arm64/crypto/poly1305-armv8.pl

47 my ($h0,$h1,$h2,$r0,$r1,$s1,$t0,$t1,$d0,$d1,$d2) = map("x$_",(4..14));
85 	and	$s1,$s1,#-4
186 	and	$t0,$d2,#-4		// final reduction
242 	tst	$d2,#-4			// see if it's carried/borrowed
262 my ($R0,$R1,$S1,$R2,$S2,$R3,$S3,$R4,$S4) = map("v$_.4s",(0..8));
300 	and	$t0,$d2,#-4		// final reduction
311 .align	4
313 	and	x12,$h0,#0x03ffffff	// base 2^64 -> base 2^26
327 	str	w13,[$ctx,#16*4]	// s2
369 	lsr	$h1,x12,#12
370 	adds	$h0,$h0,x12,lsl#52
393 	extr	x12,$h1,$h0,#52
394 	and	x12,x12,#0x03ffffff
400 .align	4
427 	extr	x12,$h1,$h0,#52
428 	and	x12,x12,#0x03ffffff
437 	fmov	${H2},x12
450 	sub	$ctx,$ctx,#4
454 	sub	$ctx,$ctx,#4
457 	bl	poly1305_mult		// r^4
458 	sub	$ctx,$ctx,#4
463 .align	4
467 	fmov	${H2},x12
472 	ldp	x8,x12,[$inp,#32]	// inp[2:3]
483 	rev	x12,x12
492 	extr	x8,x12,x8,#52
498 	ubfx	x10,x12,#14,#26
500 	add	x12,$padbit,x12,lsr#40
505 	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
508 	fmov	$IN23_4,x12
510 	ldp	x8,x12,[$inp],#16	// inp[0:1]
519 	rev	x12,x12
528 	extr	x8,x12,x8,#52
534 	ubfx	x10,x12,#14,#26
536 	add	x12,$padbit,x12,lsr#40
541 	add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
545 	fmov	$IN01_4,x12
550 .align	4
553 	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2
554 	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^3+inp[7]*r
556 	// ((inp[0]*r^4+inp[2]*r^2+inp[4])*r^4+inp[6]*r^2+inp[8])*r^2
557 	// ((inp[1]*r^4+inp[3]*r^2+inp[5])*r^4+inp[7]*r^2+inp[9])*r
574 	 ldp	x8,x12,[$in2],#16	// inp[2:3] (or zero)
580 	 rev	x12,x12
597 	 extr	x8,x12,x8,#52
610 	 ubfx	x10,x12,#14,#26
619 	 add	x12,$padbit,x12,lsr#40
625 	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
632 	// (hash+inp[0:1])*r^4 and accumulate
635 	 fmov	$IN23_4,x12
637 	 ldp	x8,x12,[$inp],#16	// inp[0:1]
645 	 rev	x12,x12
664 	 extr	x8,x12,x8,#52
677 	 ubfx	x10,x12,#14,#26
685 	 add	x12,$padbit,x12,lsr#40
692 	 add	x12,x12,x13,lsl#32	// bfi	x12,x13,#32,#32
697 	 fmov	$IN01_4,x12
793 	// (hash+inp[0:1])*r^4:r^3 and accumulate
894 .comm	OPENSSL_armcap_P,4,4
904 	(m/\bum(ul|la)l\b/ and (s/\.4s/.2s/g or 1))		or
905 	(m/\bum(ul|la)l2\b/ and (s/\.2s/.4s/g or 1))		or
906 	(m/\bst[1-4]\s+{[^}]+}\[/ and (s/\.[24]d/.s/g or 1));