powerpc/crypto/poly1305-p10le_64.S

1 /* SPDX-License-Identifier: GPL-2.0-or-later */
5 # Copyright 2023- IBM Corp. All rights reserved
10 # Poly1305 - this version mainly using vector/VSX/Scalar
11 #  - 26 bits limbs
12 #  - Handle multiple 64 byte blcok.
17 # p = 2^130 - 5
25 #  07/22/21 - this revison based on the above sum of products.  Setup r^4, r^3, r^2, r and s3, s2, …
56 #include <asm/asm-offsets.h>
57 #include <asm/asm-compat.h>
95 	stdu 1,-752(1)
101 	SAVE_GPR 18, 144, 1
106 	SAVE_GPR 23, 184, 1
120 	SAVE_VRS 23, 48, 9
134 	SAVE_VSX 18, 256, 9
139 	SAVE_VSX 23, 336, 9
155 	RESTORE_VRS 23, 48, 9
169 	RESTORE_VSX 18, 256, 9
174 	RESTORE_VSX 23, 336, 9
188 	RESTORE_GPR 18, 144, 1
193 	RESTORE_GPR 23, 184, 1
256 	vmulouw	18, 4, 30
259 	vaddudm	18, 18, 10
260 	vaddudm	18, 18, 11
263 	vaddudm	18, 18, 12
264 	vaddudm	18, 18, 13	# x4
317 	vaddudm	18, 18, 9
318 	vaddudm	18, 18, 10
319 	vaddudm	18, 18, 11
320 	vaddudm	18, 18, 12
321 	vaddudm	18, 18, 13	# x4
447 	vaddudm	18, 18, 11
448 	vsrd	12, 18, 31
452 	vand	8, 18, 25
486 	lvx	25, 0, 10	# v25 - mask
508 	extrdi	18, 10, 24, 0
510 	mtvsrdd	62, 0, 18
556 	extrdi	18, 10, 24, 0
558 	mtvsrdd	40, 0, 18
569 	vperm	15, 11, 12, 18
587 	vaddudm	23, 7, 12
596 	vperm	15, 11, 12, 18
615 	vmrgow	7, 12, 23
619 	addi	5, 5, -64	# len -= 64
633 # h3 = (h1 + m3) * r^2,	h4 = (h2 + m4) * r^2  --> (h0 + m1) r*4 + (h3 + m3) r^2, (h0 + m2) r^4 + (h…
635 # h5 = (h3 + m5) * r^2,	h6 = (h4 + m6) * r^2  -->
636 # h7 = (h5 + m7) * r^2,	h8 = (h6 + m8) * r^1  --> m5 * r^4 + m6 * r^3 + m7 * r^2 + m8 * r
649 	vaddudm	18, 18, 11
650 	vsrd	12, 18, 31
654 	vand	8, 18, 25
680 	vperm	15, 11, 12, 18
686 	vperm	18, 11, 12, 18
700 	vsld	23, 16, 13
701 	vor	22, 22, 23
703 	vand	16, 18, 25
708 	vsrd	23, 15, 13	# >> 14
709 	vsrd	24, 23, 31	# >> 26, a4
710 	vand	23, 23, 25	# a3
711 	vsrd	12, 18, 13	# >> 14
718 	vaddudm	7, 7, 23
729 	addi	5, 5, -64	# len -= 64
763 	vaddudm	8, 18, 13
857         add     19, 21, 10      # s1: r19 - (r1 >> 2) *5
906 	sldi	23, 22, 0x2
907 	add	23, 23, 22	# (h2 & 3) * 5
908 	addc	27, 27, 23	# h0
923 #   - no highbit if final leftover block (highbit = 0)
931 	stdu 1,-400(1)
937 	SAVE_GPR 18, 144, 1
942 	SAVE_GPR 23, 184, 1
1003 	RESTORE_GPR 18, 144, 1
1008 	RESTORE_GPR 23, 184, 1
1039 	# h + 5 + (-p)