1// This file is generated from a similarly-named Perl script in the BoringSSL
2// source tree. Do not edit by hand.
3
4#if !defined(__has_feature)
5#define __has_feature(x) 0
6#endif
7#if __has_feature(memory_sanitizer) && !defined(OPENSSL_NO_ASM)
8#define OPENSSL_NO_ASM
9#endif
10
11#if !defined(OPENSSL_NO_ASM)
12#if defined(__arm__)
13#include <GFp/arm_arch.h>
14
15.text
16.fpu	neon
17.code	32
18#undef	__thumb2__
19.globl	GFp_gcm_init_clmul
20.hidden	GFp_gcm_init_clmul
21.type	GFp_gcm_init_clmul,%function
22.align	4
23GFp_gcm_init_clmul:
24	AARCH64_VALID_CALL_TARGET
25	vld1.64	{q9},[r1]		@ load input H
26	vmov.i8	q11,#0xe1
27	vshl.i64	q11,q11,#57		@ 0xc2.0
28	vext.8	q3,q9,q9,#8
29	vshr.u64	q10,q11,#63
30	vdup.32	q9,d18[1]
31	vext.8	q8,q10,q11,#8		@ t0=0xc2....01
32	vshr.u64	q10,q3,#63
33	vshr.s32	q9,q9,#31		@ broadcast carry bit
34	vand	q10,q10,q8
35	vshl.i64	q3,q3,#1
36	vext.8	q10,q10,q10,#8
37	vand	q8,q8,q9
38	vorr	q3,q3,q10		@ H<<<=1
39	veor	q12,q3,q8		@ twisted H
40	vst1.64	{q12},[r0]!		@ store Htable[0]
41
42	@ calculate H^2
43	vext.8	q8,q12,q12,#8		@ Karatsuba pre-processing
44.byte	0xa8,0x0e,0xa8,0xf2	@ pmull q0,q12,q12
45	veor	q8,q8,q12
46.byte	0xa9,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q12
47.byte	0xa0,0x2e,0xa0,0xf2	@ pmull q1,q8,q8
48
49	vext.8	q9,q0,q2,#8		@ Karatsuba post-processing
50	veor	q10,q0,q2
51	veor	q1,q1,q9
52	veor	q1,q1,q10
53.byte	0x26,0x4e,0xe0,0xf2	@ pmull q10,q0,q11		@ 1st phase
54
55	vmov	d4,d3		@ Xh|Xm - 256-bit result
56	vmov	d3,d0		@ Xm is rotated Xl
57	veor	q0,q1,q10
58
59	vext.8	q10,q0,q0,#8		@ 2nd phase
60.byte	0x26,0x0e,0xa0,0xf2	@ pmull q0,q0,q11
61	veor	q10,q10,q2
62	veor	q14,q0,q10
63
64	vext.8	q9,q14,q14,#8		@ Karatsuba pre-processing
65	veor	q9,q9,q14
66	vext.8	q13,q8,q9,#8		@ pack Karatsuba pre-processed
67	vst1.64	{q13,q14},[r0]		@ store Htable[1..2]
68
69	bx	lr
70.size	GFp_gcm_init_clmul,.-GFp_gcm_init_clmul
71.globl	GFp_gcm_gmult_clmul
72.hidden	GFp_gcm_gmult_clmul
73.type	GFp_gcm_gmult_clmul,%function
74.align	4
75GFp_gcm_gmult_clmul:
76	AARCH64_VALID_CALL_TARGET
77	vld1.64	{q9},[r0]		@ load Xi
78	vmov.i8	q11,#0xe1
79	vld1.64	{q12,q13},[r1]	@ load twisted H, ...
80	vshl.u64	q11,q11,#57
81#ifndef __ARMEB__
82	vrev64.8	q9,q9
83#endif
84	vext.8	q3,q9,q9,#8
85
86.byte	0x86,0x0e,0xa8,0xf2	@ pmull q0,q12,q3		@ H.lo·Xi.lo
87	veor	q9,q9,q3		@ Karatsuba pre-processing
88.byte	0x87,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q3		@ H.hi·Xi.hi
89.byte	0xa2,0x2e,0xaa,0xf2	@ pmull q1,q13,q9		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
90
91	vext.8	q9,q0,q2,#8		@ Karatsuba post-processing
92	veor	q10,q0,q2
93	veor	q1,q1,q9
94	veor	q1,q1,q10
95.byte	0x26,0x4e,0xe0,0xf2	@ pmull q10,q0,q11		@ 1st phase of reduction
96
97	vmov	d4,d3		@ Xh|Xm - 256-bit result
98	vmov	d3,d0		@ Xm is rotated Xl
99	veor	q0,q1,q10
100
101	vext.8	q10,q0,q0,#8		@ 2nd phase of reduction
102.byte	0x26,0x0e,0xa0,0xf2	@ pmull q0,q0,q11
103	veor	q10,q10,q2
104	veor	q0,q0,q10
105
106#ifndef __ARMEB__
107	vrev64.8	q0,q0
108#endif
109	vext.8	q0,q0,q0,#8
110	vst1.64	{q0},[r0]		@ write out Xi
111
112	bx	lr
113.size	GFp_gcm_gmult_clmul,.-GFp_gcm_gmult_clmul
114.globl	GFp_gcm_ghash_clmul
115.hidden	GFp_gcm_ghash_clmul
116.type	GFp_gcm_ghash_clmul,%function
117.align	4
118GFp_gcm_ghash_clmul:
119	AARCH64_VALID_CALL_TARGET
120	vstmdb	sp!,{d8,d9,d10,d11,d12,d13,d14,d15}		@ 32-bit ABI says so
121	vld1.64	{q0},[r0]		@ load [rotated] Xi
122						@ "[rotated]" means that
123						@ loaded value would have
124						@ to be rotated in order to
125						@ make it appear as in
126						@ algorithm specification
127	subs	r3,r3,#32		@ see if r3 is 32 or larger
128	mov	r12,#16		@ r12 is used as post-
129						@ increment for input pointer;
130						@ as loop is modulo-scheduled
131						@ r12 is zeroed just in time
132						@ to preclude overstepping
133						@ inp[len], which means that
134						@ last block[s] are actually
135						@ loaded twice, but last
136						@ copy is not processed
137	vld1.64	{q12,q13},[r1]!	@ load twisted H, ..., H^2
138	vmov.i8	q11,#0xe1
139	vld1.64	{q14},[r1]
140	moveq	r12,#0			@ is it time to zero r12?
141	vext.8	q0,q0,q0,#8		@ rotate Xi
142	vld1.64	{q8},[r2]!	@ load [rotated] I[0]
143	vshl.u64	q11,q11,#57		@ compose 0xc2.0 constant
144#ifndef __ARMEB__
145	vrev64.8	q8,q8
146	vrev64.8	q0,q0
147#endif
148	vext.8	q3,q8,q8,#8		@ rotate I[0]
149	blo	.Lodd_tail_v8		@ r3 was less than 32
150	vld1.64	{q9},[r2],r12	@ load [rotated] I[1]
151#ifndef __ARMEB__
152	vrev64.8	q9,q9
153#endif
154	vext.8	q7,q9,q9,#8
155	veor	q3,q3,q0		@ I[i]^=Xi
156.byte	0x8e,0x8e,0xa8,0xf2	@ pmull q4,q12,q7		@ H·Ii+1
157	veor	q9,q9,q7		@ Karatsuba pre-processing
158.byte	0x8f,0xce,0xa9,0xf2	@ pmull2 q6,q12,q7
159	b	.Loop_mod2x_v8
160
161.align	4
162.Loop_mod2x_v8:
163	vext.8	q10,q3,q3,#8
164	subs	r3,r3,#32		@ is there more data?
165.byte	0x86,0x0e,0xac,0xf2	@ pmull q0,q14,q3		@ H^2.lo·Xi.lo
166	movlo	r12,#0			@ is it time to zero r12?
167
168.byte	0xa2,0xae,0xaa,0xf2	@ pmull q5,q13,q9
169	veor	q10,q10,q3		@ Karatsuba pre-processing
170.byte	0x87,0x4e,0xad,0xf2	@ pmull2 q2,q14,q3		@ H^2.hi·Xi.hi
171	veor	q0,q0,q4		@ accumulate
172.byte	0xa5,0x2e,0xab,0xf2	@ pmull2 q1,q13,q10		@ (H^2.lo+H^2.hi)·(Xi.lo+Xi.hi)
173	vld1.64	{q8},[r2],r12	@ load [rotated] I[i+2]
174
175	veor	q2,q2,q6
176	moveq	r12,#0			@ is it time to zero r12?
177	veor	q1,q1,q5
178
179	vext.8	q9,q0,q2,#8		@ Karatsuba post-processing
180	veor	q10,q0,q2
181	veor	q1,q1,q9
182	vld1.64	{q9},[r2],r12	@ load [rotated] I[i+3]
183#ifndef __ARMEB__
184	vrev64.8	q8,q8
185#endif
186	veor	q1,q1,q10
187.byte	0x26,0x4e,0xe0,0xf2	@ pmull q10,q0,q11		@ 1st phase of reduction
188
189#ifndef __ARMEB__
190	vrev64.8	q9,q9
191#endif
192	vmov	d4,d3		@ Xh|Xm - 256-bit result
193	vmov	d3,d0		@ Xm is rotated Xl
194	vext.8	q7,q9,q9,#8
195	vext.8	q3,q8,q8,#8
196	veor	q0,q1,q10
197.byte	0x8e,0x8e,0xa8,0xf2	@ pmull q4,q12,q7		@ H·Ii+1
198	veor	q3,q3,q2		@ accumulate q3 early
199
200	vext.8	q10,q0,q0,#8		@ 2nd phase of reduction
201.byte	0x26,0x0e,0xa0,0xf2	@ pmull q0,q0,q11
202	veor	q3,q3,q10
203	veor	q9,q9,q7		@ Karatsuba pre-processing
204	veor	q3,q3,q0
205.byte	0x8f,0xce,0xa9,0xf2	@ pmull2 q6,q12,q7
206	bhs	.Loop_mod2x_v8		@ there was at least 32 more bytes
207
208	veor	q2,q2,q10
209	vext.8	q3,q8,q8,#8		@ re-construct q3
210	adds	r3,r3,#32		@ re-construct r3
211	veor	q0,q0,q2		@ re-construct q0
212	beq	.Ldone_v8		@ is r3 zero?
213.Lodd_tail_v8:
214	vext.8	q10,q0,q0,#8
215	veor	q3,q3,q0		@ inp^=Xi
216	veor	q9,q8,q10		@ q9 is rotated inp^Xi
217
218.byte	0x86,0x0e,0xa8,0xf2	@ pmull q0,q12,q3		@ H.lo·Xi.lo
219	veor	q9,q9,q3		@ Karatsuba pre-processing
220.byte	0x87,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q3		@ H.hi·Xi.hi
221.byte	0xa2,0x2e,0xaa,0xf2	@ pmull q1,q13,q9		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
222
223	vext.8	q9,q0,q2,#8		@ Karatsuba post-processing
224	veor	q10,q0,q2
225	veor	q1,q1,q9
226	veor	q1,q1,q10
227.byte	0x26,0x4e,0xe0,0xf2	@ pmull q10,q0,q11		@ 1st phase of reduction
228
229	vmov	d4,d3		@ Xh|Xm - 256-bit result
230	vmov	d3,d0		@ Xm is rotated Xl
231	veor	q0,q1,q10
232
233	vext.8	q10,q0,q0,#8		@ 2nd phase of reduction
234.byte	0x26,0x0e,0xa0,0xf2	@ pmull q0,q0,q11
235	veor	q10,q10,q2
236	veor	q0,q0,q10
237
238.Ldone_v8:
239#ifndef __ARMEB__
240	vrev64.8	q0,q0
241#endif
242	vext.8	q0,q0,q0,#8
243	vst1.64	{q0},[r0]		@ write out Xi
244
245	vldmia	sp!,{d8,d9,d10,d11,d12,d13,d14,d15}		@ 32-bit ABI says so
246	bx	lr
247.size	GFp_gcm_ghash_clmul,.-GFp_gcm_ghash_clmul
248.byte	71,72,65,83,72,32,102,111,114,32,65,82,77,118,56,44,32,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0
249.align	2
250.align	2
251#endif
252#endif  // !OPENSSL_NO_ASM
253.section	.note.GNU-stack,"",%progbits
254