.85GHz socket.
38 open STDOUT,">$output";
40 my ($ctx,$inp,$len,$padbit,$shl,$shr) = map("%i$_",(0..5));
41 my ($r0,$r1,$r2,$r3,$s1,$s2,$s3,$h4) = map("%l$_",(0..7));
42 my ($h0,$h1,$h2,$h3, $t0,$t1,$t2) = map("%o$_",(0..5,7));
43 my ($d0,$d1,$d2,$d3) = map("%g$_",(1..4));
46 open STDOUT,">$stdout";
49 #include "sparc_arch.h"
52 .register %g2,#scratch
53 .register %g3,#scratch
60 #define LOCALS (STACK_BIAS+STACK_FRAME)
62 .section ".text",#alloc,#execinstr
71 save %sp,-STACK_FRAME-16,%sp
74 SPARC_LOAD_ADDRESS(OPENSSL_sparcv9cap_P,%g1)
77 and %g1,SPARCV9_FMADD|SPARCV9_PREFER_FPU|SPARCV9_VIS3,%g1
78 cmp %g1,SPARCV9_FMADD|SPARCV9_PREFER_FPU
79 be .Lpoly1305_init_fma
83 stx %g0,[$ctx+8] ! zero hash value
87 and $inp,7,$shr ! alignment factor
92 sethi %hi(0x0ffffffc),$t0
94 or $t0,%lo(0x0ffffffc),$t0
97 or $t0,$t1,$t1 ! 0x0ffffffc0ffffffc
98 or $t1,3,$t0 ! 0x0ffffffc0fffffff
100 ldxa [$inp+%g0]0x88,$h0 ! load little-endian key
101 brz,pt $shr,.Lkey_aligned
102 ldxa [$inp+$h1]0x88,$h1
104 ldxa [$inp+$h2]0x88,$h2
115 stx $h0,[$ctx+32+0] ! store key
118 andcc %g1,SPARCV9_VIS3,%g0
123 add %o7,poly1305_blocks_vis3-1b,%o7
125 add %o7,poly1305_emit-poly1305_blocks_vis3,%o5
127 STPTR %o5,[%i2+SIZE_T]
130 restore %g0,1,%o0 ! return 1
134 restore %g0,%g0,%o0 ! return 0
135 .size poly1305_init,.-poly1305_init
137 .globl poly1305_blocks
140 save %sp,-STACK_FRAME,%sp
143 brz,pn $len,.Lno_data
146 ld [$ctx+32+0],$r1 ! load key
151 ld [$ctx+0],$h1 ! load hash value
157 and $inp,7,$shr ! alignment factor
172 ldxa [$inp+%g0]0x88,$d0 ! load little-endian input
173 brz,pt $shr,.Linp_aligned
174 ldxa [$inp+$d1]0x88,$d1
176 ldxa [$inp+$d2]0x88,$d2
186 addcc $d0,$h0,$h0 ! accumulate input
246 srl $h4,2,$t0 ! final reduction step
258 st $h1,[$ctx+0] ! store hash value
267 .size poly1305_blocks,.-poly1305_blocks
269 ########################################################################
270 # VIS3 has umulxhi and addxc...
272 my ($H0,$H1,$H2,$R0,$R1,$S1,$T1) = map("%o$_",(0..5,7));
273 my ($D0,$D1,$D2,$T0) = map("%g$_",(1..4));
277 poly1305_blocks_vis3:
278 save %sp,-STACK_FRAME,%sp
281 brz,pn $len,.Lno_data
284 ldx [$ctx+32+0],$R0 ! load key
287 ldx [$ctx+0],$H0 ! load hash value
291 and $inp,7,$shr ! alignment factor
303 ldxa [$inp+%g0]0x88,$D0 ! load little-endian input
304 brz,pt $shr,.Linp_aligned_vis3
305 ldxa [$inp+$r1]0x88,$D1
307 ldxa [$inp+$r2]0x88,$D2
316 addcc $D0,$H0,$H0 ! accumulate input
321 mulx $R0,$H0,$D0 ! r0*h0
322 addxc $padbit,$H2,$H2
324 mulx $S1,$H1,$T0 ! s1*h1
327 mulx $R1,$H0,$T0 ! r1*h0
331 mulx $R0,$H1,$T0 ! r0*h1
335 mulx $S1,$H2,$T0 ! s1*h2
337 mulx $R0,$H2,$T1 ! r0*h2
341 srlx $D2,2,$T0 ! final reduction step
348 brnz,pt $len,.Loop_vis3
351 stx $H0,[$ctx+0] ! store hash value
357 .size poly1305_blocks_vis3,.-poly1305_blocks_vis3
360 my ($mac,$nonce) = ($inp,$len);
366 save %sp,-STACK_FRAME,%sp
368 ld [$ctx+0],$h1 ! load hash value
374 addcc $h0,5,$r0 ! compare to modulus
379 andcc $h4,4,%g0 ! did it carry/borrow?
382 ld [$nonce+0],$r0 ! load nonce
390 addcc $r0,$h0,$h0 ! accumulate nonce
396 stb $h0,[$mac+0] ! store little-endian result
429 .size poly1305_emit,.-poly1305_emit
433 my ($ctx,$inp,$len,$padbit) = map("%i$_",(0..3));
434 my ($in0,$in1,$in2,$in3,$in4) = map("%o$_",(0..4));
435 my ($i1,$step,$shr,$shl) = map("%l$_",(0..7));
438 my ($h0lo,$h0hi,$h1lo,$h1hi,$h2lo,$h2hi,$h3lo,$h3hi,
439 $two0,$two32,$two64,$two96,$two130,$five_two130,
440 $r0lo,$r0hi,$r1lo,$r1hi,$r2lo,$r2hi,
441 $s2lo,$s2hi,$s3lo,$s3hi,
442 $c0lo,$c0hi,$c1lo,$c1hi,$c2lo,$c2hi,$c3lo,$c3hi) = map("%f".2*$_,(0..31));
444 my ($r3lo,$r3hi,$s1lo,$s1hi) = ($c0lo,$c0hi,$c1lo,$c1hi);
445 my ($x0,$x1,$x2,$x3) = ($c2lo,$c2hi,$c3lo,$c3hi);
446 my ($y0,$y1,$y2,$y3) = ($c1lo,$c1hi,$c3hi,$c3lo);
451 save %sp,-STACK_FRAME-16,%sp
456 add %o7,.Lconsts_fma-1b,%o7
458 ldd [%o7+8*0],$two0 ! load constants
462 ldd [%o7+8*5],$five_two130
464 std $two0,[$ctx+8*0] ! initial hash value, biased 0
465 std $two32,[$ctx+8*1]
466 std $two64,[$ctx+8*2]
467 std $two96,[$ctx+8*3]
469 brz,pn $inp,.Lno_key_fma
472 stx %fsr,[%sp+LOCALS] ! save original %fsr
473 ldx [%o7+8*6],%fsr ! load new %fsr
475 std $two0,[$ctx+8*4] ! key "template"
476 std $two32,[$ctx+8*5]
477 std $two64,[$ctx+8*6]
478 std $two96,[$ctx+8*7]
481 andn $inp,7,$inp ! align pointer
487 ldxa [$inp+%g0]0x88,$in0 ! load little-endian key
488 ldxa [$inp+$i1]0x88,$in2
490 brz $shr,.Lkey_aligned_fma
491 sethi %hi(0xf0000000),$i1 ! 0xf0000000
493 ldxa [$inp+$i2]0x88,$in4
495 srlx $in0,$shr,$in0 ! align data
503 or $i1,3,$i2 ! 0xf0000003
505 andn $in0,$i1,$in0 ! &=0x0fffffff
506 andn $in1,$i2,$in1 ! &=0x0ffffffc
511 st $in0,[$ctx+`8*4+4`] ! fill "template"
512 st $in1,[$ctx+`8*5+4`]
513 st $in2,[$ctx+`8*6+4`]
514 st $in3,[$ctx+`8*7+4`]
516 ldd [$ctx+8*4],$h0lo ! load [biased] key
521 fsubd $h0lo,$two0, $h0lo ! r0
522 ldd [%o7+8*7],$two0 ! more constants
523 fsubd $h1lo,$two32,$h1lo ! r1
525 fsubd $h2lo,$two64,$h2lo ! r2
527 fsubd $h3lo,$two96,$h3lo ! r3
528 ldd [%o7+8*10],$two96
530 fmuld $five_two130,$h1lo,$s1lo ! s1
531 fmuld $five_two130,$h2lo,$s2lo ! s2
532 fmuld $five_two130,$h3lo,$s3lo ! s3
534 faddd $h0lo,$two0, $h0hi
535 faddd $h1lo,$two32,$h1hi
536 faddd $h2lo,$two64,$h2hi
537 faddd $h3lo,$two96,$h3hi
539 fsubd $h0hi,$two0, $h0hi
540 ldd [%o7+8*11],$two0 ! more constants
541 fsubd $h1hi,$two32,$h1hi
542 ldd [%o7+8*12],$two32
543 fsubd $h2hi,$two64,$h2hi
544 ldd [%o7+8*13],$two64
545 fsubd $h3hi,$two96,$h3hi
547 fsubd $h0lo,$h0hi,$h0lo
548 std $h0hi,[$ctx+8*5] ! r0hi
549 fsubd $h1lo,$h1hi,$h1lo
550 std $h1hi,[$ctx+8*7] ! r1hi
551 fsubd $h2lo,$h2hi,$h2lo
552 std $h2hi,[$ctx+8*9] ! r2hi
553 fsubd $h3lo,$h3hi,$h3lo
554 std $h3hi,[$ctx+8*11] ! r3hi
556 faddd $s1lo,$two0, $s1hi
557 faddd $s2lo,$two32,$s2hi
558 faddd $s3lo,$two64,$s3hi
560 fsubd $s1hi,$two0, $s1hi
561 fsubd $s2hi,$two32,$s2hi
562 fsubd $s3hi,$two64,$s3hi
564 fsubd $s1lo,$s1hi,$s1lo
565 fsubd $s2lo,$s2hi,$s2lo
566 fsubd $s3lo,$s3hi,$s3lo
568 ldx [%sp+LOCALS],%fsr ! restore %fsr
570 std $h0lo,[$ctx+8*4] ! r0lo
571 std $h1lo,[$ctx+8*6] ! r1lo
572 std $h2lo,[$ctx+8*8] ! r2lo
573 std $h3lo,[$ctx+8*10] ! r3lo
575 std $s1hi,[$ctx+8*13]
576 std $s2hi,[$ctx+8*15]
577 std $s3hi,[$ctx+8*17]
579 std $s1lo,[$ctx+8*12]
580 std $s2lo,[$ctx+8*14]
581 std $s3lo,[$ctx+8*16]
583 add %o7,poly1305_blocks_fma-.Lconsts_fma,%o0
584 add %o7,poly1305_emit_fma-.Lconsts_fma,%o1
586 STPTR %o1,[%i2+SIZE_T]
589 restore %g0,1,%o0 ! return 1
593 restore %g0,%g0,%o0 ! return 0
594 .size poly1305_init_fma,.-poly1305_init_fma
598 save %sp,-STACK_FRAME-48,%sp
605 add %o7,.Lconsts_fma-1b,%o7
607 ldd [%o7+8*0],$two0 ! load constants
611 ldd [%o7+8*4],$two130
612 ldd [%o7+8*5],$five_two130
614 ldd [$ctx+8*0],$h0lo ! load [biased] hash value
619 std $two0,[%sp+LOCALS+8*0] ! input "template"
620 sethi %hi((1023+52+96)<<20),$in3
621 std $two32,[%sp+LOCALS+8*1]
623 std $two64,[%sp+LOCALS+8*2]
624 st $in3,[%sp+LOCALS+8*3]
627 andn $inp,7,$inp ! align pointer
633 ldxa [$inp+%g0]0x88,$in0 ! load little-endian input
634 brz $shr,.Linp_aligned_fma
635 ldxa [$inp+$i1]0x88,$in2
637 ldxa [$inp+$step]0x88,$in4
640 srlx $in0,$shr,$in0 ! align data
645 srlx $in4,$shr,$in4 ! pre-shift
652 add $step,$inp,$inp ! conditional advance
654 st $in0,[%sp+LOCALS+8*0+4] ! fill "template"
655 st $in1,[%sp+LOCALS+8*1+4]
656 st $in2,[%sp+LOCALS+8*2+4]
657 st $in3,[%sp+LOCALS+8*3+4]
659 ldd [$ctx+8*4],$r0lo ! load key
665 ldd [$ctx+8*10],$r3lo
666 ldd [$ctx+8*11],$r3hi
667 ldd [$ctx+8*12],$s1lo
668 ldd [$ctx+8*13],$s1hi
669 ldd [$ctx+8*14],$s2lo
670 ldd [$ctx+8*15],$s2hi
671 ldd [$ctx+8*16],$s3lo
672 ldd [$ctx+8*17],$s3hi
674 stx %fsr,[%sp+LOCALS+8*4] ! save original %fsr
675 ldx [%o7+8*6],%fsr ! load new %fsr
680 ldd [%sp+LOCALS+8*0],$x0 ! load biased input
681 ldd [%sp+LOCALS+8*1],$x1
682 ldd [%sp+LOCALS+8*2],$x2
683 ldd [%sp+LOCALS+8*3],$x3
685 fsubd $h0lo,$two0, $h0lo ! de-bias hash value
686 fsubd $h1lo,$two32,$h1lo
687 ldxa [$inp+%g0]0x88,$in0 ! modulo-scheduled input load
688 fsubd $h2lo,$two64,$h2lo
689 fsubd $h3lo,$two96,$h3lo
690 ldxa [$inp+$i1]0x88,$in2
692 fsubd $x0,$two0, $x0 ! de-bias input
697 brz $shr,.Linp_aligned_fma2
698 add $step,$inp,$inp ! conditional advance
700 sllx $in0,$shl,$in1 ! align data
704 srlx $in2,$shr,$in4 ! pre-shift
710 faddd $h0lo,$x0,$x0 ! accumulate input
711 stw $in0,[%sp+LOCALS+8*0+4]
713 stw $in1,[%sp+LOCALS+8*1+4]
715 stw $in2,[%sp+LOCALS+8*2+4]
717 stw $in3,[%sp+LOCALS+8*3+4]
724 ldxa [$inp+%g0]0x88,$in0 ! modulo-scheduled input load
725 ldxa [$inp+$i1]0x88,$in2
728 faddd $y0,$h0lo,$h0lo ! accumulate input
729 faddd $y1,$h0hi,$h0hi
730 faddd $y2,$h2lo,$h2lo
731 faddd $y3,$h2hi,$h2hi
733 brz,pn $shr,.Linp_aligned_fma3
734 add $step,$inp,$inp ! conditional advance
736 sllx $in0,$shl,$in1 ! align data
740 srlx $in2,$shr,$in4 ! pre-shift
744 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! base 2^48 -> base 2^32
745 faddd $two64,$h1lo,$c1lo
747 faddd $two64,$h1hi,$c1hi
749 faddd $two130,$h3lo,$c3lo
750 st $in0,[%sp+LOCALS+8*0+4] ! fill "template"
751 faddd $two130,$h3hi,$c3hi
752 st $in1,[%sp+LOCALS+8*1+4]
753 faddd $two32,$h0lo,$c0lo
754 st $in2,[%sp+LOCALS+8*2+4]
755 faddd $two32,$h0hi,$c0hi
756 st $in3,[%sp+LOCALS+8*3+4]
757 faddd $two96,$h2lo,$c2lo
758 faddd $two96,$h2hi,$c2hi
760 fsubd $c1lo,$two64,$c1lo
761 fsubd $c1hi,$two64,$c1hi
762 fsubd $c3lo,$two130,$c3lo
763 fsubd $c3hi,$two130,$c3hi
764 fsubd $c0lo,$two32,$c0lo
765 fsubd $c0hi,$two32,$c0hi
766 fsubd $c2lo,$two96,$c2lo
767 fsubd $c2hi,$two96,$c2hi
769 fsubd $h1lo,$c1lo,$h1lo
770 fsubd $h1hi,$c1hi,$h1hi
771 fsubd $h3lo,$c3lo,$h3lo
772 fsubd $h3hi,$c3hi,$h3hi
773 fsubd $h2lo,$c2lo,$h2lo
774 fsubd $h2hi,$c2hi,$h2hi
775 fsubd $h0lo,$c0lo,$h0lo
776 fsubd $h0hi,$c0hi,$h0hi
778 faddd $h1lo,$c0lo,$h1lo
779 faddd $h1hi,$c0hi,$h1hi
780 faddd $h3lo,$c2lo,$h3lo
781 faddd $h3hi,$c2hi,$h3hi
782 faddd $h2lo,$c1lo,$h2lo
783 faddd $h2hi,$c1hi,$h2hi
784 fmaddd $five_two130,$c3lo,$h0lo,$h0lo
785 fmaddd $five_two130,$c3hi,$h0hi,$h0hi
787 faddd $h1lo,$h1hi,$x1
788 ldd [$ctx+8*12],$s1lo ! reload constants
789 faddd $h3lo,$h3hi,$x3
790 ldd [$ctx+8*13],$s1hi
791 faddd $h2lo,$h2hi,$x2
792 ldd [$ctx+8*10],$r3lo
793 faddd $h0lo,$h0hi,$x0
794 ldd [$ctx+8*11],$r3hi
797 fmuld $x1,$s3lo,$h0lo
798 fmuld $x1,$s3hi,$h0hi
799 fmuld $x1,$r1lo,$h2lo
800 fmuld $x1,$r1hi,$h2hi
801 fmuld $x1,$r0lo,$h1lo
802 fmuld $x1,$r0hi,$h1hi
803 fmuld $x1,$r2lo,$h3lo
804 fmuld $x1,$r2hi,$h3hi
806 fmaddd $x3,$s1lo,$h0lo,$h0lo
807 fmaddd $x3,$s1hi,$h0hi,$h0hi
808 fmaddd $x3,$s3lo,$h2lo,$h2lo
809 fmaddd $x3,$s3hi,$h2hi,$h2hi
810 fmaddd $x3,$s2lo,$h1lo,$h1lo
811 fmaddd $x3,$s2hi,$h1hi,$h1hi
812 fmaddd $x3,$r0lo,$h3lo,$h3lo
813 fmaddd $x3,$r0hi,$h3hi,$h3hi
815 fmaddd $x2,$s2lo,$h0lo,$h0lo
816 fmaddd $x2,$s2hi,$h0hi,$h0hi
817 fmaddd $x2,$r0lo,$h2lo,$h2lo
818 fmaddd $x2,$r0hi,$h2hi,$h2hi
819 fmaddd $x2,$s3lo,$h1lo,$h1lo
820 ldd [%sp+LOCALS+8*0],$y0 ! load [biased] input
821 fmaddd $x2,$s3hi,$h1hi,$h1hi
822 ldd [%sp+LOCALS+8*1],$y1
823 fmaddd $x2,$r1lo,$h3lo,$h3lo
824 ldd [%sp+LOCALS+8*2],$y2
825 fmaddd $x2,$r1hi,$h3hi,$h3hi
826 ldd [%sp+LOCALS+8*3],$y3
828 fmaddd $x0,$r0lo,$h0lo,$h0lo
829 fsubd $y0,$two0, $y0 ! de-bias input
830 fmaddd $x0,$r0hi,$h0hi,$h0hi
832 fmaddd $x0,$r2lo,$h2lo,$h2lo
834 fmaddd $x0,$r2hi,$h2hi,$h2hi
836 fmaddd $x0,$r1lo,$h1lo,$h1lo
837 fmaddd $x0,$r1hi,$h1hi,$h1hi
838 fmaddd $x0,$r3lo,$h3lo,$h3lo
839 fmaddd $x0,$r3hi,$h3hi,$h3hi
841 bcc SIZE_T_CC,.Loop_fma
844 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! base 2^48 -> base 2^32
845 faddd $h0lo,$two32,$c0lo
846 faddd $h0hi,$two32,$c0hi
847 faddd $h2lo,$two96,$c2lo
848 faddd $h2hi,$two96,$c2hi
849 faddd $h1lo,$two64,$c1lo
850 faddd $h1hi,$two64,$c1hi
851 faddd $h3lo,$two130,$c3lo
852 faddd $h3hi,$two130,$c3hi
854 fsubd $c0lo,$two32,$c0lo
855 fsubd $c0hi,$two32,$c0hi
856 fsubd $c2lo,$two96,$c2lo
857 fsubd $c2hi,$two96,$c2hi
858 fsubd $c1lo,$two64,$c1lo
859 fsubd $c1hi,$two64,$c1hi
860 fsubd $c3lo,$two130,$c3lo
861 fsubd $c3hi,$two130,$c3hi
863 fsubd $h1lo,$c1lo,$h1lo
864 fsubd $h1hi,$c1hi,$h1hi
865 fsubd $h3lo,$c3lo,$h3lo
866 fsubd $h3hi,$c3hi,$h3hi
867 fsubd $h2lo,$c2lo,$h2lo
868 fsubd $h2hi,$c2hi,$h2hi
869 fsubd $h0lo,$c0lo,$h0lo
870 fsubd $h0hi,$c0hi,$h0hi
872 faddd $h1lo,$c0lo,$h1lo
873 faddd $h1hi,$c0hi,$h1hi
874 faddd $h3lo,$c2lo,$h3lo
875 faddd $h3hi,$c2hi,$h3hi
876 faddd $h2lo,$c1lo,$h2lo
877 faddd $h2hi,$c1hi,$h2hi
878 fmaddd $five_two130,$c3lo,$h0lo,$h0lo
879 fmaddd $five_two130,$c3hi,$h0hi,$h0hi
881 faddd $h1lo,$h1hi,$x1
882 faddd $h3lo,$h3hi,$x3
883 faddd $h2lo,$h2hi,$x2
884 faddd $h0lo,$h0hi,$x0
886 faddd $x1,$two32,$x1 ! bias
891 ldx [%sp+LOCALS+8*4],%fsr ! restore saved %fsr
893 std $x1,[$ctx+8*1] ! store [biased] hash value
901 .size poly1305_blocks_fma,.-poly1305_blocks_fma
904 my ($mac,$nonce)=($inp,$len);
906 my ($h0,$h1,$h2,$h3,$h4, $d0,$d1,$d2,$d3, $mask
907 ) = (map("%l$_",(0..5)),map("%o$_",(0..4)));
912 save %sp,-STACK_FRAME,%sp
914 ld [$ctx+8*0+0],$d0 ! load hash
923 sethi %hi(0xfff00000),$mask
924 andn $d0,$mask,$d0 ! mask exponent
927 andn $d3,$mask,$d3 ! can be partially reduced...
930 srl $d3,2,$padbit ! ... so reduce
941 addcc $h0,5,$d0 ! compare to modulus
947 srl $mask,2,$mask ! did it carry/borrow?
949 sra $mask,31,$mask ! mask
956 ld [$nonce+0],$d0 ! load nonce
968 addcc $d0,$h0,$h0 ! accumulate nonce
973 stb $h0,[$mac+0] ! write little-endian result
1007 .size poly1305_emit_fma,.-poly1305_emit_fma
1014 .word 0x43300000,0x00000000 ! 2^(52+0)
1015 .word 0x45300000,0x00000000 ! 2^(52+32)
1016 .word 0x47300000,0x00000000 ! 2^(52+64)
1017 .word 0x49300000,0x00000000 ! 2^(52+96)
1018 .word 0x4b500000,0x00000000 ! 2^(52+130)
1020 .word 0x37f40000,0x00000000 ! 5/2^130
1021 .word 0,1<<30 ! fsr: truncate, no exceptions
1023 .word 0x44300000,0x00000000 ! 2^(52+16+0)
1024 .word 0x46300000,0x00000000 ! 2^(52+16+32)
1025 .word 0x48300000,0x00000000 ! 2^(52+16+64)
1026 .word 0x4a300000,0x00000000 ! 2^(52+16+96)
1027 .word 0x3e300000,0x00000000 ! 2^(52+16+0-96)
1028 .word 0x40300000,0x00000000 ! 2^(52+16+32-96)
1029 .word 0x42300000,0x00000000 ! 2^(52+16+64-96)
1030 .asciz "Poly1305 for SPARCv9/VIS3/FMA, CRYPTOGAMS by <appro\@openssl.org>"
1035 # Purpose of these subroutines is to explicitly encode VIS instructions,
1036 # so that one can compile the module without having to specify VIS
1037 # extensions on compiler command line, e.g. -xarch=v9 vs. -xarch=v9a.
1038 # Idea is to reserve for option to produce "universal" binary and let
1039 # programmer detect if current CPU is VIS capable at run-time.
1041 my ($mnemonic,$rs1,$rs2,$rd)=@_;
1042 my %bias = ( "g" => 0, "o" => 8, "l" => 16, "i" => 24 );
1044 my %visopf = ( "addxc" => 0x011,
1046 "umulxhi" => 0x016 );
1048 $ref = "$mnemonic\t$rs1,$rs2,$rd";
1050 if ($opf=$visopf{$mnemonic}) {
1051 foreach ($rs1,$rs2,$rd) {
1052 return $ref if (!/%([goli])([0-9])/);
1056 return sprintf ".word\t0x%08x !%s",
1057 0x81b00000|$rd<<25|$rs1<<14|$opf<<5|$rs2,
1065 my ($mnemonic,$rs1,$rs2,$rs3,$rd)=@_;
1067 my %fmaopf = ( "fmadds" => 0x1,
1072 $ref = "$mnemonic\t$rs1,$rs2,$rs3,$rd";
1074 if ($opf=$fmaopf{$mnemonic}) {
1075 foreach ($rs1,$rs2,$rs3,$rd) {
1076 return $ref if (!/%f([0-9]{1,2})/);
1079 return $ref if ($1&1);
1080 # re-encode for upper double register addressing
1085 return sprintf ".word\t0x%08x !%s",
1086 0x81b80000|$rd<<25|$rs1<<14|$rs3<<9|$opf<<5|$rs2,
1093 foreach (split("\n",$code)) {
1094 s/\`([^\`]*)\`/eval $1/ge;
1096 s/\b(umulxhi|addxc[c]{0,2})\s+(%[goli][0-7]),\s*(%[goli][0-7]),\s*(%[goli][0-7])/
1097 &unvis3($1,$2,$3,$4)
1099 s/\b(fmadd[sd])\s+(%f[0-9]+),\s*(%f[0-9]+),\s*(%f[0-9]+),\s*(%f[0-9]+)/
1100 &unfma($1,$2,$3,$4,$5)