generic-library/isa-l
1
0
Fork 0
mirror of https://github.com/intel/isa-l.git synced 2024-12-12 17:33:50 +01:00
Code Issues Projects Releases Wiki Activity
8c975e9cbc
isa-l/crc/crc64_ecma_refl_by8.asm
Xiaodong Liu f42fc40db3 crc64: cache prefetch optimization 
Change-Id: I4d4df3881e6fb8365cba605f7f5d010503ce73e9
Signed-off-by: Xiaodong Liu <xiaodong.liu@intel.com>
2016-12-06 15:58:38 -07:00

549 lines
17 KiB
NASM
Raw Blame History

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; Copyright(c) 2011-2016 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:
; uint64_t crc64_ecma_refl_by8(
; uint64_t init_crc, //initial CRC value, 64 bits
; const unsigned char *buf, //buffer pointer to calculate CRC on
; uint64_t len //buffer length in bytes (64-bit data)
; );
;
; Reference paper titled "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction"
; sample yasm command line:
; yasm -f x64 -f elf64 -X gnu -g dwarf2 crc64_ecma_refl_by8
%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
%else
%xdefine arg1 rdi
%xdefine arg2 rsi
%xdefine arg3 rdx
%endif
%define TMP 16*0
%ifidn __OUTPUT_FORMAT__, win64
%define XMM_SAVE 16*2
%define VARIABLE_OFFSET 16*10+8
%else
%define VARIABLE_OFFSET 16*2+8
%endif
align 16
global crc64_ecma_refl_by8:function
crc64_ecma_refl_by8:
; uint64_t c = crc ^ 0xffffffff,ffffffffL;
not arg1
sub rsp, VARIABLE_OFFSET
%ifidn __OUTPUT_FORMAT__, win64
; push the xmm registers into the stack to maintain
movdqa [rsp + XMM_SAVE + 16*0], xmm6
movdqa [rsp + XMM_SAVE + 16*1], xmm7
movdqa [rsp + XMM_SAVE + 16*2], xmm8
movdqa [rsp + XMM_SAVE + 16*3], xmm9
movdqa [rsp + XMM_SAVE + 16*4], xmm10
movdqa [rsp + XMM_SAVE + 16*5], xmm11
movdqa [rsp + XMM_SAVE + 16*6], xmm12
movdqa [rsp + XMM_SAVE + 16*7], xmm13
%endif
; check if smaller than 256B
cmp arg3, 256
; for sizes less than 256, we can't fold 128B at a time...
jl _less_than_256
; load the initial crc value
movq xmm10, arg1 ; initial crc
; receive the initial 128B data, xor the initial crc value
movdqu xmm0, [arg2+16*0]
movdqu xmm1, [arg2+16*1]
movdqu xmm2, [arg2+16*2]
movdqu xmm3, [arg2+16*3]
movdqu xmm4, [arg2+16*4]
movdqu xmm5, [arg2+16*5]
movdqu xmm6, [arg2+16*6]
movdqu xmm7, [arg2+16*7]
; XOR the initial_crc value
pxor xmm0, xmm10
movdqa xmm10, [rk3] ;xmm10 has rk3 and rk4
;imm value of pclmulqdq instruction will determine which constant to use
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; we subtract 256 instead of 128 to save one instruction from the loop
sub arg3, 256
; at this section of the code, there is 128*x+y (0<=y<128) bytes of buffer. The _fold_128_B_loop
; loop will fold 128B at a time until we have 128+y Bytes of buffer
; fold 128B at a time. This section of the code folds 8 xmm registers in parallel
_fold_128_B_loop:
; update the buffer pointer
add arg2, 128
prefetchnta [arg2+fetch_dist+0]
movdqu xmm9, [arg2+16*0]
movdqu xmm12, [arg2+16*1]
movdqa xmm8, xmm0
movdqa xmm13, xmm1
pclmulqdq xmm0, xmm10, 0x10
pclmulqdq xmm8, xmm10 , 0x1
pclmulqdq xmm1, xmm10, 0x10
pclmulqdq xmm13, xmm10 , 0x1
pxor xmm0, xmm9
xorps xmm0, xmm8
pxor xmm1, xmm12
xorps xmm1, xmm13
prefetchnta [arg2+fetch_dist+32]
movdqu xmm9, [arg2+16*2]
movdqu xmm12, [arg2+16*3]
movdqa xmm8, xmm2
movdqa xmm13, xmm3
pclmulqdq xmm2, xmm10, 0x10
pclmulqdq xmm8, xmm10 , 0x1
pclmulqdq xmm3, xmm10, 0x10
pclmulqdq xmm13, xmm10 , 0x1
pxor xmm2, xmm9
xorps xmm2, xmm8
pxor xmm3, xmm12
xorps xmm3, xmm13
prefetchnta [arg2+fetch_dist+64]
movdqu xmm9, [arg2+16*4]
movdqu xmm12, [arg2+16*5]
movdqa xmm8, xmm4
movdqa xmm13, xmm5
pclmulqdq xmm4, xmm10, 0x10
pclmulqdq xmm8, xmm10 , 0x1
pclmulqdq xmm5, xmm10, 0x10
pclmulqdq xmm13, xmm10 , 0x1
pxor xmm4, xmm9
xorps xmm4, xmm8
pxor xmm5, xmm12
xorps xmm5, xmm13
prefetchnta [arg2+fetch_dist+96]
movdqu xmm9, [arg2+16*6]
movdqu xmm12, [arg2+16*7]
movdqa xmm8, xmm6
movdqa xmm13, xmm7
pclmulqdq xmm6, xmm10, 0x10
pclmulqdq xmm8, xmm10 , 0x1
pclmulqdq xmm7, xmm10, 0x10
pclmulqdq xmm13, xmm10 , 0x1
pxor xmm6, xmm9
xorps xmm6, xmm8
pxor xmm7, xmm12
xorps xmm7, xmm13
sub arg3, 128
; check if there is another 128B in the buffer to be able to fold
jge _fold_128_B_loop
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
add arg2, 128
; at this point, the buffer pointer is pointing at the last y Bytes of the buffer, where 0 <= y < 128
; the 128B of folded data is in 8 of the xmm registers: xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
; fold the 8 xmm registers to 1 xmm register with different constants
; xmm0 to xmm7
movdqa xmm10, [rk9]
movdqa xmm8, xmm0
pclmulqdq xmm0, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
xorps xmm7, xmm0
;xmm1 to xmm7
movdqa xmm10, [rk11]
movdqa xmm8, xmm1
pclmulqdq xmm1, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
xorps xmm7, xmm1
movdqa xmm10, [rk13]
movdqa xmm8, xmm2
pclmulqdq xmm2, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
pxor xmm7, xmm2
movdqa xmm10, [rk15]
movdqa xmm8, xmm3
pclmulqdq xmm3, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
xorps xmm7, xmm3
movdqa xmm10, [rk17]
movdqa xmm8, xmm4
pclmulqdq xmm4, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
pxor xmm7, xmm4
movdqa xmm10, [rk19]
movdqa xmm8, xmm5
pclmulqdq xmm5, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
xorps xmm7, xmm5
; xmm6 to xmm7
movdqa xmm10, [rk1]
movdqa xmm8, xmm6
pclmulqdq xmm6, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
pxor xmm7, xmm6
; instead of 128, we add 128-16 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 arg3, 128-16
jl _final_reduction_for_128
; now we have 16+y bytes left to reduce. 16 Bytes is in register xmm7 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:
movdqa xmm8, xmm7
pclmulqdq xmm7, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
movdqu xmm0, [arg2]
pxor xmm7, xmm0
add arg2, 16
sub arg3, 16
; instead of a cmp instruction, we utilize the flags with the jge instruction
; equivalent of: cmp arg3, 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 xmm7 register
_final_reduction_for_128:
add arg3, 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:
movdqa xmm2, xmm7
movdqu xmm1, [arg2 - 16 + arg3]
; get rid of the extra data that was loaded before
; load the shift constant
lea rax, [pshufb_shf_table]
add rax, arg3
movdqu xmm0, [rax]
pshufb xmm7, xmm0
pxor xmm0, [mask3]
pshufb xmm2, xmm0
pblendvb xmm2, xmm1 ;xmm0 is implicit
;;;;;;;;;;
movdqa xmm8, xmm7
pclmulqdq xmm7, xmm10, 0x1
pclmulqdq xmm8, xmm10, 0x10
pxor xmm7, xmm8
pxor xmm7, xmm2
_128_done:
; compute crc of a 128-bit value
movdqa xmm10, [rk5]
movdqa xmm0, xmm7
;64b fold
pclmulqdq xmm7, xmm10, 0
psrldq xmm0, 8
pxor xmm7, xmm0
;barrett reduction
_barrett:
movdqa xmm1, xmm7
movdqa xmm10, [rk7]
pclmulqdq xmm7, xmm10, 0
movdqa xmm2, xmm7
pclmulqdq xmm7, xmm10, 0x10
pslldq xmm2, 8
pxor xmm7, xmm2
pxor xmm7, xmm1
pextrq rax, xmm7, 1
_cleanup:
; return c ^ 0xffffffff, ffffffffL;
not rax
%ifidn __OUTPUT_FORMAT__, win64
movdqa xmm6, [rsp + XMM_SAVE + 16*0]
movdqa xmm7, [rsp + XMM_SAVE + 16*1]
movdqa xmm8, [rsp + XMM_SAVE + 16*2]
movdqa xmm9, [rsp + XMM_SAVE + 16*3]
movdqa xmm10, [rsp + XMM_SAVE + 16*4]
movdqa xmm11, [rsp + XMM_SAVE + 16*5]
movdqa xmm12, [rsp + XMM_SAVE + 16*6]
movdqa xmm13, [rsp + XMM_SAVE + 16*7]
%endif
add rsp, VARIABLE_OFFSET
ret
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
align 16
_less_than_256:
; check if there is enough buffer to be able to fold 16B at a time
cmp arg3, 32
jl _less_than_32
; if there is, load the constants
movdqa xmm10, [rk1] ; rk1 and rk2 in xmm10
movq xmm0, arg1 ; get the initial crc value
movdqu xmm7, [arg2] ; load the plaintext
pxor xmm7, xmm0
; update the buffer pointer
add arg2, 16
; update the counter. subtract 32 instead of 16 to save one instruction from the loop
sub arg3, 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 rax, arg1
test arg3, arg3
je _cleanup
movq xmm0, arg1 ; get the initial crc value
cmp arg3, 16
je _exact_16_left
jl _less_than_16_left
movdqu xmm7, [arg2] ; load the plaintext
pxor xmm7, xmm0 ; xor the initial crc value
add arg2, 16
sub arg3, 16
movdqa xmm10, [rk1] ; rk1 and rk2 in xmm10
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.
pxor xmm1, xmm1
mov r11, rsp
movdqa [r11], xmm1
; backup the counter value
mov r9, arg3
cmp arg3, 8
jl _less_than_8_left
; load 8 Bytes
mov rax, [arg2]
mov [r11], rax
add r11, 8
sub arg3, 8
add arg2, 8
_less_than_8_left:
cmp arg3, 4
jl _less_than_4_left
; load 4 Bytes
mov eax, [arg2]
mov [r11], eax
add r11, 4
sub arg3, 4
add arg2, 4
_less_than_4_left:
cmp arg3, 2
jl _less_than_2_left
; load 2 Bytes
mov ax, [arg2]
mov [r11], ax
add r11, 2
sub arg3, 2
add arg2, 2
_less_than_2_left:
cmp arg3, 1
jl _zero_left
; load 1 Byte
mov al, [arg2]
mov [r11], al
_zero_left:
movdqa xmm7, [rsp]
pxor xmm7, xmm0 ; xor the initial crc value
lea rax,[pshufb_shf_table]
cmp r9, 8
jl _end_1to7
_end_8to15:
movdqu xmm0, [rax + r9]
pshufb xmm7,xmm0
jmp _128_done
_end_1to7:
; Left shift (8-length) bytes in XMM
movdqu xmm0, [rax + r9 + 8]
pshufb xmm7,xmm0
jmp _barrett
align 16
_exact_16_left:
movdqu xmm7, [arg2]
pxor xmm7, xmm0 ; xor the initial crc value
jmp _128_done
section .data
; precomputed constants
align 16
; rk7 = floor(2^128/Q)
; rk8 = Q
rk1 :
DQ 0xdabe95afc7875f40
rk2 :
DQ 0xe05dd497ca393ae4
rk3 :
DQ 0xd7d86b2af73de740
rk4 :
DQ 0x8757d71d4fcc1000
rk5 :
DQ 0xdabe95afc7875f40
rk6 :
DQ 0x0000000000000000
rk7 :
DQ 0x9c3e466c172963d5
rk8 :
DQ 0x92d8af2baf0e1e84
rk9 :
DQ 0x947874de595052cb
rk10 :
DQ 0x9e735cb59b4724da
rk11 :
DQ 0xe4ce2cd55fea0037
rk12 :
DQ 0x2fe3fd2920ce82ec
rk13 :
DQ 0xe31d519421a63a5
rk14 :
DQ 0x2e30203212cac325
rk15 :
DQ 0x81f6054a7842df4
rk16 :
DQ 0x6ae3efbb9dd441f3
rk17 :
DQ 0x69a35d91c3730254
rk18 :
DQ 0xb5ea1af9c013aca4
rk19 :
DQ 0x3be653a30fe1af51
rk20 :
DQ 0x60095b008a9efa44
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
mask:
dq 0xFFFFFFFFFFFFFFFF, 0x0000000000000000
mask2:
dq 0xFFFFFFFF00000000, 0xFFFFFFFFFFFFFFFF
mask3:
dq 0x8080808080808080, 0x8080808080808080
;;; func core, ver, snum
slversion crc64_ecma_refl_by8, 01, 00, 001d
Reference in New Issue View Git Blame Copy Permalink