Lines Matching +full:flip +full:- +full:horizontal
1 /* Do not modify. This file is auto-generated from poly1305-armv8.pl. */
34 rev x7,x7 // flip bytes
38 and x9,x9,#-4
61 .size poly1305_init,.-poly1305_init
70 ands x2,x2,#-16
115 and x10,x14,#-4 // final reduction
129 .size poly1305_blocks,.-poly1305_blocks
146 tst x14,#-4 // see if it's carried/borrowed
152 ror x10,x10,#32 // flip nonce words
158 rev x4,x4 // flip output bytes
164 .size poly1305_emit,.-poly1305_emit
192 and x10,x14,#-4 // final reduction
200 .size poly1305_mult,.-poly1305_mult
205 and x12,x4,#0x03ffffff // base 2^64 -> base 2^26
227 .size poly1305_splat,.-poly1305_splat
243 stp x29,x30,[sp,#-80]!
246 ands x2,x2,#-16
260 add x4,x10,x11,lsl#26 // base 2^26 -> base 2^64
273 and x10,x14,#-4 // ... so reduce
293 and x10,x4,#0x03ffffff // base 2^64 -> base 2^26
337 and x10,x4,#0x03ffffff // base 2^64 -> base 2^26
382 stur x4,[x0,#-24] // set is_base2_26
417 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
453 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
472 movi v31.2d,#-1
516 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
583 and x4,x8,#0x03ffffff // base 2^64 -> base 2^26
633 // [see discussion in poly1305-armv4 module]
639 add v23.2d,v23.2d,v29.2d // h3 -> h4
641 add v20.2d,v20.2d,v30.2d // h0 -> h1
648 add v21.2d,v21.2d,v30.2d // h1 -> h2
654 add v19.2d,v19.2d,v29.2d // h4 -> h0
656 add v27.2s,v27.2s,v30.2s // h2 -> h3
664 add v25.2s,v25.2s,v29.2s // h0 -> h1
665 add v28.2s,v28.2s,v30.2s // h3 -> h4
761 // horizontal add
781 add v23.2d,v23.2d,v29.2d // h3 -> h4
782 add v20.2d,v20.2d,v30.2d // h0 -> h1
788 add v21.2d,v21.2d,v30.2d // h1 -> h2
794 add v19.2d,v19.2d,v29.2d // h4 -> h0
795 add v22.2d,v22.2d,v30.2d // h2 -> h3
801 add v20.2d,v20.2d,v29.2d // h0 -> h1
802 add v23.2d,v23.2d,v30.2d // h3 -> h4
814 .size poly1305_blocks_neon,.-poly1305_blocks_neon
830 add x4,x10,x11,lsl#26 // base 2^26 -> base 2^64
841 and x12,x6,#-4 // ... so reduce
852 tst x14,#-4 // see if it's carried/borrowed
858 ror x10,x10,#32 // flip nonce words
864 rev x4,x4 // flip output bytes
870 .size poly1305_emit_neon,.-poly1305_emit_neon