1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
|
;
; jdcolmmx.asm - colorspace conversion (MMX)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
;
; Based on
; x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
; For conditions of distribution and use, see copyright notice in jsimdext.inc
;
; This file should be assembled with NASM (Netwide Assembler),
; can *not* be assembled with Microsoft's MASM or any compatible
; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; [TAB8]
%include "jsimdext.inc"
%include "jcolsamp.inc"
; --------------------------------------------------------------------------
%define SCALEBITS 16
F_0_344 equ 22554 ; FIX(0.34414)
F_0_714 equ 46802 ; FIX(0.71414)
F_1_402 equ 91881 ; FIX(1.40200)
F_1_772 equ 116130 ; FIX(1.77200)
F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1)
F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414)
F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200)
; --------------------------------------------------------------------------
SECTION SEG_CONST
alignz 16
global EXTN(jconst_ycc_rgb_convert_mmx)
EXTN(jconst_ycc_rgb_convert_mmx):
PW_F0402 times 4 dw F_0_402
PW_MF0228 times 4 dw -F_0_228
PW_MF0344_F0285 times 2 dw -F_0_344, F_0_285
PW_ONE times 4 dw 1
PD_ONEHALF times 2 dd 1 << (SCALEBITS-1)
alignz 16
; --------------------------------------------------------------------------
SECTION SEG_TEXT
BITS 32
;
; Convert some rows of samples to the output colorspace.
;
; GLOBAL(void)
; jsimd_ycc_rgb_convert_mmx (JDIMENSION out_width,
; JSAMPIMAGE input_buf, JDIMENSION input_row,
; JSAMPARRAY output_buf, int num_rows)
;
%define out_width(b) (b)+8 ; JDIMENSION out_width
%define input_buf(b) (b)+12 ; JSAMPIMAGE input_buf
%define input_row(b) (b)+16 ; JDIMENSION input_row
%define output_buf(b) (b)+20 ; JSAMPARRAY output_buf
%define num_rows(b) (b)+24 ; int num_rows
%define original_ebp ebp+0
%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM]
%define WK_NUM 2
%define gotptr wk(0)-SIZEOF_POINTER ; void * gotptr
align 16
global EXTN(jsimd_ycc_rgb_convert_mmx)
EXTN(jsimd_ycc_rgb_convert_mmx):
push ebp
mov eax,esp ; eax = original ebp
sub esp, byte 4
and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits
mov [esp],eax
mov ebp,esp ; ebp = aligned ebp
lea esp, [wk(0)]
pushpic eax ; make a room for GOT address
push ebx
; push ecx ; need not be preserved
; push edx ; need not be preserved
push esi
push edi
get_GOT ebx ; get GOT address
movpic POINTER [gotptr], ebx ; save GOT address
mov ecx, JDIMENSION [out_width(eax)] ; num_cols
test ecx,ecx
jz near .return
push ecx
mov edi, JSAMPIMAGE [input_buf(eax)]
mov ecx, JDIMENSION [input_row(eax)]
mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY]
mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY]
mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY]
lea esi, [esi+ecx*SIZEOF_JSAMPROW]
lea ebx, [ebx+ecx*SIZEOF_JSAMPROW]
lea edx, [edx+ecx*SIZEOF_JSAMPROW]
pop ecx
mov edi, JSAMPARRAY [output_buf(eax)]
mov eax, INT [num_rows(eax)]
test eax,eax
jle near .return
alignx 16,7
.rowloop:
push eax
push edi
push edx
push ebx
push esi
push ecx ; col
mov esi, JSAMPROW [esi] ; inptr0
mov ebx, JSAMPROW [ebx] ; inptr1
mov edx, JSAMPROW [edx] ; inptr2
mov edi, JSAMPROW [edi] ; outptr
movpic eax, POINTER [gotptr] ; load GOT address (eax)
alignx 16,7
.columnloop:
movq mm5, MMWORD [ebx] ; mm5=Cb(01234567)
movq mm1, MMWORD [edx] ; mm1=Cr(01234567)
pcmpeqw mm4,mm4
pcmpeqw mm7,mm7
psrlw mm4,BYTE_BIT
psllw mm7,7 ; mm7={0xFF80 0xFF80 0xFF80 0xFF80}
movq mm0,mm4 ; mm0=mm4={0xFF 0x00 0xFF 0x00 ..}
pand mm4,mm5 ; mm4=Cb(0246)=CbE
psrlw mm5,BYTE_BIT ; mm5=Cb(1357)=CbO
pand mm0,mm1 ; mm0=Cr(0246)=CrE
psrlw mm1,BYTE_BIT ; mm1=Cr(1357)=CrO
paddw mm4,mm7
paddw mm5,mm7
paddw mm0,mm7
paddw mm1,mm7
; (Original)
; R = Y + 1.40200 * Cr
; G = Y - 0.34414 * Cb - 0.71414 * Cr
; B = Y + 1.77200 * Cb
;
; (This implementation)
; R = Y + 0.40200 * Cr + Cr
; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
; B = Y - 0.22800 * Cb + Cb + Cb
movq mm2,mm4 ; mm2=CbE
movq mm3,mm5 ; mm3=CbO
paddw mm4,mm4 ; mm4=2*CbE
paddw mm5,mm5 ; mm5=2*CbO
movq mm6,mm0 ; mm6=CrE
movq mm7,mm1 ; mm7=CrO
paddw mm0,mm0 ; mm0=2*CrE
paddw mm1,mm1 ; mm1=2*CrO
pmulhw mm4,[GOTOFF(eax,PW_MF0228)] ; mm4=(2*CbE * -FIX(0.22800))
pmulhw mm5,[GOTOFF(eax,PW_MF0228)] ; mm5=(2*CbO * -FIX(0.22800))
pmulhw mm0,[GOTOFF(eax,PW_F0402)] ; mm0=(2*CrE * FIX(0.40200))
pmulhw mm1,[GOTOFF(eax,PW_F0402)] ; mm1=(2*CrO * FIX(0.40200))
paddw mm4,[GOTOFF(eax,PW_ONE)]
paddw mm5,[GOTOFF(eax,PW_ONE)]
psraw mm4,1 ; mm4=(CbE * -FIX(0.22800))
psraw mm5,1 ; mm5=(CbO * -FIX(0.22800))
paddw mm0,[GOTOFF(eax,PW_ONE)]
paddw mm1,[GOTOFF(eax,PW_ONE)]
psraw mm0,1 ; mm0=(CrE * FIX(0.40200))
psraw mm1,1 ; mm1=(CrO * FIX(0.40200))
paddw mm4,mm2
paddw mm5,mm3
paddw mm4,mm2 ; mm4=(CbE * FIX(1.77200))=(B-Y)E
paddw mm5,mm3 ; mm5=(CbO * FIX(1.77200))=(B-Y)O
paddw mm0,mm6 ; mm0=(CrE * FIX(1.40200))=(R-Y)E
paddw mm1,mm7 ; mm1=(CrO * FIX(1.40200))=(R-Y)O
movq MMWORD [wk(0)], mm4 ; wk(0)=(B-Y)E
movq MMWORD [wk(1)], mm5 ; wk(1)=(B-Y)O
movq mm4,mm2
movq mm5,mm3
punpcklwd mm2,mm6
punpckhwd mm4,mm6
pmaddwd mm2,[GOTOFF(eax,PW_MF0344_F0285)]
pmaddwd mm4,[GOTOFF(eax,PW_MF0344_F0285)]
punpcklwd mm3,mm7
punpckhwd mm5,mm7
pmaddwd mm3,[GOTOFF(eax,PW_MF0344_F0285)]
pmaddwd mm5,[GOTOFF(eax,PW_MF0344_F0285)]
paddd mm2,[GOTOFF(eax,PD_ONEHALF)]
paddd mm4,[GOTOFF(eax,PD_ONEHALF)]
psrad mm2,SCALEBITS
psrad mm4,SCALEBITS
paddd mm3,[GOTOFF(eax,PD_ONEHALF)]
paddd mm5,[GOTOFF(eax,PD_ONEHALF)]
psrad mm3,SCALEBITS
psrad mm5,SCALEBITS
packssdw mm2,mm4 ; mm2=CbE*-FIX(0.344)+CrE*FIX(0.285)
packssdw mm3,mm5 ; mm3=CbO*-FIX(0.344)+CrO*FIX(0.285)
psubw mm2,mm6 ; mm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E
psubw mm3,mm7 ; mm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O
movq mm5, MMWORD [esi] ; mm5=Y(01234567)
pcmpeqw mm4,mm4
psrlw mm4,BYTE_BIT ; mm4={0xFF 0x00 0xFF 0x00 ..}
pand mm4,mm5 ; mm4=Y(0246)=YE
psrlw mm5,BYTE_BIT ; mm5=Y(1357)=YO
paddw mm0,mm4 ; mm0=((R-Y)E+YE)=RE=(R0 R2 R4 R6)
paddw mm1,mm5 ; mm1=((R-Y)O+YO)=RO=(R1 R3 R5 R7)
packuswb mm0,mm0 ; mm0=(R0 R2 R4 R6 ** ** ** **)
packuswb mm1,mm1 ; mm1=(R1 R3 R5 R7 ** ** ** **)
paddw mm2,mm4 ; mm2=((G-Y)E+YE)=GE=(G0 G2 G4 G6)
paddw mm3,mm5 ; mm3=((G-Y)O+YO)=GO=(G1 G3 G5 G7)
packuswb mm2,mm2 ; mm2=(G0 G2 G4 G6 ** ** ** **)
packuswb mm3,mm3 ; mm3=(G1 G3 G5 G7 ** ** ** **)
paddw mm4, MMWORD [wk(0)] ; mm4=(YE+(B-Y)E)=BE=(B0 B2 B4 B6)
paddw mm5, MMWORD [wk(1)] ; mm5=(YO+(B-Y)O)=BO=(B1 B3 B5 B7)
packuswb mm4,mm4 ; mm4=(B0 B2 B4 B6 ** ** ** **)
packuswb mm5,mm5 ; mm5=(B1 B3 B5 B7 ** ** ** **)
%if RGB_PIXELSIZE == 3 ; ---------------
; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **)
punpcklbw mmA,mmC ; mmA=(00 10 02 12 04 14 06 16)
punpcklbw mmE,mmB ; mmE=(20 01 22 03 24 05 26 07)
punpcklbw mmD,mmF ; mmD=(11 21 13 23 15 25 17 27)
movq mmG,mmA
movq mmH,mmA
punpcklwd mmA,mmE ; mmA=(00 10 20 01 02 12 22 03)
punpckhwd mmG,mmE ; mmG=(04 14 24 05 06 16 26 07)
psrlq mmH,2*BYTE_BIT ; mmH=(02 12 04 14 06 16 -- --)
psrlq mmE,2*BYTE_BIT ; mmE=(22 03 24 05 26 07 -- --)
movq mmC,mmD
movq mmB,mmD
punpcklwd mmD,mmH ; mmD=(11 21 02 12 13 23 04 14)
punpckhwd mmC,mmH ; mmC=(15 25 06 16 17 27 -- --)
psrlq mmB,2*BYTE_BIT ; mmB=(13 23 15 25 17 27 -- --)
movq mmF,mmE
punpcklwd mmE,mmB ; mmE=(22 03 13 23 24 05 15 25)
punpckhwd mmF,mmB ; mmF=(26 07 17 27 -- -- -- --)
punpckldq mmA,mmD ; mmA=(00 10 20 01 11 21 02 12)
punpckldq mmE,mmG ; mmE=(22 03 13 23 04 14 24 05)
punpckldq mmC,mmF ; mmC=(15 25 06 16 26 07 17 27)
cmp ecx, byte SIZEOF_MMWORD
jb short .column_st16
movq MMWORD [edi+0*SIZEOF_MMWORD], mmA
movq MMWORD [edi+1*SIZEOF_MMWORD], mmE
movq MMWORD [edi+2*SIZEOF_MMWORD], mmC
sub ecx, byte SIZEOF_MMWORD
jz short .nextrow
add esi, byte SIZEOF_MMWORD ; inptr0
add ebx, byte SIZEOF_MMWORD ; inptr1
add edx, byte SIZEOF_MMWORD ; inptr2
add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr
jmp near .columnloop
alignx 16,7
.column_st16:
lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE
cmp ecx, byte 2*SIZEOF_MMWORD
jb short .column_st8
movq MMWORD [edi+0*SIZEOF_MMWORD], mmA
movq MMWORD [edi+1*SIZEOF_MMWORD], mmE
movq mmA,mmC
sub ecx, byte 2*SIZEOF_MMWORD
add edi, byte 2*SIZEOF_MMWORD
jmp short .column_st4
.column_st8:
cmp ecx, byte SIZEOF_MMWORD
jb short .column_st4
movq MMWORD [edi+0*SIZEOF_MMWORD], mmA
movq mmA,mmE
sub ecx, byte SIZEOF_MMWORD
add edi, byte SIZEOF_MMWORD
.column_st4:
movd eax,mmA
cmp ecx, byte SIZEOF_DWORD
jb short .column_st2
mov DWORD [edi+0*SIZEOF_DWORD], eax
psrlq mmA,DWORD_BIT
movd eax,mmA
sub ecx, byte SIZEOF_DWORD
add edi, byte SIZEOF_DWORD
.column_st2:
cmp ecx, byte SIZEOF_WORD
jb short .column_st1
mov WORD [edi+0*SIZEOF_WORD], ax
shr eax,WORD_BIT
sub ecx, byte SIZEOF_WORD
add edi, byte SIZEOF_WORD
.column_st1:
cmp ecx, byte SIZEOF_BYTE
jb short .nextrow
mov BYTE [edi+0*SIZEOF_BYTE], al
%else ; RGB_PIXELSIZE == 4 ; -----------
%ifdef RGBX_FILLER_0XFF
pcmpeqb mm6,mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **)
pcmpeqb mm7,mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **)
%else
pxor mm6,mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **)
pxor mm7,mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **)
%endif
; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **)
; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **)
; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **)
; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **)
punpcklbw mmA,mmC ; mmA=(00 10 02 12 04 14 06 16)
punpcklbw mmE,mmG ; mmE=(20 30 22 32 24 34 26 36)
punpcklbw mmB,mmD ; mmB=(01 11 03 13 05 15 07 17)
punpcklbw mmF,mmH ; mmF=(21 31 23 33 25 35 27 37)
movq mmC,mmA
punpcklwd mmA,mmE ; mmA=(00 10 20 30 02 12 22 32)
punpckhwd mmC,mmE ; mmC=(04 14 24 34 06 16 26 36)
movq mmG,mmB
punpcklwd mmB,mmF ; mmB=(01 11 21 31 03 13 23 33)
punpckhwd mmG,mmF ; mmG=(05 15 25 35 07 17 27 37)
movq mmD,mmA
punpckldq mmA,mmB ; mmA=(00 10 20 30 01 11 21 31)
punpckhdq mmD,mmB ; mmD=(02 12 22 32 03 13 23 33)
movq mmH,mmC
punpckldq mmC,mmG ; mmC=(04 14 24 34 05 15 25 35)
punpckhdq mmH,mmG ; mmH=(06 16 26 36 07 17 27 37)
cmp ecx, byte SIZEOF_MMWORD
jb short .column_st16
movq MMWORD [edi+0*SIZEOF_MMWORD], mmA
movq MMWORD [edi+1*SIZEOF_MMWORD], mmD
movq MMWORD [edi+2*SIZEOF_MMWORD], mmC
movq MMWORD [edi+3*SIZEOF_MMWORD], mmH
sub ecx, byte SIZEOF_MMWORD
jz short .nextrow
add esi, byte SIZEOF_MMWORD ; inptr0
add ebx, byte SIZEOF_MMWORD ; inptr1
add edx, byte SIZEOF_MMWORD ; inptr2
add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr
jmp near .columnloop
alignx 16,7
.column_st16:
cmp ecx, byte SIZEOF_MMWORD/2
jb short .column_st8
movq MMWORD [edi+0*SIZEOF_MMWORD], mmA
movq MMWORD [edi+1*SIZEOF_MMWORD], mmD
movq mmA,mmC
movq mmD,mmH
sub ecx, byte SIZEOF_MMWORD/2
add edi, byte 2*SIZEOF_MMWORD
.column_st8:
cmp ecx, byte SIZEOF_MMWORD/4
jb short .column_st4
movq MMWORD [edi+0*SIZEOF_MMWORD], mmA
movq mmA,mmD
sub ecx, byte SIZEOF_MMWORD/4
add edi, byte 1*SIZEOF_MMWORD
.column_st4:
cmp ecx, byte SIZEOF_MMWORD/8
jb short .nextrow
movd DWORD [edi+0*SIZEOF_DWORD], mmA
%endif ; RGB_PIXELSIZE ; ---------------
alignx 16,7
.nextrow:
pop ecx
pop esi
pop ebx
pop edx
pop edi
pop eax
add esi, byte SIZEOF_JSAMPROW
add ebx, byte SIZEOF_JSAMPROW
add edx, byte SIZEOF_JSAMPROW
add edi, byte SIZEOF_JSAMPROW ; output_buf
dec eax ; num_rows
jg near .rowloop
emms ; empty MMX state
.return:
pop edi
pop esi
; pop edx ; need not be preserved
; pop ecx ; need not be preserved
pop ebx
mov esp,ebp ; esp <- aligned ebp
pop esp ; esp <- original ebp
pop ebp
ret
|
