1 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
2 ; Copyright(c) 2011-2015 Intel Corporation All rights reserved.
4 ; Redistribution and use in source and binary forms, with or without
5 ; modification, are permitted provided that the following conditions
7 ; * Redistributions of source code must retain the above copyright
8 ; notice, this list of conditions and the following disclaimer.
9 ; * Redistributions in binary form must reproduce the above copyright
10 ; notice, this list of conditions and the following disclaimer in
11 ; the documentation and/or other materials provided with the
13 ; * Neither the name of Intel Corporation nor the names of its
14 ; contributors may be used to endorse or promote products derived
15 ; from this software without specific prior written permission.
17 ; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 ; "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 ; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 ; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21 ; OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 ; SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23 ; LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 ; DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 ; THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 ; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 ; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
31 ; UINT16 crc16_t10dif_by4(
32 ; UINT16 init_crc, //initial CRC value, 16 bits
33 ; const unsigned char *buf, //buffer pointer to calculate CRC on
34 ; UINT64 len //buffer length in bytes (64-bit data)
42 ; Reference paper titled "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction"
43 ; URL: http://download.intel.com/design/intarch/papers/323102.pdf
46 %include "reg_sizes.asm"
48 %define fetch_dist 1024
54 %ifidn __OUTPUT_FORMAT__, win64
59 %xdefine arg1_low32 ecx
65 %xdefine arg1_low32 edi
69 mk_global crc16_t10dif_by4, function
73 ; adjust the 16-bit initial_crc value, scale it to 32 bits
76 ; After this point, code flow is exactly same as a 32-bit CRC.
77 ; The only difference is before returning eax, we will shift
78 ; it right 16 bits, to scale back to 16 bits.
82 ; push the xmm registers into the stack to maintain
83 movdqa [rsp+16*2],xmm6
84 movdqa [rsp+16*3],xmm7
86 ; check if smaller than 128B
89 ; for sizes less than 128, we can't fold 64B at a time...
93 ; load the initial crc value
94 movd xmm6, arg1_low32 ; initial crc
96 ; crc value does not need to be byte-reflected, but it needs to
97 ; be moved to the high part of the register.
98 ; because data will be byte-reflected and will align with
99 ; initial crc at correct place.
102 movdqa xmm7, [SHUF_MASK]
103 ; receive the initial 64B data, xor the initial crc value
105 movdqu xmm1, [arg2+16]
106 movdqu xmm2, [arg2+32]
107 movdqu xmm3, [arg2+48]
110 ; XOR the initial_crc value
116 movdqa xmm6, [rk3] ;xmm6 has rk3 and rk4
117 ;imm value of pclmulqdq instruction
118 ;will determine which constant to use
119 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
120 ; we subtract 128 instead of 64 to save one instruction from the loop
123 ; at this section of the code, there is 64*x+y (0<=y<64) bytes of
124 ; buffer. The _fold_64_B_loop
125 ; loop will fold 64B at a time until we have 64+y Bytes of buffer
128 ; fold 64B at a time. This section of the code folds 4 xmm
129 ; registers in parallel
132 ; update the buffer pointer
133 add arg2, 64 ; buf += 64;
135 prefetchnta [arg2+fetch_dist+0]
139 pclmulqdq xmm0, xmm6 , 0x11
140 pclmulqdq xmm1, xmm6 , 0x11
142 pclmulqdq xmm4, xmm6, 0x0
143 pclmulqdq xmm5, xmm6, 0x0
148 prefetchnta [arg2+fetch_dist+32]
152 pclmulqdq xmm2, xmm6, 0x11
153 pclmulqdq xmm3, xmm6, 0x11
155 pclmulqdq xmm4, xmm6, 0x0
156 pclmulqdq xmm5, xmm6, 0x0
162 movdqu xmm5, [arg2+16]
168 movdqu xmm4, [arg2+32]
169 movdqu xmm5, [arg2+48]
178 ; check if there is another 64B in the buffer to be able to fold
180 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
184 ; at this point, the buffer pointer is pointing at the last y Bytes of the buffer
185 ; the 64B of folded data is in 4 of the xmm registers: xmm0, xmm1, xmm2, xmm3
188 ; fold the 4 xmm registers to 1 xmm register with different constants
190 movdqa xmm6, [rk1] ;xmm6 has rk1 and rk2
191 ;imm value of pclmulqdq instruction will
192 ;determine which constant to use
195 pclmulqdq xmm0, xmm6, 0x11
196 pclmulqdq xmm4, xmm6, 0x0
201 pclmulqdq xmm1, xmm6, 0x11
202 pclmulqdq xmm4, xmm6, 0x0
207 pclmulqdq xmm2, xmm6, 0x11
208 pclmulqdq xmm4, xmm6, 0x0
213 ; instead of 64, we add 48 to the loop counter to save 1 instruction from the loop
214 ; instead of a cmp instruction, we use the negative flag with the jl instruction
216 jl _final_reduction_for_128
218 ; now we have 16+y bytes left to reduce. 16 Bytes
219 ; is in register xmm3 and the rest is in memory
220 ; we can fold 16 bytes at a time if y>=16
221 ; continue folding 16B at a time
225 pclmulqdq xmm3, xmm6, 0x11
226 pclmulqdq xmm4, xmm6, 0x0
233 ; instead of a cmp instruction, we utilize the flags with the jge instruction
234 ; equivalent of: cmp arg3, 16-16
235 ; check if there is any more 16B in the buffer to be able to fold
236 jge _16B_reduction_loop
238 ;now we have 16+z bytes left to reduce, where 0<= z < 16.
239 ;first, we reduce the data in the xmm3 register
242 _final_reduction_for_128:
243 ; check if any more data to fold. If not, compute the CRC of the final 128 bits
247 ; here we are getting data that is less than 16 bytes.
248 ; since we know that there was data before the pointer,
249 ; we can offset the input pointer before the actual point,
250 ; to receive exactly 16 bytes.
251 ; after that the registers need to be adjusted.
255 movdqu xmm1, [arg2 - 16 + arg3]
258 ; get rid of the extra data that was loaded before
259 ; load the shift constant
260 lea rax, [pshufb_shf_table + 16]
264 ; shift xmm2 to the left by arg3 bytes
267 ; shift xmm3 to the right by 16-arg3 bytes
270 pblendvb xmm1, xmm2 ;xmm0 is implicit
275 pclmulqdq xmm3, xmm6, 0x11
276 pclmulqdq xmm4, xmm6, 0x0
281 ; compute crc of a 128-bit value
282 movdqa xmm6, [rk5] ; rk5 and rk6 in xmm6
286 pclmulqdq xmm3, xmm6, 0x1
296 pclmulqdq xmm3, xmm6, 0x10
301 movdqa xmm6, [rk7] ; rk7 and rk8 in xmm6
303 pclmulqdq xmm3, xmm6, 0x01
305 pclmulqdq xmm3, xmm6, 0x11
312 ; scale the result back to 16 bits
314 movdqa xmm6, [rsp+16*2]
315 movdqa xmm7, [rsp+16*3]
320 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
321 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
322 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
323 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
328 ; check if there is enough buffer to be able to fold 16B at a time
331 movdqa xmm7, [SHUF_MASK]
333 ; if there is, load the constants
334 movdqa xmm6, [rk1] ; rk1 and rk2 in xmm6
336 movd xmm0, arg1_low32 ; get the initial crc value
337 pslldq xmm0, 12 ; align it to its correct place
338 movdqu xmm3, [arg2] ; load the plaintext
339 pshufb xmm3, xmm7 ; byte-reflect the plaintext
343 ; update the buffer pointer
346 ; update the counter. subtract 32 instead of 16 to save one instruction from the loop
349 jmp _16B_reduction_loop
354 ; mov initial crc to the return value. this is necessary for zero-length buffers.
359 movdqa xmm7, [SHUF_MASK]
361 movd xmm0, arg1_low32 ; get the initial crc value
362 pslldq xmm0, 12 ; align it to its correct place
366 jl _less_than_16_left
368 movdqu xmm3, [arg2] ; load the plaintext
369 pshufb xmm3, xmm7 ; byte-reflect the plaintext
370 pxor xmm3, xmm0 ; xor the initial crc value
373 movdqa xmm6, [rk1] ; rk1 and rk2 in xmm6
374 jmp _get_last_two_xmms
379 ; use stack space to load data less than 16 bytes, zero-out the 16B in memory first.
388 ; backup the counter value
431 pxor xmm3, xmm0 ; xor the initial crc value
434 lea rax, [pshufb_shf_table + 16]
446 pxor xmm3, xmm0 ; xor the initial crc value
466 pxor xmm3, xmm0 ; xor the initial crc value
484 pxor xmm3, xmm0 ; xor the initial crc value
497 pxor xmm3, xmm0 ; xor the initial crc value
505 ; precomputed constants
506 ; these constants are precomputed from the poly: 0x8bb70000 (0x8bb7 scaled to 32 bits)
509 ; rk1 = 2^(32*3) mod Q << 32
510 ; rk2 = 2^(32*5) mod Q << 32
511 ; rk3 = 2^(32*15) mod Q << 32
512 ; rk4 = 2^(32*17) mod Q << 32
513 ; rk5 = 2^(32*3) mod Q << 32
514 ; rk6 = 2^(32*2) mod Q << 32
515 ; rk7 = floor(2^64/Q)
518 DQ 0x2d56000000000000
520 DQ 0x06df000000000000
522 DQ 0x044c000000000000
524 DQ 0xe658000000000000
526 DQ 0x2d56000000000000
528 DQ 0x1368000000000000
530 DQ 0x00000001f65a57f8
532 DQ 0x000000018bb70000
534 dq 0x8080808080808080, 0x8080808080808080
536 dq 0xFFFFFFFFFFFFFFFF, 0x00000000FFFFFFFF
539 dq 0x08090A0B0C0D0E0F, 0x0001020304050607
542 ; use these values for shift constants for the pshufb instruction
543 ; different alignments result in values as shown:
544 ; dq 0x8887868584838281, 0x008f8e8d8c8b8a89 ; shl 15 (16-1) / shr1
545 ; dq 0x8988878685848382, 0x01008f8e8d8c8b8a ; shl 14 (16-3) / shr2
546 ; dq 0x8a89888786858483, 0x0201008f8e8d8c8b ; shl 13 (16-4) / shr3
547 ; dq 0x8b8a898887868584, 0x030201008f8e8d8c ; shl 12 (16-4) / shr4
548 ; dq 0x8c8b8a8988878685, 0x04030201008f8e8d ; shl 11 (16-5) / shr5
549 ; dq 0x8d8c8b8a89888786, 0x0504030201008f8e ; shl 10 (16-6) / shr6
550 ; dq 0x8e8d8c8b8a898887, 0x060504030201008f ; shl 9 (16-7) / shr7
551 ; dq 0x8f8e8d8c8b8a8988, 0x0706050403020100 ; shl 8 (16-8) / shr8
552 ; dq 0x008f8e8d8c8b8a89, 0x0807060504030201 ; shl 7 (16-9) / shr9
553 ; dq 0x01008f8e8d8c8b8a, 0x0908070605040302 ; shl 6 (16-10) / shr10
554 ; dq 0x0201008f8e8d8c8b, 0x0a09080706050403 ; shl 5 (16-11) / shr11
555 ; dq 0x030201008f8e8d8c, 0x0b0a090807060504 ; shl 4 (16-12) / shr12
556 ; dq 0x04030201008f8e8d, 0x0c0b0a0908070605 ; shl 3 (16-13) / shr13
557 ; dq 0x0504030201008f8e, 0x0d0c0b0a09080706 ; shl 2 (16-14) / shr14
558 ; dq 0x060504030201008f, 0x0e0d0c0b0a090807 ; shl 1 (16-15) / shr15
559 dq 0x8786858483828100, 0x8f8e8d8c8b8a8988
560 dq 0x0706050403020100, 0x000e0d0c0b0a0908
562 ;;; func core, ver, snum
563 slversion crc16_t10dif_by4, 05, 02, 0016