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
13
; This file compile with Microsoft Macro Assembler (x64) for AMD64
15
; ml64.exe is given with Visual Studio 2005/2008/2010 and Windows WDK
17
; (you can get Windows WDK with ml64 for AMD64 from
18
; http://www.microsoft.com/whdc/Devtools/wdk/default.mspx for low price)
25
; see http://weblogs.asp.net/oldnewthing/archive/2004/01/14/58579.aspx and
26
; http://msdn.microsoft.com/library/en-us/kmarch/hh/kmarch/64bitAMD_8e951dd2-ee77-4728-8702-55ce4b5dd24a.xml.asp
28
; All registers must be preserved across the call, except for
29
; rax, rcx, rdx, r8, r-9, r10, and r11, which are scratch.
42
mov [rax+8], rbp ; /* save regs rbp and rsp */
45
mov rsp, rax ; /* make rsp point to &ar */
47
mov rsi, [rsp+16] ; /* rsi = in */
48
mov rdi, [rsp+32] ; /* rdi = out */
49
mov r9, [rsp+24] ; /* r9 = last */
50
mov r10, [rsp+48] ; /* r10 = end */
51
mov rbp, [rsp+64] ; /* rbp = lcode */
52
mov r11, [rsp+72] ; /* r11 = dcode */
53
mov rdx, [rsp+80] ; /* rdx = hold */
54
mov ebx, [rsp+88] ; /* ebx = bits */
55
mov r12d, [rsp+100] ; /* r12d = lmask */
56
mov r13d, [rsp+104] ; /* r13d = dmask */
63
je L_one_time ; /* if only one decode left */
70
mov r8, r12 ; /* r8 = lmask */
72
ja L_get_length_code_one_time
74
lodsd ; /* eax = *(uint *)in++ */
75
mov cl, bl ; /* cl = bits, needs it for shifting */
76
add bl, 32 ; /* bits += 32 */
78
or rdx, rax ; /* hold |= *((uint *)in)++ << bits */
79
jmp L_get_length_code_one_time
89
mov r8, r12 ; /* r8 = lmask */
91
ja L_get_length_code ; /* if (32 < bits) */
93
lodsd ; /* eax = *(uint *)in++ */
94
mov cl, bl ; /* cl = bits, needs it for shifting */
95
add bl, 32 ; /* bits += 32 */
97
or rdx, rax ; /* hold |= *((uint *)in)++ << bits */
100
and r8, rdx ; /* r8 &= hold */
101
mov eax, [rbp+r8*4] ; /* eax = lcode[hold & lmask] */
103
mov cl, ah ; /* cl = this.bits */
104
sub bl, ah ; /* bits -= this.bits */
105
shr rdx, cl ; /* hold >>= this.bits */
108
jnz L_test_for_length_base ; /* if (op != 0) 45.7% */
110
mov r8, r12 ; /* r8 = lmask */
111
shr eax, 16 ; /* output this.val char */
114
L_get_length_code_one_time:
115
and r8, rdx ; /* r8 &= hold */
116
mov eax, [rbp+r8*4] ; /* eax = lcode[hold & lmask] */
119
mov cl, ah ; /* cl = this.bits */
120
sub bl, ah ; /* bits -= this.bits */
121
shr rdx, cl ; /* hold >>= this.bits */
124
jnz L_test_for_length_base ; /* if (op != 0) 45.7% */
126
shr eax, 16 ; /* output this.val char */
131
L_test_for_length_base:
132
mov r14d, eax ; /* len = this */
133
shr r14d, 16 ; /* len = this.val */
137
jz L_test_for_second_level_length ; /* if ((op & 16) == 0) 8% */
138
and cl, 15 ; /* op &= 15 */
139
jz L_decode_distance ; /* if (!op) */
147
and eax, edx ; /* eax &= hold */
149
add r14d, eax ; /* len += hold & mask[op] */
152
mov r8, r13 ; /* r8 = dmask */
154
ja L_get_distance_code ; /* if (32 < bits) */
156
lodsd ; /* eax = *(uint *)in++ */
157
mov cl, bl ; /* cl = bits, needs it for shifting */
158
add bl, 32 ; /* bits += 32 */
160
or rdx, rax ; /* hold |= *((uint *)in)++ << bits */
163
and r8, rdx ; /* r8 &= hold */
164
mov eax, [r11+r8*4] ; /* eax = dcode[hold & dmask] */
167
mov r15d, eax ; /* dist = this */
168
shr r15d, 16 ; /* dist = this.val */
170
sub bl, ah ; /* bits -= this.bits */
171
shr rdx, cl ; /* hold >>= this.bits */
172
mov cl, al ; /* cl = this.op */
174
test al, 16 ; /* if ((op & 16) == 0) */
175
jz L_test_for_second_level_dist
176
and cl, 15 ; /* op &= 15 */
184
dec eax ; /* (1 << op) - 1 */
185
and eax, edx ; /* eax &= hold */
187
add r15d, eax ; /* dist += hold & ((1 << op) - 1) */
190
mov r8, rsi ; /* save in so from can use it's reg */
192
sub rax, [rsp+40] ; /* nbytes = out - beg */
195
jb L_clip_window ; /* if (dist > nbytes) 4.2% */
197
mov ecx, r14d ; /* ecx = len */
199
sub rsi, r15 ; /* from = out - dist */
202
jnc L_copy_two ; /* if len % 2 == 0 */
209
mov rsi, r8 ; /* move in back to %rsi, toss from */
214
mov rsi, r8 ; /* move in back to %rsi, toss from */
219
cmp r15d, 1 ; /* if dist 1, is a memset */
221
cmp [rsp+40], rdi ; /* if out == beg, outside window */
224
mov ecx, r14d ; /* ecx = len */
238
L_test_for_second_level_length:
240
jnz L_test_for_end_of_block ; /* if ((op & 64) != 0) */
246
and eax, edx ; /* eax &= hold */
247
add eax, r14d ; /* eax += len */
248
mov eax, [rbp+rax*4] ; /* eax = lcode[val+(hold&mask[op])]*/
252
L_test_for_second_level_dist:
254
jnz L_invalid_distance_code ; /* if ((op & 64) != 0) */
260
and eax, edx ; /* eax &= hold */
261
add eax, r15d ; /* eax += dist */
262
mov eax, [r11+rax*4] ; /* eax = dcode[val+(hold&mask[op])]*/
267
mov ecx, eax ; /* ecx = nbytes */
268
mov eax, [rsp+92] ; /* eax = wsize, prepare for dist cmp */
269
neg ecx ; /* nbytes = -nbytes */
272
jb L_invalid_distance_too_far ; /* if (dist > wsize) */
274
add ecx, r15d ; /* nbytes = dist - nbytes */
275
cmp dword ptr [rsp+96], 0
276
jne L_wrap_around_window ; /* if (write != 0) */
278
mov rsi, [rsp+56] ; /* from = window */
279
sub eax, ecx ; /* eax -= nbytes */
280
add rsi, rax ; /* from += wsize - nbytes */
282
mov eax, r14d ; /* eax = len */
284
jbe L_do_copy ; /* if (nbytes >= len) */
286
sub eax, ecx ; /* eax -= nbytes */
289
sub rsi, r15 ; /* from = &out[ -dist ] */
293
L_wrap_around_window:
294
mov eax, [rsp+96] ; /* eax = write */
296
jbe L_contiguous_in_window ; /* if (write >= nbytes) */
298
mov esi, [rsp+92] ; /* from = wsize */
299
add rsi, [rsp+56] ; /* from += window */
300
add rsi, rax ; /* from += write */
301
sub rsi, rcx ; /* from -= nbytes */
302
sub ecx, eax ; /* nbytes -= write */
304
mov eax, r14d ; /* eax = len */
306
jbe L_do_copy ; /* if (nbytes >= len) */
308
sub eax, ecx ; /* len -= nbytes */
310
mov rsi, [rsp+56] ; /* from = window */
311
mov ecx, [rsp+96] ; /* nbytes = write */
313
jbe L_do_copy ; /* if (nbytes >= len) */
315
sub eax, ecx ; /* len -= nbytes */
318
sub rsi, r15 ; /* from = out - dist */
322
L_contiguous_in_window:
323
mov rsi, [rsp+56] ; /* rsi = window */
325
sub rsi, rcx ; /* from += write - nbytes */
327
mov eax, r14d ; /* eax = len */
329
jbe L_do_copy ; /* if (nbytes >= len) */
331
sub eax, ecx ; /* len -= nbytes */
334
sub rsi, r15 ; /* from = out - dist */
335
jmp L_do_copy ; /* if (nbytes >= len) */
339
mov ecx, eax ; /* ecx = len */
342
mov rsi, r8 ; /* move in back to %esi, toss from */
345
L_test_for_end_of_block:
347
jz L_invalid_literal_length_code
348
mov dword ptr [rsp+116], 1
349
jmp L_break_loop_with_status
351
L_invalid_literal_length_code:
352
mov dword ptr [rsp+116], 2
353
jmp L_break_loop_with_status
355
L_invalid_distance_code:
356
mov dword ptr [rsp+116], 3
357
jmp L_break_loop_with_status
359
L_invalid_distance_too_far:
360
mov dword ptr [rsp+116], 4
361
jmp L_break_loop_with_status
364
mov dword ptr [rsp+116], 0
366
L_break_loop_with_status:
367
; /* put in, out, bits, and hold back into ar and pop esp */
368
mov [rsp+16], rsi ; /* in */
369
mov [rsp+32], rdi ; /* out */
370
mov [rsp+88], ebx ; /* bits */
371
mov [rsp+80], rdx ; /* hold */
373
mov rax, [rsp] ; /* restore rbp and rsp */
390
; : "memory", "%rax", "%rbx", "%rcx", "%rdx", "%rsi", "%rdi",
391
; "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15"