/
mmxp2_32.asm
387 lines (293 loc) · 8.99 KB
/
mmxp2_32.asm
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
;
; pII-optimised MMX format converters for HERMES
; Copyright (c) 1998 Christian Nentwich (c.nentwich@cs.ucl.ac.uk)
; and (c) 1999 Jonathan Matthew (jmatthew@uq.net.au)
; This source code is licensed under the GNU LGPL
;
; Please refer to the file COPYING.LIB contained in the distribution for
; licensing conditions
;
; COPYRIGHT NOTICE
;
; This file partly contains code that is (c) Intel Corporation, specifically
; the mode detection routine, and the converter to 15 bit (8 pixel
; conversion routine from the mmx programming tutorial pages).
;
;
; These routines aren't exactly pII optimised - it's just that as they
; are, they're terrible on p5 MMXs, but less so on pIIs. Someone needs to
; optimise them for p5 MMXs..
BITS 32
GLOBAL _ConvertMMXpII32_24RGB888
GLOBAL _ConvertMMXpII32_16RGB565
GLOBAL _ConvertMMXpII32_16BGR565
GLOBAL _ConvertMMXpII32_16RGB555
GLOBAL _ConvertMMXpII32_16BGR555
EXTERN _mmxreturn
SECTION .note.GNU-stack noalloc progbits noexec nowrite
SECTION .data
ALIGN 8
;; Constants for conversion routines
mmx32_rgb888_mask dd 00ffffffh,00ffffffh
mmx32_rgb565_b dd 000000f8h, 000000f8h
mmx32_rgb565_g dd 0000fc00h, 0000fc00h
mmx32_rgb565_r dd 00f80000h, 00f80000h
mmx32_rgb555_rb dd 00f800f8h,00f800f8h
mmx32_rgb555_g dd 0000f800h,0000f800h
mmx32_rgb555_mul dd 20000008h,20000008h
mmx32_bgr555_mul dd 00082000h,00082000h
SECTION .text
_ConvertMMXpII32_24RGB888:
; set up mm6 as the mask, mm7 as zero
movq mm6, qword [mmx32_rgb888_mask]
pxor mm7, mm7
mov edx, ecx ; save ecx
and ecx, 0fffffffch ; clear lower two bits
jnz .L1
jmp .L2
.L1:
movq mm0, [esi] ; A R G B a r g b
pand mm0, mm6 ; 0 R G B 0 r g b
movq mm1, [esi+8] ; A R G B a r g b
pand mm1, mm6 ; 0 R G B 0 r g b
movq mm2, mm0 ; 0 R G B 0 r g b
punpckhdq mm2, mm7 ; 0 0 0 0 0 R G B
punpckldq mm0, mm7 ; 0 0 0 0 0 r g b
psllq mm2, 24 ; 0 0 R G B 0 0 0
por mm0, mm2 ; 0 0 R G B r g b
movq mm3, mm1 ; 0 R G B 0 r g b
psllq mm3, 48 ; g b 0 0 0 0 0 0
por mm0, mm3 ; g b R G B r g b
movq mm4, mm1 ; 0 R G B 0 r g b
punpckhdq mm4, mm7 ; 0 0 0 0 0 R G B
punpckldq mm1, mm7 ; 0 0 0 0 0 r g b
psrlq mm1, 16 ; 0 0 0 R G B 0 r
psllq mm4, 8 ; 0 0 0 0 R G B 0
por mm1, mm4 ; 0 0 0 0 R G B r
movq [edi], mm0
add esi, BYTE 16
movd [edi+8], mm1
add edi, BYTE 12
sub ecx, BYTE 4
jnz .L1
.L2:
mov ecx, edx
and ecx, BYTE 3
jz .L4
.L3:
mov al, [esi]
mov bl, [esi+1]
mov dl, [esi+2]
mov [edi], al
mov [edi+1], bl
mov [edi+2], dl
add esi, BYTE 4
add edi, BYTE 3
dec ecx
jnz .L3
.L4:
jmp _mmxreturn
_ConvertMMXpII32_16RGB565:
; set up masks
movq mm5, [mmx32_rgb565_b]
movq mm6, [mmx32_rgb565_g]
movq mm7, [mmx32_rgb565_r]
mov edx, ecx
shr ecx, 2
jnz .L1
jmp .L2 ; not necessary at the moment, but doesn't hurt (much)
.L1:
movq mm0, [esi] ; argb
movq mm1, mm0 ; argb
pand mm0, mm6 ; 00g0
movq mm3, mm1 ; argb
pand mm1, mm5 ; 000b
pand mm3, mm7 ; 0r00
pslld mm1, 2 ; 0 0 000000bb bbb00000
por mm0, mm1 ; 0 0 ggggggbb bbb00000
psrld mm0, 5 ; 0 0 00000ggg gggbbbbb
movq mm4, [esi+8] ; argb
movq mm2, mm4 ; argb
pand mm4, mm6 ; 00g0
movq mm1, mm2 ; argb
pand mm2, mm5 ; 000b
pand mm1, mm7 ; 0r00
pslld mm2, 2 ; 0 0 000000bb bbb00000
por mm4, mm2 ; 0 0 ggggggbb bbb00000
psrld mm4, 5 ; 0 0 00000ggg gggbbbbb
packuswb mm3, mm1 ; R 0 r 0
packssdw mm0, mm4 ; as above.. ish
por mm0, mm3 ; done.
movq [edi], mm0
add esi, 16
add edi, 8
dec ecx
jnz .L1
.L2:
mov ecx, edx
and ecx, BYTE 3
jz .L4
.L3:
mov al, [esi]
mov bh, [esi+1]
mov ah, [esi+2]
shr al, 3
and eax, 0F81Fh ; BYTE?
shr ebx, 5
and ebx, 07E0h ; BYTE?
add eax, ebx
mov [edi], al
mov [edi+1], ah
add esi, BYTE 4
add edi, BYTE 2
dec ecx
jnz .L3
.L4:
jmp _mmxreturn
_ConvertMMXpII32_16BGR565:
movq mm5, [mmx32_rgb565_r]
movq mm6, [mmx32_rgb565_g]
movq mm7, [mmx32_rgb565_b]
mov edx, ecx
shr ecx, 2
jnz .L1
jmp .L2
.L1:
movq mm0, [esi] ; a r g b
movq mm1, mm0 ; a r g b
pand mm0, mm6 ; 0 0 g 0
movq mm3, mm1 ; a r g b
pand mm1, mm5 ; 0 r 0 0
pand mm3, mm7 ; 0 0 0 b
psllq mm3, 16 ; 0 b 0 0
psrld mm1, 14 ; 0 0 000000rr rrr00000
por mm0, mm1 ; 0 0 ggggggrr rrr00000
psrld mm0, 5 ; 0 0 00000ggg gggrrrrr
movq mm4, [esi+8] ; a r g b
movq mm2, mm4 ; a r g b
pand mm4, mm6 ; 0 0 g 0
movq mm1, mm2 ; a r g b
pand mm2, mm5 ; 0 r 0 0
pand mm1, mm7 ; 0 0 0 b
psllq mm1, 16 ; 0 b 0 0
psrld mm2, 14 ; 0 0 000000rr rrr00000
por mm4, mm2 ; 0 0 ggggggrr rrr00000
psrld mm4, 5 ; 0 0 00000ggg gggrrrrr
packuswb mm3, mm1 ; BBBBB000 00000000 bbbbb000 00000000
packssdw mm0, mm4 ; 00000GGG GGGRRRRR 00000GGG GGGRRRRR
por mm0, mm3 ; BBBBBGGG GGGRRRRR bbbbbggg gggrrrrr
movq [edi], mm0
add esi, BYTE 16
add edi, BYTE 8
dec ecx
jnz .L1
.L2:
and edx, BYTE 3
jz .L4
.L3:
mov al, [esi+2]
mov bh, [esi+1]
mov ah, [esi]
shr al, 3
and eax, 0F81Fh ; BYTE ?
shr ebx, 5
and ebx, 07E0h ; BYTE ?
add eax, ebx
mov [edi], al
mov [edi+1], ah
add esi, BYTE 4
add edi, BYTE 2
dec edx
jnz .L3
.L4:
jmp _mmxreturn
_ConvertMMXpII32_16BGR555:
; the 16BGR555 converter is identical to the RGB555 one,
; except it uses a different multiplier for the pmaddwd
; instruction. cool huh.
movq mm7, qword [mmx32_bgr555_mul]
jmp _convert_bgr555_cheat
; This is the same as the Intel version.. they obviously went to
; much more trouble to expand/coil the loop than I did, so theirs
; would almost certainly be faster, even if only a little.
; I did rename 'mmx32_rgb555_add' to 'mmx32_rgb555_mul', which is
; (I think) a more accurate name..
_ConvertMMXpII32_16RGB555:
movq mm7,qword [mmx32_rgb555_mul]
_convert_bgr555_cheat:
movq mm6,qword [mmx32_rgb555_g]
mov edx,ecx ; Save ecx
and ecx,BYTE 0fffffff8h ; clear lower three bits
jnz .L_OK
jmp near .L2
.L_OK:
movq mm2,[esi+8]
movq mm0,[esi]
movq mm3,mm2
pand mm3,qword [mmx32_rgb555_rb]
movq mm1,mm0
pand mm1,qword [mmx32_rgb555_rb]
pmaddwd mm3,mm7
pmaddwd mm1,mm7
pand mm2,mm6
.L1:
movq mm4,[esi+24]
pand mm0,mm6
movq mm5,[esi+16]
por mm3,mm2
psrld mm3,6
por mm1,mm0
movq mm0,mm4
psrld mm1,6
pand mm0,qword [mmx32_rgb555_rb]
packssdw mm1,mm3
movq mm3,mm5
pmaddwd mm0,mm7
pand mm3,qword [mmx32_rgb555_rb]
pand mm4,mm6
movq [edi],mm1
pmaddwd mm3,mm7
add esi,BYTE 32
por mm4,mm0
pand mm5,mm6
psrld mm4,6
movq mm2,[esi+8]
por mm5,mm3
movq mm0,[esi]
psrld mm5,6
movq mm3,mm2
movq mm1,mm0
pand mm3,qword [mmx32_rgb555_rb]
packssdw mm5,mm4
pand mm1,qword [mmx32_rgb555_rb]
pand mm2,mm6
movq [edi+8],mm5
pmaddwd mm3,mm7
pmaddwd mm1,mm7
add edi,BYTE 16
sub ecx,BYTE 8
jz .L2
jmp .L1
.L2:
mov ecx,edx
and ecx,BYTE 7
jz .L4
.L3:
mov ebx,[esi]
add esi,BYTE 4
mov eax,ebx
mov edx,ebx
shr eax,3
shr edx,6
and eax,BYTE 0000000000011111b
and edx, 0000001111100000b
shr ebx,9
or eax,edx
and ebx, 0111110000000000b
or eax,ebx
mov [edi],ax
add edi,BYTE 2
dec ecx
jnz .L3
.L4:
jmp _mmxreturn