diff mbox series

[3/5] An optimized Poly1305 implementation with 4-way unrolling for ppc64le.

Message ID 20230424184726.2091-4-dtsen@linux.ibm.com (mailing list archive)
State Superseded
Delegated to: Herbert Xu
Headers show
Series crypto: Accelerated Chacha20/Poly1305 implementation | expand

Commit Message

Danny Tsen April 24, 2023, 6:47 p.m. UTC
Improve overall performance of Poly1305 for Power10 or later CPU.

Signed-off-by: Danny Tsen <dtsen@linux.ibm.com>
---
 arch/powerpc/crypto/poly1305-p10le_64.S | 1075 +++++++++++++++++++++++
 1 file changed, 1075 insertions(+)
 create mode 100644 arch/powerpc/crypto/poly1305-p10le_64.S
diff mbox series

Patch

diff --git a/arch/powerpc/crypto/poly1305-p10le_64.S b/arch/powerpc/crypto/poly1305-p10le_64.S
new file mode 100644
index 000000000000..22fd255af87e
--- /dev/null
+++ b/arch/powerpc/crypto/poly1305-p10le_64.S
@@ -0,0 +1,1075 @@ 
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#
+# Accelerated poly1305 implementation for ppc64le.
+#
+# Copyright 2023- IBM Inc. All rights reserved
+#
+#===================================================================================
+# Written by Danny Tsen <dtsen@us.ibm.com>
+#
+# Poly1305 - this version mainly using vector/VSX/Scalar
+#  - 26 bits limbs
+#  - Handle multiple 64 byte blcok.
+#
+# Block size 16 bytes
+# key = (r, s)
+# clamp r &= 0x0FFFFFFC0FFFFFFC 0x0FFFFFFC0FFFFFFF
+# p = 2^130 - 5
+# a += m
+# a = (r + a) % p
+# a += s
+#
+# Improve performance by breaking down polynominal to the sum of products with
+#     h4 = m1 * r⁴ + m2 * r³ + m3 * r² + m4 * r
+#
+#  07/22/21 - this revison based on the above sum of products.  Setup r^4, r^3, r^2, r and s3, s2, s1, s0
+#             to 9 vectors for multiplications.
+#
+# setup r^4, r^3, r^2, r vectors
+#    vs    [r^1, r^3, r^2, r^4]
+#    vs0 = [r0,.....]
+#    vs1 = [r1,.....]
+#    vs2 = [r2,.....]
+#    vs3 = [r3,.....]
+#    vs4 = [r4,.....]
+#    vs5 = [r1*5,...]
+#    vs6 = [r2*5,...]
+#    vs7 = [r2*5,...]
+#    vs8 = [r4*5,...]
+#
+#  Each word in a vector consists a member of a "r/s" in [a * r/s].
+#
+# r0, r4*5, r3*5, r2*5, r1*5;
+# r1, r0,   r4*5, r3*5, r2*5;
+# r2, r1,   r0,   r4*5, r3*5;
+# r3, r2,   r1,   r0,   r4*5;
+# r4, r3,   r2,   r1,   r0  ;
+#
+#
+# poly1305_p10le_4blocks( uint8_t *k, uint32_t mlen, uint8_t *m)
+#  k = 32 bytes key
+#  r3 = k (r, s)
+#  r4 = mlen
+#  r5 = m
+#
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+#include <asm/asm-compat.h>
+#include <linux/linkage.h>
+
+.machine "any"
+
+.text
+
+.macro	SAVE_GPR GPR OFFSET FRAME
+	std	\GPR,\OFFSET(\FRAME)
+.endm
+
+.macro	SAVE_VRS VRS OFFSET FRAME
+	li	16, \OFFSET
+	stvx	\VRS, 16, \FRAME
+.endm
+
+.macro	SAVE_VSX VSX OFFSET FRAME
+	li	16, \OFFSET
+	stxvx	\VSX, 16, \FRAME
+.endm
+
+.macro	RESTORE_GPR GPR OFFSET FRAME
+	ld	\GPR,\OFFSET(\FRAME)
+.endm
+
+.macro	RESTORE_VRS VRS OFFSET FRAME
+	li	16, \OFFSET
+	lvx	\VRS, 16, \FRAME
+.endm
+
+.macro	RESTORE_VSX VSX OFFSET FRAME
+	li	16, \OFFSET
+	lxvx	\VSX, 16, \FRAME
+.endm
+
+.macro SAVE_REGS
+	mflr 0
+	std 0, 16(1)
+	stdu 1,-752(1)
+
+	SAVE_GPR 14, 112, 1
+	SAVE_GPR 15, 120, 1
+	SAVE_GPR 16, 128, 1
+	SAVE_GPR 17, 136, 1
+	SAVE_GPR 18, 144, 1
+	SAVE_GPR 19, 152, 1
+	SAVE_GPR 20, 160, 1
+	SAVE_GPR 21, 168, 1
+	SAVE_GPR 22, 176, 1
+	SAVE_GPR 23, 184, 1
+	SAVE_GPR 24, 192, 1
+	SAVE_GPR 25, 200, 1
+	SAVE_GPR 26, 208, 1
+	SAVE_GPR 27, 216, 1
+	SAVE_GPR 28, 224, 1
+	SAVE_GPR 29, 232, 1
+	SAVE_GPR 30, 240, 1
+	SAVE_GPR 31, 248, 1
+
+	addi	9, 1, 256
+	SAVE_VRS 20, 0, 9
+	SAVE_VRS 21, 16, 9
+	SAVE_VRS 22, 32, 9
+	SAVE_VRS 23, 48, 9
+	SAVE_VRS 24, 64, 9
+	SAVE_VRS 25, 80, 9
+	SAVE_VRS 26, 96, 9
+	SAVE_VRS 27, 112, 9
+	SAVE_VRS 28, 128, 9
+	SAVE_VRS 29, 144, 9
+	SAVE_VRS 30, 160, 9
+	SAVE_VRS 31, 176, 9
+
+	SAVE_VSX 14, 192, 9
+	SAVE_VSX 15, 208, 9
+	SAVE_VSX 16, 224, 9
+	SAVE_VSX 17, 240, 9
+	SAVE_VSX 18, 256, 9
+	SAVE_VSX 19, 272, 9
+	SAVE_VSX 20, 288, 9
+	SAVE_VSX 21, 304, 9
+	SAVE_VSX 22, 320, 9
+	SAVE_VSX 23, 336, 9
+	SAVE_VSX 24, 352, 9
+	SAVE_VSX 25, 368, 9
+	SAVE_VSX 26, 384, 9
+	SAVE_VSX 27, 400, 9
+	SAVE_VSX 28, 416, 9
+	SAVE_VSX 29, 432, 9
+	SAVE_VSX 30, 448, 9
+	SAVE_VSX 31, 464, 9
+.endm # SAVE_REGS
+
+.macro RESTORE_REGS
+	addi	9, 1, 256
+	RESTORE_VRS 20, 0, 9
+	RESTORE_VRS 21, 16, 9
+	RESTORE_VRS 22, 32, 9
+	RESTORE_VRS 23, 48, 9
+	RESTORE_VRS 24, 64, 9
+	RESTORE_VRS 25, 80, 9
+	RESTORE_VRS 26, 96, 9
+	RESTORE_VRS 27, 112, 9
+	RESTORE_VRS 28, 128, 9
+	RESTORE_VRS 29, 144, 9
+	RESTORE_VRS 30, 160, 9
+	RESTORE_VRS 31, 176, 9
+
+	RESTORE_VSX 14, 192, 9
+	RESTORE_VSX 15, 208, 9
+	RESTORE_VSX 16, 224, 9
+	RESTORE_VSX 17, 240, 9
+	RESTORE_VSX 18, 256, 9
+	RESTORE_VSX 19, 272, 9
+	RESTORE_VSX 20, 288, 9
+	RESTORE_VSX 21, 304, 9
+	RESTORE_VSX 22, 320, 9
+	RESTORE_VSX 23, 336, 9
+	RESTORE_VSX 24, 352, 9
+	RESTORE_VSX 25, 368, 9
+	RESTORE_VSX 26, 384, 9
+	RESTORE_VSX 27, 400, 9
+	RESTORE_VSX 28, 416, 9
+	RESTORE_VSX 29, 432, 9
+	RESTORE_VSX 30, 448, 9
+	RESTORE_VSX 31, 464, 9
+
+	RESTORE_GPR 14, 112, 1
+	RESTORE_GPR 15, 120, 1
+	RESTORE_GPR 16, 128, 1
+	RESTORE_GPR 17, 136, 1
+	RESTORE_GPR 18, 144, 1
+	RESTORE_GPR 19, 152, 1
+	RESTORE_GPR 20, 160, 1
+	RESTORE_GPR 21, 168, 1
+	RESTORE_GPR 22, 176, 1
+	RESTORE_GPR 23, 184, 1
+	RESTORE_GPR 24, 192, 1
+	RESTORE_GPR 25, 200, 1
+	RESTORE_GPR 26, 208, 1
+	RESTORE_GPR 27, 216, 1
+	RESTORE_GPR 28, 224, 1
+	RESTORE_GPR 29, 232, 1
+	RESTORE_GPR 30, 240, 1
+	RESTORE_GPR 31, 248, 1
+
+	addi    1, 1, 752
+	ld 0, 16(1)
+	mtlr 0
+.endm # RESTORE_REGS
+
+#
+# p[0] = a0*r0 + a1*r4*5 + a2*r3*5 + a3*r2*5 + a4*r1*5;
+# p[1] = a0*r1 + a1*r0   + a2*r4*5 + a3*r3*5 + a4*r2*5;
+# p[2] = a0*r2 + a1*r1   + a2*r0   + a3*r4*5 + a4*r3*5;
+# p[3] = a0*r3 + a1*r2   + a2*r1   + a3*r0   + a4*r4*5;
+# p[4] = a0*r4 + a1*r3   + a2*r2   + a3*r1   + a4*r0  ;
+#
+#    [r^2, r^3, r^1, r^4]
+#    [m3,  m2,  m4,  m1]
+#
+# multiply odd and even words
+.macro mul_odd
+	vmulouw	14, 4, 26
+	vmulouw	10, 5, 3
+	vmulouw	11, 6, 2
+	vmulouw	12, 7, 1
+	vmulouw	13, 8, 0
+	vmulouw	15, 4, 27
+	vaddudm	14, 14, 10
+	vaddudm	14, 14, 11
+	vmulouw	10, 5, 26
+	vmulouw	11, 6, 3
+	vaddudm	14, 14, 12
+	vaddudm	14, 14, 13	# x0
+	vaddudm	15, 15, 10
+	vaddudm	15, 15, 11
+	vmulouw	12, 7, 2
+	vmulouw	13, 8, 1
+	vaddudm	15, 15, 12
+	vaddudm	15, 15, 13	# x1
+	vmulouw	16, 4, 28
+	vmulouw	10, 5, 27
+	vmulouw	11, 6, 26
+	vaddudm	16, 16, 10
+	vaddudm	16, 16, 11
+	vmulouw	12, 7, 3
+	vmulouw	13, 8, 2
+	vaddudm	16, 16, 12
+	vaddudm	16, 16, 13	# x2
+	vmulouw	17, 4, 29
+	vmulouw	10, 5, 28
+	vmulouw	11, 6, 27
+	vaddudm	17, 17, 10
+	vaddudm	17, 17, 11
+	vmulouw	12, 7, 26
+	vmulouw	13, 8, 3
+	vaddudm	17, 17, 12
+	vaddudm	17, 17, 13	# x3
+	vmulouw	18, 4, 30
+	vmulouw	10, 5, 29
+	vmulouw	11, 6, 28
+	vaddudm	18, 18, 10
+	vaddudm	18, 18, 11
+	vmulouw	12, 7, 27
+	vmulouw	13, 8, 26
+	vaddudm	18, 18, 12
+	vaddudm	18, 18, 13	# x4
+.endm
+
+.macro mul_even
+	vmuleuw	9, 4, 26
+	vmuleuw	10, 5, 3
+	vmuleuw	11, 6, 2
+	vmuleuw	12, 7, 1
+	vmuleuw	13, 8, 0
+	vaddudm	14, 14, 9
+	vaddudm	14, 14, 10
+	vaddudm	14, 14, 11
+	vaddudm	14, 14, 12
+	vaddudm	14, 14, 13	# x0
+
+	vmuleuw	9, 4, 27
+	vmuleuw	10, 5, 26
+	vmuleuw	11, 6, 3
+	vmuleuw	12, 7, 2
+	vmuleuw	13, 8, 1
+	vaddudm	15, 15, 9
+	vaddudm	15, 15, 10
+	vaddudm	15, 15, 11
+	vaddudm	15, 15, 12
+	vaddudm	15, 15, 13	# x1
+
+	vmuleuw	9, 4, 28
+	vmuleuw	10, 5, 27
+	vmuleuw	11, 6, 26
+	vmuleuw	12, 7, 3
+	vmuleuw	13, 8, 2
+	vaddudm	16, 16, 9
+	vaddudm	16, 16, 10
+	vaddudm	16, 16, 11
+	vaddudm	16, 16, 12
+	vaddudm	16, 16, 13	# x2
+
+	vmuleuw	9, 4, 29
+	vmuleuw	10, 5, 28
+	vmuleuw	11, 6, 27
+	vmuleuw	12, 7, 26
+	vmuleuw	13, 8, 3
+	vaddudm	17, 17, 9
+	vaddudm	17, 17, 10
+	vaddudm	17, 17, 11
+	vaddudm	17, 17, 12
+	vaddudm	17, 17, 13	# x3
+
+	vmuleuw	9, 4, 30
+	vmuleuw	10, 5, 29
+	vmuleuw	11, 6, 28
+	vmuleuw	12, 7, 27
+	vmuleuw	13, 8, 26
+	vaddudm	18, 18, 9
+	vaddudm	18, 18, 10
+	vaddudm	18, 18, 11
+	vaddudm	18, 18, 12
+	vaddudm	18, 18, 13	# x4
+.endm
+
+#
+# poly1305_setup_r
+#
+# setup r^4, r^3, r^2, r vectors
+#    [r, r^3, r^2, r^4]
+#    vs0 = [r0,...]
+#    vs1 = [r1,...]
+#    vs2 = [r2,...]
+#    vs3 = [r3,...]
+#    vs4 = [r4,...]
+#    vs5 = [r4*5,...]
+#    vs6 = [r3*5,...]
+#    vs7 = [r2*5,...]
+#    vs8 = [r1*5,...]
+#
+# r0, r4*5, r3*5, r2*5, r1*5;
+# r1, r0,   r4*5, r3*5, r2*5;
+# r2, r1,   r0,   r4*5, r3*5;
+# r3, r2,   r1,   r0,   r4*5;
+# r4, r3,   r2,   r1,   r0  ;
+#
+.macro poly1305_setup_r
+
+	# save r
+	xxlor	26, 58, 58
+	xxlor	27, 59, 59
+	xxlor	28, 60, 60
+	xxlor	29, 61, 61
+	xxlor	30, 62, 62
+
+	xxlxor	31, 31, 31
+
+#    [r, r^3, r^2, r^4]
+	# compute r^2
+	vmr	4, 26
+	vmr	5, 27
+	vmr	6, 28
+	vmr	7, 29
+	vmr	8, 30
+	bl	do_mul		# r^2 r^1
+	xxpermdi 58, 58, 36, 0x3		# r0
+	xxpermdi 59, 59, 37, 0x3		# r1
+	xxpermdi 60, 60, 38, 0x3		# r2
+	xxpermdi 61, 61, 39, 0x3		# r3
+	xxpermdi 62, 62, 40, 0x3		# r4
+	xxpermdi 36, 36, 36, 0x3
+	xxpermdi 37, 37, 37, 0x3
+	xxpermdi 38, 38, 38, 0x3
+	xxpermdi 39, 39, 39, 0x3
+	xxpermdi 40, 40, 40, 0x3
+	vspltisb 13, 2
+	vsld	9, 27, 13
+	vsld	10, 28, 13
+	vsld	11, 29, 13
+	vsld	12, 30, 13
+	vaddudm	0, 9, 27
+	vaddudm	1, 10, 28
+	vaddudm	2, 11, 29
+	vaddudm	3, 12, 30
+
+	bl	do_mul		# r^4 r^3
+	vmrgow	26, 26, 4
+	vmrgow	27, 27, 5
+	vmrgow	28, 28, 6
+	vmrgow	29, 29, 7
+	vmrgow	30, 30, 8
+	vspltisb 13, 2
+	vsld	9, 27, 13
+	vsld	10, 28, 13
+	vsld	11, 29, 13
+	vsld	12, 30, 13
+	vaddudm	0, 9, 27
+	vaddudm	1, 10, 28
+	vaddudm	2, 11, 29
+	vaddudm	3, 12, 30
+
+	# r^2 r^4
+	xxlor	0, 58, 58
+	xxlor	1, 59, 59
+	xxlor	2, 60, 60
+	xxlor	3, 61, 61
+	xxlor	4, 62, 62
+	xxlor	5, 32, 32
+	xxlor	6, 33, 33
+	xxlor	7, 34, 34
+	xxlor	8, 35, 35
+
+	vspltw	9, 26, 3
+	vspltw	10, 26, 2
+	vmrgow	26, 10, 9
+	vspltw	9, 27, 3
+	vspltw	10, 27, 2
+	vmrgow	27, 10, 9
+	vspltw	9, 28, 3
+	vspltw	10, 28, 2
+	vmrgow	28, 10, 9
+	vspltw	9, 29, 3
+	vspltw	10, 29, 2
+	vmrgow	29, 10, 9
+	vspltw	9, 30, 3
+	vspltw	10, 30, 2
+	vmrgow	30, 10, 9
+
+	vsld	9, 27, 13
+	vsld	10, 28, 13
+	vsld	11, 29, 13
+	vsld	12, 30, 13
+	vaddudm	0, 9, 27
+	vaddudm	1, 10, 28
+	vaddudm	2, 11, 29
+	vaddudm	3, 12, 30
+.endm
+
+SYM_FUNC_START_LOCAL(do_mul)
+	mul_odd
+
+	# do reduction ( h %= p )
+	# carry reduction
+	vspltisb 9, 2
+	vsrd	10, 14, 31
+	vsrd	11, 17, 31
+	vand	7, 17, 25
+	vand	4, 14, 25
+	vaddudm	18, 18, 11
+	vsrd	12, 18, 31
+	vaddudm	15, 15, 10
+
+	vsrd	11, 15, 31
+	vand	8, 18, 25
+	vand	5, 15, 25
+	vaddudm	4, 4, 12
+	vsld	10, 12, 9
+	vaddudm	6, 16, 11
+
+	vsrd	13, 6, 31
+	vand	6, 6, 25
+	vaddudm	4, 4, 10
+	vsrd	10, 4, 31
+	vaddudm	7, 7, 13
+
+	vsrd	11, 7, 31
+	vand	7, 7, 25
+	vand	4, 4, 25
+	vaddudm	5, 5, 10
+	vaddudm	8, 8, 11
+	blr
+SYM_FUNC_END(do_mul)
+
+#
+# init key
+#
+.macro do_poly1305_init
+	addis	10, 2, rmask@toc@ha
+	addi	10, 10, rmask@toc@l
+
+	ld	11, 0(10)
+	ld	12, 8(10)
+
+	li	14, 16
+	li	15, 32
+	addis	10, 2, cnum@toc@ha
+	addi	10, 10, cnum@toc@l
+	lvx	25, 0, 10	# v25 - mask
+	lvx	31, 14, 10	# v31 = 1a
+	lvx	19, 15, 10	# v19 = 1 << 24
+	lxv	24, 48(10)	# vs24
+	lxv	25, 64(10)	# vs25
+
+	# initialize
+	# load key from r3 to vectors
+	ld	9, 24(3)
+	ld	10, 32(3)
+	and.	9, 9, 11
+	and.	10, 10, 12
+
+	# break 26 bits
+	extrdi	14, 9, 26, 38
+	extrdi	15, 9, 26, 12
+	extrdi	16, 9, 12, 0
+	mtvsrdd	58, 0, 14
+	insrdi	16, 10, 14, 38
+	mtvsrdd	59, 0, 15
+	extrdi	17, 10, 26, 24
+	mtvsrdd	60, 0, 16
+	extrdi	18, 10, 24, 0
+	mtvsrdd	61, 0, 17
+	mtvsrdd	62, 0, 18
+
+	# r1 = r1 * 5, r2 = r2 * 5, r3 = r3 * 5, r4 = r4 * 5
+	li	9, 5
+	mtvsrdd	36, 0, 9
+	vmulouw	0, 27, 4		# v0 = rr0
+	vmulouw	1, 28, 4		# v1 = rr1
+	vmulouw	2, 29, 4		# v2 = rr2
+	vmulouw	3, 30, 4		# v3 = rr3
+.endm
+
+#
+# poly1305_p10le_4blocks( uint8_t *k, uint32_t mlen, uint8_t *m)
+#  k = 32 bytes key
+#  r3 = k (r, s)
+#  r4 = mlen
+#  r5 = m
+#
+SYM_FUNC_START(poly1305_p10le_4blocks)
+.align 5
+	cmpdi	5, 64
+	blt	Out_no_poly1305
+
+	SAVE_REGS
+
+	do_poly1305_init
+
+	li	21, 0	# counter to message
+
+	poly1305_setup_r
+
+	# load previous H state
+	# break/convert r6 to 26 bits
+	ld	9, 0(3)
+	ld	10, 8(3)
+	ld	19, 16(3)
+	sldi	19, 19, 24
+	mtvsrdd	41, 0, 19
+	extrdi	14, 9, 26, 38
+	extrdi	15, 9, 26, 12
+	extrdi	16, 9, 12, 0
+	mtvsrdd	36, 0, 14
+	insrdi	16, 10, 14, 38
+	mtvsrdd	37, 0, 15
+	extrdi	17, 10, 26, 24
+	mtvsrdd	38, 0, 16
+	extrdi	18, 10, 24, 0
+	mtvsrdd	39, 0, 17
+	mtvsrdd	40, 0, 18
+	vor	8, 8, 9
+
+	# input m1 m2
+	add	20, 4, 21
+	xxlor	49, 24, 24
+	xxlor	50, 25, 25
+	lxvw4x	43, 0, 20
+	addi	17, 20, 16
+	lxvw4x	44, 0, 17
+	vperm	14, 11, 12, 17
+	vperm	15, 11, 12, 18
+	vand	9, 14, 25	# a0
+	vsrd	10, 14, 31	# >> 26
+	vsrd	11, 10, 31	# 12 bits left
+	vand	10, 10, 25	# a1
+	vspltisb 13, 12
+	vand	16, 15, 25
+	vsld	12, 16, 13
+	vor	11, 11, 12
+	vand	11, 11, 25	# a2
+	vspltisb 13, 14
+	vsrd	12, 15, 13	# >> 14
+	vsrd	13, 12, 31	# >> 26, a4
+	vand	12, 12, 25	# a3
+
+	vaddudm	20, 4, 9
+	vaddudm	21, 5, 10
+	vaddudm	22, 6, 11
+	vaddudm	23, 7, 12
+	vaddudm	24, 8, 13
+
+	# m3 m4
+	addi	17, 17, 16
+	lxvw4x	43, 0, 17
+	addi	17, 17, 16
+	lxvw4x	44, 0, 17
+	vperm	14, 11, 12, 17
+	vperm	15, 11, 12, 18
+	vand	9, 14, 25	# a0
+	vsrd	10, 14, 31	# >> 26
+	vsrd	11, 10, 31	# 12 bits left
+	vand	10, 10, 25	# a1
+	vspltisb 13, 12
+	vand	16, 15, 25
+	vsld	12, 16, 13
+	vspltisb 13, 14
+	vor	11, 11, 12
+	vand	11, 11, 25	# a2
+	vsrd	12, 15, 13	# >> 14
+	vsrd	13, 12, 31	# >> 26, a4
+	vand	12, 12, 25	# a3
+
+	# Smash 4 message blocks into 5 vectors of [m4,  m2,  m3,  m1]
+	vmrgow	4, 9, 20
+	vmrgow	5, 10, 21
+	vmrgow	6, 11, 22
+	vmrgow	7, 12, 23
+	vmrgow	8, 13, 24
+	vaddudm	8, 8, 19
+
+	addi	5, 5, -64	# len -= 64
+	addi	21, 21, 64	# offset += 64
+
+	li      9, 64
+	divdu   31, 5, 9
+
+	cmpdi	31, 0
+	ble	Skip_block_loop
+
+	mtctr	31
+
+# h4 =   m1 * r⁴ + m2 * r³ + m3 * r² + m4 * r
+# Rewrite the polynominal sum of product as follows,
+# h1 = (h0 + m1) * r^2,	h2 = (h0 + m2) * r^2
+# h3 = (h1 + m3) * r^2,	h4 = (h2 + m4) * r^2  --> (h0 + m1) r*4 + (h3 + m3) r^2, (h0 + m2) r^4 + (h0 + m4) r^2
+#  .... Repeat
+# h5 = (h3 + m5) * r^2,	h6 = (h4 + m6) * r^2  -->
+# h7 = (h5 + m7) * r^2,	h8 = (h6 + m8) * r^1  --> m5 * r^4 + m6 * r^3 + m7 * r^2 + m8 * r
+#
+loop_4blocks:
+
+	# Multiply odd words and even words
+	mul_odd
+	mul_even
+	# carry reduction
+	vspltisb 9, 2
+	vsrd	10, 14, 31
+	vsrd	11, 17, 31
+	vand	7, 17, 25
+	vand	4, 14, 25
+	vaddudm	18, 18, 11
+	vsrd	12, 18, 31
+	vaddudm	15, 15, 10
+
+	vsrd	11, 15, 31
+	vand	8, 18, 25
+	vand	5, 15, 25
+	vaddudm	4, 4, 12
+	vsld	10, 12, 9
+	vaddudm	6, 16, 11
+
+	vsrd	13, 6, 31
+	vand	6, 6, 25
+	vaddudm	4, 4, 10
+	vsrd	10, 4, 31
+	vaddudm	7, 7, 13
+
+	vsrd	11, 7, 31
+	vand	7, 7, 25
+	vand	4, 4, 25
+	vaddudm	5, 5, 10
+	vaddudm	8, 8, 11
+
+	# input m1  m2  m3  m4
+	add	20, 4, 21
+	xxlor	49, 24, 24
+	xxlor	50, 25, 25
+	lxvw4x	43, 0, 20
+	addi	17, 20, 16
+	lxvw4x	44, 0, 17
+	vperm	14, 11, 12, 17
+	vperm	15, 11, 12, 18
+	addi	17, 17, 16
+	lxvw4x	43, 0, 17
+	addi	17, 17, 16
+	lxvw4x	44, 0, 17
+	vperm	17, 11, 12, 17
+	vperm	18, 11, 12, 18
+
+	vand	20, 14, 25	# a0
+	vand	9, 17, 25	# a0
+	vsrd	21, 14, 31	# >> 26
+	vsrd	22, 21, 31	# 12 bits left
+	vsrd	10, 17, 31	# >> 26
+	vsrd	11, 10, 31	# 12 bits left
+
+	vand	21, 21, 25	# a1
+	vand	10, 10, 25	# a1
+
+	vspltisb 13, 12
+	vand	16, 15, 25
+	vsld	23, 16, 13
+	vor	22, 22, 23
+	vand	22, 22, 25	# a2
+	vand	16, 18, 25
+	vsld	12, 16, 13
+	vor	11, 11, 12
+	vand	11, 11, 25	# a2
+	vspltisb 13, 14
+	vsrd	23, 15, 13	# >> 14
+	vsrd	24, 23, 31	# >> 26, a4
+	vand	23, 23, 25	# a3
+	vsrd	12, 18, 13	# >> 14
+	vsrd	13, 12, 31	# >> 26, a4
+	vand	12, 12, 25	# a3
+
+	vaddudm	4, 4, 20
+	vaddudm	5, 5, 21
+	vaddudm	6, 6, 22
+	vaddudm	7, 7, 23
+	vaddudm	8, 8, 24
+
+	# Smash 4 message blocks into 5 vectors of [m4,  m2,  m3,  m1]
+	vmrgow	4, 9, 4
+	vmrgow	5, 10, 5
+	vmrgow	6, 11, 6
+	vmrgow	7, 12, 7
+	vmrgow	8, 13, 8
+	vaddudm	8, 8, 19
+
+	addi	5, 5, -64	# len -= 64
+	addi	21, 21, 64	# offset += 64
+
+	bdnz	loop_4blocks
+
+Skip_block_loop:
+	xxlor	58, 0, 0
+	xxlor	59, 1, 1
+	xxlor	60, 2, 2
+	xxlor	61, 3, 3
+	xxlor	62, 4, 4
+	xxlor	32, 5, 5
+	xxlor	33, 6, 6
+	xxlor	34, 7, 7
+	xxlor	35, 8, 8
+
+	# Multiply odd words and even words
+	mul_odd
+	mul_even
+
+	# Sum the products.
+	xxpermdi 41, 31, 46, 0
+	xxpermdi 42, 31, 47, 0
+	vaddudm	4, 14, 9
+	xxpermdi 36, 31, 36, 3
+	vaddudm	5, 15, 10
+	xxpermdi 37, 31, 37, 3
+	xxpermdi 43, 31, 48, 0
+	vaddudm	6, 16, 11
+	xxpermdi 38, 31, 38, 3
+	xxpermdi 44, 31, 49, 0
+	vaddudm	7, 17, 12
+	xxpermdi 39, 31, 39, 3
+	xxpermdi 45, 31, 50, 0
+	vaddudm	8, 18, 13
+	xxpermdi 40, 31, 40, 3
+
+	# carry reduction
+	vspltisb 9, 2
+	vsrd	10, 4, 31
+	vsrd	11, 7, 31
+	vand	7, 7, 25
+	vand	4, 4, 25
+	vaddudm	8, 8, 11
+	vsrd	12, 8, 31
+	vaddudm	5, 5, 10
+
+	vsrd	11, 5, 31
+	vand	8, 8, 25
+	vand	5, 5, 25
+	vaddudm	4, 4, 12
+	vsld	10, 12, 9
+	vaddudm	6, 6, 11
+
+	vsrd	13, 6, 31
+	vand	6, 6, 25
+	vaddudm	4, 4, 10
+	vsrd	10, 4, 31
+	vaddudm	7, 7, 13
+
+	vsrd	11, 7, 31
+	vand	7, 7, 25
+	vand	4, 4, 25
+	vaddudm	5, 5, 10
+	vsrd	10, 5, 31
+	vand	5, 5, 25
+	vaddudm	6, 6, 10
+	vaddudm	8, 8, 11
+
+	b	do_final_update
+
+do_final_update:
+	# combine 26 bit limbs
+	# v4, v5, v6, v7 and v8 are 26 bit vectors
+	vsld	5, 5, 31
+	vor	20, 4, 5
+	vspltisb 11, 12
+	vsrd	12, 6, 11
+	vsld	6, 6, 31
+	vsld	6, 6, 31
+	vor	20, 20, 6
+	vspltisb 11, 14
+	vsld	7, 7, 11
+	vor	21, 7, 12
+	mfvsrld	16, 40		# save last 2 bytes
+	vsld	8, 8, 11
+	vsld	8, 8, 31
+	vor	21, 21, 8
+	mfvsrld	17, 52
+	mfvsrld	19, 53
+	srdi	16, 16, 24
+
+	std	17, 0(3)
+	std	19, 8(3)
+	stw	16, 16(3)
+
+Out_loop:
+	li	3, 0
+
+	RESTORE_REGS
+
+	blr
+
+Out_no_poly1305:
+	li	3, 0
+	blr
+SYM_FUNC_END(poly1305_p10le_4blocks)
+
+#
+# =======================================================================
+# The following functions implement 64 x 64 bits multiplication poly1305.
+#
+SYM_FUNC_START_LOCAL(Poly1305_init_64)
+	#  mask 0x0FFFFFFC0FFFFFFC
+	#  mask 0x0FFFFFFC0FFFFFFF
+	addis	10, 2, rmask@toc@ha
+	addi	10, 10, rmask@toc@l
+	ld	11, 0(10)
+	ld	12, 8(10)
+
+	# initialize
+	# load key from r3
+	ld	9, 24(3)
+	ld	10, 32(3)
+	and.	9, 9, 11	# cramp mask r0
+	and.	10, 10, 12	# cramp mask r1
+
+        srdi    21, 10, 2
+        add     19, 21, 10      # s1: r19 - (r1 >> 2) *5
+
+        # setup r and s
+        li      25, 0
+	mtvsrdd 32+0, 9, 19	# r0, s1
+	mtvsrdd 32+1, 10, 9	# r1, r0
+	mtvsrdd 32+2, 19, 25	# s1
+	mtvsrdd 32+3, 9, 25	# r0
+
+	blr
+SYM_FUNC_END(Poly1305_init_64)
+
+# Poly1305_mult
+# v6 = (h0, h1), v8 = h2
+# v0 = (r0, s1), v1 = (r1, r0), v2 = s1, v3 = r0
+#
+# Output: v7, v10, v11
+#
+SYM_FUNC_START_LOCAL(Poly1305_mult)
+	#
+	#	d0 = h0 * r0 + h1 * s1
+	vmsumudm	7, 6, 0, 9		# h0 * r0, h1 * s1
+
+	#	d1 = h0 * r1 + h1 * r0 + h2 * s1
+	vmsumudm	11, 6, 1, 9		# h0 * r1, h1 * r0
+	vmsumudm	10, 8, 2, 11		# d1 += h2 * s1
+
+	#       d2 = r0
+	vmsumudm	11, 8, 3, 9		# d2 = h2 * r0
+	blr
+SYM_FUNC_END(Poly1305_mult)
+
+#
+# carry reduction
+# h %=p
+#
+# Input: v7, v10, v11
+# Output: r27, r28, r29
+#
+SYM_FUNC_START_LOCAL(Carry_reduction)
+	mfvsrld	27, 32+7
+	mfvsrld	28, 32+10
+	mfvsrld	29, 32+11
+	mfvsrd	20, 32+7	# h0.h
+	mfvsrd	21, 32+10	# h1.h
+
+	addc	28, 28, 20
+	adde	29, 29, 21
+	srdi	22, 29, 0x2
+	sldi	23, 22, 0x2
+	add	23, 23, 22	# (h2 & 3) * 5
+	addc	27, 27, 23	# h0
+	addze	28, 28		# h1
+	andi.	29, 29, 0x3	# h2
+	blr
+SYM_FUNC_END(Carry_reduction)
+
+#
+# poly1305 multiplication
+# h *= r, h %= p
+#	d0 = h0 * r0 + h1 * s1
+#	d1 = h0 * r1 + h1 * r0 + h2 * s1
+#       d2 = h0 * r0
+#
+#
+# unsigned int poly1305_test_64s(unisgned char *state, const byte *src, size_t len, highbit)
+#   - no highbit if final leftover block (highbit = 0)
+#
+SYM_FUNC_START(poly1305_64s)
+	cmpdi	5, 0
+	ble	Out_no_poly1305_64
+
+	mflr 0
+	std 0, 16(1)
+	stdu 1,-400(1)
+
+	SAVE_GPR 14, 112, 1
+	SAVE_GPR 15, 120, 1
+	SAVE_GPR 16, 128, 1
+	SAVE_GPR 17, 136, 1
+	SAVE_GPR 18, 144, 1
+	SAVE_GPR 19, 152, 1
+	SAVE_GPR 20, 160, 1
+	SAVE_GPR 21, 168, 1
+	SAVE_GPR 22, 176, 1
+	SAVE_GPR 23, 184, 1
+	SAVE_GPR 24, 192, 1
+	SAVE_GPR 25, 200, 1
+	SAVE_GPR 26, 208, 1
+	SAVE_GPR 27, 216, 1
+	SAVE_GPR 28, 224, 1
+	SAVE_GPR 29, 232, 1
+	SAVE_GPR 30, 240, 1
+	SAVE_GPR 31, 248, 1
+
+	# Init poly1305
+	bl Poly1305_init_64
+
+	li 25, 0			# offset to inp and outp
+
+	add 11, 25, 4
+
+	# load h
+	# h0, h1, h2?
+        ld	27, 0(3)
+        ld	28, 8(3)
+        lwz	29, 16(3)
+
+        li      30, 16
+        divdu   31, 5, 30
+
+        mtctr   31
+
+        mr      24, 6		# highbit
+
+Loop_block_64:
+	vxor	9, 9, 9
+
+	ld	20, 0(11)
+	ld	21, 8(11)
+	addi	11, 11, 16
+
+	addc	27, 27, 20
+	adde	28, 28, 21
+	adde	29, 29, 24
+
+	li	22, 0
+	mtvsrdd	32+6, 27, 28	# h0, h1
+	mtvsrdd	32+8, 29, 22	# h2
+
+	bl	Poly1305_mult
+
+	bl	Carry_reduction
+
+	bdnz	Loop_block_64
+
+	std	27, 0(3)
+	std	28, 8(3)
+	stw	29, 16(3)
+
+	li	3, 0
+
+	RESTORE_GPR 14, 112, 1
+	RESTORE_GPR 15, 120, 1
+	RESTORE_GPR 16, 128, 1
+	RESTORE_GPR 17, 136, 1
+	RESTORE_GPR 18, 144, 1
+	RESTORE_GPR 19, 152, 1
+	RESTORE_GPR 20, 160, 1
+	RESTORE_GPR 21, 168, 1
+	RESTORE_GPR 22, 176, 1
+	RESTORE_GPR 23, 184, 1
+	RESTORE_GPR 24, 192, 1
+	RESTORE_GPR 25, 200, 1
+	RESTORE_GPR 26, 208, 1
+	RESTORE_GPR 27, 216, 1
+	RESTORE_GPR 28, 224, 1
+	RESTORE_GPR 29, 232, 1
+	RESTORE_GPR 30, 240, 1
+	RESTORE_GPR 31, 248, 1
+
+	addi    1, 1, 400
+	ld 0, 16(1)
+	mtlr 0
+
+	blr
+
+Out_no_poly1305_64:
+	li	3, 0
+	blr
+SYM_FUNC_END(poly1305_64s)
+
+#
+# Input: r3 = h, r4 = s, r5 = mac
+# mac = h + s
+#
+SYM_FUNC_START(poly1305_emit_64)
+	ld	10, 0(3)
+	ld	11, 8(3)
+	ld	12, 16(3)
+
+	# compare modulus
+	# h + 5 + (-p)
+	mr	6, 10
+	mr	7, 11
+	mr	8, 12
+	addic.	6, 6, 5
+	addze	7, 7
+	addze	8, 8
+	srdi	9, 8, 2		# overflow?
+	cmpdi	9, 0
+	beq	Skip_h64
+	mr	10, 6
+	mr	11, 7
+	mr	12, 8
+
+Skip_h64:
+	ld	6, 0(4)
+	ld	7, 8(4)
+	addc	10, 10, 6
+	adde	11, 11, 7
+	addze	12, 12
+
+	std	10, 0(5)
+	std	11, 8(5)
+	blr
+SYM_FUNC_END(poly1305_emit_64)
+
+SYM_DATA_START_LOCAL(RMASK)
+.align 5
+rmask:
+.byte	0xff, 0xff, 0xff, 0x0f, 0xfc, 0xff, 0xff, 0x0f, 0xfc, 0xff, 0xff, 0x0f, 0xfc, 0xff, 0xff, 0x0f
+cnum:
+.long	0x03ffffff, 0x00000000, 0x03ffffff, 0x00000000
+.long	0x1a, 0x00, 0x1a, 0x00
+.long	0x01000000, 0x01000000, 0x01000000, 0x01000000
+.long	0x00010203, 0x04050607, 0x10111213, 0x14151617
+.long	0x08090a0b, 0x0c0d0e0f, 0x18191a1b, 0x1c1d1e1f
+SYM_DATA_END(RMASK)