generic-library/isa-l
1
0
Fork 0
mirror of https://github.com/intel/isa-l.git synced 2025-11-04 04:09:56 +01:00
Code Issues Projects Releases Wiki Activity

igzip: Implement multibyte decode

Browse Source 
Change-Id: I923a57a01f696f2082945fafcc2773655b9a5beb
Signed-off-by: Roy Oursler <roy.j.oursler@intel.com>
This commit is contained in:
Roy Oursler
2018-02-22 17:05:16 -05:00
committed by Greg Tucker
parent 0b8fe87648
commit 222a68f760
4 changed files with 337 additions and 184 deletions
Download Patch File Download Diff File Expand all files Collapse all files

131
igzip/igzip_decode_block_stateless.asm
View File

@@ -24,10 +24,25 @@ default rel
extern rfc1951_lookup_table extern rfc1951_lookup_table
%define LARGE_CODE_LENGTH_OFFSET 9
%define LARGE_FLAG_BIT_OFFSET 15 %define LARGE_SHORT_SYM_LEN 25
%define LARGE_SHORT_SYM_MASK ((1 << LARGE_SHORT_SYM_LEN) - 1)
%define LARGE_LONG_SYM_LEN 9
%define LARGE_LONG_SYM_MASK ((1 << LARGE_LONG_SYM_LEN) - 1)
%define LARGE_SHORT_CODE_LEN_OFFSET 28
%define LARGE_LONG_CODE_LEN_OFFSET 9
%define LARGE_FLAG_BIT_OFFSET 27
%define LARGE_FLAG_BIT (1 << LARGE_FLAG_BIT_OFFSET) %define LARGE_FLAG_BIT (1 << LARGE_FLAG_BIT_OFFSET)
%define SMALL_CODE_LENGTH_OFFSET 9 %define LARGE_SYM_COUNT_OFFSET 25
%define LARGE_SYM_COUNT_LEN 2
%define LARGE_SYM_COUNT_MASK ((1 << LARGE_SYM_COUNT_LEN) - 1)
%define SMALL_SHORT_SYM_LEN 9
%define SMALL_SHORT_SYM_MASK ((1 << SMALL_SHORT_SYM_LEN) - 1)
%define SMALL_LONG_SYM_LEN 9
%define SMALL_LONG_SYM_MASK ((1 << SMALL_LONG_SYM_LEN) - 1)
%define SMALL_SHORT_CODE_LEN_OFFSET 9
%define SMALL_LONG_CODE_LEN_OFFSET 9
%define SMALL_FLAG_BIT_OFFSET 15 %define SMALL_FLAG_BIT_OFFSET 15
%define SMALL_FLAG_BIT (1 << SMALL_FLAG_BIT_OFFSET) %define SMALL_FLAG_BIT (1 << SMALL_FLAG_BIT_OFFSET)
@@ -50,6 +65,7 @@ extern rfc1951_lookup_table
;; rsi arg2 ;; rsi arg2
%define tmp2 rsi %define tmp2 rsi
%define next_sym_num rsi
%define next_bits rsi %define next_bits rsi
;; rbx ; Saved ;; rbx ; Saved
@@ -85,8 +101,9 @@ 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
gpr_save_mem_offset equ 24 next_out_mem_offset equ 24
stack_size equ 3 * 8 + 8 * 8 gpr_save_mem_offset equ 32
stack_size equ 4 * 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
@@ -262,41 +279,49 @@ stack_size equ 3 * 8 + 8 * 8
;; Decode next symbol ;; Decode next symbol
;; Clobber rcx ;; Clobber rcx
%macro decode_next_large 7 %macro decode_next_large 8
%define %%state %1 ; State structure associated with compressed stream %define %%state %1 ; State structure associated with compressed stream
%define %%lookup_size %2 ; Number of bits used for small lookup %define %%lookup_size %2 ; Number of bits used for small lookup
%define %%state_offset %3 ; Type of huff code, should be either LIT or DIST %define %%state_offset %3 ; Type of huff code, should be either LIT or DIST
%define %%read_in %4 ; Bits read in from compressed stream %define %%read_in %4 ; Bits read in from compressed stream
%define %%read_in_length %5 ; Number of valid bits in read_in %define %%read_in_length %5 ; Number of valid bits in read_in
%define %%next_sym %6 ; Returned symobl %define %%next_sym %6 ; Returned symbols
%define %%next_bits %7 %define %%next_sym_num %7 ; Returned symbols count
%define %%next_bits %8
;; Save length associated with symbol ;; Save length associated with symbol
mov %%next_bits, %%read_in mov %%next_bits, %%read_in
shr %%next_bits, %%lookup_size shr %%next_bits, %%lookup_size
mov rcx, %%next_sym mov rcx, %%next_sym
shr rcx, LARGE_CODE_LENGTH_OFFSET shr rcx, LARGE_SHORT_CODE_LEN_OFFSET
jz invalid_symbol jz invalid_symbol
mov %%next_sym_num, %%next_sym
shr %%next_sym_num, LARGE_SYM_COUNT_OFFSET
and %%next_sym_num, LARGE_SYM_COUNT_MASK
;; Check if symbol or hint was looked up ;; Check if symbol or hint was looked up
and %%next_sym, 0x81FF and %%next_sym, LARGE_FLAG_BIT | LARGE_SHORT_SYM_MASK
cmp %%next_sym, LARGE_FLAG_BIT test %%next_sym, LARGE_FLAG_BIT
jl %%end jz %%end
;; Extract the 15-DECODE_LOOKUP_SIZE bits beyond the first %%lookup_size bits. ;; Extract the 15-DECODE_LOOKUP_SIZE bits beyond the first %%lookup_size bits.
lea %%next_sym, [%%state + LARGE_LONG_CODE_SIZE * %%next_sym]
CLEAR_HIGH_BITS %%next_bits, rcx, %%lookup_size CLEAR_HIGH_BITS %%next_bits, rcx, %%lookup_size
and %%next_sym, LARGE_SHORT_SYM_MASK
add %%next_sym, %%next_bits
lea %%next_sym, [%%state + LARGE_LONG_CODE_SIZE * %%next_sym]
;; Lookup actual next symbol ;; Lookup actual next symbol
movzx %%next_sym, word [%%next_sym + %%state_offset + LARGE_LONG_CODE_SIZE * %%next_bits + LARGE_SHORT_CODE_SIZE * (1 << %%lookup_size) - LARGE_LONG_CODE_SIZE * LARGE_FLAG_BIT] movzx %%next_sym, word [%%next_sym + %%state_offset + LARGE_SHORT_CODE_SIZE * (1 << %%lookup_size)]
mov %%next_sym_num, 1
;; Save length associated with symbol ;; Save length associated with symbol
mov rcx, %%next_sym mov rcx, %%next_sym
shr rcx, LARGE_CODE_LENGTH_OFFSET shr rcx, LARGE_LONG_CODE_LEN_OFFSET
jz invalid_symbol jz invalid_symbol
and %%next_sym, 0x1FF and %%next_sym, LARGE_LONG_SYM_MASK
%%end: %%end:
;; Updated read_in to reflect the bits which were decoded ;; Updated read_in to reflect the bits which were decoded
@@ -306,21 +331,22 @@ stack_size equ 3 * 8 + 8 * 8
;; Decode next symbol ;; Decode next symbol
;; Clobber rcx ;; Clobber rcx
%macro decode_next_large_with_load 7 %macro decode_next_large_with_load 8
%define %%state %1 ; State structure associated with compressed stream %define %%state %1 ; State structure associated with compressed stream
%define %%lookup_size %2 ; Number of bits used for small lookup %define %%lookup_size %2 ; Number of bits used for small lookup
%define %%state_offset %3 %define %%state_offset %3
%define %%read_in %4 ; Bits read in from compressed stream %define %%read_in %4 ; Bits read in from compressed stream
%define %%read_in_length %5 ; Number of valid bits in read_in %define %%read_in_length %5 ; Number of valid bits in read_in
%define %%next_sym %6 ; Returned symobl %define %%next_sym %6 ; Returned symbols
%define %%next_bits %7 %define %%next_sym_num %7 ; Returned symbols count
%define %%next_bits %8
;; Lookup possible next symbol ;; Lookup possible next symbol
mov %%next_bits, %%read_in mov %%next_bits, %%read_in
and %%next_bits, (1 << %%lookup_size) - 1 and %%next_bits, (1 << %%lookup_size) - 1
mov %%next_sym %+ d, dword [%%state + %%state_offset + LARGE_SHORT_CODE_SIZE * %%next_bits] mov %%next_sym %+ d, dword [%%state + %%state_offset + LARGE_SHORT_CODE_SIZE * %%next_bits]
decode_next_large %%state, %%lookup_size, %%state_offset, %%read_in, %%read_in_length, %%next_sym, %%next_bits decode_next_large %%state, %%lookup_size, %%state_offset, %%read_in, %%read_in_length, %%next_sym, %%next_sym_num, %%next_bits
%endm %endm
;; Decode next symbol ;; Decode next symbol
@@ -339,13 +365,13 @@ stack_size equ 3 * 8 + 8 * 8
shr %%next_bits, %%lookup_size shr %%next_bits, %%lookup_size
mov rcx, %%next_sym mov rcx, %%next_sym
shr rcx, SMALL_CODE_LENGTH_OFFSET shr rcx, SMALL_SHORT_CODE_LEN_OFFSET
jz invalid_symbol jz invalid_symbol
;; Check if symbol or hint was looked up ;; Check if symbol or hint was looked up
and %%next_sym, 0x81FF and %%next_sym, SMALL_FLAG_BIT | SMALL_SHORT_SYM_MASK
cmp %%next_sym, SMALL_FLAG_BIT test %%next_sym, SMALL_FLAG_BIT
jl %%end jz %%end
;; Extract the 15-DECODE_LOOKUP_SIZE bits beyond the first %%lookup_size bits. ;; Extract the 15-DECODE_LOOKUP_SIZE bits beyond the first %%lookup_size bits.
lea %%next_sym, [%%state + SMALL_LONG_CODE_SIZE * %%next_sym] lea %%next_sym, [%%state + SMALL_LONG_CODE_SIZE * %%next_sym]
@@ -357,9 +383,9 @@ stack_size equ 3 * 8 + 8 * 8
;; Save length associated with symbol ;; Save length associated with symbol
mov rcx, %%next_sym mov rcx, %%next_sym
shr rcx, SMALL_CODE_LENGTH_OFFSET shr rcx, SMALL_LONG_CODE_LEN_OFFSET
jz invalid_symbol jz invalid_symbol
and %%next_sym, 0x1FF and %%next_sym, SMALL_LONG_SYM_MASK
%%end: %%end:
;; Updated read_in to reflect the bits which were decoded ;; Updated read_in to reflect the bits which were decoded
@@ -440,10 +466,13 @@ loop_block:
jg end_loop_block_pre jg end_loop_block_pre
;; Decode next symbol and reload the read_in buffer ;; Decode next symbol and reload the read_in buffer
decode_next_large state, ISAL_DECODE_LONG_BITS, _lit_huff_code, read_in, read_in_length, next_sym, tmp1 decode_next_large state, ISAL_DECODE_LONG_BITS, _lit_huff_code, read_in, read_in_length, next_sym, next_sym_num, tmp1
;; Save next_sym in next_sym2 so next_sym can be preloaded ;; Specutively write next_sym2 if it is a literal
mov next_sym2, next_sym mov [next_out], next_sym
add next_out, next_sym_num
lea next_sym_num, [8 * next_sym_num - 8]
SHRX next_sym2, next_sym, next_sym_num
;; Find index to specutively preload next_sym from ;; Find index to specutively preload next_sym from
mov tmp3, read_in mov tmp3, read_in
@@ -478,10 +507,6 @@ loop_block:
and next_bits2, (1 << ISAL_DECODE_SHORT_BITS) - 1 and next_bits2, (1 << ISAL_DECODE_SHORT_BITS) - 1
movzx next_sym3, word [state + _dist_huff_code + SMALL_SHORT_CODE_SIZE * next_bits2] movzx next_sym3, word [state + _dist_huff_code + SMALL_SHORT_CODE_SIZE * next_bits2]
;; Specutively write next_sym2 if it is a literal
mov [next_out], next_sym2
add next_out, 1
;; Check if next_sym2 is a literal, length, or end of block symbol ;; Check if next_sym2 is a literal, length, or end of block symbol
cmp next_sym2, 256 cmp next_sym2, 256
jl loop_block jl loop_block
@@ -581,13 +606,18 @@ end_loop_block:
inflate_in_small_load next_in, end_in, read_in, read_in_length, tmp1, tmp2 inflate_in_small_load next_in, end_in, read_in, read_in_length, tmp1, tmp2
mov [rsp + read_in_mem_offset], read_in mov [rsp + read_in_mem_offset], read_in
mov [rsp + read_in_length_mem_offset], read_in_length mov [rsp + read_in_length_mem_offset], read_in_length
mov [rsp + next_out_mem_offset], next_out
decode_next_large_with_load state, ISAL_DECODE_LONG_BITS, _lit_huff_code, read_in, read_in_length, next_sym, tmp1 decode_next_large_with_load state, ISAL_DECODE_LONG_BITS, _lit_huff_code, read_in, read_in_length, next_sym, next_sym_num, tmp1
;; Check that enough input was available to decode symbol ;; Check that enough input was available to decode symbol
cmp read_in_length, 0 cmp read_in_length, 0
jl end_of_input jl end_of_input
multi_symbol_start:
cmp next_sym_num, 1
jg decode_literal
cmp next_sym, 256 cmp next_sym, 256
jl decode_literal jl decode_literal
je end_symbol je end_symbol
@@ -649,12 +679,19 @@ decode_literal:
jge out_buffer_overflow_lit jge out_buffer_overflow_lit
add next_out, 1 add next_out, 1
mov byte [next_out - 1], next_sym %+ b mov byte [next_out - 1], next_sym %+ b
jmp end_loop_block sub next_sym_num, 1
jz end_loop_block
shr next_sym, 8
jmp multi_symbol_start
;; Set exit codes ;; Set exit codes
end_of_input: end_of_input:
mov read_in, [rsp + read_in_mem_offset] mov read_in, [rsp + read_in_mem_offset]
mov read_in_length, [rsp + read_in_length_mem_offset] mov read_in_length, [rsp + read_in_length_mem_offset]
mov next_out, [rsp + next_out_mem_offset]
xor tmp1, tmp1
mov dword [state + _write_overflow_lits], tmp1 %+ d
mov dword [state + _write_overflow_len], tmp1 %+ d
mov rax, END_INPUT mov rax, END_INPUT
jmp end jmp end
@@ -674,10 +711,18 @@ out_buffer_overflow_repeat:
jmp end jmp end
out_buffer_overflow_lit: out_buffer_overflow_lit:
mov read_in, [rsp + read_in_mem_offset] mov dword [state + _write_overflow_lits], next_sym %+ d
mov read_in_length, [rsp + read_in_length_mem_offset] mov dword [state + _write_overflow_len], next_sym_num %+ d
sub next_sym_num, 1
shl next_sym_num, 3
SHRX next_sym, next_sym, next_sym_num
mov rax, OUT_OVERFLOW mov rax, OUT_OVERFLOW
jmp end shr next_sym_num, 3
cmp next_sym, 256
jl end
mov dword [state + _write_overflow_len], next_sym_num %+ d
jg decode_len_dist_2
jmp end_state
invalid_look_back_distance: invalid_look_back_distance:
mov rax, INVALID_LOOKBACK mov rax, INVALID_LOOKBACK
@@ -689,12 +734,18 @@ invalid_symbol:
end_symbol_pre: end_symbol_pre:
;; Fix up in buffer and out buffer to reflect the actual buffer ;; Fix up in buffer and out buffer to reflect the actual buffer
sub next_out, 1
add end_out, OUT_BUFFER_SLOP add end_out, OUT_BUFFER_SLOP
add end_in, IN_BUFFER_SLOP add end_in, IN_BUFFER_SLOP
end_symbol: end_symbol:
xor rax, rax
end_state:
;; Set flag identifying a new block is required ;; Set flag identifying a new block is required
mov byte [state + _block_state], ISAL_BLOCK_NEW_HDR mov byte [state + _block_state], ISAL_BLOCK_NEW_HDR
xor rax, rax cmp dword [state + _bfinal], 0
je end
mov byte [state + _block_state], ISAL_BLOCK_INPUT_DONE
end: end:
;; Save current buffer states ;; Save current buffer states
mov [state + _read_in], read_in mov [state + _read_in], read_in

238
igzip/igzip_inflate.c
View File

@@ -34,6 +34,28 @@
extern int decode_huffman_code_block_stateless(struct inflate_state *); extern int decode_huffman_code_block_stateless(struct inflate_state *);
#define LARGE_SHORT_SYM_LEN 25
#define LARGE_SHORT_SYM_MASK ((1 << LARGE_SHORT_SYM_LEN) - 1)
#define LARGE_LONG_SYM_LEN 9
#define LARGE_LONG_SYM_MASK ((1 << LARGE_LONG_SYM_LEN) - 1)
#define LARGE_SHORT_CODE_LEN_OFFSET 28
#define LARGE_LONG_CODE_LEN_OFFSET 9
#define LARGE_FLAG_BIT_OFFSET 27
#define LARGE_FLAG_BIT (1 << LARGE_FLAG_BIT_OFFSET)
#define LARGE_SYM_COUNT_OFFSET 25
#define LARGE_SYM_COUNT_LEN 2
#define LARGE_SYM_COUNT_MASK ((1 << LARGE_SYM_COUNT_LEN) - 1)
#define SMALL_SHORT_SYM_LEN 9
#define SMALL_SHORT_SYM_MASK ((1 << SMALL_SHORT_SYM_LEN) - 1)
#define SMALL_LONG_SYM_LEN 9
#define SMALL_LONG_SYM_MASK ((1 << SMALL_LONG_SYM_LEN) - 1)
#define SMALL_SHORT_CODE_LEN_OFFSET 9
#define SMALL_LONG_CODE_LEN_OFFSET 9
#define SMALL_FLAG_BIT_OFFSET 15
#define SMALL_FLAG_BIT (1 << SMALL_FLAG_BIT_OFFSET)
#define INVALID_SYMBOL 0x1FF
/* structure contain lookup data based on RFC 1951 */ /* structure contain lookup data based on RFC 1951 */
struct rfc1951_tables { struct rfc1951_tables {
uint8_t dist_extra_bit_count[32]; uint8_t dist_extra_bit_count[32];
@@ -275,7 +297,8 @@ static void inline make_inflate_huff_code_large(struct inflate_huff_code_large *
if (i < max_symbol) if (i < max_symbol)
short_code_lookup[huff_code_table[i].code] = short_code_lookup[huff_code_table[i].code] =
i | (huff_code_table[i].length) << 9; i | (huff_code_table[i].length) << LARGE_SHORT_CODE_LEN_OFFSET |
(1 << LARGE_SYM_COUNT_OFFSET);
else else
short_code_lookup[huff_code_table[i].code] = 0; short_code_lookup[huff_code_table[i].code] = 0;
@@ -339,12 +362,13 @@ static void inline make_inflate_huff_code_large(struct inflate_huff_code_large *
for (; long_bits < (1 << (max_length - ISAL_DECODE_LONG_BITS)); for (; long_bits < (1 << (max_length - ISAL_DECODE_LONG_BITS));
long_bits += min_increment) { long_bits += min_increment) {
result->long_code_lookup[long_code_lookup_length + long_bits] = result->long_code_lookup[long_code_lookup_length + long_bits] =
temp_code_list[j] | (code_length << 9); temp_code_list[j] |
(code_length << LARGE_LONG_CODE_LEN_OFFSET);
} }
huff_code_table[temp_code_list[j]].code = 0xFFFF; huff_code_table[temp_code_list[j]].code = 0xFFFF;
} }
result->short_code_lookup[first_bits] = result->short_code_lookup[first_bits] = long_code_lookup_length |
long_code_lookup_length | (max_length << 9) | 0x8000; (max_length << LARGE_SHORT_CODE_LEN_OFFSET) | LARGE_FLAG_BIT;
long_code_lookup_length += 1 << (max_length - ISAL_DECODE_LONG_BITS); long_code_lookup_length += 1 << (max_length - ISAL_DECODE_LONG_BITS);
} }
@@ -434,7 +458,7 @@ static void inline make_inflate_huff_code_small(struct inflate_huff_code_small *
* bit 15 is a flag representing this is a symbol*/ * bit 15 is a flag representing this is a symbol*/
if (i < max_symbol) if (i < max_symbol)
short_code_lookup[huff_code_table[i].code] = short_code_lookup[huff_code_table[i].code] =
i | (huff_code_table[i].length) << 9; i | (huff_code_table[i].length) << SMALL_SHORT_CODE_LEN_OFFSET;
else else
short_code_lookup[huff_code_table[i].code] = 0; short_code_lookup[huff_code_table[i].code] = 0;
} }
@@ -496,12 +520,13 @@ static void inline make_inflate_huff_code_small(struct inflate_huff_code_small *
for (; long_bits < (1 << (max_length - ISAL_DECODE_SHORT_BITS)); for (; long_bits < (1 << (max_length - ISAL_DECODE_SHORT_BITS));
long_bits += min_increment) { long_bits += min_increment) {
result->long_code_lookup[long_code_lookup_length + long_bits] = result->long_code_lookup[long_code_lookup_length + long_bits] =
temp_code_list[j] | (code_length << 9); temp_code_list[j] |
(code_length << SMALL_LONG_CODE_LEN_OFFSET);
} }
huff_code_table[temp_code_list[j]].code = 0xFFFF; huff_code_table[temp_code_list[j]].code = 0xFFFF;
} }
result->short_code_lookup[first_bits] = result->short_code_lookup[first_bits] = long_code_lookup_length |
long_code_lookup_length | (max_length << 9) | 0x8000; (max_length << SMALL_SHORT_CODE_LEN_OFFSET) | SMALL_FLAG_BIT;
long_code_lookup_length += 1 << (max_length - ISAL_DECODE_SHORT_BITS); long_code_lookup_length += 1 << (max_length - ISAL_DECODE_SHORT_BITS);
} }
@@ -556,12 +581,13 @@ static int inline setup_static_header(struct inflate_state *state)
} }
/* Decodes the next symbol symbol in in_buffer using the huff code defined by /* Decodes the next symbol symbol in in_buffer using the huff code defined by
* huff_code */ * huff_code and returns the value in next_lits and sym_count */
static uint16_t inline decode_next_large(struct inflate_state *state, static void inline decode_next_large(uint32_t * next_lits, uint32_t * sym_count,
struct inflate_state *state,
struct inflate_huff_code_large *huff_code) struct inflate_huff_code_large *huff_code)
{ {
uint16_t next_bits; uint16_t next_bits;
uint16_t next_sym; uint32_t next_sym;
uint32_t bit_count; uint32_t bit_count;
uint32_t bit_mask; uint32_t bit_mask;
@@ -578,36 +604,37 @@ static uint16_t inline decode_next_large(struct inflate_state *state,
* first actual symbol in the long code list.*/ * first actual symbol in the long code list.*/
next_sym = huff_code->short_code_lookup[next_bits]; next_sym = huff_code->short_code_lookup[next_bits];
if (next_sym < 0x8000) { if ((next_sym & LARGE_FLAG_BIT) == 0) {
/* Return symbol found if next_code is a complete huffman code /* Return symbol found if next_code is a complete huffman code
* and shift in buffer over by the length of the next_code */ * and shift in buffer over by the length of the next_code */
bit_count = next_sym >> 9; bit_count = next_sym >> LARGE_SHORT_CODE_LEN_OFFSET;
state->read_in >>= bit_count; state->read_in >>= bit_count;
state->read_in_length -= bit_count; state->read_in_length -= bit_count;
if (bit_count == 0) if (bit_count == 0)
next_sym = 0x1FF; next_sym = INVALID_SYMBOL;
return next_sym & 0x1FF; *sym_count = (next_sym >> LARGE_SYM_COUNT_OFFSET) & LARGE_SYM_COUNT_MASK;
*next_lits = next_sym & LARGE_SHORT_SYM_MASK;
} else { } else {
/* If a symbol is not found, perform a linear search of the long code /* If a symbol is not found, do a lookup in the long code
* list starting from the hint in next_sym */ * list starting from the hint in next_sym */
bit_mask = (next_sym - 0x8000) >> 9; bit_mask = (next_sym - LARGE_FLAG_BIT) >> LARGE_SHORT_CODE_LEN_OFFSET;
bit_mask = (1 << bit_mask) - 1; bit_mask = (1 << bit_mask) - 1;
next_bits = state->read_in & bit_mask; next_bits = state->read_in & bit_mask;
next_sym = next_sym =
huff_code->long_code_lookup[(next_sym & 0x1FF) + huff_code->long_code_lookup[(next_sym & LARGE_SHORT_SYM_MASK) +
(next_bits >> ISAL_DECODE_LONG_BITS)]; (next_bits >> ISAL_DECODE_LONG_BITS)];
bit_count = next_sym >> 9; bit_count = next_sym >> LARGE_LONG_CODE_LEN_OFFSET;
state->read_in >>= bit_count; state->read_in >>= bit_count;
state->read_in_length -= bit_count; state->read_in_length -= bit_count;
if (bit_count == 0) if (bit_count == 0)
next_sym = 0x1FF; next_sym = INVALID_SYMBOL;
return next_sym & 0x1FF;
*sym_count = 1;
*next_lits = next_sym & LARGE_LONG_SYM_MASK;
} }
} }
@@ -632,31 +659,31 @@ static uint16_t inline decode_next_small(struct inflate_state *state,
* first actual symbol in the long code list.*/ * first actual symbol in the long code list.*/
next_sym = huff_code->short_code_lookup[next_bits]; next_sym = huff_code->short_code_lookup[next_bits];
if (next_sym < 0x8000) { if ((next_sym & SMALL_FLAG_BIT) == 0) {
/* Return symbol found if next_code is a complete huffman code /* Return symbol found if next_code is a complete huffman code
* and shift in buffer over by the length of the next_code */ * and shift in buffer over by the length of the next_code */
bit_count = next_sym >> 9; bit_count = next_sym >> SMALL_SHORT_CODE_LEN_OFFSET;
state->read_in >>= bit_count; state->read_in >>= bit_count;
state->read_in_length -= bit_count; state->read_in_length -= bit_count;
if (bit_count == 0) if (bit_count == 0)
next_sym = 0x1FF; next_sym = INVALID_SYMBOL;
return next_sym & 0x1FF; return next_sym & SMALL_SHORT_SYM_MASK;
} else { } else {
/* If a symbol is not found, perform a linear search of the long code /* If a symbol is not found, perform a linear search of the long code
* list starting from the hint in next_sym */ * list starting from the hint in next_sym */
bit_mask = (next_sym - 0x8000) >> 9; bit_mask = (next_sym - SMALL_FLAG_BIT) >> SMALL_SHORT_CODE_LEN_OFFSET;
bit_mask = (1 << bit_mask) - 1; bit_mask = (1 << bit_mask) - 1;
next_bits = state->read_in & bit_mask; next_bits = state->read_in & bit_mask;
next_sym = next_sym =
huff_code->long_code_lookup[(next_sym & 0x1FF) + huff_code->long_code_lookup[(next_sym & SMALL_SHORT_SYM_MASK) +
(next_bits >> ISAL_DECODE_SHORT_BITS)]; (next_bits >> ISAL_DECODE_SHORT_BITS)];
bit_count = next_sym >> 9; bit_count = next_sym >> SMALL_LONG_CODE_LEN_OFFSET;
state->read_in >>= bit_count; state->read_in >>= bit_count;
state->read_in_length -= bit_count; state->read_in_length -= bit_count;
return next_sym & 0x1FF; return next_sym & SMALL_LONG_SYM_MASK;
} }
} }
@@ -865,13 +892,6 @@ static int read_header(struct inflate_state *state)
state->read_in >>= 16; state->read_in >>= 16;
state->read_in_length -= 32; state->read_in_length -= 32;
bytes = state->read_in_length / 8;
state->next_in -= bytes;
state->avail_in += bytes;
state->read_in = 0;
state->read_in_length = 0;
/* Check if len and nlen match */ /* Check if len and nlen match */
if (len != (~nlen & 0xffff)) if (len != (~nlen & 0xffff))
return ISAL_INVALID_BLOCK; return ISAL_INVALID_BLOCK;
@@ -948,21 +968,48 @@ static int read_header_stateful(struct inflate_state *state)
static int inline decode_literal_block(struct inflate_state *state) static int inline decode_literal_block(struct inflate_state *state)
{ {
uint32_t len = state->type0_block_len; uint32_t len = state->type0_block_len;
uint32_t bytes = state->read_in_length / 8;
/* If the block is uncompressed, perform a memcopy while /* If the block is uncompressed, perform a memcopy while
* updating state data */ * updating state data */
state->block_state = state->bfinal ? ISAL_BLOCK_INPUT_DONE : ISAL_BLOCK_NEW_HDR;
state->block_state = ISAL_BLOCK_NEW_HDR;
if (state->avail_out < len) { if (state->avail_out < len) {
len = state->avail_out; len = state->avail_out;
state->block_state = ISAL_BLOCK_TYPE0; state->block_state = ISAL_BLOCK_TYPE0;
} }
if (state->avail_in < len) { if (state->avail_in + bytes < len) {
len = state->avail_in; len = state->avail_in + bytes;
state->block_state = ISAL_BLOCK_TYPE0; state->block_state = ISAL_BLOCK_TYPE0;
} }
if (state->read_in_length) {
if (len >= bytes) {
memcpy(state->next_out, &state->read_in, bytes);
state->next_out += bytes;
state->avail_out -= bytes;
state->total_out += bytes;
state->type0_block_len -= bytes;
state->read_in = 0;
state->read_in_length = 0;
len -= bytes;
bytes = 0;
} else {
memcpy(state->next_out, &state->read_in, len);
state->next_out += len;
state->avail_out -= len;
state->total_out += len;
state->type0_block_len -= len;
state->read_in >>= 8 * len;
state->read_in_length -= 8 * len;
bytes -= len;
len = 0;
}
}
memcpy(state->next_out, state->next_in, len); memcpy(state->next_out, state->next_in, len);
state->next_out += len; state->next_out += len;
@@ -972,7 +1019,7 @@ static int inline decode_literal_block(struct inflate_state *state)
state->avail_in -= len; state->avail_in -= len;
state->type0_block_len -= len; state->type0_block_len -= len;
if (state->avail_in == 0 && state->block_state != ISAL_BLOCK_NEW_HDR) if (state->avail_in + bytes == 0 && state->block_state != ISAL_BLOCK_INPUT_DONE)
return ISAL_END_INPUT; return ISAL_END_INPUT;
if (state->avail_out == 0 && state->type0_block_len > 0) if (state->avail_out == 0 && state->type0_block_len > 0)
@@ -991,8 +1038,9 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
uint32_t look_back_dist; uint32_t look_back_dist;
uint64_t read_in_tmp; uint64_t read_in_tmp;
int32_t read_in_length_tmp; int32_t read_in_length_tmp;
uint8_t *next_in_tmp; uint8_t *next_in_tmp, *next_out_tmp;
uint32_t avail_in_tmp; uint32_t avail_in_tmp, avail_out_tmp, total_out_tmp;
uint32_t next_lits, sym_count;
state->copy_overflow_length = 0; state->copy_overflow_length = 0;
state->copy_overflow_distance = 0; state->copy_overflow_distance = 0;
@@ -1006,8 +1054,14 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
read_in_length_tmp = state->read_in_length; read_in_length_tmp = state->read_in_length;
next_in_tmp = state->next_in; next_in_tmp = state->next_in;
avail_in_tmp = state->avail_in; avail_in_tmp = state->avail_in;
next_out_tmp = state->next_out;
avail_out_tmp = state->avail_out;
total_out_tmp = state->total_out;
next_lit = decode_next_large(state, &state->lit_huff_code); decode_next_large(&next_lits, &sym_count, state, &state->lit_huff_code);
if (sym_count == 0)
return ISAL_INVALID_SYMBOL;
if (state->read_in_length < 0) { if (state->read_in_length < 0) {
state->read_in = read_in_tmp; state->read_in = read_in_tmp;
@@ -1017,16 +1071,29 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
return ISAL_END_INPUT; return ISAL_END_INPUT;
} }
if (next_lit < 256) { while (sym_count > 0) {
next_lit = next_lits & 0xffff;
if (next_lit < 256 || sym_count > 1) {
/* If the next symbol is a literal, /* If the next symbol is a literal,
* write out the symbol and update state * write out the symbol and update state
* data accordingly. */ * data accordingly. */
if (state->avail_out < 1) { if (state->avail_out < 1) {
state->read_in = read_in_tmp; state->write_overflow_lits = next_lits;
state->read_in_length = read_in_length_tmp; state->write_overflow_len = sym_count;
state->next_in = next_in_tmp; next_lits = next_lits >> (8 * (sym_count - 1));
state->avail_in = avail_in_tmp; sym_count = 1;
if (next_lits < 256)
return ISAL_OUT_OVERFLOW; return ISAL_OUT_OVERFLOW;
else if (next_lits == 256) {
state->write_overflow_len -= 1;
state->block_state = state->bfinal ?
ISAL_BLOCK_INPUT_DONE : ISAL_BLOCK_NEW_HDR;
return ISAL_OUT_OVERFLOW;
} else {
state->write_overflow_len -= 1;
continue;
}
} }
*state->next_out = next_lit; *state->next_out = next_lit;
@@ -1038,7 +1105,8 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
/* If the next symbol is the end of /* If the next symbol is the end of
* block, update the state data * block, update the state data
* accordingly */ * accordingly */
state->block_state = ISAL_BLOCK_NEW_HDR; state->block_state = state->bfinal ?
ISAL_BLOCK_INPUT_DONE : ISAL_BLOCK_NEW_HDR;
} else if (next_lit < 286) { } else if (next_lit < 286) {
/* Else if the next symbol is a repeat /* Else if the next symbol is a repeat
@@ -1050,8 +1118,8 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
repeat_length = repeat_length =
rfc_lookup_table.len_start[next_lit - 257] + rfc_lookup_table.len_start[next_lit - 257] +
inflate_in_read_bits(state, inflate_in_read_bits(state,
rfc_lookup_table.len_extra_bit_count[next_lit rfc_lookup_table.len_extra_bit_count
- 257]); [next_lit - 257]);
next_dist = decode_next_small(state, &state->dist_huff_code); next_dist = decode_next_small(state, &state->dist_huff_code);
if (next_dist >= DIST_LEN) if (next_dist >= DIST_LEN)
@@ -1067,6 +1135,11 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
state->read_in_length = read_in_length_tmp; state->read_in_length = read_in_length_tmp;
state->next_in = next_in_tmp; state->next_in = next_in_tmp;
state->avail_in = avail_in_tmp; state->avail_in = avail_in_tmp;
state->next_out = next_out_tmp;
state->avail_out = avail_out_tmp;
state->total_out = total_out_tmp;
state->write_overflow_lits = 0;
state->write_overflow_len = 0;
return ISAL_END_INPUT; return ISAL_END_INPUT;
} }
@@ -1074,16 +1147,19 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
return ISAL_INVALID_LOOKBACK; return ISAL_INVALID_LOOKBACK;
if (state->avail_out < repeat_length) { if (state->avail_out < repeat_length) {
state->copy_overflow_length = repeat_length - state->avail_out; state->copy_overflow_length =
repeat_length - state->avail_out;
state->copy_overflow_distance = look_back_dist; state->copy_overflow_distance = look_back_dist;
repeat_length = state->avail_out; repeat_length = state->avail_out;
} }
if (look_back_dist > repeat_length) if (look_back_dist > repeat_length)
memcpy(state->next_out, memcpy(state->next_out,
state->next_out - look_back_dist, repeat_length); state->next_out - look_back_dist,
repeat_length);
else else
byte_copy(state->next_out, look_back_dist, repeat_length); byte_copy(state->next_out, look_back_dist,
repeat_length);
state->next_out += repeat_length; state->next_out += repeat_length;
state->avail_out -= repeat_length; state->avail_out -= repeat_length;
@@ -1095,6 +1171,11 @@ int decode_huffman_code_block_stateless_base(struct inflate_state *state)
/* Else the read in bits do not /* Else the read in bits do not
* correspond to any valid symbol */ * correspond to any valid symbol */
return ISAL_INVALID_SYMBOL; return ISAL_INVALID_SYMBOL;
next_lits >>= 8;
sym_count--;
}
} }
return 0; return 0;
} }
@@ -1115,6 +1196,8 @@ void isal_inflate_init(struct inflate_state *state)
state->crc_flag = 0; state->crc_flag = 0;
state->crc = 0; state->crc = 0;
state->type0_block_len = 0; state->type0_block_len = 0;
state->write_overflow_lits = 0;
state->write_overflow_len = 0;
state->copy_overflow_length = 0; state->copy_overflow_length = 0;
state->copy_overflow_distance = 0; state->copy_overflow_distance = 0;
state->tmp_in_size = 0; state->tmp_in_size = 0;
@@ -1132,6 +1215,8 @@ void isal_inflate_reset(struct inflate_state *state)
state->bfinal = 0; state->bfinal = 0;
state->crc = 0; state->crc = 0;
state->type0_block_len = 0; state->type0_block_len = 0;
state->write_overflow_lits = 0;
state->write_overflow_len = 0;
state->copy_overflow_length = 0; state->copy_overflow_length = 0;
state->copy_overflow_distance = 0; state->copy_overflow_distance = 0;
state->tmp_in_size = 0; state->tmp_in_size = 0;
@@ -1187,7 +1272,7 @@ int isal_inflate_stateless(struct inflate_state *state)
if (ret) if (ret)
break; break;
if (state->bfinal != 0 && state->block_state == ISAL_BLOCK_NEW_HDR) if (state->block_state == ISAL_BLOCK_INPUT_DONE)
state->block_state = ISAL_BLOCK_FINISH; state->block_state = ISAL_BLOCK_FINISH;
} }
@@ -1242,13 +1327,17 @@ int isal_inflate(struct inflate_state *state)
if (ret) if (ret)
break; break;
if (state->bfinal != 0
&& state->block_state == ISAL_BLOCK_NEW_HDR) {
state->block_state = ISAL_BLOCK_INPUT_DONE;
}
} }
/* Copy valid data from internal buffer into out_buffer */ /* Copy valid data from internal buffer into out_buffer */
if (state->write_overflow_len != 0) {
*(uint32_t *) state->next_out = state->write_overflow_lits;
state->next_out += state->write_overflow_len;
state->total_out += state->write_overflow_len;
state->write_overflow_lits = 0;
state->write_overflow_len = 0;
}
if (state->copy_overflow_length != 0) { if (state->copy_overflow_length != 0) {
byte_copy(state->next_out, state->copy_overflow_distance, byte_copy(state->next_out, state->copy_overflow_distance,
state->copy_overflow_length); state->copy_overflow_length);
@@ -1304,16 +1393,15 @@ int isal_inflate(struct inflate_state *state)
ret = decode_huffman_code_block_stateless(state); ret = decode_huffman_code_block_stateless(state);
if (ret) if (ret)
break; break;
if (state->bfinal != 0
&& state->block_state == ISAL_BLOCK_NEW_HDR)
state->block_state = ISAL_BLOCK_INPUT_DONE;
} }
} }
if (state->crc_flag) if (state->crc_flag)
update_checksum(state, start_out, state->next_out - start_out); update_checksum(state, start_out, state->next_out - start_out);
if (state->block_state != ISAL_BLOCK_INPUT_DONE) { if (state->block_state != ISAL_BLOCK_INPUT_DONE
|| state->copy_overflow_length + state->write_overflow_len +
state->tmp_out_valid > sizeof(state->tmp_out_buffer)) {
/* Save decompression history in tmp_out buffer */ /* Save decompression history in tmp_out buffer */
if (state->tmp_out_valid == state->tmp_out_processed if (state->tmp_out_valid == state->tmp_out_processed
&& avail_out - state->avail_out >= ISAL_DEF_HIST_SIZE) { && avail_out - state->avail_out >= ISAL_DEF_HIST_SIZE) {
@@ -1335,11 +1423,21 @@ int isal_inflate(struct inflate_state *state)
state->tmp_out_processed -= shift_size; state->tmp_out_processed -= shift_size;
} }
}
/* Write overflow data into tmp buffer */
if (state->write_overflow_len != 0) {
*(uint32_t *) & state->tmp_out_buffer[state->tmp_out_valid] =
state->write_overflow_lits;
state->tmp_out_valid += state->write_overflow_len;
state->total_out += state->write_overflow_len;
state->write_overflow_lits = 0;
state->write_overflow_len = 0;
}
if (state->copy_overflow_length != 0) { if (state->copy_overflow_length != 0) {
byte_copy(&state->tmp_out_buffer[state->tmp_out_valid], byte_copy(&state->tmp_out_buffer[state->tmp_out_valid],
state->copy_overflow_distance, state->copy_overflow_distance, state->copy_overflow_length);
state->copy_overflow_length);
state->tmp_out_valid += state->copy_overflow_length; state->tmp_out_valid += state->copy_overflow_length;
state->total_out += state->copy_overflow_length; state->total_out += state->copy_overflow_length;
state->copy_overflow_distance = 0; state->copy_overflow_distance = 0;
@@ -1348,12 +1446,12 @@ int isal_inflate(struct inflate_state *state)
if (ret == ISAL_INVALID_LOOKBACK || ret == ISAL_INVALID_BLOCK if (ret == ISAL_INVALID_LOOKBACK || ret == ISAL_INVALID_BLOCK
|| ret == ISAL_INVALID_SYMBOL) { || ret == ISAL_INVALID_SYMBOL) {
state->total_out -= state->total_out -= state->tmp_out_valid - state->tmp_out_processed;
state->tmp_out_valid - state->tmp_out_processed;
return ret; return ret;
} }
} else if (state->tmp_out_valid == state->tmp_out_processed) { if (state->block_state == ISAL_BLOCK_INPUT_DONE
&& state->tmp_out_valid == state->tmp_out_processed) {
state->block_state = ISAL_BLOCK_FINISH; state->block_state = ISAL_BLOCK_FINISH;
if (state->crc_flag == ISAL_ZLIB if (state->crc_flag == ISAL_ZLIB
|| state->crc_flag == ISAL_ZLIB_NO_HDR) || state->crc_flag == ISAL_ZLIB_NO_HDR)

4
igzip/inflate_data_structs.asm
View File

@@ -119,6 +119,8 @@ FIELD _bfinal, 4, 4
FIELD _crc_flag, 4, 4 FIELD _crc_flag, 4, 4
FIELD _crc, 4, 4 FIELD _crc, 4, 4
FIELD _type0_block_len, 4, 4 FIELD _type0_block_len, 4, 4
FIELD _write_overflow_lits, 4, 4
FIELD _write_overflow_len, 4, 4
FIELD _copy_overflow_len, 4, 4 FIELD _copy_overflow_len, 4, 4
FIELD _copy_overflow_dist, 4, 4 FIELD _copy_overflow_dist, 4, 4
@@ -135,7 +137,7 @@ ISAL_BLOCK_NEW_HDR equ 0
ISAL_BLOCK_HDR equ 1 ISAL_BLOCK_HDR equ 1
ISAL_BLOCK_TYPE0 equ 2 ISAL_BLOCK_TYPE0 equ 2
ISAL_BLOCK_CODED equ 3 ISAL_BLOCK_CODED equ 3
ISAL_BLOCK_END equ 4 ISAL_BLOCK_INPUT_DONE equ 4
ISAL_BLOCK_FINISH equ 5 ISAL_BLOCK_FINISH equ 5
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

2
include/igzip_lib.h
View File

@@ -472,6 +472,8 @@ struct inflate_state {
uint32_t crc_flag; //!< Flag identifying whether to track of crc uint32_t crc_flag; //!< Flag identifying whether to track of crc
uint32_t crc; //!< Contains crc of output if crc_flag is set uint32_t crc; //!< Contains crc of output if crc_flag is set
int32_t type0_block_len; //!< Length left to read of type 0 block when outbuffer overflow occured int32_t type0_block_len; //!< Length left to read of type 0 block when outbuffer overflow occured
int32_t write_overflow_lits;
int32_t write_overflow_len;
int32_t copy_overflow_length; //!< Length left to copy when outbuffer overflow occured int32_t copy_overflow_length; //!< Length left to copy when outbuffer overflow occured
int32_t copy_overflow_distance; //!< Lookback distance when outbuffer overlow occured int32_t copy_overflow_distance; //!< Lookback distance when outbuffer overlow occured
int32_t tmp_in_size; //!< Number of bytes in tmp_in_buffer int32_t tmp_in_size; //!< Number of bytes in tmp_in_buffer