1 |
;/************************************************************************** |
;/**************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * mmx sum of absolute difference |
; * - K7 optimized SAD operators - |
5 |
; * |
; * |
6 |
; * This program is free software; you can redistribute it and/or modify |
; * Copyright(C) 2001 Peter Ross <pross@xvid.org> |
7 |
; * it under the terms of the GNU General Public License as published by |
; * 2002 Pascal Massimino <skal@planet-d.net> |
8 |
|
; * |
9 |
|
; * This program is free software; you can redistribute it and/or modify it |
10 |
|
; * under the terms of the GNU General Public License as published by |
11 |
; * the Free Software Foundation; either version 2 of the License, or |
; * the Free Software Foundation; either version 2 of the License, or |
12 |
; * (at your option) any later version. |
; * (at your option) any later version. |
13 |
; * |
; * |
18 |
; * |
; * |
19 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
20 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
21 |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
|
; * |
|
|
; *************************************************************************/ |
|
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
22 |
; * |
; * |
23 |
; * 23.07.2002 sad[16,8]bi_xmm; <pross@xvid.org> |
; * $Id: sad_mmx.asm,v 1.13 2004-04-12 15:49:56 edgomez Exp $ |
|
; * 04.06.2002 cleanup -Skal- |
|
|
; * 12.11.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
|
24 |
; * |
; * |
25 |
; *************************************************************************/ |
; ***************************************************************************/ |
26 |
|
|
27 |
bits 32 |
BITS 32 |
28 |
|
|
29 |
%macro cglobal 1 |
%macro cglobal 1 |
30 |
%ifdef PREFIX |
%ifdef PREFIX |
35 |
%endif |
%endif |
36 |
%endmacro |
%endmacro |
37 |
|
|
38 |
section .data |
;============================================================================= |
39 |
|
; Read only data |
40 |
|
;============================================================================= |
41 |
|
|
42 |
align 16 |
%ifdef FORMAT_COFF |
43 |
mmx_one times 4 dw 1 |
SECTION .rodata data |
44 |
|
%else |
45 |
section .text |
SECTION .rodata data align=16 |
46 |
|
%endif |
|
cglobal sad16_mmx |
|
|
cglobal sad16v_mmx |
|
|
cglobal sad8_mmx |
|
|
cglobal sad16bi_mmx |
|
|
cglobal sad8bi_mmx |
|
|
cglobal dev16_mmx |
|
47 |
|
|
48 |
;=========================================================================== |
ALIGN 16 |
49 |
; |
mmx_one: |
50 |
; uint32_t sad16_mmx(const uint8_t * const cur, |
times 4 dw 1 |
51 |
; const uint8_t * const ref, |
|
52 |
; const uint32_t stride, |
;============================================================================= |
53 |
; const uint32_t best_sad); |
; Helper macros |
54 |
; |
;============================================================================= |
|
; (early termination ignore; slows this down) |
|
|
; |
|
|
;=========================================================================== |
|
55 |
|
|
56 |
%macro SAD_16x16_MMX 0 |
%macro SAD_16x16_MMX 0 |
57 |
movq mm0, [eax] |
movq mm0, [eax] |
86 |
paddusw mm6,mm2 |
paddusw mm6,mm2 |
87 |
%endmacro |
%endmacro |
88 |
|
|
89 |
align 16 |
%macro SAD_8x8_MMX 0 |
90 |
|
movq mm0, [eax] |
91 |
|
movq mm1, [edx] |
92 |
|
|
93 |
|
movq mm2, [eax+ecx] |
94 |
|
movq mm3, [edx+ecx] |
95 |
|
|
96 |
|
lea eax,[eax+2*ecx] |
97 |
|
lea edx,[edx+2*ecx] |
98 |
|
|
99 |
|
movq mm4, mm0 |
100 |
|
psubusb mm0, mm1 |
101 |
|
movq mm5, mm2 |
102 |
|
psubusb mm2, mm3 |
103 |
|
|
104 |
|
psubusb mm1, mm4 |
105 |
|
por mm0, mm1 |
106 |
|
psubusb mm3, mm5 |
107 |
|
por mm2, mm3 |
108 |
|
|
109 |
|
movq mm1,mm0 |
110 |
|
movq mm3,mm2 |
111 |
|
|
112 |
|
punpcklbw mm0,mm7 |
113 |
|
punpckhbw mm1,mm7 |
114 |
|
punpcklbw mm2,mm7 |
115 |
|
punpckhbw mm3,mm7 |
116 |
|
|
117 |
|
paddusw mm0,mm1 |
118 |
|
paddusw mm6,mm0 |
119 |
|
paddusw mm2,mm3 |
120 |
|
paddusw mm6,mm2 |
121 |
|
%endmacro |
122 |
|
|
123 |
|
%macro SADV_16x16_MMX 0 |
124 |
|
movq mm0, [eax] |
125 |
|
movq mm1, [edx] |
126 |
|
|
127 |
|
movq mm2, [eax+8] |
128 |
|
movq mm3, [edx+8] |
129 |
|
|
130 |
|
movq mm4, mm0 |
131 |
|
psubusb mm0, mm1 |
132 |
|
|
133 |
|
psubusb mm1, mm4 |
134 |
|
por mm0, mm1 |
135 |
|
lea eax,[eax+ecx] |
136 |
|
|
137 |
|
movq mm4, mm2 |
138 |
|
psubusb mm2, mm3 |
139 |
|
|
140 |
|
psubusb mm3, mm4 |
141 |
|
por mm2, mm3 |
142 |
|
lea edx,[edx+ecx] |
143 |
|
|
144 |
|
movq mm1,mm0 |
145 |
|
movq mm3,mm2 |
146 |
|
|
147 |
|
punpcklbw mm0,mm7 |
148 |
|
punpckhbw mm1,mm7 |
149 |
|
punpcklbw mm2,mm7 |
150 |
|
punpckhbw mm3,mm7 |
151 |
|
|
152 |
|
paddusw mm0,mm1 |
153 |
|
paddusw mm2,mm3 |
154 |
|
|
155 |
|
paddusw mm5, mm0 |
156 |
|
paddusw mm6, mm2 |
157 |
|
%endmacro |
158 |
|
|
159 |
|
%macro SADBI_16x16_MMX 2 ; SADBI_16x16_MMX( int_ptr_offset, bool_increment_ptr ); |
160 |
|
|
161 |
|
movq mm0, [edx+%1] |
162 |
|
movq mm2, [ebx+%1] |
163 |
|
movq mm1, mm0 |
164 |
|
movq mm3, mm2 |
165 |
|
|
166 |
|
%if %2 != 0 |
167 |
|
add edx, ecx |
168 |
|
%endif |
169 |
|
|
170 |
|
punpcklbw mm0, mm7 |
171 |
|
punpckhbw mm1, mm7 |
172 |
|
punpcklbw mm2, mm7 |
173 |
|
punpckhbw mm3, mm7 |
174 |
|
|
175 |
|
%if %2 != 0 |
176 |
|
add ebx, ecx |
177 |
|
%endif |
178 |
|
|
179 |
|
paddusw mm0, mm2 ; mm01 = ref1 + ref2 |
180 |
|
paddusw mm1, mm3 |
181 |
|
paddusw mm0, [mmx_one] ; mm01 += 1 |
182 |
|
paddusw mm1, [mmx_one] |
183 |
|
psrlw mm0, 1 ; mm01 >>= 1 |
184 |
|
psrlw mm1, 1 |
185 |
|
|
186 |
|
movq mm2, [eax+%1] |
187 |
|
movq mm3, mm2 |
188 |
|
punpcklbw mm2, mm7 ; mm23 = src |
189 |
|
punpckhbw mm3, mm7 |
190 |
|
|
191 |
|
%if %2 != 0 |
192 |
|
add eax, ecx |
193 |
|
%endif |
194 |
|
|
195 |
|
movq mm4, mm0 |
196 |
|
movq mm5, mm1 |
197 |
|
psubusw mm0, mm2 |
198 |
|
psubusw mm1, mm3 |
199 |
|
psubusw mm2, mm4 |
200 |
|
psubusw mm3, mm5 |
201 |
|
por mm0, mm2 ; mm01 = ABS(mm01 - mm23) |
202 |
|
por mm1, mm3 |
203 |
|
|
204 |
|
paddusw mm6, mm0 ; mm6 += mm01 |
205 |
|
paddusw mm6, mm1 |
206 |
|
|
207 |
|
%endmacro |
208 |
|
|
209 |
|
%macro MEAN_16x16_MMX 0 |
210 |
|
movq mm0, [eax] |
211 |
|
movq mm2, [eax+8] |
212 |
|
lea eax, [eax+ecx] |
213 |
|
movq mm1, mm0 |
214 |
|
movq mm3, mm2 |
215 |
|
punpcklbw mm0, mm7 |
216 |
|
punpcklbw mm2, mm7 |
217 |
|
punpckhbw mm1, mm7 |
218 |
|
punpckhbw mm3, mm7 |
219 |
|
paddw mm5, mm0 |
220 |
|
paddw mm6, mm1 |
221 |
|
paddw mm5, mm2 |
222 |
|
paddw mm6, mm3 |
223 |
|
%endmacro |
224 |
|
|
225 |
|
%macro ABS_16x16_MMX 0 |
226 |
|
movq mm0, [eax] |
227 |
|
movq mm2, [eax+8] |
228 |
|
lea eax, [eax+ecx] |
229 |
|
movq mm1, mm0 |
230 |
|
movq mm3, mm2 |
231 |
|
punpcklbw mm0, mm7 |
232 |
|
punpcklbw mm2, mm7 |
233 |
|
punpckhbw mm1, mm7 |
234 |
|
punpckhbw mm3, mm7 |
235 |
|
movq mm4, mm6 |
236 |
|
psubusw mm4, mm0 |
237 |
|
|
238 |
|
psubusw mm0, mm6 |
239 |
|
por mm0, mm4 |
240 |
|
movq mm4, mm6 |
241 |
|
psubusw mm4, mm1 |
242 |
|
psubusw mm1, mm6 |
243 |
|
por mm1, mm4 |
244 |
|
|
245 |
|
movq mm4, mm6 |
246 |
|
psubusw mm4, mm2 |
247 |
|
psubusw mm2, mm6 |
248 |
|
por mm2, mm4 |
249 |
|
movq mm4, mm6 |
250 |
|
psubusw mm4, mm3 |
251 |
|
psubusw mm3, mm6 |
252 |
|
por mm3, mm4 |
253 |
|
|
254 |
|
paddw mm0, mm1 |
255 |
|
paddw mm2, mm3 |
256 |
|
paddw mm5, mm0 |
257 |
|
paddw mm5, mm2 |
258 |
|
%endmacro |
259 |
|
|
260 |
|
;============================================================================= |
261 |
|
; Code |
262 |
|
;============================================================================= |
263 |
|
|
264 |
|
SECTION .text |
265 |
|
|
266 |
|
cglobal sad16_mmx |
267 |
|
cglobal sad16v_mmx |
268 |
|
cglobal sad8_mmx |
269 |
|
cglobal sad16bi_mmx |
270 |
|
cglobal sad8bi_mmx |
271 |
|
cglobal dev16_mmx |
272 |
|
cglobal sse8_16bit_mmx |
273 |
|
cglobal sse8_8bit_mmx |
274 |
|
|
275 |
|
;----------------------------------------------------------------------------- |
276 |
|
; |
277 |
|
; uint32_t sad16_mmx(const uint8_t * const cur, |
278 |
|
; const uint8_t * const ref, |
279 |
|
; const uint32_t stride, |
280 |
|
; const uint32_t best_sad); |
281 |
|
; |
282 |
|
; (early termination ignore; slows this down) |
283 |
|
; |
284 |
|
;----------------------------------------------------------------------------- |
285 |
|
|
286 |
|
ALIGN 16 |
287 |
sad16_mmx: |
sad16_mmx: |
288 |
|
|
289 |
mov eax, [esp+ 4] ; Src1 |
mov eax, [esp+ 4] ; Src1 |
320 |
|
|
321 |
ret |
ret |
322 |
|
|
323 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
324 |
; |
; |
325 |
; uint32_t sad8_mmx(const uint8_t * const cur, |
; uint32_t sad8_mmx(const uint8_t * const cur, |
326 |
; const uint8_t * const ref, |
; const uint8_t * const ref, |
327 |
; const uint32_t stride); |
; const uint32_t stride); |
328 |
; |
; |
329 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
330 |
|
|
331 |
%macro SAD_8x8_MMX 0 |
ALIGN 16 |
|
movq mm0, [eax] |
|
|
movq mm1, [edx] |
|
|
|
|
|
movq mm2, [eax+ecx] |
|
|
movq mm3, [edx+ecx] |
|
|
|
|
|
lea eax,[eax+2*ecx] |
|
|
lea edx,[edx+2*ecx] |
|
|
|
|
|
movq mm4, mm0 |
|
|
psubusb mm0, mm1 |
|
|
movq mm5, mm2 |
|
|
psubusb mm2, mm3 |
|
|
|
|
|
psubusb mm1, mm4 |
|
|
por mm0, mm1 |
|
|
psubusb mm3, mm5 |
|
|
por mm2, mm3 |
|
|
|
|
|
movq mm1,mm0 |
|
|
movq mm3,mm2 |
|
|
|
|
|
punpcklbw mm0,mm7 |
|
|
punpckhbw mm1,mm7 |
|
|
punpcklbw mm2,mm7 |
|
|
punpckhbw mm3,mm7 |
|
|
|
|
|
paddusw mm0,mm1 |
|
|
paddusw mm6,mm0 |
|
|
paddusw mm2,mm3 |
|
|
paddusw mm6,mm2 |
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
332 |
sad8_mmx: |
sad8_mmx: |
333 |
|
|
334 |
mov eax, [esp+ 4] ; Src1 |
mov eax, [esp+ 4] ; Src1 |
352 |
|
|
353 |
ret |
ret |
354 |
|
|
355 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
356 |
; |
; |
357 |
; uint32_t sad16v_mmx(const uint8_t * const cur, |
; uint32_t sad16v_mmx(const uint8_t * const cur, |
358 |
; const uint8_t * const ref, |
; const uint8_t * const ref, |
359 |
; const uint32_t stride, |
; const uint32_t stride, |
360 |
; int32_t *sad); |
; int32_t *sad); |
361 |
; |
; |
362 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro SADV_16x16_MMX 0 |
|
|
movq mm0, [eax] |
|
|
movq mm1, [edx] |
|
|
|
|
|
movq mm2, [eax+8] |
|
|
movq mm3, [edx+8] |
|
|
|
|
|
movq mm4, mm0 |
|
|
psubusb mm0, mm1 |
|
363 |
|
|
364 |
psubusb mm1, mm4 |
ALIGN 16 |
|
por mm0, mm1 |
|
|
lea eax,[eax+ecx] |
|
|
|
|
|
movq mm4, mm2 |
|
|
psubusb mm2, mm3 |
|
|
|
|
|
psubusb mm3, mm4 |
|
|
por mm2, mm3 |
|
|
lea edx,[edx+ecx] |
|
|
|
|
|
movq mm1,mm0 |
|
|
movq mm3,mm2 |
|
|
|
|
|
punpcklbw mm0,mm7 |
|
|
punpckhbw mm1,mm7 |
|
|
punpcklbw mm2,mm7 |
|
|
punpckhbw mm3,mm7 |
|
|
|
|
|
paddusw mm0,mm1 |
|
|
paddusw mm2,mm3 |
|
|
|
|
|
paddusw mm5, mm0 |
|
|
paddusw mm6, mm2 |
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
365 |
sad16v_mmx: |
sad16v_mmx: |
366 |
|
|
367 |
push ebx |
push ebx |
439 |
|
|
440 |
pop edi |
pop edi |
441 |
pop ebx |
pop ebx |
|
ret |
|
|
|
|
442 |
|
|
443 |
|
ret |
444 |
|
|
445 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
446 |
; |
; |
447 |
; uint32_t sad16bi_mmx(const uint8_t * const cur, |
; uint32_t sad16bi_mmx(const uint8_t * const cur, |
448 |
; const uint8_t * const ref1, |
; const uint8_t * const ref1, |
449 |
; const uint8_t * const ref2, |
; const uint8_t * const ref2, |
450 |
; const uint32_t stride); |
; const uint32_t stride); |
451 |
; |
; |
452 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
%macro SADBI_16x16_MMX 2 ; SADBI_16x16_MMX( int_ptr_offset, bool_increment_ptr ); |
|
|
|
|
|
movq mm0, [edx+%1] |
|
|
movq mm2, [ebx+%1] |
|
|
movq mm1, mm0 |
|
|
movq mm3, mm2 |
|
|
|
|
|
%if %2 != 0 |
|
|
add edx, ecx |
|
|
%endif |
|
|
|
|
|
punpcklbw mm0, mm7 |
|
|
punpckhbw mm1, mm7 |
|
|
punpcklbw mm2, mm7 |
|
|
punpckhbw mm3, mm7 |
|
|
|
|
|
%if %2 != 0 |
|
|
add ebx, ecx |
|
|
%endif |
|
|
|
|
|
paddusw mm0, mm2 ; mm01 = ref1 + ref2 |
|
|
paddusw mm1, mm3 |
|
|
paddusw mm0, [mmx_one] ; mm01 += 1 |
|
|
paddusw mm1, [mmx_one] |
|
|
psrlw mm0, 1 ; mm01 >>= 1 |
|
|
psrlw mm1, 1 |
|
453 |
|
|
454 |
movq mm2, [eax+%1] |
ALIGN 16 |
|
movq mm3, mm2 |
|
|
punpcklbw mm2, mm7 ; mm23 = src |
|
|
punpckhbw mm3, mm7 |
|
|
|
|
|
%if %2 != 0 |
|
|
add eax, ecx |
|
|
%endif |
|
|
|
|
|
movq mm4, mm0 |
|
|
movq mm5, mm1 |
|
|
psubusw mm0, mm2 |
|
|
psubusw mm1, mm3 |
|
|
psubusw mm2, mm4 |
|
|
psubusw mm3, mm5 |
|
|
por mm0, mm2 ; mm01 = ABS(mm01 - mm23) |
|
|
por mm1, mm3 |
|
|
|
|
|
paddusw mm6,mm0 ; mm6 += mm01 |
|
|
paddusw mm6,mm1 |
|
|
|
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
455 |
sad16bi_mmx: |
sad16bi_mmx: |
456 |
push ebx |
push ebx |
457 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |
503 |
|
|
504 |
movd eax, mm6 |
movd eax, mm6 |
505 |
pop ebx |
pop ebx |
506 |
|
|
507 |
ret |
ret |
508 |
|
|
509 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
510 |
; |
; |
511 |
; uint32_t sad8bi_mmx(const uint8_t * const cur, |
; uint32_t sad8bi_mmx(const uint8_t * const cur, |
512 |
; const uint8_t * const ref1, |
; const uint8_t * const ref1, |
513 |
; const uint8_t * const ref2, |
; const uint8_t * const ref2, |
514 |
; const uint32_t stride); |
; const uint32_t stride); |
515 |
; |
; |
516 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
517 |
align 16 |
|
518 |
|
ALIGN 16 |
519 |
sad8bi_mmx: |
sad8bi_mmx: |
520 |
push ebx |
push ebx |
521 |
mov eax, [esp+4+ 4] ; Src |
mov eax, [esp+4+ 4] ; Src |
544 |
pop ebx |
pop ebx |
545 |
ret |
ret |
546 |
|
|
547 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
|
|
;=========================================================================== |
|
548 |
; |
; |
549 |
; uint32_t dev16_mmx(const uint8_t * const cur, |
; uint32_t dev16_mmx(const uint8_t * const cur, |
550 |
; const uint32_t stride); |
; const uint32_t stride); |
551 |
; |
; |
552 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro MEAN_16x16_MMX 0 |
|
|
movq mm0, [eax] |
|
|
movq mm2, [eax+8] |
|
|
lea eax,[eax+ecx] |
|
|
movq mm1, mm0 |
|
|
movq mm3, mm2 |
|
|
punpcklbw mm0,mm7 |
|
|
punpcklbw mm2,mm7 |
|
|
punpckhbw mm1,mm7 |
|
|
punpckhbw mm3,mm7 |
|
|
paddw mm5, mm0 |
|
|
paddw mm6, mm1 |
|
|
paddw mm5, mm2 |
|
|
paddw mm6, mm3 |
|
|
%endmacro |
|
|
|
|
|
%macro ABS_16x16_MMX 0 |
|
|
movq mm0, [eax] |
|
|
movq mm2, [eax+8] |
|
|
lea eax,[eax+ecx] |
|
|
movq mm1, mm0 |
|
|
movq mm3, mm2 |
|
|
punpcklbw mm0, mm7 |
|
|
punpcklbw mm2, mm7 |
|
|
punpckhbw mm1, mm7 |
|
|
punpckhbw mm3, mm7 |
|
|
movq mm4, mm6 |
|
|
psubusw mm4, mm0 |
|
|
|
|
|
psubusw mm0, mm6 |
|
|
por mm0, mm4 |
|
|
movq mm4, mm6 |
|
|
psubusw mm4, mm1 |
|
|
psubusw mm1, mm6 |
|
|
por mm1, mm4 |
|
|
|
|
|
movq mm4, mm6 |
|
|
psubusw mm4, mm2 |
|
|
psubusw mm2, mm6 |
|
|
por mm2, mm4 |
|
|
movq mm4, mm6 |
|
|
psubusw mm4, mm3 |
|
|
psubusw mm3, mm6 |
|
|
por mm3, mm4 |
|
|
|
|
|
paddw mm0, mm1 |
|
|
paddw mm2, mm3 |
|
|
paddw mm5, mm0 |
|
|
paddw mm5, mm2 |
|
|
%endmacro |
|
553 |
|
|
554 |
align 16 |
ALIGN 16 |
555 |
dev16_mmx: |
dev16_mmx: |
556 |
mov eax, [esp+ 4] ; Src |
mov eax, [esp+ 4] ; Src |
557 |
mov ecx, [esp+ 8] ; Stride |
mov ecx, [esp+ 8] ; Stride |
620 |
paddd mm6, mm5 |
paddd mm6, mm5 |
621 |
|
|
622 |
movd eax, mm6 |
movd eax, mm6 |
623 |
|
|
624 |
|
ret |
625 |
|
|
626 |
|
;----------------------------------------------------------------------------- |
627 |
|
; |
628 |
|
; uint32_t sse8_16bit_mmx(const int16_t *b1, |
629 |
|
; const int16_t *b2, |
630 |
|
; const uint32_t stride); |
631 |
|
; |
632 |
|
;----------------------------------------------------------------------------- |
633 |
|
|
634 |
|
%macro ROW_SSE_16bit_MMX 2 |
635 |
|
movq mm0, [%1] |
636 |
|
movq mm1, [%1+8] |
637 |
|
psubw mm0, [%2] |
638 |
|
psubw mm1, [%2+8] |
639 |
|
pmaddwd mm0, mm0 |
640 |
|
pmaddwd mm1, mm1 |
641 |
|
paddd mm2, mm0 |
642 |
|
paddd mm2, mm1 |
643 |
|
%endmacro |
644 |
|
|
645 |
|
sse8_16bit_mmx: |
646 |
|
push esi |
647 |
|
push edi |
648 |
|
|
649 |
|
;; Load the function params |
650 |
|
mov esi, [esp+8+4] |
651 |
|
mov edi, [esp+8+8] |
652 |
|
mov edx, [esp+8+12] |
653 |
|
|
654 |
|
;; Reset the sse accumulator |
655 |
|
pxor mm2, mm2 |
656 |
|
|
657 |
|
;; Let's go |
658 |
|
%rep 8 |
659 |
|
ROW_SSE_16bit_MMX esi, edi |
660 |
|
lea esi, [esi+edx] |
661 |
|
lea edi, [edi+edx] |
662 |
|
%endrep |
663 |
|
|
664 |
|
;; Finish adding each dword of the accumulator |
665 |
|
movq mm3, mm2 |
666 |
|
psrlq mm2, 32 |
667 |
|
paddd mm2, mm3 |
668 |
|
movd eax, mm2 |
669 |
|
|
670 |
|
;; All done |
671 |
|
pop edi |
672 |
|
pop esi |
673 |
|
ret |
674 |
|
|
675 |
|
;----------------------------------------------------------------------------- |
676 |
|
; |
677 |
|
; uint32_t sse8_8bit_mmx(const int8_t *b1, |
678 |
|
; const int8_t *b2, |
679 |
|
; const uint32_t stride); |
680 |
|
; |
681 |
|
;----------------------------------------------------------------------------- |
682 |
|
|
683 |
|
%macro ROW_SSE_8bit_MMX 2 |
684 |
|
movq mm0, [%1] ; load a row |
685 |
|
movq mm2, [%2] ; load a row |
686 |
|
|
687 |
|
movq mm1, mm0 ; copy row |
688 |
|
movq mm3, mm2 ; copy row |
689 |
|
|
690 |
|
punpcklbw mm0, mm7 ; turn the 4low elements into 16bit |
691 |
|
punpckhbw mm1, mm7 ; turn the 4high elements into 16bit |
692 |
|
|
693 |
|
punpcklbw mm2, mm7 ; turn the 4low elements into 16bit |
694 |
|
punpckhbw mm3, mm7 ; turn the 4high elements into 16bit |
695 |
|
|
696 |
|
psubw mm0, mm2 ; low part of src-dst |
697 |
|
psubw mm1, mm3 ; high part of src-dst |
698 |
|
|
699 |
|
pmaddwd mm0, mm0 ; compute the square sum |
700 |
|
pmaddwd mm1, mm1 ; compute the square sum |
701 |
|
|
702 |
|
paddd mm6, mm0 ; add to the accumulator |
703 |
|
paddd mm6, mm1 ; add to the accumulator |
704 |
|
%endmacro |
705 |
|
|
706 |
|
sse8_8bit_mmx: |
707 |
|
push esi |
708 |
|
push edi |
709 |
|
|
710 |
|
;; Load the function params |
711 |
|
mov esi, [esp+8+4] |
712 |
|
mov edi, [esp+8+8] |
713 |
|
mov edx, [esp+8+12] |
714 |
|
|
715 |
|
;; Reset the sse accumulator |
716 |
|
pxor mm6, mm6 |
717 |
|
|
718 |
|
;; Used to interleave 8bit data with 0x00 values |
719 |
|
pxor mm7, mm7 |
720 |
|
|
721 |
|
;; Let's go |
722 |
|
%rep 8 |
723 |
|
ROW_SSE_8bit_MMX esi, edi |
724 |
|
lea esi, [esi+edx] |
725 |
|
lea edi, [edi+edx] |
726 |
|
%endrep |
727 |
|
|
728 |
|
;; Finish adding each dword of the accumulator |
729 |
|
movq mm7, mm6 |
730 |
|
psrlq mm6, 32 |
731 |
|
paddd mm6, mm7 |
732 |
|
movd eax, mm6 |
733 |
|
|
734 |
|
;; All done |
735 |
|
pop edi |
736 |
|
pop esi |
737 |
ret |
ret |