powerpc/crypto/aesp10-ppc.pl

2 # SPDX-License-Identifier: GPL-2.0
12 # Copyright (c) 2006-2017, CRYPTOGAMS by <appro@openssl.org>
58 # The module is endian-agnostic in sense that it supports both big-
59 # and little-endian cases. Data alignment in parallelizable modes is
64 # is aligned programmatically, which in turn guarantees exception-
72 # Add XTS subroutine, 9x on little- and 12x improvement on big-endian
76 # Current large-block performance in cycles per byte processed with
77 # 128-bit key (less is better).
79 #		CBC en-/decrypt	CTR	XTS
106 ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
107 ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
108 die "can't locate ppc-xlate.pl";
139 	addi	$ptr,$ptr,-0x48
151 	li		$ptr,-1
153 	beq-		Lenc_key_abort		# if ($inp==0) return -1;
155 	beq-		Lenc_key_abort		# if ($out==0) return -1;
156 	li		$ptr,-2
158 	blt-		Lenc_key_abort
160 	bgt-		Lenc_key_abort
162 	bne-		Lenc_key_abort
189 	vspltisb	$outmask,-1
201 	vperm		$key,$in0,$in0,$mask	# rotate-n-splat
203 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
221 	vperm		$key,$in0,$in0,$mask	# rotate-n-splat
223 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
238 	vperm		$key,$in0,$in0,$mask	# rotate-n-splat
240 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
253 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
268 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
279 	vperm		$key,$in1,$in1,$mask	# roate-n-splat
299 	vperm		$key,$in1,$in1,$mask	# rotate-n-splat
301 	 vperm		$outtail,$stage,$stage,$outperm	# rotate
311 	 vperm		$outtail,$stage,$stage,$outperm	# rotate
327 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
344 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
353 	vperm		$key,$in1,$in1,$mask	# rotate-n-splat
355 	 vperm		$outtail,$in1,$in1,$outperm	# rotate
369 	 vperm		$outtail,$in0,$in0,$outperm	# rotate
405 .size	.${prefix}_set_encrypt_key,.-.${prefix}_set_encrypt_key
408 	$STU		$sp,-$FRAME($sp)
415 	bne-		Ldec_key_abort
438 	stw		r9, -16($inp)
439 	stw		r10,-12($inp)
440 	stw		r11,-8($inp)
441 	stw		r12,-4($inp)
451 .size	.${prefix}_set_decrypt_key,.-.${prefix}_set_decrypt_key
455 {{{	# Single block en- and decrypt procedures			#
508 	vspltisb	v2,-1
514 	vperm		v0,v0,v0,v3		# rotate [and byte swap in LE]
526 .size	.${prefix}_${dir}crypt,.-.${prefix}_${dir}crypt
537 	# constants table endian-specific conversion
538 	if ($consts && m/\.(long|byte)\s+(.+)\s+(\?[a-z]*)$/o) {
542 	    # convert to endian-agnostic format
552 	    # little-endian conversion
566 	# instructions prefixed with '?' are endian-specific and need
568 	if ($flavour =~ /le$/o) {	# little-endian
573 	    s/\?(vperm\s+v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+)/$1$3$2$4/o or
574 	    s/\?(vsldoi\s+v[0-9]+,\s*)(v[0-9]+,)\s*(v[0-9]+,\s*)([0-9]+)/$1$3$2 16-$4/o or
575 	    s/\?(vspltw\s+v[0-9]+,\s*)(v[0-9]+,)\s*([0-9])/$1$2 3-$3/o;
576 	} else {			# big-endian
579 	    s/\?([a-z]+)/$1/o;