generic-library/isa-l
1
0
Fork 0
mirror of https://github.com/intel/isa-l.git synced 2024-12-13 09:52:56 +01:00
Code Issues Projects Releases Wiki Activity

crc: Add vec version of crc16_t10dif_copy

Browse Source 
Change-Id: I5f73e8a38efd1ff50d30a39689d9d85da702e809
Signed-off-by: Greg Tucker <greg.b.tucker@intel.com>
This commit is contained in:
Greg Tucker 2020-01-29 14:16:37 -07:00
parent 02a41e0653
commit 4217930338
4 changed files with 599 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
Makefile.nmake
View File

@ -92,6 +92,7 @@ objs = \
bin\crc16_t10dif_by4.obj \
bin\crc16_t10dif_02.obj \
bin\crc16_t10dif_copy_by4.obj \
bin\crc16_t10dif_copy_by4_02.obj \
bin\crc32_ieee_01.obj \
bin\crc32_ieee_02.obj \
bin\crc32_ieee_by4.obj \

1
crc/Makefile.am
View File

@ -42,6 +42,7 @@ lsrc_x86_64 += \
crc/crc16_t10dif_by4.asm \
crc/crc16_t10dif_02.asm \
crc/crc16_t10dif_copy_by4.asm \
crc/crc16_t10dif_copy_by4_02.asm \
crc/crc32_ieee_01.asm \
crc/crc32_ieee_02.asm \
crc/crc32_ieee_by4.asm \

595
crc/crc16_t10dif_copy_by4_02.asm Normal file
View File

@ -0,0 +1,595 @@
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; Copyright(c) 2011-2020 Intel Corporation All rights reserved.
;
; Redistribution and use in source and binary forms, with or without
; modification, are permitted provided that the following conditions
; are met:
; * Redistributions of source code must retain the above copyright
; notice, this list of conditions and the following disclaimer.
; * Redistributions in binary form must reproduce the above copyright
; notice, this list of conditions and the following disclaimer in
; the documentation and/or other materials provided with the
; distribution.
; * Neither the name of Intel Corporation nor the names of its
; contributors may be used to endorse or promote products derived
; from this software without specific prior written permission.
;
; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
; "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
; OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
; SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
; LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
; DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
; THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
; OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;
; Function API:
; UINT16 crc16_t10dif_copy_by4_02(
; UINT16 init_crc, //initial CRC value, 16 bits
; unsigned char *dst, //buffer pointer destination for copy
; const unsigned char *src, //buffer pointer to calculate CRC on
; UINT64 len //buffer length in bytes (64-bit data)
; );
;
; Authors:
; Erdinc Ozturk
; Vinodh Gopal
; James Guilford
;
; Reference paper titled "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction"
; URL: http://download.intel.com/design/intarch/papers/323102.pdf
;
%include "reg_sizes.asm"
%define fetch_dist 1024
[bits 64]
default rel
section .text
%ifidn __OUTPUT_FORMAT__, win64
%xdefine arg1 rcx
%xdefine arg2 rdx
%xdefine arg3 r8
%xdefine arg4 r9
%xdefine tmp1 r10
%xdefine arg1_low32 ecx
%else
%xdefine arg1 rdi
%xdefine arg2 rsi
%xdefine arg3 rdx
%xdefine arg4 rcx
%xdefine tmp1 r10
%xdefine arg1_low32 edi
%endif
align 16
global crc16_t10dif_copy_by4_02:ISAL_SYM_TYPE_FUNCTION
crc16_t10dif_copy_by4_02:
; adjust the 16-bit initial_crc value, scale it to 32 bits
shl arg1_low32, 16
; After this point, code flow is exactly same as a 32-bit CRC.
; The only difference is before returning eax, we will shift
; it right 16 bits, to scale back to 16 bits.
sub rsp,16*4+8
; push the xmm registers into the stack to maintain
movdqa [rsp+16*2],xmm6
movdqa [rsp+16*3],xmm7
; check if smaller than 128B
cmp arg4, 128
; for sizes less than 128, we can't fold 64B at a time...
jl _less_than_128
; load the initial crc value
vmovd xmm6, arg1_low32 ; initial crc
; crc value does not need to be byte-reflected, but it needs to
; be moved to the high part of the register.
; because data will be byte-reflected and will align with
; initial crc at correct place.
vpslldq xmm6, 12
vmovdqa xmm7, [SHUF_MASK]
; receive the initial 64B data, xor the initial crc value
vmovdqu xmm0, [arg3]
vmovdqu xmm1, [arg3+16]
vmovdqu xmm2, [arg3+32]
vmovdqu xmm3, [arg3+48]
; copy initial data
vmovdqu [arg2], xmm0
vmovdqu [arg2+16], xmm1
vmovdqu [arg2+32], xmm2
vmovdqu [arg2+48], xmm3
vpshufb xmm0, xmm7
; XOR the initial_crc value
vpxor xmm0, xmm6
vpshufb xmm1, xmm7
vpshufb xmm2, xmm7
vpshufb xmm3, xmm7
vmovdqa xmm6, [rk3] ;xmm6 has rk3 and rk4
;imm value of pclmulqdq instruction
;will determine which constant to use
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; we subtract 128 instead of 64 to save one instruction from the loop
sub arg4, 128
; at this section of the code, there is 64*x+y (0<=y<64) bytes of
; buffer. The _fold_64_B_loop
; loop will fold 64B at a time until we have 64+y Bytes of buffer
; fold 64B at a time. This section of the code folds 4 xmm
; registers in parallel
_fold_64_B_loop:
; update the buffer pointer
add arg3, 64 ; buf += 64;
add arg2, 64
prefetchnta [arg3+fetch_dist+0]
vmovdqu xmm4, xmm0
vmovdqu xmm5, xmm1
vpclmulqdq xmm0, xmm6 , 0x11
vpclmulqdq xmm1, xmm6 , 0x11
vpclmulqdq xmm4, xmm6, 0x0
vpclmulqdq xmm5, xmm6, 0x0
vpxor xmm0, xmm4
vpxor xmm1, xmm5
prefetchnta [arg3+fetch_dist+32]
vmovdqu xmm4, xmm2
vmovdqu xmm5, xmm3
vpclmulqdq xmm2, xmm6, 0x11
vpclmulqdq xmm3, xmm6, 0x11
vpclmulqdq xmm4, xmm6, 0x0
vpclmulqdq xmm5, xmm6, 0x0
vpxor xmm2, xmm4
vpxor xmm3, xmm5
vmovdqu xmm4, [arg3]
vmovdqu xmm5, [arg3+16]
vmovdqu [arg2], xmm4
vmovdqu [arg2+16], xmm5
vpshufb xmm4, xmm7
vpshufb xmm5, xmm7
vpxor xmm0, xmm4
vpxor xmm1, xmm5
vmovdqu xmm4, [arg3+32]
vmovdqu xmm5, [arg3+48]
vmovdqu [arg2+32], xmm4
vmovdqu [arg2+48], xmm5
vpshufb xmm4, xmm7
vpshufb xmm5, xmm7
vpxor xmm2, xmm4
vpxor xmm3, xmm5
sub arg4, 64
; check if there is another 64B in the buffer to be able to fold
jge _fold_64_B_loop
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
add arg3, 64
add arg2, 64
; at this point, the buffer pointer is pointing at the last y Bytes of the buffer
; the 64B of folded data is in 4 of the xmm registers: xmm0, xmm1, xmm2, xmm3
; fold the 4 xmm registers to 1 xmm register with different constants
vmovdqa xmm6, [rk1] ;xmm6 has rk1 and rk2
;imm value of pclmulqdq instruction will
;determine which constant to use
vmovdqa xmm4, xmm0
vpclmulqdq xmm0, xmm6, 0x11
vpclmulqdq xmm4, xmm6, 0x0
vpxor xmm1, xmm4
vpxor xmm1, xmm0
vmovdqa xmm4, xmm1
vpclmulqdq xmm1, xmm6, 0x11
vpclmulqdq xmm4, xmm6, 0x0
vpxor xmm2, xmm4
vpxor xmm2, xmm1
vmovdqa xmm4, xmm2
vpclmulqdq xmm2, xmm6, 0x11
vpclmulqdq xmm4, xmm6, 0x0
vpxor xmm3, xmm4
vpxor xmm3, xmm2
; instead of 64, we add 48 to the loop counter to save 1 instruction from the loop
; instead of a cmp instruction, we use the negative flag with the jl instruction
add arg4, 64-16
jl _final_reduction_for_128
; now we have 16+y bytes left to reduce. 16 Bytes
; is in register xmm3 and the rest is in memory
; we can fold 16 bytes at a time if y>=16
; continue folding 16B at a time
_16B_reduction_loop:
vmovdqa xmm4, xmm3
vpclmulqdq xmm3, xmm6, 0x11
vpclmulqdq xmm4, xmm6, 0x0
vpxor xmm3, xmm4
vmovdqu xmm0, [arg3]
vmovdqu [arg2], xmm0
vpshufb xmm0, xmm7
vpxor xmm3, xmm0
add arg3, 16
add arg2, 16
sub arg4, 16
; instead of a cmp instruction, we utilize the flags with the jge instruction
; equivalent of: cmp arg4, 16-16
; check if there is any more 16B in the buffer to be able to fold
jge _16B_reduction_loop
;now we have 16+z bytes left to reduce, where 0<= z < 16.
;first, we reduce the data in the xmm3 register
_final_reduction_for_128:
; check if any more data to fold. If not, compute the CRC of the final 128 bits
add arg4, 16
je _128_done
; here we are getting data that is less than 16 bytes.
; since we know that there was data before the pointer,
; we can offset the input pointer before the actual point,
; to receive exactly 16 bytes.
; after that the registers need to be adjusted.
_get_last_two_xmms:
vmovdqa xmm2, xmm3
vmovdqu xmm1, [arg3 - 16 + arg4]
vmovdqu [arg2 - 16 + arg4], xmm1
vpshufb xmm1, xmm7
; get rid of the extra data that was loaded before
; load the shift constant
lea rax, [pshufb_shf_table + 16]
sub rax, arg4
vmovdqu xmm0, [rax]
; shift xmm2 to the left by arg4 bytes
vpshufb xmm2, xmm0
; shift xmm3 to the right by 16-arg4 bytes
vpxor xmm0, [mask1]
vpshufb xmm3, xmm0
vpblendvb xmm1, xmm1, xmm2, xmm0
; fold 16 Bytes
vmovdqa xmm2, xmm1
vmovdqa xmm4, xmm3
vpclmulqdq xmm3, xmm6, 0x11
vpclmulqdq xmm4, xmm6, 0x0
vpxor xmm3, xmm4
vpxor xmm3, xmm2
_128_done:
; compute crc of a 128-bit value
vmovdqa xmm6, [rk5] ; rk5 and rk6 in xmm6
vmovdqa xmm0, xmm3
;64b fold
vpclmulqdq xmm3, xmm6, 0x1
vpslldq xmm0, 8
vpxor xmm3, xmm0
;32b fold
vmovdqa xmm0, xmm3
vpand xmm0, [mask2]
vpsrldq xmm3, 12
vpclmulqdq xmm3, xmm6, 0x10
vpxor xmm3, xmm0
;barrett reduction
_barrett:
vmovdqa xmm6, [rk7] ; rk7 and rk8 in xmm6
vmovdqa xmm0, xmm3
vpclmulqdq xmm3, xmm6, 0x01
vpslldq xmm3, 4
vpclmulqdq xmm3, xmm6, 0x11
vpslldq xmm3, 4
vpxor xmm3, xmm0
vpextrd eax, xmm3,1
_cleanup:
; scale the result back to 16 bits
shr eax, 16
vmovdqa xmm6, [rsp+16*2]
vmovdqa xmm7, [rsp+16*3]
add rsp,16*4+8
ret
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
align 16
_less_than_128:
; check if there is enough buffer to be able to fold 16B at a time
cmp arg4, 32
jl _less_than_32
vmovdqa xmm7, [SHUF_MASK]
; if there is, load the constants
vmovdqa xmm6, [rk1] ; rk1 and rk2 in xmm6
vmovd xmm0, arg1_low32 ; get the initial crc value
vpslldq xmm0, 12 ; align it to its correct place
vmovdqu xmm3, [arg3] ; load the plaintext
vmovdqu [arg2], xmm3 ; store copy
vpshufb xmm3, xmm7 ; byte-reflect the plaintext
vpxor xmm3, xmm0
; update the buffer pointer
add arg3, 16
add arg2, 16
; update the counter. subtract 32 instead of 16 to save one instruction from the loop
sub arg4, 32
jmp _16B_reduction_loop
align 16
_less_than_32:
; mov initial crc to the return value. this is necessary for zero-length buffers.
mov eax, arg1_low32
test arg4, arg4
je _cleanup
vmovdqa xmm7, [SHUF_MASK]
vmovd xmm0, arg1_low32 ; get the initial crc value
vpslldq xmm0, 12 ; align it to its correct place
cmp arg4, 16
je _exact_16_left
jl _less_than_16_left
vmovdqu xmm3, [arg3] ; load the plaintext
vmovdqu [arg2], xmm3 ; store the copy
vpshufb xmm3, xmm7 ; byte-reflect the plaintext
vpxor xmm3, xmm0 ; xor the initial crc value
add arg3, 16
add arg2, 16
sub arg4, 16
vmovdqa xmm6, [rk1] ; rk1 and rk2 in xmm6
jmp _get_last_two_xmms
align 16
_less_than_16_left:
; use stack space to load data less than 16 bytes, zero-out the 16B in memory first.
vpxor xmm1, xmm1
mov r11, rsp
vmovdqa [r11], xmm1
cmp arg4, 4
jl _only_less_than_4
; backup the counter value
mov tmp1, arg4
cmp arg4, 8
jl _less_than_8_left
; load 8 Bytes
mov rax, [arg3]
mov [arg2], rax
mov [r11], rax
add r11, 8
sub arg4, 8
add arg3, 8
add arg2, 8
_less_than_8_left:
cmp arg4, 4
jl _less_than_4_left
; load 4 Bytes
mov eax, [arg3]
mov [arg2], eax
mov [r11], eax
add r11, 4
sub arg4, 4
add arg3, 4
add arg2, 4
_less_than_4_left:
cmp arg4, 2
jl _less_than_2_left
; load 2 Bytes
mov ax, [arg3]
mov [arg2], ax
mov [r11], ax
add r11, 2
sub arg4, 2
add arg3, 2
add arg2, 2
_less_than_2_left:
cmp arg4, 1
jl _zero_left
; load 1 Byte
mov al, [arg3]
mov [arg2], al
mov [r11], al
_zero_left:
vmovdqa xmm3, [rsp]
vpshufb xmm3, xmm7
vpxor xmm3, xmm0 ; xor the initial crc value
; shl tmp1, 4
lea rax, [pshufb_shf_table + 16]
sub rax, tmp1
vmovdqu xmm0, [rax]
vpxor xmm0, [mask1]
vpshufb xmm3, xmm0
jmp _128_done
align 16
_exact_16_left:
vmovdqu xmm3, [arg3]
vmovdqu [arg2], xmm3
vpshufb xmm3, xmm7
vpxor xmm3, xmm0 ; xor the initial crc value
jmp _128_done
_only_less_than_4:
cmp arg4, 3
jl _only_less_than_3
; load 3 Bytes
mov al, [arg3]
mov [arg2], al
mov [r11], al
mov al, [arg3+1]
mov [arg2+1], al
mov [r11+1], al
mov al, [arg3+2]
mov [arg2+2], al
mov [r11+2], al
vmovdqa xmm3, [rsp]
vpshufb xmm3, xmm7
vpxor xmm3, xmm0 ; xor the initial crc value
vpsrldq xmm3, 5
jmp _barrett
_only_less_than_3:
cmp arg4, 2
jl _only_less_than_2
; load 2 Bytes
mov al, [arg3]
mov [arg2], al
mov [r11], al
mov al, [arg3+1]
mov [arg2+1], al
mov [r11+1], al
vmovdqa xmm3, [rsp]
vpshufb xmm3, xmm7
vpxor xmm3, xmm0 ; xor the initial crc value
vpsrldq xmm3, 6
jmp _barrett
_only_less_than_2:
; load 1 Byte
mov al, [arg3]
mov [arg2],al
mov [r11], al
vmovdqa xmm3, [rsp]
vpshufb xmm3, xmm7
vpxor xmm3, xmm0 ; xor the initial crc value
vpsrldq xmm3, 7
jmp _barrett
section .data
; precomputed constants
; these constants are precomputed from the poly: 0x8bb70000 (0x8bb7 scaled to 32 bits)
align 16
; Q = 0x18BB70000
; rk1 = 2^(32*3) mod Q << 32
; rk2 = 2^(32*5) mod Q << 32
; rk3 = 2^(32*15) mod Q << 32
; rk4 = 2^(32*17) mod Q << 32
; rk5 = 2^(32*3) mod Q << 32
; rk6 = 2^(32*2) mod Q << 32
; rk7 = floor(2^64/Q)
; rk8 = Q
rk1:
DQ 0x2d56000000000000
rk2:
DQ 0x06df000000000000
rk3:
DQ 0x044c000000000000
rk4:
DQ 0xe658000000000000
rk5:
DQ 0x2d56000000000000
rk6:
DQ 0x1368000000000000
rk7:
DQ 0x00000001f65a57f8
rk8:
DQ 0x000000018bb70000
mask1:
dq 0x8080808080808080, 0x8080808080808080
mask2:
dq 0xFFFFFFFFFFFFFFFF, 0x00000000FFFFFFFF
SHUF_MASK:
dq 0x08090A0B0C0D0E0F, 0x0001020304050607
pshufb_shf_table:
; use these values for shift constants for the pshufb instruction
; different alignments result in values as shown:
; dq 0x8887868584838281, 0x008f8e8d8c8b8a89 ; shl 15 (16-1) / shr1
; dq 0x8988878685848382, 0x01008f8e8d8c8b8a ; shl 14 (16-3) / shr2
; dq 0x8a89888786858483, 0x0201008f8e8d8c8b ; shl 13 (16-4) / shr3
; dq 0x8b8a898887868584, 0x030201008f8e8d8c ; shl 12 (16-4) / shr4
; dq 0x8c8b8a8988878685, 0x04030201008f8e8d ; shl 11 (16-5) / shr5
; dq 0x8d8c8b8a89888786, 0x0504030201008f8e ; shl 10 (16-6) / shr6
; dq 0x8e8d8c8b8a898887, 0x060504030201008f ; shl 9 (16-7) / shr7
; dq 0x8f8e8d8c8b8a8988, 0x0706050403020100 ; shl 8 (16-8) / shr8
; dq 0x008f8e8d8c8b8a89, 0x0807060504030201 ; shl 7 (16-9) / shr9
; dq 0x01008f8e8d8c8b8a, 0x0908070605040302 ; shl 6 (16-10) / shr10
; dq 0x0201008f8e8d8c8b, 0x0a09080706050403 ; shl 5 (16-11) / shr11
; dq 0x030201008f8e8d8c, 0x0b0a090807060504 ; shl 4 (16-12) / shr12
; dq 0x04030201008f8e8d, 0x0c0b0a0908070605 ; shl 3 (16-13) / shr13
; dq 0x0504030201008f8e, 0x0d0c0b0a09080706 ; shl 2 (16-14) / shr14
; dq 0x060504030201008f, 0x0e0d0c0b0a090807 ; shl 1 (16-15) / shr15
dq 0x8786858483828100, 0x8f8e8d8c8b8a8988
dq 0x0706050403020100, 0x000e0d0c0b0a0908

3
crc/crc_multibinary.asm
View File

@ -51,6 +51,7 @@ extern crc32_gzip_refl_by8_02
extern crc32_gzip_refl_base
extern crc16_t10dif_copy_by4
extern crc16_t10dif_copy_by4_02
extern crc16_t10dif_copy_base
%include "multibinary.asm"
@ -215,7 +216,7 @@ mbin_interface crc32_gzip_refl
mbin_dispatch_init_clmul crc32_gzip_refl, crc32_gzip_refl_base, crc32_gzip_refl_by8, crc32_gzip_refl_by8_02
mbin_interface crc16_t10dif_copy
mbin_dispatch_init_clmul crc16_t10dif_copy, crc16_t10dif_copy_base, crc16_t10dif_copy_by4, crc16_t10dif_copy_by4
mbin_dispatch_init_clmul crc16_t10dif_copy, crc16_t10dif_copy_base, crc16_t10dif_copy_by4, crc16_t10dif_copy_by4_02
;;; func core, ver, snum
slversion crc16_t10dif, 00, 03, 011a