1
; inffasx64.asm is a hand tuned assembler version of inffast.c - fast decoding
2
; version for AMD64 on Windows using Microsoft C compiler
4
; inffasx64.asm is automatically convert from AMD64 portion of inffas86.c
5
; inffasx64.asm is called by inffas8664.c, which contain more info.
8
; to compile this file, I use option
9
; ml64.exe /Flinffasx64 /c /Zi inffasx64.asm
10
; with Microsoft Macro Assembler (x64) for AMD64
12
; ml64.exe is given with Visual Studio 2005, Windows 2003 server DDK
14
; (you can get Windows 2003 server DDK with ml64 and cl.exe for AMD64 from
15
; http://www.microsoft.com/whdc/devtools/ddk/default.mspx for low price)
21
; see http://weblogs.asp.net/oldnewthing/archive/2004/01/14/58579.aspx and
22
; http://msdn.microsoft.com/library/en-us/kmarch/hh/kmarch/64bitAMD_8e951dd2-ee77-4728-8702-55ce4b5dd24a.xml.asp
24
; All registers must be preserved across the call, except for
25
; rax, rcx, rdx, r8, r-9, r10, and r11, which are scratch.
38
mov [rax+8], rbp ; /* save regs rbp and rsp */
41
mov rsp, rax ; /* make rsp point to &ar */
43
mov rsi, [rsp+16] ; /* rsi = in */
44
mov rdi, [rsp+32] ; /* rdi = out */
45
mov r9, [rsp+24] ; /* r9 = last */
46
mov r10, [rsp+48] ; /* r10 = end */
47
mov rbp, [rsp+64] ; /* rbp = lcode */
48
mov r11, [rsp+72] ; /* r11 = dcode */
49
mov rdx, [rsp+80] ; /* rdx = hold */
50
mov ebx, [rsp+88] ; /* ebx = bits */
51
mov r12d, [rsp+100] ; /* r12d = lmask */
52
mov r13d, [rsp+104] ; /* r13d = dmask */
59
je L_one_time ; /* if only one decode left */
66
mov r8, r12 ; /* r8 = lmask */
68
ja L_get_length_code_one_time
70
lodsd ; /* eax = *(uint *)in++ */
71
mov cl, bl ; /* cl = bits, needs it for shifting */
72
add bl, 32 ; /* bits += 32 */
74
or rdx, rax ; /* hold |= *((uint *)in)++ << bits */
75
jmp L_get_length_code_one_time
85
mov r8, r12 ; /* r8 = lmask */
87
ja L_get_length_code ; /* if (32 < bits) */
89
lodsd ; /* eax = *(uint *)in++ */
90
mov cl, bl ; /* cl = bits, needs it for shifting */
91
add bl, 32 ; /* bits += 32 */
93
or rdx, rax ; /* hold |= *((uint *)in)++ << bits */
96
and r8, rdx ; /* r8 &= hold */
97
mov eax, [rbp+r8*4] ; /* eax = lcode[hold & lmask] */
99
mov cl, ah ; /* cl = this.bits */
100
sub bl, ah ; /* bits -= this.bits */
101
shr rdx, cl ; /* hold >>= this.bits */
104
jnz L_test_for_length_base ; /* if (op != 0) 45.7% */
106
mov r8, r12 ; /* r8 = lmask */
107
shr eax, 16 ; /* output this.val char */
110
L_get_length_code_one_time:
111
and r8, rdx ; /* r8 &= hold */
112
mov eax, [rbp+r8*4] ; /* eax = lcode[hold & lmask] */
115
mov cl, ah ; /* cl = this.bits */
116
sub bl, ah ; /* bits -= this.bits */
117
shr rdx, cl ; /* hold >>= this.bits */
120
jnz L_test_for_length_base ; /* if (op != 0) 45.7% */
122
shr eax, 16 ; /* output this.val char */
127
L_test_for_length_base:
128
mov r14d, eax ; /* len = this */
129
shr r14d, 16 ; /* len = this.val */
133
jz L_test_for_second_level_length ; /* if ((op & 16) == 0) 8% */
134
and cl, 15 ; /* op &= 15 */
135
jz L_decode_distance ; /* if (!op) */
143
and eax, edx ; /* eax &= hold */
145
add r14d, eax ; /* len += hold & mask[op] */
148
mov r8, r13 ; /* r8 = dmask */
150
ja L_get_distance_code ; /* if (32 < bits) */
152
lodsd ; /* eax = *(uint *)in++ */
153
mov cl, bl ; /* cl = bits, needs it for shifting */
154
add bl, 32 ; /* bits += 32 */
156
or rdx, rax ; /* hold |= *((uint *)in)++ << bits */
159
and r8, rdx ; /* r8 &= hold */
160
mov eax, [r11+r8*4] ; /* eax = dcode[hold & dmask] */
163
mov r15d, eax ; /* dist = this */
164
shr r15d, 16 ; /* dist = this.val */
166
sub bl, ah ; /* bits -= this.bits */
167
shr rdx, cl ; /* hold >>= this.bits */
168
mov cl, al ; /* cl = this.op */
170
test al, 16 ; /* if ((op & 16) == 0) */
171
jz L_test_for_second_level_dist
172
and cl, 15 ; /* op &= 15 */
180
dec eax ; /* (1 << op) - 1 */
181
and eax, edx ; /* eax &= hold */
183
add r15d, eax ; /* dist += hold & ((1 << op) - 1) */
186
mov r8, rsi ; /* save in so from can use it's reg */
188
sub rax, [rsp+40] ; /* nbytes = out - beg */
191
jb L_clip_window ; /* if (dist > nbytes) 4.2% */
193
mov ecx, r14d ; /* ecx = len */
195
sub rsi, r15 ; /* from = out - dist */
198
jnc L_copy_two ; /* if len % 2 == 0 */
205
mov rsi, r8 ; /* move in back to %rsi, toss from */
210
mov rsi, r8 ; /* move in back to %rsi, toss from */
215
cmp r15d, 1 ; /* if dist 1, is a memset */
217
cmp [rsp+40], rdi ; /* if out == beg, outside window */
220
mov ecx, r14d ; /* ecx = len */
234
L_test_for_second_level_length:
236
jnz L_test_for_end_of_block ; /* if ((op & 64) != 0) */
242
and eax, edx ; /* eax &= hold */
243
add eax, r14d ; /* eax += len */
244
mov eax, [rbp+rax*4] ; /* eax = lcode[val+(hold&mask[op])]*/
248
L_test_for_second_level_dist:
250
jnz L_invalid_distance_code ; /* if ((op & 64) != 0) */
256
and eax, edx ; /* eax &= hold */
257
add eax, r15d ; /* eax += dist */
258
mov eax, [r11+rax*4] ; /* eax = dcode[val+(hold&mask[op])]*/
263
mov ecx, eax ; /* ecx = nbytes */
264
mov eax, [rsp+92] ; /* eax = wsize, prepare for dist cmp */
265
neg ecx ; /* nbytes = -nbytes */
268
jb L_invalid_distance_too_far ; /* if (dist > wsize) */
270
add ecx, r15d ; /* nbytes = dist - nbytes */
271
cmp dword ptr [rsp+96], 0
272
jne L_wrap_around_window ; /* if (write != 0) */
274
mov rsi, [rsp+56] ; /* from = window */
275
sub eax, ecx ; /* eax -= nbytes */
276
add rsi, rax ; /* from += wsize - nbytes */
278
mov eax, r14d ; /* eax = len */
280
jbe L_do_copy ; /* if (nbytes >= len) */
282
sub eax, ecx ; /* eax -= nbytes */
285
sub rsi, r15 ; /* from = &out[ -dist ] */
289
L_wrap_around_window:
290
mov eax, [rsp+96] ; /* eax = write */
292
jbe L_contiguous_in_window ; /* if (write >= nbytes) */
294
mov esi, [rsp+92] ; /* from = wsize */
295
add rsi, [rsp+56] ; /* from += window */
296
add rsi, rax ; /* from += write */
297
sub rsi, rcx ; /* from -= nbytes */
298
sub ecx, eax ; /* nbytes -= write */
300
mov eax, r14d ; /* eax = len */
302
jbe L_do_copy ; /* if (nbytes >= len) */
304
sub eax, ecx ; /* len -= nbytes */
306
mov rsi, [rsp+56] ; /* from = window */
307
mov ecx, [rsp+96] ; /* nbytes = write */
309
jbe L_do_copy ; /* if (nbytes >= len) */
311
sub eax, ecx ; /* len -= nbytes */
314
sub rsi, r15 ; /* from = out - dist */
318
L_contiguous_in_window:
319
mov rsi, [rsp+56] ; /* rsi = window */
321
sub rsi, rcx ; /* from += write - nbytes */
323
mov eax, r14d ; /* eax = len */
325
jbe L_do_copy ; /* if (nbytes >= len) */
327
sub eax, ecx ; /* len -= nbytes */
330
sub rsi, r15 ; /* from = out - dist */
331
jmp L_do_copy ; /* if (nbytes >= len) */
335
mov ecx, eax ; /* ecx = len */
338
mov rsi, r8 ; /* move in back to %esi, toss from */
341
L_test_for_end_of_block:
343
jz L_invalid_literal_length_code
344
mov dword ptr [rsp+116], 1
345
jmp L_break_loop_with_status
347
L_invalid_literal_length_code:
348
mov dword ptr [rsp+116], 2
349
jmp L_break_loop_with_status
351
L_invalid_distance_code:
352
mov dword ptr [rsp+116], 3
353
jmp L_break_loop_with_status
355
L_invalid_distance_too_far:
356
mov dword ptr [rsp+116], 4
357
jmp L_break_loop_with_status
360
mov dword ptr [rsp+116], 0
362
L_break_loop_with_status:
363
; /* put in, out, bits, and hold back into ar and pop esp */
364
mov [rsp+16], rsi ; /* in */
365
mov [rsp+32], rdi ; /* out */
366
mov [rsp+88], ebx ; /* bits */
367
mov [rsp+80], rdx ; /* hold */
369
mov rax, [rsp] ; /* restore rbp and rsp */
386
; : "memory", "%rax", "%rbx", "%rcx", "%rdx", "%rsi", "%rdi",
387
; "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15"