2017-06-10 02:03:41 +02:00
|
|
|
%include "reg_sizes.asm"
|
|
|
|
|
%include "lz0a_const.asm"
|
|
|
|
|
%include "data_struct2.asm"
|
|
|
|
|
|
|
|
|
|
%ifdef HAVE_AS_KNOWS_AVX512
|
|
|
|
|
%ifidn __OUTPUT_FORMAT__, win64
|
|
|
|
|
%define arg1 rcx
|
|
|
|
|
%define arg2 rdx
|
|
|
|
|
%define arg3 r8
|
|
|
|
|
%define hash rsi
|
|
|
|
|
%define next_in rdi
|
|
|
|
|
%else
|
|
|
|
|
%define arg1 rdi
|
|
|
|
|
%define arg2 rsi
|
|
|
|
|
%define arg3 rdx
|
|
|
|
|
%define hash r8
|
|
|
|
|
%define next_in rcx
|
|
|
|
|
%endif
|
|
|
|
|
|
|
|
|
|
%define stream arg1
|
|
|
|
|
%define level_buf arg1
|
|
|
|
|
%define matches_next arg2
|
|
|
|
|
%define f_i_end arg3
|
|
|
|
|
|
|
|
|
|
%define f_i rax
|
|
|
|
|
%define file_start rbp
|
|
|
|
|
%define next_byte r9
|
|
|
|
|
%define encode_size r10
|
|
|
|
|
%define prev_len r11
|
|
|
|
|
%define prev_dist r12
|
|
|
|
|
|
2017-11-30 21:07:22 +01:00
|
|
|
%define hash_table level_buf + _hash_map_hash_table
|
2017-06-10 02:03:41 +02:00
|
|
|
|
|
|
|
|
%define datas zmm0
|
|
|
|
|
%define datas_lookup zmm1
|
|
|
|
|
%define zhashes zmm2
|
|
|
|
|
%define zdists zmm3
|
|
|
|
|
%define zdists_lookup zmm4
|
|
|
|
|
%define zscatter zmm5
|
|
|
|
|
%define zdists2 zmm6
|
|
|
|
|
%define zlens1 zmm7
|
|
|
|
|
%define zlens2 zmm8
|
|
|
|
|
%define zlookup zmm9
|
|
|
|
|
%define zlookup2 zmm10
|
|
|
|
|
%define match_lookups zmm11
|
|
|
|
|
%define zindex zmm12
|
|
|
|
|
%define zdist_extra zmm13
|
|
|
|
|
%define zdists_tmp zmm14
|
|
|
|
|
%define znull_dist_syms zmm15
|
|
|
|
|
%define zcode zmm16
|
|
|
|
|
%define zthirty zmm17
|
|
|
|
|
%define zdist_mask zmm18
|
|
|
|
|
%define zshortest_matches zmm19
|
|
|
|
|
%define zrot_left zmm20
|
|
|
|
|
%define zdatas_perm zmm21
|
|
|
|
|
%define zdatas_perm2 zmm22
|
|
|
|
|
%define zdatas_perm3 zmm23
|
|
|
|
|
%define zdatas_shuf zmm24
|
|
|
|
|
%define zhash_prod zmm25
|
|
|
|
|
%define zhash_mask zmm26
|
|
|
|
|
%define zincrement zmm27
|
|
|
|
|
%define zqword_shuf zmm28
|
|
|
|
|
%define zones zmm29
|
|
|
|
|
%define ztwofiftyfour zmm30
|
|
|
|
|
%define zbswap zmm31
|
|
|
|
|
|
|
|
|
|
%ifidn __OUTPUT_FORMAT__, win64
|
|
|
|
|
%define stack_size 10*16 + 4 * 8 + 8
|
|
|
|
|
%macro FUNC_SAVE 0
|
|
|
|
|
alloc_stack stack_size
|
|
|
|
|
vmovdqa [rsp + 0*16], xmm6
|
|
|
|
|
vmovdqa [rsp + 1*16], xmm7
|
|
|
|
|
vmovdqa [rsp + 2*16], xmm8
|
|
|
|
|
vmovdqa [rsp + 3*16], xmm9
|
|
|
|
|
vmovdqa [rsp + 4*16], xmm10
|
|
|
|
|
vmovdqa [rsp + 5*16], xmm11
|
|
|
|
|
vmovdqa [rsp + 6*16], xmm12
|
|
|
|
|
vmovdqa [rsp + 7*16], xmm13
|
|
|
|
|
vmovdqu [rsp + 8*16], xmm14
|
|
|
|
|
vmovdqa [rsp + 9*16], xmm15
|
|
|
|
|
save_reg rsi, 10*16 + 0*8
|
|
|
|
|
save_reg rdi, 10*16 + 1*8
|
|
|
|
|
save_reg rbp, 10*16 + 2*8
|
|
|
|
|
save_reg r12, 10*16 + 3*8
|
|
|
|
|
end_prolog
|
|
|
|
|
%endm
|
|
|
|
|
|
|
|
|
|
%macro FUNC_RESTORE 0
|
|
|
|
|
vmovdqa xmm6, [rsp + 0*16]
|
|
|
|
|
vmovdqa xmm7, [rsp + 1*16]
|
|
|
|
|
vmovdqa xmm8, [rsp + 2*16]
|
|
|
|
|
vmovdqa xmm9, [rsp + 3*16]
|
|
|
|
|
vmovdqa xmm10, [rsp + 4*16]
|
|
|
|
|
vmovdqa xmm11, [rsp + 5*16]
|
|
|
|
|
vmovdqa xmm12, [rsp + 6*16]
|
|
|
|
|
vmovdqa xmm13, [rsp + 7*16]
|
|
|
|
|
vmovdqa xmm14, [rsp + 8*16]
|
|
|
|
|
vmovdqa xmm15, [rsp + 9*16]
|
|
|
|
|
|
|
|
|
|
mov 10*16 + 0*8, rsi
|
|
|
|
|
mov 10*16 + 1*8, rdi
|
|
|
|
|
mov 10*16 + 2*8, rbp
|
|
|
|
|
mov 10*16 + 3*8, r12
|
|
|
|
|
add rsp, stack_size
|
|
|
|
|
%endm
|
|
|
|
|
%else
|
|
|
|
|
%macro FUNC_SAVE 0
|
|
|
|
|
push rbp
|
|
|
|
|
push r12
|
|
|
|
|
%endm
|
|
|
|
|
|
|
|
|
|
%macro FUNC_RESTORE 0
|
|
|
|
|
pop r12
|
|
|
|
|
pop rbp
|
|
|
|
|
%endm
|
|
|
|
|
%endif
|
|
|
|
|
|
|
|
|
|
%define VECT_SIZE 16
|
|
|
|
|
%define HASH_BYTES 2
|
|
|
|
|
|
|
|
|
|
global gen_icf_map_lh1_06
|
|
|
|
|
gen_icf_map_lh1_06:
|
|
|
|
|
FUNC_SAVE
|
|
|
|
|
|
|
|
|
|
mov file_start, [stream + _next_in]
|
|
|
|
|
mov f_i %+ d, dword [stream + _total_in]
|
|
|
|
|
|
|
|
|
|
sub file_start, f_i
|
|
|
|
|
add f_i_end, f_i
|
|
|
|
|
cmp f_i, f_i_end
|
|
|
|
|
jge end_main
|
|
|
|
|
|
|
|
|
|
;; Prep for main loop
|
|
|
|
|
mov level_buf, [stream + _level_buf]
|
|
|
|
|
sub f_i_end, LA
|
|
|
|
|
vmovdqu64 zdatas_perm, [datas_perm]
|
|
|
|
|
vmovdqu64 zdatas_shuf, [datas_shuf]
|
|
|
|
|
vmovdqu64 zhash_prod, [hash_prod]
|
|
|
|
|
vmovdqu64 zhash_mask, [hash_mask]
|
|
|
|
|
vmovdqu64 zincrement, [increment]
|
|
|
|
|
vmovdqu64 zqword_shuf, [qword_shuf]
|
|
|
|
|
vmovdqu64 zdatas_perm2, [datas_perm2]
|
|
|
|
|
vmovdqu64 zdatas_perm3, [datas_perm3]
|
|
|
|
|
vmovdqu64 zones, [ones]
|
|
|
|
|
vmovdqu64 zbswap, [bswap_shuf]
|
|
|
|
|
vmovdqu64 zthirty, [thirty]
|
|
|
|
|
vmovdqu64 zrot_left, [drot_left]
|
|
|
|
|
vmovdqu64 zdist_mask, [dist_mask]
|
|
|
|
|
vmovdqu64 zshortest_matches, [shortest_matches]
|
|
|
|
|
vmovdqu64 ztwofiftyfour, [twofiftyfour]
|
|
|
|
|
vmovdqu64 znull_dist_syms, [null_dist_syms]
|
|
|
|
|
kxorq k0, k0, k0
|
|
|
|
|
kmovq k1, [k_mask_1]
|
|
|
|
|
kmovq k2, [k_mask_2]
|
|
|
|
|
|
|
|
|
|
xor prev_len, prev_len
|
|
|
|
|
xor prev_dist, prev_dist
|
|
|
|
|
|
|
|
|
|
;; Process first byte
|
|
|
|
|
vmovd zhashes %+ x, dword [f_i + file_start]
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpandd zhashes, zhashes, zhash_mask
|
|
|
|
|
vmovd hash %+ d, zhashes %+ x
|
|
|
|
|
mov word [hash_table + HASH_BYTES * hash], f_i %+ w
|
|
|
|
|
|
|
|
|
|
add f_i, 1
|
|
|
|
|
cmp f_i, f_i_end
|
|
|
|
|
jg end_main
|
|
|
|
|
|
|
|
|
|
;;hash
|
|
|
|
|
vmovdqu64 datas %+ y, [f_i + file_start]
|
|
|
|
|
vpermq zhashes, zdatas_perm, datas
|
|
|
|
|
vpshufb zhashes, zhashes, zdatas_shuf
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpandd zhashes, zhashes, zhash_mask
|
|
|
|
|
|
|
|
|
|
vpermq zlookup, zdatas_perm2, datas
|
|
|
|
|
vpshufb zlookup, zlookup, zqword_shuf
|
|
|
|
|
vpermq zlookup2, zdatas_perm3, datas
|
|
|
|
|
vpshufb zlookup2, zlookup2, zqword_shuf
|
|
|
|
|
|
|
|
|
|
;;gather/scatter hashes
|
|
|
|
|
knotq k6, k0
|
|
|
|
|
vpgatherdd zdists_lookup {k6}, [hash_table + HASH_BYTES * zhashes]
|
|
|
|
|
|
|
|
|
|
vpbroadcastd zindex, f_i %+ d
|
|
|
|
|
vpaddd zindex, zindex, zincrement
|
|
|
|
|
vpblendmw zscatter {k1}, zindex, zdists_lookup
|
|
|
|
|
|
|
|
|
|
knotq k6, k0
|
|
|
|
|
vpscatterdd [hash_table + HASH_BYTES * zhashes] {k6}, zscatter
|
|
|
|
|
|
|
|
|
|
;; Compute hash for next loop
|
|
|
|
|
vmovdqu64 datas %+ y, [f_i + file_start + VECT_SIZE]
|
|
|
|
|
vpermq zhashes, zdatas_perm, datas
|
|
|
|
|
vpshufb zhashes, zhashes, zdatas_shuf
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpandd zhashes, zhashes, zhash_mask
|
|
|
|
|
|
|
|
|
|
vmovdqu64 datas_lookup %+ y, [f_i + file_start + 2 * VECT_SIZE]
|
|
|
|
|
|
|
|
|
|
sub f_i_end, VECT_SIZE
|
|
|
|
|
cmp f_i, f_i_end
|
|
|
|
|
jg loop1_end
|
|
|
|
|
|
|
|
|
|
loop1:
|
|
|
|
|
lea next_in, [f_i + file_start]
|
|
|
|
|
|
|
|
|
|
;; Calculate look back dists
|
|
|
|
|
vpaddd zdists, zdists_lookup, zones
|
|
|
|
|
vpsubd zdists, zindex, zdists
|
|
|
|
|
vpandd zdists, zdists, zdist_mask
|
|
|
|
|
vpaddd zdists, zdists, zones
|
|
|
|
|
vpsubd zdists, zincrement, zdists
|
|
|
|
|
|
|
|
|
|
;;gather/scatter hashes
|
|
|
|
|
add f_i, VECT_SIZE
|
|
|
|
|
|
|
|
|
|
kxnorq k6, k6, k6
|
|
|
|
|
kxnorq k7, k7, k7
|
|
|
|
|
vpgatherdd zdists_lookup {k6}, [hash_table + HASH_BYTES * zhashes]
|
|
|
|
|
|
|
|
|
|
vpbroadcastd zindex, f_i %+ d
|
|
|
|
|
vpaddd zindex, zindex, zincrement
|
|
|
|
|
vpblendmw zscatter {k1}, zindex, zdists_lookup
|
|
|
|
|
|
|
|
|
|
vpscatterdd [hash_table + HASH_BYTES * zhashes] {k7}, zscatter
|
|
|
|
|
|
|
|
|
|
;; Compute hash for next loop
|
|
|
|
|
vpermq zhashes, zdatas_perm, datas_lookup
|
|
|
|
|
vpshufb zhashes, zhashes, zdatas_shuf
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpmaddwd zhashes, zhashes, zhash_prod
|
|
|
|
|
vpandd zhashes, zhashes, zhash_mask
|
|
|
|
|
|
|
|
|
|
;;lookup old codes
|
|
|
|
|
vextracti32x8 zdists2 %+ y, zdists, 1
|
|
|
|
|
kxnorq k6, k6, k6
|
|
|
|
|
kxnorq k7, k7, k7
|
|
|
|
|
vpgatherdq zlens1 {k6}, [next_in + zdists %+ y]
|
|
|
|
|
vpgatherdq zlens2 {k7}, [next_in + zdists2 %+ y]
|
|
|
|
|
|
|
|
|
|
;; Calculate dist_icf_code
|
|
|
|
|
vpaddd zdists, zdists, zones
|
|
|
|
|
vpsubd zdists, zincrement, zdists
|
|
|
|
|
vpcmpgtd k5, zdists, zones
|
|
|
|
|
vplzcntd zdist_extra, zdists
|
|
|
|
|
vpsubd zdist_extra {k5}{z}, zthirty, zdist_extra
|
|
|
|
|
vpsllvd zcode, zones, zdist_extra
|
|
|
|
|
vpsubd zcode, zcode, zones
|
|
|
|
|
vpandd zcode {k5}{z}, zdists, zcode
|
|
|
|
|
vpsrlvd zdists, zdists, zdist_extra
|
|
|
|
|
vpslld zdist_extra, zdist_extra, 1
|
|
|
|
|
vpaddd zdists, zdists, zdist_extra
|
|
|
|
|
vpslld zcode, zcode, EXTRA_BITS_OFFSET - DIST_OFFSET
|
|
|
|
|
vpaddd zdists, zdists, zcode
|
|
|
|
|
|
|
|
|
|
;; Setup zdists for combining with zlens
|
|
|
|
|
vpslld zdists, zdists, DIST_OFFSET
|
|
|
|
|
|
|
|
|
|
;; xor current data with lookback dist
|
|
|
|
|
vpxorq zlens1, zlens1, zlookup
|
|
|
|
|
vpxorq zlens2, zlens2, zlookup2
|
|
|
|
|
|
|
|
|
|
;; Setup registers for next loop
|
|
|
|
|
vpermq zlookup, zdatas_perm2, datas
|
|
|
|
|
vpshufb zlookup, zlookup, zqword_shuf
|
|
|
|
|
vpermq zlookup2, zdatas_perm3, datas
|
|
|
|
|
vpshufb zlookup2, zlookup2, zqword_shuf
|
|
|
|
|
|
|
|
|
|
;; Compute match length
|
|
|
|
|
vpshufb zlens1, zlens1, zbswap
|
|
|
|
|
vpshufb zlens2, zlens2, zbswap
|
|
|
|
|
vplzcntq zlens1, zlens1
|
|
|
|
|
vplzcntq zlens2, zlens2
|
|
|
|
|
vpmovqd zlens1 %+ y, zlens1
|
|
|
|
|
vpmovqd zlens2 %+ y, zlens2
|
|
|
|
|
vinserti32x8 zlens1, zlens2 %+ y, 1
|
|
|
|
|
vpsrld zlens1, zlens1, 3
|
|
|
|
|
|
|
|
|
|
;; Preload for next loops
|
|
|
|
|
vmovdqu64 datas, datas_lookup
|
|
|
|
|
vmovdqu64 datas_lookup %+ y, [f_i + file_start + 2 * VECT_SIZE]
|
|
|
|
|
|
|
|
|
|
;; Zero out matches which should not be taken
|
|
|
|
|
kshiftrw k3, k1, 15
|
|
|
|
|
vpermd zlens2, zrot_left, zlens1
|
|
|
|
|
vpermd zdists, zrot_left, zdists
|
|
|
|
|
|
|
|
|
|
vmovd zdists_tmp %+ x, prev_len %+ d
|
|
|
|
|
vmovd prev_len %+ d, zlens2 %+ x
|
|
|
|
|
vmovdqu32 zlens2 {k3}, zdists_tmp
|
|
|
|
|
|
|
|
|
|
vmovd zdists_tmp %+ x, prev_dist %+ d
|
|
|
|
|
vmovd prev_dist %+ d, zdists %+ x
|
|
|
|
|
vmovdqu32 zdists {k3}, zdists_tmp
|
|
|
|
|
|
|
|
|
|
vpcmpgtd k3, zlens2, zshortest_matches
|
|
|
|
|
vpcmpgtd k4, zlens1, zlens2
|
|
|
|
|
|
|
|
|
|
knotq k3, k3
|
|
|
|
|
korq k3, k3, k4
|
|
|
|
|
knotq k4, k3
|
|
|
|
|
vmovdqu32 zlens1 {k4}{z}, zlens2
|
|
|
|
|
|
|
|
|
|
;; Update zdists to match zlens1
|
|
|
|
|
vpaddd zdists, zdists, zlens1
|
|
|
|
|
vpaddd zdists, zdists, ztwofiftyfour
|
|
|
|
|
vpmovzxbd zdists {k3}, [f_i + file_start - VECT_SIZE - 1]
|
|
|
|
|
vpaddd zdists {k3}, zdists, znull_dist_syms
|
|
|
|
|
|
|
|
|
|
;;Store zdists
|
|
|
|
|
vmovdqu64 [matches_next], zdists
|
|
|
|
|
add matches_next, ICF_CODE_BYTES * VECT_SIZE
|
|
|
|
|
|
|
|
|
|
cmp f_i, f_i_end
|
|
|
|
|
jle loop1
|
|
|
|
|
|
|
|
|
|
loop1_end:
|
|
|
|
|
lea next_in, [f_i + file_start]
|
|
|
|
|
|
|
|
|
|
;; Calculate look back dists
|
|
|
|
|
vpaddd zdists, zdists_lookup, zones
|
|
|
|
|
vpsubd zdists, zindex, zdists
|
|
|
|
|
vpandd zdists, zdists, zdist_mask
|
|
|
|
|
vpaddd zdists, zdists, zones
|
|
|
|
|
vpsubd zdists, zincrement, zdists
|
|
|
|
|
|
|
|
|
|
;;lookup old codes
|
|
|
|
|
vextracti32x8 zdists2 %+ y, zdists, 1
|
|
|
|
|
kxnorq k6, k6, k6
|
|
|
|
|
kxnorq k7, k7, k7
|
|
|
|
|
vpgatherdq zlens1 {k6}, [next_in + zdists %+ y]
|
|
|
|
|
vpgatherdq zlens2 {k7}, [next_in + zdists2 %+ y]
|
|
|
|
|
|
|
|
|
|
;; Calculate dist_icf_code
|
|
|
|
|
vpaddd zdists, zdists, zones
|
|
|
|
|
vpsubd zdists, zincrement, zdists
|
|
|
|
|
vpcmpgtd k5, zdists, zones
|
|
|
|
|
vplzcntd zdist_extra, zdists
|
|
|
|
|
vpsubd zdist_extra {k5}{z}, zthirty, zdist_extra
|
|
|
|
|
vpsllvd zcode, zones, zdist_extra
|
|
|
|
|
vpsubd zcode, zcode, zones
|
|
|
|
|
vpandd zcode {k5}{z}, zdists, zcode
|
|
|
|
|
vpsrlvd zdists, zdists, zdist_extra
|
|
|
|
|
vpslld zdist_extra, zdist_extra, 1
|
|
|
|
|
vpaddd zdists, zdists, zdist_extra
|
|
|
|
|
vpslld zcode, zcode, EXTRA_BITS_OFFSET - DIST_OFFSET
|
|
|
|
|
vpaddd zdists, zdists, zcode
|
|
|
|
|
|
|
|
|
|
;; Setup zdists for combining with zlens
|
|
|
|
|
vpslld zdists, zdists, DIST_OFFSET
|
|
|
|
|
|
|
|
|
|
;; xor current data with lookback dist
|
|
|
|
|
vpxorq zlens1, zlens1, zlookup
|
|
|
|
|
vpxorq zlens2, zlens2, zlookup2
|
|
|
|
|
|
|
|
|
|
;; Compute match length
|
|
|
|
|
vpshufb zlens1, zlens1, zbswap
|
|
|
|
|
vpshufb zlens2, zlens2, zbswap
|
|
|
|
|
vplzcntq zlens1, zlens1
|
|
|
|
|
vplzcntq zlens2, zlens2
|
|
|
|
|
vpmovqd zlens1 %+ y, zlens1
|
|
|
|
|
vpmovqd zlens2 %+ y, zlens2
|
|
|
|
|
vinserti32x8 zlens1, zlens2 %+ y, 1
|
|
|
|
|
vpsrld zlens1, zlens1, 3
|
|
|
|
|
|
|
|
|
|
;; Zero out matches which should not be taken
|
|
|
|
|
kshiftrw k3, k1, 15
|
|
|
|
|
vpermd zlens2, zrot_left, zlens1
|
|
|
|
|
vpermd zdists, zrot_left, zdists
|
|
|
|
|
|
|
|
|
|
vmovd zdists_tmp %+ x, prev_len %+ d
|
|
|
|
|
vmovd prev_len %+ d, zlens2 %+ x
|
|
|
|
|
vmovdqu32 zlens2 {k3}, zdists_tmp
|
|
|
|
|
|
|
|
|
|
vmovd zdists_tmp %+ x, prev_dist %+ d
|
|
|
|
|
vmovd prev_dist %+ d, zdists %+ x
|
|
|
|
|
vmovdqu32 zdists {k3}, zdists_tmp
|
|
|
|
|
|
|
|
|
|
vpcmpgtd k3, zlens2, zshortest_matches
|
|
|
|
|
vpcmpgtd k4, zlens1, zlens2
|
|
|
|
|
|
|
|
|
|
knotq k3, k3
|
|
|
|
|
korq k3, k3, k4
|
|
|
|
|
knotq k4, k3
|
|
|
|
|
vmovdqu32 zlens1 {k4}{z}, zlens2
|
|
|
|
|
|
|
|
|
|
;; Update zdists to match zlens1
|
|
|
|
|
vpaddd zdists, zdists, zlens1
|
|
|
|
|
vpaddd zdists, zdists, ztwofiftyfour
|
|
|
|
|
vpmovzxbd zdists {k3}, [f_i + file_start - 1]
|
|
|
|
|
vpaddd zdists {k3}, zdists, znull_dist_syms
|
|
|
|
|
|
|
|
|
|
;;Store zdists
|
|
|
|
|
vmovdqu64 [matches_next], zdists
|
|
|
|
|
|
|
|
|
|
end_main:
|
|
|
|
|
FUNC_RESTORE
|
|
|
|
|
ret
|
|
|
|
|
|
|
|
|
|
section .data
|
|
|
|
|
align 64
|
|
|
|
|
datas_perm:
|
|
|
|
|
dq 0x0, 0x1, 0x0, 0x1, 0x1, 0x2, 0x1, 0x2
|
|
|
|
|
datas_perm2:
|
|
|
|
|
dq 0x0, 0x1, 0x0, 0x1, 0x0, 0x1, 0x0, 0x1
|
|
|
|
|
datas_perm3:
|
|
|
|
|
dq 0x1, 0x2, 0x1, 0x2, 0x1, 0x2, 0x1, 0x2
|
|
|
|
|
drot_left:
|
|
|
|
|
dd 0xf, 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6
|
|
|
|
|
dd 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe
|
|
|
|
|
datas_shuf:
|
|
|
|
|
db 0x0, 0x1, 0x2, 0x3
|
|
|
|
|
db 0x1, 0x2, 0x3, 0x4
|
|
|
|
|
db 0x2, 0x3, 0x4, 0x5
|
|
|
|
|
db 0x3, 0x4, 0x5, 0x6
|
|
|
|
|
db 0x4, 0x5, 0x6, 0x7
|
|
|
|
|
db 0x5, 0x6, 0x7, 0x8
|
|
|
|
|
db 0x6, 0x7, 0x8, 0x9
|
|
|
|
|
db 0x7, 0x8, 0x9, 0xa
|
|
|
|
|
db 0x0, 0x1, 0x2, 0x3
|
|
|
|
|
db 0x1, 0x2, 0x3, 0x4
|
|
|
|
|
db 0x2, 0x3, 0x4, 0x5
|
|
|
|
|
db 0x3, 0x4, 0x5, 0x6
|
|
|
|
|
db 0x4, 0x5, 0x6, 0x7
|
|
|
|
|
db 0x5, 0x6, 0x7, 0x8
|
|
|
|
|
db 0x6, 0x7, 0x8, 0x9
|
|
|
|
|
db 0x7, 0x8, 0x9, 0xa
|
|
|
|
|
bswap_shuf:
|
|
|
|
|
db 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00
|
|
|
|
|
db 0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08
|
|
|
|
|
db 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00
|
|
|
|
|
db 0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08
|
|
|
|
|
db 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00
|
|
|
|
|
db 0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08
|
|
|
|
|
db 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00
|
|
|
|
|
db 0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08
|
|
|
|
|
|
|
|
|
|
qword_shuf:
|
|
|
|
|
db 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7
|
|
|
|
|
db 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8
|
|
|
|
|
db 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9
|
|
|
|
|
db 0x3, 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa
|
|
|
|
|
db 0x4, 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb
|
|
|
|
|
db 0x5, 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc
|
|
|
|
|
db 0x6, 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd
|
|
|
|
|
db 0x7, 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe
|
|
|
|
|
db 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf
|
|
|
|
|
|
|
|
|
|
%define PROD1 0xE84B
|
|
|
|
|
%define PROD2 0x97B1
|
|
|
|
|
|
|
|
|
|
hash_prod:
|
|
|
|
|
dw PROD1, PROD2, PROD1, PROD2, PROD1, PROD2, PROD1, PROD2
|
|
|
|
|
dw PROD1, PROD2, PROD1, PROD2, PROD1, PROD2, PROD1, PROD2
|
|
|
|
|
dw PROD1, PROD2, PROD1, PROD2, PROD1, PROD2, PROD1, PROD2
|
|
|
|
|
dw PROD1, PROD2, PROD1, PROD2, PROD1, PROD2, PROD1, PROD2
|
|
|
|
|
null_dist_syms:
|
|
|
|
|
dd LIT, LIT, LIT, LIT, LIT, LIT, LIT, LIT
|
|
|
|
|
dd LIT, LIT, LIT, LIT, LIT, LIT, LIT, LIT
|
|
|
|
|
increment:
|
|
|
|
|
dd 0x0, 0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7
|
|
|
|
|
dd 0x8, 0x9, 0xa, 0xb, 0xc, 0xd, 0xe, 0xf
|
|
|
|
|
ones:
|
|
|
|
|
dd 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1
|
|
|
|
|
dd 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1, 0x1
|
|
|
|
|
thirty:
|
|
|
|
|
dd 0x1e, 0x1e, 0x1e, 0x1e, 0x1e, 0x1e, 0x1e, 0x1e
|
|
|
|
|
dd 0x1e, 0x1e, 0x1e, 0x1e, 0x1e, 0x1e, 0x1e, 0x1e
|
|
|
|
|
twofiftyfour:
|
|
|
|
|
dd 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe
|
|
|
|
|
dd 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe
|
|
|
|
|
dist_mask:
|
|
|
|
|
dd D-1, D-1, D-1, D-1, D-1, D-1, D-1, D-1
|
|
|
|
|
dd D-1, D-1, D-1, D-1, D-1, D-1, D-1, D-1
|
|
|
|
|
hash_mask:
|
2017-11-30 21:07:22 +01:00
|
|
|
dd HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK
|
|
|
|
|
dd HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK
|
|
|
|
|
dd HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK
|
|
|
|
|
dd HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK, HASH_MAP_HASH_MASK
|
2017-06-10 02:03:41 +02:00
|
|
|
lit_len_mask:
|
|
|
|
|
dd LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK
|
|
|
|
|
dd LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK
|
|
|
|
|
dd LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK
|
|
|
|
|
dd LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK, LIT_LEN_MASK
|
|
|
|
|
shortest_matches:
|
|
|
|
|
dd MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH
|
|
|
|
|
dd MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH
|
|
|
|
|
dd MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH
|
|
|
|
|
dd MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH, MIN_DEF_MATCH
|
|
|
|
|
|
|
|
|
|
k_mask_1:
|
|
|
|
|
dq 0xaaaaaaaaaaaaaaaa
|
|
|
|
|
k_mask_2:
|
|
|
|
|
dq 0x7fff
|
|
|
|
|
%endif
|
