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
6b2bedfe28
isa-l/igzip/igzip_stateless.asm
Roy Oursler eb1b7788d0 igzip: Move code in igzip_stateless to hide latencies more in ivybridge. 
Signed-off-by: Roy Oursler <roy.j.oursler@intel.com>
Reviewed-by: Greg Tucker <greg.b.tucker@intel.com>
2016-12-02 17:23:29 -07:00

645 lines
16 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.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
%include "options.asm"
%include "lz0a_const.asm"
%include "data_struct2.asm"
%include "bitbuf2.asm"
%include "huffman.asm"
%include "igzip_compare_types.asm"
%include "reg_sizes.asm"
%include "stdmac.asm"
%define LAST_BYTES_COUNT 3 ; Bytes to prevent reading out of array bounds
%define LA_STATELESS 280 ; Max number of bytes read in loop2 rounded up to 8 byte boundary
%ifdef DEBUG
%macro MARK 1
global %1
%1:
%endm
%else
%macro MARK 1
%endm
%endif
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
%define tmp2 rcx
%define hash2 rcx
%define curr_data rax
%define code rax
%define tmp5 rax
%define tmp4 rbx
%define dist rbx
%define code2 rbx
%define hash rdx
%define len rdx
%define code_len3 rdx
%define tmp8 rdx
%define tmp1 rsi
%define code_len2 rsi
%define file_start rdi
%define m_bit_count rbp
%define curr_data2 r8
%define len2 r8
%define tmp6 r8
%define m_bits r9
%define f_i r10
%define m_out_buf r11
%define f_end_i r12
%define dist2 r12
%define tmp7 r12
%define code4 r12
%define tmp3 r13
%define code3 r13
%define stream r14
%define hufftables r15
;; GPR r8 & r15 can be used
%define xtmp0 xmm0 ; tmp
%define xtmp1 xmm1 ; tmp
%define xhash xmm2
%define xmask xmm3
%define xdata xmm4
%define ytmp0 ymm0 ; tmp
%define ytmp1 ymm1 ; tmp
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
blen_mem_offset equ 0 ; local variable (8 bytes)
f_end_i_mem_offset equ 8
gpr_save_mem_offset equ 16 ; gpr save area (8*8 bytes)
xmm_save_mem_offset equ 16 + 8*8 ; xmm save area (4*16 bytes) (16 byte aligned)
stack_size equ 2*8 + 8*8 + 4*16 + 8
;;; 8 because stack address is odd multiple of 8 after a function call and
;;; we want it aligned to 16 bytes
; void isal_deflate_body_stateless ( isal_zstream *stream )
; arg 1: rcx: addr of stream
global isal_deflate_body_stateless_ %+ ARCH
isal_deflate_body_stateless_ %+ ARCH %+ :
%ifidn __OUTPUT_FORMAT__, elf64
mov rcx, rdi
%endif
;; do nothing if (avail_in == 0)
cmp dword [rcx + _avail_in], 0
jne skip1
ret
skip1:
%ifdef ALIGN_STACK
push rbp
mov rbp, rsp
sub rsp, stack_size
and rsp, ~15
%else
sub rsp, stack_size
%endif
mov [rsp + gpr_save_mem_offset + 0*8], rbx
mov [rsp + gpr_save_mem_offset + 1*8], rsi
mov [rsp + gpr_save_mem_offset + 2*8], rdi
mov [rsp + gpr_save_mem_offset + 3*8], rbp
mov [rsp + gpr_save_mem_offset + 4*8], r12
mov [rsp + gpr_save_mem_offset + 5*8], r13
mov [rsp + gpr_save_mem_offset + 6*8], r14
mov [rsp + gpr_save_mem_offset + 7*8], r15
mov stream, rcx
mov dword [stream + _internal_state_has_eob], 0
MOVDQU xmask, [mask]
; state->bitbuf.set_buf(stream->next_out, stream->avail_out);
mov m_out_buf, [stream + _next_out]
mov [stream + _internal_state_bitbuf_m_out_start], m_out_buf
mov tmp1 %+ d, [stream + _avail_out]
add tmp1, m_out_buf
sub tmp1, SLOP
skip_SLOP:
mov [stream + _internal_state_bitbuf_m_out_end], tmp1
mov m_bits, [stream + _internal_state_bitbuf_m_bits]
mov m_bit_count %+ d, [stream + _internal_state_bitbuf_m_bit_count]
mov hufftables, [stream + _hufftables]
; state->b_bytes_valid = stream->avail_in;
mov f_end_i %+ d, [stream + _avail_in]
mov [stream + _internal_state_b_bytes_valid], f_end_i %+ d
mov f_i, 0
mov file_start, [stream + _next_in]
mov [stream + _internal_state_file_start], file_start
; f_end_i -= LA;
sub f_end_i, LA_STATELESS
mov [rsp + f_end_i_mem_offset], f_end_i
; if (f_end_i <= 0) continue;
cmp f_end_i, 0
jle end_loop_2
; for (f_i = f_start_i; f_i < f_end_i; f_i++) {
MARK __stateless_compute_hash_ %+ ARCH
mov curr_data, [file_start + f_i]
cmp m_out_buf, [stream + _internal_state_bitbuf_m_out_end]
ja end
;; Encode first byte in the stream as a literal
compute_hash hash, curr_data
and hash %+ d, HASH_MASK
mov [stream + _internal_state_head + 2 * hash], f_i %+ w
and curr_data, 0xff
get_lit_code curr_data, code2, code_len2, hufftables
mov tmp3, [file_start + f_i + 1]
mov tmp6, tmp3
compute_hash hash, tmp3
shr tmp6, 8
compute_hash hash2, tmp6
MOVD xhash, hash %+ d
PINSRD xhash, hash2 %+ d, 1
PAND xhash, xhash, xmask
jmp write_lit_bits
align 16
loop2:
; if (state->bitbuf.is_full()) {
cmp m_out_buf, [stream + _internal_state_bitbuf_m_out_end]
ja end
xor dist, dist
xor dist2, dist2
xor tmp3, tmp3
lea tmp1, [file_start + f_i]
mov dist %+ w, f_i %+ w
sub dist %+ w, word [stream + _internal_state_head + 2 * hash]
mov [stream + _internal_state_head + 2 * hash], f_i %+ w
dec dist
inc f_i
MOVQ tmp6, xdata
shr tmp5, 16
mov tmp8, tmp5
compute_hash tmp6, tmp5
mov dist2 %+ w, f_i %+ w
sub dist2 %+ w, word [stream + _internal_state_head + 2 * hash2]
mov [stream + _internal_state_head + 2 * hash2], f_i %+ w
dec dist2
; if ((dist-1) < (D-1)) {
cmp dist %+ d, (D-1)
cmovae dist, tmp3
add dist, 1
neg dist
shr tmp8, 8
compute_hash tmp2, tmp8
cmp dist2 %+ d, (D-1)
cmovae dist2, tmp3
add dist2, 1
neg dist2
MARK __stateless_compare_ %+ ARCH
;; Check for long len/dist match (>7) with first literal
MOVQ len, xdata
mov curr_data, len
PSRLDQ xdata, 1
xor len, [tmp1 + dist]
jz compare_loop
MOVD xhash, tmp6 %+ d
PINSRD xhash, tmp2 %+ d, 1
PAND xhash, xhash, xmask
;; Check for len/dist match (>7) with second literal
MOVQ len2, xdata
xor len2, [tmp1 + dist2 + 1]
jz compare_loop2
;; Specutively load the code for the first literal
movzx tmp1, curr_data %+ b
get_lit_code tmp1, code3, rcx, hufftables
;; Check for len/dist match for first literal
test len %+ d, 0xFFFFFFFF
jz len_dist_huffman_pre
;; Specutively load the code for the second literal
shr curr_data, 8
and curr_data, 0xff
get_lit_code curr_data, code2, code_len2, hufftables
shl code2, cl
or code2, code3
add code_len2, rcx
;; Check for len/dist match for second literal
test len2 %+ d, 0xFFFFFFFF
jnz write_lit_bits
MARK __stateless_len_dist_lit_huffman_ %+ ARCH
len_dist_lit_huffman_pre:
mov code_len3, rcx
bsf len2, len2
shr len2, 3
len_dist_lit_huffman:
neg dist2
%ifndef LONGER_HUFFTABLE
mov tmp4, dist2
get_dist_code tmp4, code4, code_len2, hufftables ;; clobbers dist, rcx
%else
get_dist_code dist2, code4, code_len2, hufftables
%endif
get_len_code len2, code, rcx, hufftables ;; rcx is code_len
SHLX code4, code4, rcx
or code4, code
add code_len2, rcx
mov rcx, code_len3
MOVQ tmp5, xdata
shr tmp5, 24
compute_hash tmp4, tmp5
and tmp4, HASH_MASK
SHLX code4, code4, rcx
or code4, code3
add code_len2, rcx
;; Setup for updating hash
lea tmp3, [f_i + 1] ; tmp3 <= k
add f_i, len2
MOVDQU xdata, [file_start + f_i]
mov curr_data, [file_start + f_i]
mov curr_data2, curr_data
MOVD hash %+ d, xhash
PEXTRD hash2 %+ d, xhash, 1
mov [stream + _internal_state_head + 2 * hash], tmp3 %+ w
compute_hash hash, curr_data
add tmp3,1
mov [stream + _internal_state_head + 2 * hash2], tmp3 %+ w
add tmp3, 1
mov [stream + _internal_state_head + 2 * tmp4], tmp3 %+ w
write_bits m_bits, m_bit_count, code4, code_len2, m_out_buf, tmp4
mov f_end_i, [rsp + f_end_i_mem_offset]
shr curr_data2, 8
compute_hash hash2, curr_data2
%ifdef NO_LIMIT_HASH_UPDATE
loop3:
add tmp3,1
cmp tmp3, f_i
jae loop3_done
mov tmp6, [file_start + tmp3]
compute_hash tmp4, tmp6
and tmp4 %+ d, HASH_MASK
; state->head[hash] = k;
mov [stream + _internal_state_head + 2 * tmp4], tmp3 %+ w
jmp loop3
loop3_done:
%endif
; hash = compute_hash(state->file_start + f_i) & HASH_MASK;
and hash %+ d, HASH_MASK
and hash2 %+ d, HASH_MASK
; continue
cmp f_i, f_end_i
jl loop2
jmp end_loop_2
;; encode as dist/len
MARK __stateless_len_dist_huffman_ %+ ARCH
len_dist_huffman_pre:
bsf len, len
shr len, 3
len_dist_huffman:
dec f_i
neg dist
; get_dist_code(dist, &code2, &code_len2);
%ifndef LONGER_HUFFTABLE
mov tmp3, dist ; since code2 and dist are rbx
get_dist_code tmp3, code2, code_len2, hufftables ;; clobbers dist, rcx
%else
get_dist_code dist, code2, code_len2, hufftables
%endif
; get_len_code(len, &code, &code_len);
get_len_code len, code, rcx, hufftables ;; rcx is code_len
; code2 <<= code_len
; code2 |= code
; code_len2 += code_len
SHLX code2, code2, rcx
or code2, code
add code_len2, rcx
;; Setup for updateing hash
lea tmp3, [f_i + 2] ; tmp3 <= k
add f_i, len
MOVD hash %+ d, xhash
PEXTRD hash2 %+ d, xhash, 1
mov [stream + _internal_state_head + 2 * hash], tmp3 %+ w
add tmp3,1
mov [stream + _internal_state_head + 2 * hash2], tmp3 %+ w
MOVDQU xdata, [file_start + f_i]
mov curr_data, [file_start + f_i]
mov curr_data2, curr_data
compute_hash hash, curr_data
write_bits m_bits, m_bit_count, code2, code_len2, m_out_buf, tmp7
mov f_end_i, [rsp + f_end_i_mem_offset]
shr curr_data2, 8
compute_hash hash2, curr_data2
%ifdef NO_LIMIT_HASH_UPDATE
loop4:
add tmp3,1
cmp tmp3, f_i
jae loop4_done
mov tmp6, [file_start + tmp3]
compute_hash tmp4, tmp6
and tmp4, HASH_MASK
mov [stream + _internal_state_head + 2 * tmp4], tmp3 %+ w
jmp loop4
loop4_done:
%endif
; hash = compute_hash(state->file_start + f_i) & HASH_MASK;
and hash %+ d, HASH_MASK
and hash2 %+ d, HASH_MASK
; continue
cmp f_i, f_end_i
jl loop2
jmp end_loop_2
MARK __stateless_write_lit_bits_ %+ ARCH
write_lit_bits:
MOVDQU xdata, [file_start + f_i + 1]
mov f_end_i, [rsp + f_end_i_mem_offset]
add f_i, 1
mov curr_data, [file_start + f_i]
MOVD hash %+ d, xhash
write_bits m_bits, m_bit_count, code2, code_len2, m_out_buf, tmp3
PEXTRD hash2 %+ d, xhash, 1
; continue
cmp f_i, f_end_i
jl loop2
MARK __stateless_end_loops_ %+ ARCH
end_loop_2:
;; Handle the last bytes (at most LA_statless bytes)
add f_end_i, LA_STATELESS - LAST_BYTES_COUNT
cmp f_i, f_end_i
jge end_loop_2_finish
loop2_finish:
;; Check for space in out buffer
cmp m_out_buf, [stream + _internal_state_bitbuf_m_out_end]
ja end
mov curr_data %+ d, dword [file_start + f_i]
compute_hash hash, curr_data
and hash %+ d, HASH_MASK
;; Calculate possible distance for length/dist pair.
xor dist, dist
mov dist %+ w, f_i %+ w
sub dist %+ w, word [stream + _internal_state_head + 2 * hash]
mov [stream + _internal_state_head + 2 * hash], f_i %+ w
;; Check if look back distance is valid (the dec is to handle when dist = 0)
dec dist
cmp dist %+ d, (D-1)
jae encode_literal_finish
inc dist
;; Check if look back distance is a match
lea tmp6, [f_end_i + LAST_BYTES_COUNT]
sub tmp6, f_i
lea tmp1, [file_start + f_i]
mov tmp2, tmp1
sub tmp2, dist
compare tmp6, tmp1, tmp2, len, tmp3
;; Limit len to maximum value of 258
mov tmp2, 258
cmp len, 258
cmova len, tmp2
cmp len, SHORTEST_MATCH
jb encode_literal_finish
;; Encode len/dist pair
%ifndef LONGER_HUFFTABLE
mov tmp3, dist
get_dist_code tmp3, code2, code_len2, hufftables ;; clobbers dist, rcx
%else
get_dist_code dist, code2, code_len2, hufftables ;; clobbers dist, rcx
%endif
get_len_code len, code, rcx, hufftables ;; rcx is code_len
;; Combine length and distance code for writing it out
%ifdef USE_HSWNI
shlx code2, code2, rcx
%else
shl code2, cl
%endif
or code2, code
add code_len2, rcx
write_bits m_bits, m_bit_count, code2, code_len2, m_out_buf, tmp3
;; Setup for next loop
add f_i, len
cmp f_i, f_end_i
jl loop2_finish
jmp end_loop_2_finish
encode_literal_finish:
;; Encode literal
and curr_data %+ d, 0xFF
get_lit_code curr_data, code2, code_len2, hufftables
write_bits m_bits, m_bit_count, code2, code_len2, m_out_buf, tmp3
;; Setup for next loop
add f_i, 1
cmp f_i, f_end_i
jl loop2_finish
end_loop_2_finish:
cmp m_out_buf, [stream + _internal_state_bitbuf_m_out_end]
ja end
;; Check if any bytes left (at most LAST_BYTES_COUNT bytes)
add f_end_i, LAST_BYTES_COUNT
cmp f_i, f_end_i
jz write_eob
;; Handle encoding last few bytes by encoding them as literals
xor curr_data, curr_data
final_bytes:
movzx curr_data, byte [file_start + f_i]
get_lit_code curr_data, code2, code_len2, hufftables
write_bits m_bits, m_bit_count, code2, code_len2, m_out_buf, tmp3
cmp m_out_buf, [stream + _internal_state_bitbuf_m_out_end]
ja end
inc f_i
cmp f_i, f_end_i
jl final_bytes
write_eob:
;; Write out end of block
get_lit_code 256, code2, code_len2, hufftables
write_bits m_bits, m_bit_count, code2, code_len2, m_out_buf, tmp3
mov dword [stream + _internal_state_has_eob], 1
end:
;; update input buffer
add [stream + _total_in], f_i %+ d
add [stream + _next_in], f_i
sub [stream + _avail_in], f_i %+ d
;; update output buffer
mov [stream + _next_out], m_out_buf
sub m_out_buf, [stream + _internal_state_bitbuf_m_out_start]
sub [stream + _avail_out], m_out_buf %+ d
add [stream + _total_out], m_out_buf %+ d
mov [stream + _internal_state_bitbuf_m_bits], m_bits
mov [stream + _internal_state_bitbuf_m_bit_count], m_bit_count %+ d
mov rbx, [rsp + gpr_save_mem_offset + 0*8]
mov rsi, [rsp + gpr_save_mem_offset + 1*8]
mov rdi, [rsp + gpr_save_mem_offset + 2*8]
mov rbp, [rsp + gpr_save_mem_offset + 3*8]
mov r12, [rsp + gpr_save_mem_offset + 4*8]
mov r13, [rsp + gpr_save_mem_offset + 5*8]
mov r14, [rsp + gpr_save_mem_offset + 6*8]
mov r15, [rsp + gpr_save_mem_offset + 7*8]
%ifndef ALIGN_STACK
add rsp, stack_size
%else
mov rsp, rbp
pop rbp
%endif
ret
MARK __stateless_compare_loops_ %+ ARCH
compare_loop:
MOVD xhash, tmp6 %+ d
PINSRD xhash, tmp2 %+ d, 1
PAND xhash, xhash, xmask
lea tmp2, [tmp1 + dist]
%if (COMPARE_TYPE == 1)
compare250 tmp1, tmp2, len, tmp3
%elif (COMPARE_TYPE == 2)
compare250_x tmp1, tmp2, len, tmp3, xtmp0, xtmp1
%elif (COMPARE_TYPE == 3)
compare250_y tmp1, tmp2, len, tmp3, ytmp0, ytmp1
%else
%error Unknown Compare type COMPARE_TYPE
% error
%endif
jmp len_dist_huffman
compare_loop2:
add tmp1, 1
lea tmp2, [tmp1 + dist2]
%if (COMPARE_TYPE == 1)
compare250 tmp1, tmp2, len2, tmp3
%elif (COMPARE_TYPE == 2)
compare250_x tmp1, tmp2, len2, tmp3, xtmp0, xtmp1
%elif (COMPARE_TYPE == 3)
compare250_y tmp1, tmp2, len2, tmp3, ytmp0, ytmp1
%else
%error Unknown Compare type COMPARE_TYPE
% error
%endif
and curr_data, 0xff
get_lit_code curr_data, code3, code_len3, hufftables
jmp len_dist_lit_huffman
section .data
align 16
mask: dd HASH_MASK, HASH_MASK, HASH_MASK, HASH_MASK
const_D: dq D
Reference in New Issue View Git Blame Copy Permalink