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 11.9(**)
21 # SPARC T4 1.67(***) 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$_",(1..4));
46 #include "sparc_arch.h"
49 .register %g2,#scratch
50 .register %g3,#scratch
57 #define LOCALS (STACK_BIAS+STACK_FRAME)
59 .section ".text",#alloc,#execinstr
68 save %sp,-STACK_FRAME-16,%sp
71 SPARC_LOAD_ADDRESS(OPENSSL_sparcv9cap_P,%g1)
74 and %g1,SPARCV9_FMADD|SPARCV9_PREFER_FPU|SPARCV9_VIS3,%g1
75 cmp %g1,SPARCV9_FMADD|SPARCV9_PREFER_FPU
76 be .Lpoly1305_init_fma
80 stx %g0,[$ctx+8] ! zero hash value
84 and $inp,7,$shr ! alignment factor
89 sethi %hi(0x0ffffffc),$t0
91 or $t0,%lo(0x0ffffffc),$t0
94 or $t0,$t1,$t1 ! 0x0ffffffc0ffffffc
95 or $t1,3,$t0 ! 0x0ffffffc0fffffff
97 ldxa [$inp+%g0]0x88,$h0 ! load little-endian key
98 brz,pt $shr,.Lkey_aligned
99 ldxa [$inp+$h1]0x88,$h1
101 ldxa [$inp+$h2]0x88,$h2
112 stx $h0,[$ctx+32+0] ! store key
115 andcc %g1,SPARCV9_VIS3,%g0
120 add %o7,poly1305_blocks_vis3-1b,%o7
122 add %o7,poly1305_emit-poly1305_blocks_vis3,%o5
124 STPTR %o5,[%i2+SIZE_T]
127 restore %g0,1,%o0 ! return 1
131 restore %g0,%g0,%o0 ! return 0
132 .size poly1305_init,.-poly1305_init
134 .globl poly1305_blocks
137 save %sp,-STACK_FRAME,%sp
140 brz,pn $len,.Lno_data
143 ld [$ctx+32+0],$r1 ! load key
148 ld [$ctx+0],$h1 ! load hash value
154 and $inp,7,$shr ! alignment factor
169 ldxa [$inp+%g0]0x88,$d0 ! load little-endian input
170 brz,pt $shr,.Linp_aligned
171 ldxa [$inp+$d1]0x88,$d1
173 ldxa [$inp+$d2]0x88,$d2
183 addcc $d0,$h0,$h0 ! accumulate input
243 srl $h4,2,$t0 ! final reduction step
254 st $h1,[$ctx+0] ! store hash value
263 .size poly1305_blocks,.-poly1305_blocks
265 ########################################################################
266 # VIS3 has umulxhi and addxc...
268 my ($H0,$H1,$H2,$R0,$R1,$S1,$T1) = map("%o$_",(0..5,7));
269 my ($D0,$D1,$D2,$T0) = map("%g$_",(1..4));
273 poly1305_blocks_vis3:
274 save %sp,-STACK_FRAME,%sp
277 brz,pn $len,.Lno_data
280 ldx [$ctx+32+0],$R0 ! load key
283 ldx [$ctx+0],$H0 ! load hash value
287 and $inp,7,$shr ! alignment factor
298 ldxa [$inp+%g0]0x88,$D0 ! load little-endian input
299 brz,pt $shr,.Linp_aligned_vis3
300 ldxa [$inp+$r1]0x88,$D1
302 ldxa [$inp+$r2]0x88,$D2
311 addcc $D0,$H0,$H0 ! accumulate input
316 mulx $R0,$H0,$D0 ! r0*h0
317 addxc $padbit,$H2,$H2
319 mulx $S1,$H1,$T0 ! s1*h1
322 mulx $R1,$H0,$T0 ! r1*h0
326 mulx $R0,$H1,$T0 ! r0*h1
330 mulx $S1,$H2,$T0 ! s1*h2
332 mulx $R0,$H2,$T1 ! r0*h2
336 srlx $D2,2,$T0 ! final reduction step
342 brnz,pt $len,.Loop_vis3
345 stx $H0,[$ctx+0] ! store hash value
351 .size poly1305_blocks_vis3,.-poly1305_blocks_vis3
354 my ($mac,$nonce) = ($inp,$len);
360 save %sp,-STACK_FRAME,%sp
362 ld [$ctx+0],$h1 ! load hash value
368 addcc $h0,5,$r0 ! compare to modulus
373 andcc $h4,4,%g0 ! did it carry/borrow?
376 ld [$nonce+0],$r0 ! load nonce
384 addcc $r0,$h0,$h0 ! accumulate nonce
390 stb $h0,[$mac+0] ! store little-endian result
423 .size poly1305_emit,.-poly1305_emit
427 my ($ctx,$inp,$len,$padbit) = map("%i$_",(0..3));
428 my ($in0,$in1,$in2,$in3,$in4) = map("%o$_",(0..4));
429 my ($i1,$step,$shr,$shl) = map("%l$_",(0..7));
432 my ($h0lo,$h0hi,$h1lo,$h1hi,$h2lo,$h2hi,$h3lo,$h3hi,
433 $two0,$two32,$two64,$two96,$two130,$five_two130,
434 $r0lo,$r0hi,$r1lo,$r1hi,$r2lo,$r2hi,
435 $s2lo,$s2hi,$s3lo,$s3hi,
436 $c0lo,$c0hi,$c1lo,$c1hi,$c2lo,$c2hi,$c3lo,$c3hi) = map("%f".2*$_,(0..31));
438 my ($r3lo,$r3hi,$s1lo,$s1hi) = ($c0lo,$c0hi,$c1lo,$c1hi);
439 my ($x0,$x1,$x2,$x3) = ($c2lo,$c2hi,$c3lo,$c3hi);
440 my ($y0,$y1,$y2,$y3) = ($c1lo,$c1hi,$c3hi,$c3lo);
445 save %sp,-STACK_FRAME-16,%sp
450 add %o7,.Lconsts_fma-1b,%o7
452 ldd [%o7+8*0],$two0 ! load constants
456 ldd [%o7+8*5],$five_two130
458 std $two0,[$ctx+8*0] ! initial hash value, biased 0
459 std $two32,[$ctx+8*1]
460 std $two64,[$ctx+8*2]
461 std $two96,[$ctx+8*3]
463 brz,pn $inp,.Lno_key_fma
466 stx %fsr,[%sp+LOCALS] ! save original %fsr
467 ldx [%o7+8*6],%fsr ! load new %fsr
469 std $two0,[$ctx+8*4] ! key "template"
470 std $two32,[$ctx+8*5]
471 std $two64,[$ctx+8*6]
472 std $two96,[$ctx+8*7]
475 andn $inp,7,$inp ! align pointer
481 ldxa [$inp+%g0]0x88,$in0 ! load little-endian key
482 ldxa [$inp+$i1]0x88,$in2
484 brz $shr,.Lkey_aligned_fma
485 sethi %hi(0xf0000000),$i1 ! 0xf0000000
487 ldxa [$inp+$i2]0x88,$in4
489 srlx $in0,$shr,$in0 ! align data
497 or $i1,3,$i2 ! 0xf0000003
499 andn $in0,$i1,$in0 ! &=0x0fffffff
500 andn $in1,$i2,$in1 ! &=0x0ffffffc
505 st $in0,[$ctx+`8*4+4`] ! fill "template"
506 st $in1,[$ctx+`8*5+4`]
507 st $in2,[$ctx+`8*6+4`]
508 st $in3,[$ctx+`8*7+4`]
510 ldd [$ctx+8*4],$h0lo ! load [biased] key
515 fsubd $h0lo,$two0, $h0lo ! r0
516 ldd [%o7+8*7],$two0 ! more constants
517 fsubd $h1lo,$two32,$h1lo ! r1
519 fsubd $h2lo,$two64,$h2lo ! r2
521 fsubd $h3lo,$two96,$h3lo ! r3
522 ldd [%o7+8*10],$two96
524 fmuld $five_two130,$h1lo,$s1lo ! s1
525 fmuld $five_two130,$h2lo,$s2lo ! s2
526 fmuld $five_two130,$h3lo,$s3lo ! s3
528 faddd $h0lo,$two0, $h0hi
529 faddd $h1lo,$two32,$h1hi
530 faddd $h2lo,$two64,$h2hi
531 faddd $h3lo,$two96,$h3hi
533 fsubd $h0hi,$two0, $h0hi
534 ldd [%o7+8*11],$two0 ! more constants
535 fsubd $h1hi,$two32,$h1hi
536 ldd [%o7+8*12],$two32
537 fsubd $h2hi,$two64,$h2hi
538 ldd [%o7+8*13],$two64
539 fsubd $h3hi,$two96,$h3hi
541 fsubd $h0lo,$h0hi,$h0lo
542 std $h0hi,[$ctx+8*5] ! r0hi
543 fsubd $h1lo,$h1hi,$h1lo
544 std $h1hi,[$ctx+8*7] ! r1hi
545 fsubd $h2lo,$h2hi,$h2lo
546 std $h2hi,[$ctx+8*9] ! r2hi
547 fsubd $h3lo,$h3hi,$h3lo
548 std $h3hi,[$ctx+8*11] ! r3hi
550 faddd $s1lo,$two0, $s1hi
551 faddd $s2lo,$two32,$s2hi
552 faddd $s3lo,$two64,$s3hi
554 fsubd $s1hi,$two0, $s1hi
555 fsubd $s2hi,$two32,$s2hi
556 fsubd $s3hi,$two64,$s3hi
558 fsubd $s1lo,$s1hi,$s1lo
559 fsubd $s2lo,$s2hi,$s2lo
560 fsubd $s3lo,$s3hi,$s3lo
562 ldx [%sp+LOCALS],%fsr ! restore %fsr
564 std $h0lo,[$ctx+8*4] ! r0lo
565 std $h1lo,[$ctx+8*6] ! r1lo
566 std $h2lo,[$ctx+8*8] ! r2lo
567 std $h3lo,[$ctx+8*10] ! r3lo
569 std $s1hi,[$ctx+8*13]
570 std $s2hi,[$ctx+8*15]
571 std $s3hi,[$ctx+8*17]
573 std $s1lo,[$ctx+8*12]
574 std $s2lo,[$ctx+8*14]
575 std $s3lo,[$ctx+8*16]
577 add %o7,poly1305_blocks_fma-.Lconsts_fma,%o0
578 add %o7,poly1305_emit_fma-.Lconsts_fma,%o1
580 STPTR %o1,[%i2+SIZE_T]
583 restore %g0,1,%o0 ! return 1
587 restore %g0,%g0,%o0 ! return 0
588 .size poly1305_init_fma,.-poly1305_init_fma
592 save %sp,-STACK_FRAME-48,%sp
599 add %o7,.Lconsts_fma-1b,%o7
601 ldd [%o7+8*0],$two0 ! load constants
605 ldd [%o7+8*4],$two130
606 ldd [%o7+8*5],$five_two130
608 ldd [$ctx+8*0],$h0lo ! load [biased] hash value
613 std $two0,[%sp+LOCALS+8*0] ! input "template"
614 sethi %hi((1023+52+96)<<20),$in3
615 std $two32,[%sp+LOCALS+8*1]
617 std $two64,[%sp+LOCALS+8*2]
618 st $in3,[%sp+LOCALS+8*3]
621 andn $inp,7,$inp ! align pointer
627 ldxa [$inp+%g0]0x88,$in0 ! load little-endian input
628 brz $shr,.Linp_aligned_fma
629 ldxa [$inp+$i1]0x88,$in2
631 ldxa [$inp+$step]0x88,$in4
634 srlx $in0,$shr,$in0 ! align data
639 srlx $in4,$shr,$in4 ! pre-shift
646 add $step,$inp,$inp ! conditional advance
648 st $in0,[%sp+LOCALS+8*0+4] ! fill "template"
649 st $in1,[%sp+LOCALS+8*1+4]
650 st $in2,[%sp+LOCALS+8*2+4]
651 st $in3,[%sp+LOCALS+8*3+4]
653 ldd [$ctx+8*4],$r0lo ! load key
659 ldd [$ctx+8*10],$r3lo
660 ldd [$ctx+8*11],$r3hi
661 ldd [$ctx+8*12],$s1lo
662 ldd [$ctx+8*13],$s1hi
663 ldd [$ctx+8*14],$s2lo
664 ldd [$ctx+8*15],$s2hi
665 ldd [$ctx+8*16],$s3lo
666 ldd [$ctx+8*17],$s3hi
668 stx %fsr,[%sp+LOCALS+8*4] ! save original %fsr
669 ldx [%o7+8*6],%fsr ! load new %fsr
674 ldd [%sp+LOCALS+8*0],$x0 ! load biased input
675 ldd [%sp+LOCALS+8*1],$x1
676 ldd [%sp+LOCALS+8*2],$x2
677 ldd [%sp+LOCALS+8*3],$x3
679 fsubd $h0lo,$two0, $h0lo ! de-bias hash value
680 fsubd $h1lo,$two32,$h1lo
681 ldxa [$inp+%g0]0x88,$in0 ! modulo-scheduled input load
682 fsubd $h2lo,$two64,$h2lo
683 fsubd $h3lo,$two96,$h3lo
684 ldxa [$inp+$i1]0x88,$in2
686 fsubd $x0,$two0, $x0 ! de-bias input
691 brz $shr,.Linp_aligned_fma2
692 add $step,$inp,$inp ! conditional advance
694 sllx $in0,$shl,$in1 ! align data
698 srlx $in2,$shr,$in4 ! pre-shift
704 faddd $h0lo,$x0,$x0 ! accumulate input
705 stw $in0,[%sp+LOCALS+8*0+4]
707 stw $in1,[%sp+LOCALS+8*1+4]
709 stw $in2,[%sp+LOCALS+8*2+4]
711 stw $in3,[%sp+LOCALS+8*3+4]
718 ldxa [$inp+%g0]0x88,$in0 ! modulo-scheduled input load
719 ldxa [$inp+$i1]0x88,$in2
722 faddd $y0,$h0lo,$h0lo ! accumulate input
723 faddd $y1,$h0hi,$h0hi
724 faddd $y2,$h2lo,$h2lo
725 faddd $y3,$h2hi,$h2hi
727 brz,pn $shr,.Linp_aligned_fma3
728 add $step,$inp,$inp ! conditional advance
730 sllx $in0,$shl,$in1 ! align data
734 srlx $in2,$shr,$in4 ! pre-shift
738 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! base 2^48 -> base 2^32
739 faddd $two64,$h1lo,$c1lo
741 faddd $two64,$h1hi,$c1hi
743 faddd $two130,$h3lo,$c3lo
744 st $in0,[%sp+LOCALS+8*0+4] ! fill "template"
745 faddd $two130,$h3hi,$c3hi
746 st $in1,[%sp+LOCALS+8*1+4]
747 faddd $two32,$h0lo,$c0lo
748 st $in2,[%sp+LOCALS+8*2+4]
749 faddd $two32,$h0hi,$c0hi
750 st $in3,[%sp+LOCALS+8*3+4]
751 faddd $two96,$h2lo,$c2lo
752 faddd $two96,$h2hi,$c2hi
754 fsubd $c1lo,$two64,$c1lo
755 fsubd $c1hi,$two64,$c1hi
756 fsubd $c3lo,$two130,$c3lo
757 fsubd $c3hi,$two130,$c3hi
758 fsubd $c0lo,$two32,$c0lo
759 fsubd $c0hi,$two32,$c0hi
760 fsubd $c2lo,$two96,$c2lo
761 fsubd $c2hi,$two96,$c2hi
763 fsubd $h1lo,$c1lo,$h1lo
764 fsubd $h1hi,$c1hi,$h1hi
765 fsubd $h3lo,$c3lo,$h3lo
766 fsubd $h3hi,$c3hi,$h3hi
767 fsubd $h2lo,$c2lo,$h2lo
768 fsubd $h2hi,$c2hi,$h2hi
769 fsubd $h0lo,$c0lo,$h0lo
770 fsubd $h0hi,$c0hi,$h0hi
772 faddd $h1lo,$c0lo,$h1lo
773 faddd $h1hi,$c0hi,$h1hi
774 faddd $h3lo,$c2lo,$h3lo
775 faddd $h3hi,$c2hi,$h3hi
776 faddd $h2lo,$c1lo,$h2lo
777 faddd $h2hi,$c1hi,$h2hi
778 fmaddd $five_two130,$c3lo,$h0lo,$h0lo
779 fmaddd $five_two130,$c3hi,$h0hi,$h0hi
781 faddd $h1lo,$h1hi,$x1
782 ldd [$ctx+8*12],$s1lo ! reload constants
783 faddd $h3lo,$h3hi,$x3
784 ldd [$ctx+8*13],$s1hi
785 faddd $h2lo,$h2hi,$x2
786 ldd [$ctx+8*10],$r3lo
787 faddd $h0lo,$h0hi,$x0
788 ldd [$ctx+8*11],$r3hi
791 fmuld $x1,$s3lo,$h0lo
792 fmuld $x1,$s3hi,$h0hi
793 fmuld $x1,$r1lo,$h2lo
794 fmuld $x1,$r1hi,$h2hi
795 fmuld $x1,$r0lo,$h1lo
796 fmuld $x1,$r0hi,$h1hi
797 fmuld $x1,$r2lo,$h3lo
798 fmuld $x1,$r2hi,$h3hi
800 fmaddd $x3,$s1lo,$h0lo,$h0lo
801 fmaddd $x3,$s1hi,$h0hi,$h0hi
802 fmaddd $x3,$s3lo,$h2lo,$h2lo
803 fmaddd $x3,$s3hi,$h2hi,$h2hi
804 fmaddd $x3,$s2lo,$h1lo,$h1lo
805 fmaddd $x3,$s2hi,$h1hi,$h1hi
806 fmaddd $x3,$r0lo,$h3lo,$h3lo
807 fmaddd $x3,$r0hi,$h3hi,$h3hi
809 fmaddd $x2,$s2lo,$h0lo,$h0lo
810 fmaddd $x2,$s2hi,$h0hi,$h0hi
811 fmaddd $x2,$r0lo,$h2lo,$h2lo
812 fmaddd $x2,$r0hi,$h2hi,$h2hi
813 fmaddd $x2,$s3lo,$h1lo,$h1lo
814 ldd [%sp+LOCALS+8*0],$y0 ! load [biased] input
815 fmaddd $x2,$s3hi,$h1hi,$h1hi
816 ldd [%sp+LOCALS+8*1],$y1
817 fmaddd $x2,$r1lo,$h3lo,$h3lo
818 ldd [%sp+LOCALS+8*2],$y2
819 fmaddd $x2,$r1hi,$h3hi,$h3hi
820 ldd [%sp+LOCALS+8*3],$y3
822 fmaddd $x0,$r0lo,$h0lo,$h0lo
823 fsubd $y0,$two0, $y0 ! de-bias input
824 fmaddd $x0,$r0hi,$h0hi,$h0hi
826 fmaddd $x0,$r2lo,$h2lo,$h2lo
828 fmaddd $x0,$r2hi,$h2hi,$h2hi
830 fmaddd $x0,$r1lo,$h1lo,$h1lo
831 fmaddd $x0,$r1hi,$h1hi,$h1hi
832 fmaddd $x0,$r3lo,$h3lo,$h3lo
833 fmaddd $x0,$r3hi,$h3hi,$h3hi
835 bcc SIZE_T_CC,.Loop_fma
838 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! base 2^48 -> base 2^32
839 faddd $h0lo,$two32,$c0lo
840 faddd $h0hi,$two32,$c0hi
841 faddd $h2lo,$two96,$c2lo
842 faddd $h2hi,$two96,$c2hi
843 faddd $h1lo,$two64,$c1lo
844 faddd $h1hi,$two64,$c1hi
845 faddd $h3lo,$two130,$c3lo
846 faddd $h3hi,$two130,$c3hi
848 fsubd $c0lo,$two32,$c0lo
849 fsubd $c0hi,$two32,$c0hi
850 fsubd $c2lo,$two96,$c2lo
851 fsubd $c2hi,$two96,$c2hi
852 fsubd $c1lo,$two64,$c1lo
853 fsubd $c1hi,$two64,$c1hi
854 fsubd $c3lo,$two130,$c3lo
855 fsubd $c3hi,$two130,$c3hi
857 fsubd $h1lo,$c1lo,$h1lo
858 fsubd $h1hi,$c1hi,$h1hi
859 fsubd $h3lo,$c3lo,$h3lo
860 fsubd $h3hi,$c3hi,$h3hi
861 fsubd $h2lo,$c2lo,$h2lo
862 fsubd $h2hi,$c2hi,$h2hi
863 fsubd $h0lo,$c0lo,$h0lo
864 fsubd $h0hi,$c0hi,$h0hi
866 faddd $h1lo,$c0lo,$h1lo
867 faddd $h1hi,$c0hi,$h1hi
868 faddd $h3lo,$c2lo,$h3lo
869 faddd $h3hi,$c2hi,$h3hi
870 faddd $h2lo,$c1lo,$h2lo
871 faddd $h2hi,$c1hi,$h2hi
872 fmaddd $five_two130,$c3lo,$h0lo,$h0lo
873 fmaddd $five_two130,$c3hi,$h0hi,$h0hi
875 faddd $h1lo,$h1hi,$x1
876 faddd $h3lo,$h3hi,$x3
877 faddd $h2lo,$h2hi,$x2
878 faddd $h0lo,$h0hi,$x0
880 faddd $x1,$two32,$x1 ! bias
885 ldx [%sp+LOCALS+8*4],%fsr ! restore saved %fsr
887 std $x1,[$ctx+8*1] ! store [biased] hash value
895 .size poly1305_blocks_fma,.-poly1305_blocks_fma
898 my ($mac,$nonce)=($inp,$len);
900 my ($h0,$h1,$h2,$h3,$h4, $d0,$d1,$d2,$d3, $mask
901 ) = (map("%l$_",(0..5)),map("%o$_",(0..4)));
906 save %sp,-STACK_FRAME,%sp
908 ld [$ctx+8*0+0],$d0 ! load hash
917 sethi %hi(0xfff00000),$mask
918 andn $d0,$mask,$d0 ! mask exponent
921 andn $d3,$mask,$d3 ! can be partially reduced...
924 srl $d3,2,$padbit ! ... so reduce
935 addcc $h0,5,$d0 ! compare to modulus
941 srl $mask,2,$mask ! did it carry/borrow?
943 sra $mask,31,$mask ! mask
950 ld [$nonce+0],$d0 ! load nonce
962 addcc $d0,$h0,$h0 ! accumulate nonce
967 stb $h0,[$mac+0] ! write little-endian result
1001 .size poly1305_emit_fma,.-poly1305_emit_fma
1008 .word 0x43300000,0x00000000 ! 2^(52+0)
1009 .word 0x45300000,0x00000000 ! 2^(52+32)
1010 .word 0x47300000,0x00000000 ! 2^(52+64)
1011 .word 0x49300000,0x00000000 ! 2^(52+96)
1012 .word 0x4b500000,0x00000000 ! 2^(52+130)
1014 .word 0x37f40000,0x00000000 ! 5/2^130
1015 .word 0,1<<30 ! fsr: truncate, no exceptions
1017 .word 0x44300000,0x00000000 ! 2^(52+16+0)
1018 .word 0x46300000,0x00000000 ! 2^(52+16+32)
1019 .word 0x48300000,0x00000000 ! 2^(52+16+64)
1020 .word 0x4a300000,0x00000000 ! 2^(52+16+96)
1021 .word 0x3e300000,0x00000000 ! 2^(52+16+0-96)
1022 .word 0x40300000,0x00000000 ! 2^(52+16+32-96)
1023 .word 0x42300000,0x00000000 ! 2^(52+16+64-96)
1024 .asciz "Poly1305 for SPARCv9/VIS3/FMA, CRYPTOGAMS by <appro\@openssl.org>"
1029 # Purpose of these subroutines is to explicitly encode VIS instructions,
1030 # so that one can compile the module without having to specify VIS
1031 # extensions on compiler command line, e.g. -xarch=v9 vs. -xarch=v9a.
1032 # Idea is to reserve for option to produce "universal" binary and let
1033 # programmer detect if current CPU is VIS capable at run-time.
1035 my ($mnemonic,$rs1,$rs2,$rd)=@_;
1036 my %bias = ( "g" => 0, "o" => 8, "l" => 16, "i" => 24 );
1038 my %visopf = ( "addxc" => 0x011,
1040 "umulxhi" => 0x016 );
1042 $ref = "$mnemonic\t$rs1,$rs2,$rd";
1044 if ($opf=$visopf{$mnemonic}) {
1045 foreach ($rs1,$rs2,$rd) {
1046 return $ref if (!/%([goli])([0-9])/);
1050 return sprintf ".word\t0x%08x !%s",
1051 0x81b00000|$rd<<25|$rs1<<14|$opf<<5|$rs2,
1059 my ($mnemonic,$rs1,$rs2,$rs3,$rd)=@_;
1061 my %fmaopf = ( "fmadds" => 0x1,
1066 $ref = "$mnemonic\t$rs1,$rs2,$rs3,$rd";
1068 if ($opf=$fmaopf{$mnemonic}) {
1069 foreach ($rs1,$rs2,$rs3,$rd) {
1070 return $ref if (!/%f([0-9]{1,2})/);
1073 return $ref if ($1&1);
1074 # re-encode for upper double register addressing
1079 return sprintf ".word\t0x%08x !%s",
1080 0x81b80000|$rd<<25|$rs1<<14|$rs3<<9|$opf<<5|$rs2,
1087 foreach (split("\n",$code)) {
1088 s/\`([^\`]*)\`/eval $1/ge;
1090 s/\b(umulxhi|addxc[c]{0,2})\s+(%[goli][0-7]),\s*(%[goli][0-7]),\s*(%[goli][0-7])/
1091 &unvis3($1,$2,$3,$4)
1093 s/\b(fmadd[sd])\s+(%f[0-9]+),\s*(%f[0-9]+),\s*(%f[0-9]+),\s*(%f[0-9]+)/
1094 &unfma($1,$2,$3,$4,$5)