generic-library/isa-l
1
0
Fork 0
mirror of https://github.com/intel/isa-l.git synced 2025-07-04 17:40:07 +02:00
Code Issues Projects Releases Wiki Activity

igzip: Port update_histogram and decode_huffman_code_block to windows

Browse Source 
Signed-off-by: Roy Oursler <roy.j.oursler@intel.com>
Reviewed-by: Greg Tucker <greg.b.tucker@intel.com>
This commit is contained in:
Roy Oursler 2016-07-26 17:02:23 -07:00 committed by Greg Tucker
parent 340e0f99df
commit d06e14b937
5 changed files with 212 additions and 48 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
igzip/huff_codes.c
View File

@ -42,6 +42,18 @@
static const uint8_t code_length_code_order[] = static const uint8_t code_length_code_order[] =
{ 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 }; { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 };
struct slver {
uint16_t snum;
uint8_t ver;
uint8_t core;
};
/* Version info */
struct slver isal_update_histogram_slver_00010085;
struct slver isal_update_histogram_slver = { 0x0085, 0x01, 0x00 };
struct slver isal_create_hufftables_slver_00010086;
struct slver isal_create_hufftables_slver = { 0x0086, 0x01, 0x00 };
int heap_push(struct huff_tree element, struct histheap *heap) int heap_push(struct huff_tree element, struct histheap *heap)
{ {
uint16_t index; uint16_t index;

3
igzip/igzip.c
View File

@ -97,6 +97,9 @@ struct slver {
struct slver isal_deflate_init_slver_01030081; struct slver isal_deflate_init_slver_01030081;
struct slver isal_deflate_init_slver = { 0x0081, 0x03, 0x01 }; struct slver isal_deflate_init_slver = { 0x0081, 0x03, 0x01 };
struct slver isal_deflate_stateless_init_slver_00010084;
struct slver isal_deflate_stateless_init_slver = { 0x0084, 0x01, 0x00 };
struct slver isal_deflate_slver_01030082; struct slver isal_deflate_slver_01030082;
struct slver isal_deflate_slver = { 0x0082, 0x03, 0x01 }; struct slver isal_deflate_slver = { 0x0082, 0x03, 0x01 };

101
igzip/igzip_decode_block_stateless.asm
View File

@ -81,13 +81,93 @@ extern rfc1951_lookup_table
start_out_mem_offset equ 0 start_out_mem_offset equ 0
read_in_mem_offset equ 8 read_in_mem_offset equ 8
read_in_length_mem_offset equ 16 read_in_length_mem_offset equ 16
stack_size equ 4 * 8 + 8 gpr_save_mem_offset equ 24
stack_size equ 3 * 8 + 8 * 8
%define _dist_extra_bit_count 264 %define _dist_extra_bit_count 264
%define _dist_start _dist_extra_bit_count + 1*32 %define _dist_start _dist_extra_bit_count + 1*32
%define _len_extra_bit_count _dist_start + 4*32 %define _len_extra_bit_count _dist_start + 4*32
%define _len_start _len_extra_bit_count + 1*32 %define _len_start _len_extra_bit_count + 1*32
%ifidn __OUTPUT_FORMAT__, elf64
%define arg0 rdi
%macro FUNC_SAVE 0
%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], rbp
mov [rsp + gpr_save_mem_offset + 2*8], r12
mov [rsp + gpr_save_mem_offset + 3*8], r13
mov [rsp + gpr_save_mem_offset + 4*8], r14
mov [rsp + gpr_save_mem_offset + 5*8], r15
%endm
%macro FUNC_RESTORE 0
mov rbx, [rsp + gpr_save_mem_offset + 0*8]
mov rbp, [rsp + gpr_save_mem_offset + 1*8]
mov r12, [rsp + gpr_save_mem_offset + 2*8]
mov r13, [rsp + gpr_save_mem_offset + 3*8]
mov r14, [rsp + gpr_save_mem_offset + 4*8]
mov r15, [rsp + gpr_save_mem_offset + 5*8]
%ifndef ALIGN_STACK
add rsp, stack_size
%else
mov rsp, rbp
pop rbp
%endif
%endm
%endif
%ifidn __OUTPUT_FORMAT__, win64
%define arg0 rcx
%macro FUNC_SAVE 0
%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
%endm
%macro FUNC_RESTORE 0
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
%endm
%endif
;; Load read_in and updated in_buffer accordingly ;; Load read_in and updated in_buffer accordingly
;; when there are at least 8 bytes in the in buffer ;; when there are at least 8 bytes in the in buffer
;; Clobbers rcx, unless rcx is %%read_in_length ;; Clobbers rcx, unless rcx is %%read_in_length
@ -261,16 +341,9 @@ stack_size equ 4 * 8 + 8
global decode_huffman_code_block_stateless_ %+ ARCH global decode_huffman_code_block_stateless_ %+ ARCH
decode_huffman_code_block_stateless_ %+ ARCH %+ : decode_huffman_code_block_stateless_ %+ ARCH %+ :
push rbx FUNC_SAVE
push rbp
push r12
push r13
push r14
push r15
sub rsp, stack_size mov state, arg0
mov state, rdi
lea rfc_lookup, [rfc1951_lookup_table] lea rfc_lookup, [rfc1951_lookup_table]
mov read_in,[state + _read_in] mov read_in,[state + _read_in]
@ -581,12 +654,6 @@ end:
sub end_in, next_in sub end_in, next_in
mov [state + _avail_in], end_in %+ d mov [state + _avail_in], end_in %+ d
add rsp, stack_size FUNC_RESTORE
pop r15
pop r14
pop r13
pop r12
pop rbp
pop rbx
ret ret

16
igzip/igzip_inflate.c
View File

@ -43,6 +43,22 @@ static struct rfc1951_tables rfc_lookup_table = {
0x0083, 0x00a3, 0x00c3, 0x00e3, 0x0102, 0x0000, 0x0000, 0x0000} 0x0083, 0x00a3, 0x00c3, 0x00e3, 0x0102, 0x0000, 0x0000, 0x0000}
}; };
struct slver {
uint16_t snum;
uint8_t ver;
uint8_t core;
};
/* Version info */
struct slver isal_inflate_init_slver_00010088;
struct slver isal_inflate_init_slver = { 0x0088, 0x01, 0x00 };
struct slver isal_inflate_stateless_slver_00010089;
struct slver isal_inflate_stateless_slver = { 0x0089, 0x01, 0x00 };
struct slver isal_inflate_slver_0001008a;
struct slver isal_inflate_slver = { 0x008a, 0x01, 0x00 };
/*Performs a copy of length repeat_length data starting at dest - /*Performs a copy of length repeat_length data starting at dest -
* lookback_distance into dest. This copy copies data previously copied when the * lookback_distance into dest. This copy copies data previously copied when the
* src buffer and the dest buffer overlap. */ * src buffer and the dest buffer overlap. */

128
igzip/igzip_update_histogram.asm
View File

@ -83,8 +83,6 @@ global %1
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
_eob_count_offset equ 0 ; local variable (8 bytes) _eob_count_offset equ 0 ; local variable (8 bytes)
f_end_i_mem_offset equ 8 f_end_i_mem_offset equ 8
gpr_save_mem_offset equ 16 ; gpr save area (8*8 bytes) gpr_save_mem_offset equ 16 ; gpr save area (8*8 bytes)
@ -92,10 +90,97 @@ 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 stack_size equ 2*8 + 8*8 + 4*16 + 8
;;; 8 because stack address is odd multiple of 8 after a function call and ;;; 8 because stack address is odd multiple of 8 after a function call and
;;; we want it aligned to 16 bytes ;;; we want it aligned to 16 bytes
%ifidn __OUTPUT_FORMAT__, elf64
%define arg0 rdi
%define arg1 rsi
%define arg2 rdx
%macro FUNC_SAVE 0
%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], rbp
mov [rsp + gpr_save_mem_offset + 2*8], r12
mov [rsp + gpr_save_mem_offset + 3*8], r13
mov [rsp + gpr_save_mem_offset + 4*8], r14
mov [rsp + gpr_save_mem_offset + 5*8], r15
%endm
%macro FUNC_RESTORE 0
mov rbx, [rsp + gpr_save_mem_offset + 0*8]
mov rbp, [rsp + gpr_save_mem_offset + 1*8]
mov r12, [rsp + gpr_save_mem_offset + 2*8]
mov r13, [rsp + gpr_save_mem_offset + 3*8]
mov r14, [rsp + gpr_save_mem_offset + 4*8]
mov r15, [rsp + gpr_save_mem_offset + 5*8]
%ifndef ALIGN_STACK
add rsp, stack_size
%else
mov rsp, rbp
pop rbp
%endif
%endm
%endif
%ifidn __OUTPUT_FORMAT__, win64
%define arg0 rcx
%define arg1 rdx
%define arg2 r8
%macro FUNC_SAVE 0
%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
%endm
%macro FUNC_RESTORE 0
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
%endm
%endif
_lit_len_offset equ 0 _lit_len_offset equ 0
_dist_offset equ (8 * LIT_LEN) _dist_offset equ (8 * LIT_LEN)
_hash_offset equ (_dist_offset + 8 * DIST_LEN) _hash_offset equ (_dist_offset + 8 * DIST_LEN)
%macro len_to_len_code 3 %macro len_to_len_code 3
%define %%len_code %1 ; Output %define %%len_code %1 ; Output
%define %%len %2 ; Input %define %%len %2 ; Input
@ -145,24 +230,18 @@ isal_update_histogram_ %+ ARCH %+ :
jne skip1 jne skip1
ret ret
skip1: skip1:
FUNC_SAVE
%ifdef ALIGN_STACK %ifnidn file_start, arg0
push rbp mov file_start, arg0
mov rbp, rsp %endif
sub rsp, stack_size %ifnidn file_length, arg1
and rsp, ~15 mov file_length, arg1
%else %endif
sub rsp, stack_size %ifnidn histogram, arg2
mov histogram, arg2
%endif %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 f_i, 0 mov f_i, 0
mov tmp1, qword [histogram + _lit_len_offset + 8*256] mov tmp1, qword [histogram + _lit_len_offset + 8*256]
@ -436,21 +515,8 @@ end:
mov tmp1, [rsp + _eob_count_offset] mov tmp1, [rsp + _eob_count_offset]
mov qword [histogram + _lit_len_offset + HIST_ELEM_SIZE * 256], tmp1 mov qword [histogram + _lit_len_offset + HIST_ELEM_SIZE * 256], tmp1
mov rbx, [rsp + gpr_save_mem_offset + 0*8] FUNC_RESTORE
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 ret
compare_loop: compare_loop: