1 |
;/***************************************************************************** |
;/***************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * mmx 8x8 block-based halfpel interpolation |
; * - mmx 8x8 block-based halfpel interpolation - |
5 |
; * |
; * |
6 |
; * Copyright(C) 2002 Peter Ross <pross@xvid.org> |
; * Copyright(C) 2001 Peter Ross <pross@xvid.org> |
7 |
|
; * 2002-2008 Michael Militzer <michael@xvid.org> |
8 |
; * |
; * |
9 |
; * This file is part of XviD, a free MPEG-4 video encoder/decoder |
; * This program is free software ; you can redistribute it and/or modify |
10 |
; * |
; * it under the terms of the GNU General Public License as published by |
|
; * XviD is free software; you can redistribute it and/or modify it |
|
|
; * 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 |
; * |
; * |
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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
22 |
; * |
; * |
|
; * Under section 8 of the GNU General Public License, the copyright |
|
|
; * holders of XVID explicitly forbid distribution in the following |
|
|
; * countries: |
|
|
; * |
|
|
; * - Japan |
|
|
; * - United States of America |
|
|
; * |
|
|
; * Linking XviD statically or dynamically with other modules is making a |
|
|
; * combined work based on XviD. Thus, the terms and conditions of the |
|
|
; * GNU General Public License cover the whole combination. |
|
|
; * |
|
|
; * As a special exception, the copyright holders of XviD give you |
|
|
; * permission to link XviD with independent modules that communicate with |
|
|
; * XviD solely through the VFW1.1 and DShow interfaces, regardless of the |
|
|
; * license terms of these independent modules, and to copy and distribute |
|
|
; * the resulting combined work under terms of your choice, provided that |
|
|
; * every copy of the combined work is accompanied by a complete copy of |
|
|
; * the source code of XviD (the version of XviD used to produce the |
|
|
; * combined work), being distributed under the terms of the GNU General |
|
|
; * Public License plus this exception. An independent module is a module |
|
|
; * which is not derived from or based on XviD. |
|
|
; * |
|
|
; * Note that people who make modified versions of XviD are not obligated |
|
|
; * to grant this special exception for their modified versions; it is |
|
|
; * their choice whether to do so. The GNU General Public License gives |
|
|
; * permission to release a modified version without this exception; this |
|
|
; * exception also makes it possible to release a modified version which |
|
|
; * carries forward this exception. |
|
|
; * |
|
|
; * $Id: interpolate8x8_mmx.asm,v 1.11 2002-11-17 00:20:30 edgomez Exp $ |
|
|
; * |
|
23 |
; ****************************************************************************/ |
; ****************************************************************************/ |
24 |
|
|
25 |
bits 32 |
%include "nasm.inc" |
26 |
|
|
27 |
%macro cglobal 1 |
;============================================================================= |
28 |
%ifdef PREFIX |
; Read only data |
29 |
global _%1 |
;============================================================================= |
|
%define %1 _%1 |
|
|
%else |
|
|
global %1 |
|
|
%endif |
|
|
%endmacro |
|
30 |
|
|
31 |
section .data |
DATA |
32 |
|
|
33 |
align 16 |
;----------------------------------------------------------------------------- |
34 |
|
; (16 - r) rounding table |
35 |
|
;----------------------------------------------------------------------------- |
36 |
|
|
37 |
;=========================================================================== |
ALIGN SECTION_ALIGN |
38 |
|
rounding_lowpass_mmx: |
39 |
|
times 4 dw 16 |
40 |
|
times 4 dw 15 |
41 |
|
|
42 |
|
;----------------------------------------------------------------------------- |
43 |
; (1 - r) rounding table |
; (1 - r) rounding table |
44 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
45 |
|
|
46 |
rounding1_mmx |
rounding1_mmx: |
47 |
times 4 dw 1 |
times 4 dw 1 |
48 |
times 4 dw 0 |
times 4 dw 0 |
49 |
|
|
50 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
51 |
; (2 - r) rounding table |
; (2 - r) rounding table |
52 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
53 |
|
|
54 |
rounding2_mmx |
rounding2_mmx: |
55 |
times 4 dw 2 |
times 4 dw 2 |
56 |
times 4 dw 1 |
times 4 dw 1 |
57 |
|
|
58 |
mmx_one |
mmx_one: |
59 |
times 8 db 1 |
times 8 db 1 |
60 |
|
|
61 |
section .text |
mmx_two: |
62 |
|
times 8 db 2 |
63 |
|
|
64 |
|
mmx_three: |
65 |
|
times 8 db 3 |
66 |
|
|
67 |
|
mmx_five: |
68 |
|
times 4 dw 5 |
69 |
|
|
70 |
|
mmx_mask: |
71 |
|
times 8 db 254 |
72 |
|
|
73 |
|
mmx_mask2: |
74 |
|
times 8 db 252 |
75 |
|
|
76 |
|
;============================================================================= |
77 |
|
; Code |
78 |
|
;============================================================================= |
79 |
|
|
80 |
|
SECTION .rotext align=SECTION_ALIGN |
81 |
|
|
82 |
|
cglobal interpolate8x8_halfpel_h_mmx |
83 |
|
cglobal interpolate8x8_halfpel_v_mmx |
84 |
|
cglobal interpolate8x8_halfpel_hv_mmx |
85 |
|
|
86 |
|
cglobal interpolate8x4_halfpel_h_mmx |
87 |
|
cglobal interpolate8x4_halfpel_v_mmx |
88 |
|
cglobal interpolate8x4_halfpel_hv_mmx |
89 |
|
|
90 |
|
cglobal interpolate8x8_avg4_mmx |
91 |
|
cglobal interpolate8x8_avg2_mmx |
92 |
|
|
93 |
|
cglobal interpolate8x8_6tap_lowpass_h_mmx |
94 |
|
cglobal interpolate8x8_6tap_lowpass_v_mmx |
95 |
|
|
96 |
|
cglobal interpolate8x8_halfpel_add_mmx |
97 |
|
cglobal interpolate8x8_halfpel_h_add_mmx |
98 |
|
cglobal interpolate8x8_halfpel_v_add_mmx |
99 |
|
cglobal interpolate8x8_halfpel_hv_add_mmx |
100 |
|
|
101 |
%macro CALC_AVG 6 |
%macro CALC_AVG 6 |
102 |
punpcklbw %3, %6 |
punpcklbw %3, %6 |
109 |
|
|
110 |
psrlw %1, 1 ; mm01 >>= 1 |
psrlw %1, 1 ; mm01 >>= 1 |
111 |
psrlw %2, 1 |
psrlw %2, 1 |
|
|
|
112 |
%endmacro |
%endmacro |
113 |
|
|
114 |
|
|
115 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
116 |
; |
; |
117 |
; void interpolate8x8_halfpel_h_mmx(uint8_t * const dst, |
; void interpolate8x8_halfpel_h_mmx(uint8_t * const dst, |
118 |
; const uint8_t * const src, |
; const uint8_t * const src, |
119 |
; const uint32_t stride, |
; const uint32_t stride, |
120 |
; const uint32_t rounding); |
; const uint32_t rounding); |
121 |
; |
; |
122 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
123 |
|
|
124 |
%macro COPY_H_MMX 0 |
%macro COPY_H_MMX 0 |
125 |
movq mm0, [esi] |
movq mm0, [TMP0] |
126 |
movq mm2, [esi + 1] |
movq mm2, [TMP0 + 1] |
127 |
movq mm1, mm0 |
movq mm1, mm0 |
128 |
movq mm3, mm2 |
movq mm3, mm2 |
129 |
|
|
133 |
CALC_AVG mm0, mm1, mm2, mm3, mm7, mm6 |
CALC_AVG mm0, mm1, mm2, mm3, mm7, mm6 |
134 |
|
|
135 |
packuswb mm0, mm1 |
packuswb mm0, mm1 |
136 |
movq [edi], mm0 ; [dst] = mm01 |
movq [_EAX], mm0 ; [dst] = mm01 |
137 |
|
|
138 |
add esi, edx ; src += stride |
add TMP0, TMP1 ; src += stride |
139 |
add edi, edx ; dst += stride |
add _EAX, TMP1 ; dst += stride |
140 |
%endmacro |
%endmacro |
141 |
|
|
142 |
align 16 |
ALIGN SECTION_ALIGN |
143 |
cglobal interpolate8x8_halfpel_h_mmx |
interpolate8x8_halfpel_h_mmx: |
|
interpolate8x8_halfpel_h_mmx |
|
|
|
|
|
push esi |
|
|
push edi |
|
|
|
|
|
mov eax, [esp + 8 + 16] ; rounding |
|
|
|
|
|
interpolate8x8_halfpel_h_mmx.start |
|
|
movq mm7, [rounding1_mmx + eax * 8] |
|
144 |
|
|
145 |
mov edi, [esp + 8 + 4] ; dst |
mov _EAX, prm4 ; rounding |
146 |
mov esi, [esp + 8 + 8] ; src |
lea TMP0, [rounding1_mmx] |
147 |
mov edx, [esp + 8 + 12] ; stride |
movq mm7, [TMP0 + _EAX * 8] |
148 |
|
|
149 |
|
mov _EAX, prm1 ; dst |
150 |
|
mov TMP0, prm2 ; src |
151 |
|
mov TMP1, prm3 ; stride |
152 |
|
|
153 |
pxor mm6, mm6 ; zero |
pxor mm6, mm6 ; zero |
154 |
|
|
161 |
COPY_H_MMX |
COPY_H_MMX |
162 |
COPY_H_MMX |
COPY_H_MMX |
163 |
|
|
|
pop edi |
|
|
pop esi |
|
|
|
|
164 |
ret |
ret |
165 |
|
ENDFUNC |
166 |
|
|
167 |
|
|
168 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
169 |
; |
; |
170 |
; void interpolate8x8_halfpel_v_mmx(uint8_t * const dst, |
; void interpolate8x8_halfpel_v_mmx(uint8_t * const dst, |
171 |
; const uint8_t * const src, |
; const uint8_t * const src, |
172 |
; const uint32_t stride, |
; const uint32_t stride, |
173 |
; const uint32_t rounding); |
; const uint32_t rounding); |
174 |
; |
; |
175 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
176 |
|
|
177 |
%macro COPY_V_MMX 0 |
%macro COPY_V_MMX 0 |
178 |
movq mm0, [esi] |
movq mm0, [TMP0] |
179 |
movq mm2, [esi + edx] |
movq mm2, [TMP0 + TMP1] |
180 |
movq mm1, mm0 |
movq mm1, mm0 |
181 |
movq mm3, mm2 |
movq mm3, mm2 |
182 |
|
|
186 |
CALC_AVG mm0, mm1, mm2, mm3, mm7, mm6 |
CALC_AVG mm0, mm1, mm2, mm3, mm7, mm6 |
187 |
|
|
188 |
packuswb mm0, mm1 |
packuswb mm0, mm1 |
189 |
movq [edi], mm0 ; [dst] = mm01 |
movq [_EAX], mm0 ; [dst] = mm01 |
190 |
|
|
191 |
add esi, edx ; src += stride |
add TMP0, TMP1 ; src += stride |
192 |
add edi, edx ; dst += stride |
add _EAX, TMP1 ; dst += stride |
193 |
%endmacro |
%endmacro |
194 |
|
|
195 |
align 16 |
ALIGN SECTION_ALIGN |
196 |
cglobal interpolate8x8_halfpel_v_mmx |
interpolate8x8_halfpel_v_mmx: |
|
interpolate8x8_halfpel_v_mmx |
|
|
|
|
|
push esi |
|
|
push edi |
|
|
|
|
|
mov eax, [esp + 8 + 16] ; rounding |
|
|
|
|
|
interpolate8x8_halfpel_v_mmx.start |
|
|
movq mm7, [rounding1_mmx + eax * 8] |
|
197 |
|
|
198 |
mov edi, [esp + 8 + 4] ; dst |
mov _EAX, prm4 ; rounding |
199 |
mov esi, [esp + 8 + 8] ; src |
lea TMP0, [rounding1_mmx] |
200 |
mov edx, [esp + 8 + 12] ; stride |
movq mm7, [TMP0 + _EAX * 8] |
201 |
|
|
202 |
|
mov _EAX, prm1 ; dst |
203 |
|
mov TMP0, prm2 ; src |
204 |
|
mov TMP1, prm3 ; stride |
205 |
|
|
206 |
pxor mm6, mm6 ; zero |
pxor mm6, mm6 ; zero |
207 |
|
|
215 |
COPY_V_MMX |
COPY_V_MMX |
216 |
COPY_V_MMX |
COPY_V_MMX |
217 |
|
|
|
pop edi |
|
|
pop esi |
|
|
|
|
218 |
ret |
ret |
219 |
|
ENDFUNC |
220 |
|
|
221 |
|
|
222 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
223 |
; |
; |
224 |
; void interpolate8x8_halfpel_hv_mmx(uint8_t * const dst, |
; void interpolate8x8_halfpel_hv_mmx(uint8_t * const dst, |
225 |
; const uint8_t * const src, |
; const uint8_t * const src, |
227 |
; const uint32_t rounding); |
; const uint32_t rounding); |
228 |
; |
; |
229 |
; |
; |
230 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
231 |
|
|
232 |
%macro COPY_HV_MMX 0 |
%macro COPY_HV_MMX 0 |
233 |
; current row |
; current row |
234 |
|
movq mm0, [TMP0] |
235 |
movq mm0, [esi] |
movq mm2, [TMP0 + 1] |
|
movq mm2, [esi + 1] |
|
236 |
|
|
237 |
movq mm1, mm0 |
movq mm1, mm0 |
238 |
movq mm3, mm2 |
movq mm3, mm2 |
246 |
paddusw mm1, mm3 |
paddusw mm1, mm3 |
247 |
|
|
248 |
; next row |
; next row |
249 |
|
movq mm4, [TMP0 + TMP1] |
250 |
movq mm4, [esi + edx] |
movq mm2, [TMP0 + TMP1 + 1] |
|
movq mm2, [esi + edx + 1] |
|
251 |
|
|
252 |
movq mm5, mm4 |
movq mm5, mm4 |
253 |
movq mm3, mm2 |
movq mm3, mm2 |
261 |
paddusw mm5, mm3 |
paddusw mm5, mm3 |
262 |
|
|
263 |
; add current + next row |
; add current + next row |
|
|
|
264 |
paddusw mm0, mm4 ; mm01 += mm45 |
paddusw mm0, mm4 ; mm01 += mm45 |
265 |
paddusw mm1, mm5 |
paddusw mm1, mm5 |
266 |
paddusw mm0, mm7 ; mm01 += rounding2 |
paddusw mm0, mm7 ; mm01 += rounding2 |
270 |
psrlw mm1, 2 |
psrlw mm1, 2 |
271 |
|
|
272 |
packuswb mm0, mm1 |
packuswb mm0, mm1 |
273 |
movq [edi], mm0 ; [dst] = mm01 |
movq [_EAX], mm0 ; [dst] = mm01 |
274 |
|
|
275 |
add esi, edx ; src += stride |
add TMP0, TMP1 ; src += stride |
276 |
add edi, edx ; dst += stride |
add _EAX, TMP1 ; dst += stride |
277 |
%endmacro |
%endmacro |
278 |
|
|
279 |
align 16 |
ALIGN SECTION_ALIGN |
280 |
cglobal interpolate8x8_halfpel_hv_mmx |
interpolate8x8_halfpel_hv_mmx: |
|
interpolate8x8_halfpel_hv_mmx |
|
|
|
|
|
push esi |
|
|
push edi |
|
|
|
|
|
mov eax, [esp + 8 + 16] ; rounding |
|
|
interpolate8x8_halfpel_hv_mmx.start |
|
|
|
|
|
movq mm7, [rounding2_mmx + eax * 8] |
|
281 |
|
|
282 |
mov edi, [esp + 8 + 4] ; dst |
mov _EAX, prm4 ; rounding |
283 |
mov esi, [esp + 8 + 8] ; src |
lea TMP0, [rounding2_mmx] |
284 |
|
movq mm7, [TMP0 + _EAX * 8] |
285 |
|
|
286 |
mov eax, 8 |
mov _EAX, prm1 ; dst |
287 |
|
mov TMP0, prm2 ; src |
288 |
|
|
289 |
pxor mm6, mm6 ; zero |
pxor mm6, mm6 ; zero |
290 |
|
|
291 |
mov edx, [esp + 8 + 12] ; stride |
mov TMP1, prm3 ; stride |
292 |
|
|
293 |
COPY_HV_MMX |
COPY_HV_MMX |
294 |
COPY_HV_MMX |
COPY_HV_MMX |
299 |
COPY_HV_MMX |
COPY_HV_MMX |
300 |
COPY_HV_MMX |
COPY_HV_MMX |
301 |
|
|
302 |
pop edi |
ret |
303 |
pop esi |
ENDFUNC |
304 |
|
|
305 |
|
;----------------------------------------------------------------------------- |
306 |
|
; |
307 |
|
; void interpolate8x4_halfpel_h_mmx(uint8_t * const dst, |
308 |
|
; const uint8_t * const src, |
309 |
|
; const uint32_t stride, |
310 |
|
; const uint32_t rounding); |
311 |
|
; |
312 |
|
;----------------------------------------------------------------------------- |
313 |
|
|
314 |
|
ALIGN SECTION_ALIGN |
315 |
|
interpolate8x4_halfpel_h_mmx: |
316 |
|
|
317 |
|
mov _EAX, prm4 ; rounding |
318 |
|
lea TMP0, [rounding1_mmx] |
319 |
|
movq mm7, [TMP0 + _EAX * 8] |
320 |
|
|
321 |
|
mov _EAX, prm1 ; dst |
322 |
|
mov TMP0, prm2 ; src |
323 |
|
mov TMP1, prm3 ; stride |
324 |
|
|
325 |
|
pxor mm6, mm6 ; zero |
326 |
|
|
327 |
|
COPY_H_MMX |
328 |
|
COPY_H_MMX |
329 |
|
COPY_H_MMX |
330 |
|
COPY_H_MMX |
331 |
|
|
332 |
|
ret |
333 |
|
ENDFUNC |
334 |
|
|
335 |
|
|
336 |
|
;----------------------------------------------------------------------------- |
337 |
|
; |
338 |
|
; void interpolate8x4_halfpel_v_mmx(uint8_t * const dst, |
339 |
|
; const uint8_t * const src, |
340 |
|
; const uint32_t stride, |
341 |
|
; const uint32_t rounding); |
342 |
|
; |
343 |
|
;----------------------------------------------------------------------------- |
344 |
|
|
345 |
|
ALIGN SECTION_ALIGN |
346 |
|
interpolate8x4_halfpel_v_mmx: |
347 |
|
|
348 |
|
mov _EAX, prm4 ; rounding |
349 |
|
lea TMP0, [rounding1_mmx] |
350 |
|
movq mm7, [TMP0 + _EAX * 8] |
351 |
|
|
352 |
|
mov _EAX, prm1 ; dst |
353 |
|
mov TMP0, prm2 ; src |
354 |
|
mov TMP1, prm3 ; stride |
355 |
|
|
356 |
|
pxor mm6, mm6 ; zero |
357 |
|
|
358 |
|
|
359 |
|
COPY_V_MMX |
360 |
|
COPY_V_MMX |
361 |
|
COPY_V_MMX |
362 |
|
COPY_V_MMX |
363 |
|
|
364 |
ret |
ret |
365 |
|
ENDFUNC |
366 |
|
|
367 |
|
|
368 |
|
;----------------------------------------------------------------------------- |
369 |
|
; |
370 |
|
; void interpolate8x4_halfpel_hv_mmx(uint8_t * const dst, |
371 |
|
; const uint8_t * const src, |
372 |
|
; const uint32_t stride, |
373 |
|
; const uint32_t rounding); |
374 |
|
; |
375 |
|
; |
376 |
|
;----------------------------------------------------------------------------- |
377 |
|
|
378 |
|
ALIGN SECTION_ALIGN |
379 |
|
interpolate8x4_halfpel_hv_mmx: |
380 |
|
|
381 |
|
mov _EAX, prm4 ; rounding |
382 |
|
lea TMP0, [rounding2_mmx] |
383 |
|
movq mm7, [TMP0 + _EAX * 8] |
384 |
|
|
385 |
|
mov _EAX, prm1 ; dst |
386 |
|
mov TMP0, prm2 ; src |
387 |
|
|
388 |
|
pxor mm6, mm6 ; zero |
389 |
|
|
390 |
|
mov TMP1, prm3 ; stride |
391 |
|
|
392 |
|
COPY_HV_MMX |
393 |
|
COPY_HV_MMX |
394 |
|
COPY_HV_MMX |
395 |
|
COPY_HV_MMX |
396 |
|
|
397 |
|
ret |
398 |
|
ENDFUNC |
399 |
|
|
400 |
|
;----------------------------------------------------------------------------- |
401 |
|
; |
402 |
|
; void interpolate8x8_avg2_mmx(uint8_t const *dst, |
403 |
|
; const uint8_t * const src1, |
404 |
|
; const uint8_t * const src2, |
405 |
|
; const uint32_t stride, |
406 |
|
; const uint32_t rounding, |
407 |
|
; const uint32_t height); |
408 |
|
; |
409 |
|
;----------------------------------------------------------------------------- |
410 |
|
|
411 |
|
%macro AVG2_MMX_RND0 0 |
412 |
|
movq mm0, [_EAX] ; src1 -> mm0 |
413 |
|
movq mm1, [_EBX] ; src2 -> mm1 |
414 |
|
|
415 |
|
movq mm4, [_EAX+TMP1] |
416 |
|
movq mm5, [_EBX+TMP1] |
417 |
|
|
418 |
|
movq mm2, mm0 ; src1 -> mm2 |
419 |
|
movq mm3, mm1 ; src2 -> mm3 |
420 |
|
|
421 |
|
pand mm2, mm7 ; isolate the lsb |
422 |
|
pand mm3, mm7 ; isolate the lsb |
423 |
|
|
424 |
|
por mm2, mm3 ; ODD(src1) OR ODD(src2) -> mm2 |
425 |
|
|
426 |
|
movq mm3, mm4 |
427 |
|
movq mm6, mm5 |
428 |
|
|
429 |
|
pand mm3, mm7 |
430 |
|
pand mm6, mm7 |
431 |
|
|
432 |
|
por mm3, mm6 |
433 |
|
|
434 |
|
pand mm0, [mmx_mask] |
435 |
|
pand mm1, [mmx_mask] |
436 |
|
pand mm4, [mmx_mask] |
437 |
|
pand mm5, [mmx_mask] |
438 |
|
|
439 |
|
psrlq mm0, 1 ; src1 / 2 |
440 |
|
psrlq mm1, 1 ; src2 / 2 |
441 |
|
|
442 |
|
psrlq mm4, 1 |
443 |
|
psrlq mm5, 1 |
444 |
|
|
445 |
|
paddb mm0, mm1 ; src1/2 + src2/2 -> mm0 |
446 |
|
paddb mm0, mm2 ; correct rounding error |
447 |
|
|
448 |
|
paddb mm4, mm5 |
449 |
|
paddb mm4, mm3 |
450 |
|
|
451 |
|
lea _EAX, [_EAX+2*TMP1] |
452 |
|
lea _EBX, [_EBX+2*TMP1] |
453 |
|
|
454 |
|
movq [TMP0], mm0 ; (src1 + src2 + 1) / 2 -> dst |
455 |
|
movq [TMP0+TMP1], mm4 |
456 |
|
%endmacro |
457 |
|
|
458 |
|
%macro AVG2_MMX_RND1 0 |
459 |
|
movq mm0, [_EAX] ; src1 -> mm0 |
460 |
|
movq mm1, [_EBX] ; src2 -> mm1 |
461 |
|
|
462 |
|
movq mm4, [_EAX+TMP1] |
463 |
|
movq mm5, [_EBX+TMP1] |
464 |
|
|
465 |
|
movq mm2, mm0 ; src1 -> mm2 |
466 |
|
movq mm3, mm1 ; src2 -> mm3 |
467 |
|
|
468 |
|
pand mm2, mm7 ; isolate the lsb |
469 |
|
pand mm3, mm7 ; isolate the lsb |
470 |
|
|
471 |
|
pand mm2, mm3 ; ODD(src1) AND ODD(src2) -> mm2 |
472 |
|
|
473 |
|
movq mm3, mm4 |
474 |
|
movq mm6, mm5 |
475 |
|
|
476 |
|
pand mm3, mm7 |
477 |
|
pand mm6, mm7 |
478 |
|
|
479 |
|
pand mm3, mm6 |
480 |
|
|
481 |
|
pand mm0, [mmx_mask] |
482 |
|
pand mm1, [mmx_mask] |
483 |
|
pand mm4, [mmx_mask] |
484 |
|
pand mm5, [mmx_mask] |
485 |
|
|
486 |
|
psrlq mm0, 1 ; src1 / 2 |
487 |
|
psrlq mm1, 1 ; src2 / 2 |
488 |
|
|
489 |
|
psrlq mm4, 1 |
490 |
|
psrlq mm5, 1 |
491 |
|
|
492 |
|
paddb mm0, mm1 ; src1/2 + src2/2 -> mm0 |
493 |
|
paddb mm0, mm2 ; correct rounding error |
494 |
|
|
495 |
|
paddb mm4, mm5 |
496 |
|
paddb mm4, mm3 |
497 |
|
|
498 |
|
lea _EAX, [_EAX+2*TMP1] |
499 |
|
lea _EBX, [_EBX+2*TMP1] |
500 |
|
|
501 |
|
movq [TMP0], mm0 ; (src1 + src2 + 1) / 2 -> dst |
502 |
|
movq [TMP0+TMP1], mm4 |
503 |
|
%endmacro |
504 |
|
|
505 |
|
ALIGN SECTION_ALIGN |
506 |
|
interpolate8x8_avg2_mmx: |
507 |
|
|
508 |
|
mov eax, prm5d ; rounding |
509 |
|
test _EAX, _EAX |
510 |
|
|
511 |
|
jnz near .rounding1 |
512 |
|
|
513 |
|
mov eax, prm6d ; height -> _EAX |
514 |
|
sub _EAX, 8 |
515 |
|
|
516 |
|
mov TMP0, prm1 ; dst -> edi |
517 |
|
mov _EAX, prm2 ; src1 -> esi |
518 |
|
mov TMP1, prm4 ; stride -> TMP1 |
519 |
|
|
520 |
|
push _EBX |
521 |
|
%ifdef ARCH_IS_X86_64 |
522 |
|
mov _EBX, prm3 |
523 |
|
%else |
524 |
|
mov _EBX, [esp + 4 + 12] ; src2 -> eax |
525 |
|
%endif |
526 |
|
|
527 |
|
movq mm7, [mmx_one] |
528 |
|
|
529 |
|
jz near .start0 |
530 |
|
|
531 |
|
AVG2_MMX_RND0 |
532 |
|
lea TMP0, [TMP0+2*TMP1] |
533 |
|
|
534 |
|
.start0: |
535 |
|
|
536 |
|
AVG2_MMX_RND0 |
537 |
|
lea TMP0, [TMP0+2*TMP1] |
538 |
|
AVG2_MMX_RND0 |
539 |
|
lea TMP0, [TMP0+2*TMP1] |
540 |
|
AVG2_MMX_RND0 |
541 |
|
lea TMP0, [TMP0+2*TMP1] |
542 |
|
AVG2_MMX_RND0 |
543 |
|
|
544 |
|
pop _EBX |
545 |
|
ret |
546 |
|
|
547 |
|
.rounding1: |
548 |
|
mov eax, prm6d ; height -> _EAX |
549 |
|
sub _EAX, 8 |
550 |
|
|
551 |
|
mov TMP0, prm1 ; dst -> edi |
552 |
|
mov _EAX, prm2 ; src1 -> esi |
553 |
|
mov TMP1, prm4 ; stride -> TMP1 |
554 |
|
|
555 |
|
push _EBX |
556 |
|
%ifdef ARCH_IS_X86_64 |
557 |
|
mov _EBX, prm3 |
558 |
|
%else |
559 |
|
mov _EBX, [esp + 4 + 12] ; src2 -> eax |
560 |
|
%endif |
561 |
|
|
562 |
|
movq mm7, [mmx_one] |
563 |
|
|
564 |
|
jz near .start1 |
565 |
|
|
566 |
|
AVG2_MMX_RND1 |
567 |
|
lea TMP0, [TMP0+2*TMP1] |
568 |
|
|
569 |
|
.start1: |
570 |
|
|
571 |
|
AVG2_MMX_RND1 |
572 |
|
lea TMP0, [TMP0+2*TMP1] |
573 |
|
AVG2_MMX_RND1 |
574 |
|
lea TMP0, [TMP0+2*TMP1] |
575 |
|
AVG2_MMX_RND1 |
576 |
|
lea TMP0, [TMP0+2*TMP1] |
577 |
|
AVG2_MMX_RND1 |
578 |
|
|
579 |
|
pop _EBX |
580 |
|
ret |
581 |
|
ENDFUNC |
582 |
|
|
583 |
|
|
584 |
|
;----------------------------------------------------------------------------- |
585 |
|
; |
586 |
|
; void interpolate8x8_avg4_mmx(uint8_t const *dst, |
587 |
|
; const uint8_t * const src1, |
588 |
|
; const uint8_t * const src2, |
589 |
|
; const uint8_t * const src3, |
590 |
|
; const uint8_t * const src4, |
591 |
|
; const uint32_t stride, |
592 |
|
; const uint32_t rounding); |
593 |
|
; |
594 |
|
;----------------------------------------------------------------------------- |
595 |
|
|
596 |
|
%macro AVG4_MMX_RND0 0 |
597 |
|
movq mm0, [_EAX] ; src1 -> mm0 |
598 |
|
movq mm1, [_EBX] ; src2 -> mm1 |
599 |
|
|
600 |
|
movq mm2, mm0 |
601 |
|
movq mm3, mm1 |
602 |
|
|
603 |
|
pand mm2, [mmx_three] |
604 |
|
pand mm3, [mmx_three] |
605 |
|
|
606 |
|
pand mm0, [mmx_mask2] |
607 |
|
pand mm1, [mmx_mask2] |
608 |
|
|
609 |
|
psrlq mm0, 2 |
610 |
|
psrlq mm1, 2 |
611 |
|
|
612 |
|
lea _EAX, [_EAX+TMP1] |
613 |
|
lea _EBX, [_EBX+TMP1] |
614 |
|
|
615 |
|
paddb mm0, mm1 |
616 |
|
paddb mm2, mm3 |
617 |
|
|
618 |
|
movq mm4, [_ESI] ; src3 -> mm0 |
619 |
|
movq mm5, [_EDI] ; src4 -> mm1 |
620 |
|
|
621 |
|
movq mm1, mm4 |
622 |
|
movq mm3, mm5 |
623 |
|
|
624 |
|
pand mm1, [mmx_three] |
625 |
|
pand mm3, [mmx_three] |
626 |
|
|
627 |
|
pand mm4, [mmx_mask2] |
628 |
|
pand mm5, [mmx_mask2] |
629 |
|
|
630 |
|
psrlq mm4, 2 |
631 |
|
psrlq mm5, 2 |
632 |
|
|
633 |
|
paddb mm4, mm5 |
634 |
|
paddb mm0, mm4 |
635 |
|
|
636 |
|
paddb mm1, mm3 |
637 |
|
paddb mm2, mm1 |
638 |
|
|
639 |
|
paddb mm2, [mmx_two] |
640 |
|
pand mm2, [mmx_mask2] |
641 |
|
|
642 |
|
psrlq mm2, 2 |
643 |
|
paddb mm0, mm2 |
644 |
|
|
645 |
|
lea _ESI, [_ESI+TMP1] |
646 |
|
lea _EDI, [_EDI+TMP1] |
647 |
|
|
648 |
|
movq [TMP0], mm0 ; (src1 + src2 + src3 + src4 + 2) / 4 -> dst |
649 |
|
%endmacro |
650 |
|
|
651 |
|
%macro AVG4_MMX_RND1 0 |
652 |
|
movq mm0, [_EAX] ; src1 -> mm0 |
653 |
|
movq mm1, [_EBX] ; src2 -> mm1 |
654 |
|
|
655 |
|
movq mm2, mm0 |
656 |
|
movq mm3, mm1 |
657 |
|
|
658 |
|
pand mm2, [mmx_three] |
659 |
|
pand mm3, [mmx_three] |
660 |
|
|
661 |
|
pand mm0, [mmx_mask2] |
662 |
|
pand mm1, [mmx_mask2] |
663 |
|
|
664 |
|
psrlq mm0, 2 |
665 |
|
psrlq mm1, 2 |
666 |
|
|
667 |
|
lea _EAX,[_EAX+TMP1] |
668 |
|
lea _EBX,[_EBX+TMP1] |
669 |
|
|
670 |
|
paddb mm0, mm1 |
671 |
|
paddb mm2, mm3 |
672 |
|
|
673 |
|
movq mm4, [_ESI] ; src3 -> mm0 |
674 |
|
movq mm5, [_EDI] ; src4 -> mm1 |
675 |
|
|
676 |
|
movq mm1, mm4 |
677 |
|
movq mm3, mm5 |
678 |
|
|
679 |
|
pand mm1, [mmx_three] |
680 |
|
pand mm3, [mmx_three] |
681 |
|
|
682 |
|
pand mm4, [mmx_mask2] |
683 |
|
pand mm5, [mmx_mask2] |
684 |
|
|
685 |
|
psrlq mm4, 2 |
686 |
|
psrlq mm5, 2 |
687 |
|
|
688 |
|
paddb mm4, mm5 |
689 |
|
paddb mm0, mm4 |
690 |
|
|
691 |
|
paddb mm1, mm3 |
692 |
|
paddb mm2, mm1 |
693 |
|
|
694 |
|
paddb mm2, [mmx_one] |
695 |
|
pand mm2, [mmx_mask2] |
696 |
|
|
697 |
|
psrlq mm2, 2 |
698 |
|
paddb mm0, mm2 |
699 |
|
|
700 |
|
lea _ESI,[_ESI+TMP1] |
701 |
|
lea _EDI,[_EDI+TMP1] |
702 |
|
|
703 |
|
movq [TMP0], mm0 ; (src1 + src2 + src3 + src4 + 2) / 4 -> dst |
704 |
|
%endmacro |
705 |
|
|
706 |
|
ALIGN SECTION_ALIGN |
707 |
|
interpolate8x8_avg4_mmx: |
708 |
|
|
709 |
|
mov eax, prm7d ; rounding |
710 |
|
test _EAX, _EAX |
711 |
|
|
712 |
|
mov TMP0, prm1 ; dst -> edi |
713 |
|
mov _EAX, prm5 ; src4 -> edi |
714 |
|
mov TMP1d, prm6d ; stride -> TMP1 |
715 |
|
|
716 |
|
|
717 |
|
push _EBX |
718 |
|
push _EDI |
719 |
|
push _ESI |
720 |
|
|
721 |
|
mov _EDI, _EAX |
722 |
|
|
723 |
|
%ifdef ARCH_IS_X86_64 |
724 |
|
mov _EAX, prm2 |
725 |
|
mov _EBX, prm3 |
726 |
|
mov _ESI, prm4 |
727 |
|
%else |
728 |
|
mov _EAX, [esp + 12 + 8] ; src1 -> esi |
729 |
|
mov _EBX, [esp + 12 + 12] ; src2 -> _EAX |
730 |
|
mov _ESI, [esp + 12 + 16] ; src3 -> esi |
731 |
|
%endif |
732 |
|
|
733 |
|
movq mm7, [mmx_one] |
734 |
|
|
735 |
|
jnz near .rounding1 |
736 |
|
|
737 |
|
AVG4_MMX_RND0 |
738 |
|
lea TMP0, [TMP0+TMP1] |
739 |
|
AVG4_MMX_RND0 |
740 |
|
lea TMP0, [TMP0+TMP1] |
741 |
|
AVG4_MMX_RND0 |
742 |
|
lea TMP0, [TMP0+TMP1] |
743 |
|
AVG4_MMX_RND0 |
744 |
|
lea TMP0, [TMP0+TMP1] |
745 |
|
AVG4_MMX_RND0 |
746 |
|
lea TMP0, [TMP0+TMP1] |
747 |
|
AVG4_MMX_RND0 |
748 |
|
lea TMP0, [TMP0+TMP1] |
749 |
|
AVG4_MMX_RND0 |
750 |
|
lea TMP0, [TMP0+TMP1] |
751 |
|
AVG4_MMX_RND0 |
752 |
|
|
753 |
|
pop _ESI |
754 |
|
pop _EDI |
755 |
|
pop _EBX |
756 |
|
ret |
757 |
|
|
758 |
|
.rounding1: |
759 |
|
AVG4_MMX_RND1 |
760 |
|
lea TMP0, [TMP0+TMP1] |
761 |
|
AVG4_MMX_RND1 |
762 |
|
lea TMP0, [TMP0+TMP1] |
763 |
|
AVG4_MMX_RND1 |
764 |
|
lea TMP0, [TMP0+TMP1] |
765 |
|
AVG4_MMX_RND1 |
766 |
|
lea TMP0, [TMP0+TMP1] |
767 |
|
AVG4_MMX_RND1 |
768 |
|
lea TMP0, [TMP0+TMP1] |
769 |
|
AVG4_MMX_RND1 |
770 |
|
lea TMP0, [TMP0+TMP1] |
771 |
|
AVG4_MMX_RND1 |
772 |
|
lea TMP0, [TMP0+TMP1] |
773 |
|
AVG4_MMX_RND1 |
774 |
|
|
775 |
|
pop _ESI |
776 |
|
pop _EDI |
777 |
|
pop _EBX |
778 |
|
ret |
779 |
|
ENDFUNC |
780 |
|
|
781 |
|
|
782 |
|
;----------------------------------------------------------------------------- |
783 |
|
; |
784 |
|
; void interpolate8x8_6tap_lowpass_h_mmx(uint8_t const *dst, |
785 |
|
; const uint8_t * const src, |
786 |
|
; const uint32_t stride, |
787 |
|
; const uint32_t rounding); |
788 |
|
; |
789 |
|
;----------------------------------------------------------------------------- |
790 |
|
|
791 |
|
%macro LOWPASS_6TAP_H_MMX 0 |
792 |
|
movq mm0, [_EAX] |
793 |
|
movq mm2, [_EAX+1] |
794 |
|
|
795 |
|
movq mm1, mm0 |
796 |
|
movq mm3, mm2 |
797 |
|
|
798 |
|
punpcklbw mm0, mm7 |
799 |
|
punpcklbw mm2, mm7 |
800 |
|
|
801 |
|
punpckhbw mm1, mm7 |
802 |
|
punpckhbw mm3, mm7 |
803 |
|
|
804 |
|
paddw mm0, mm2 |
805 |
|
paddw mm1, mm3 |
806 |
|
|
807 |
|
psllw mm0, 2 |
808 |
|
psllw mm1, 2 |
809 |
|
|
810 |
|
movq mm2, [_EAX-1] |
811 |
|
movq mm4, [_EAX+2] |
812 |
|
|
813 |
|
movq mm3, mm2 |
814 |
|
movq mm5, mm4 |
815 |
|
|
816 |
|
punpcklbw mm2, mm7 |
817 |
|
punpcklbw mm4, mm7 |
818 |
|
|
819 |
|
punpckhbw mm3, mm7 |
820 |
|
punpckhbw mm5, mm7 |
821 |
|
|
822 |
|
paddw mm2, mm4 |
823 |
|
paddw mm3, mm5 |
824 |
|
|
825 |
|
psubsw mm0, mm2 |
826 |
|
psubsw mm1, mm3 |
827 |
|
|
828 |
|
pmullw mm0, [mmx_five] |
829 |
|
pmullw mm1, [mmx_five] |
830 |
|
|
831 |
|
movq mm2, [_EAX-2] |
832 |
|
movq mm4, [_EAX+3] |
833 |
|
|
834 |
|
movq mm3, mm2 |
835 |
|
movq mm5, mm4 |
836 |
|
|
837 |
|
punpcklbw mm2, mm7 |
838 |
|
punpcklbw mm4, mm7 |
839 |
|
|
840 |
|
punpckhbw mm3, mm7 |
841 |
|
punpckhbw mm5, mm7 |
842 |
|
|
843 |
|
paddw mm2, mm4 |
844 |
|
paddw mm3, mm5 |
845 |
|
|
846 |
|
paddsw mm0, mm2 |
847 |
|
paddsw mm1, mm3 |
848 |
|
|
849 |
|
paddsw mm0, mm6 |
850 |
|
paddsw mm1, mm6 |
851 |
|
|
852 |
|
psraw mm0, 5 |
853 |
|
psraw mm1, 5 |
854 |
|
|
855 |
|
lea _EAX, [_EAX+TMP1] |
856 |
|
packuswb mm0, mm1 |
857 |
|
movq [TMP0], mm0 |
858 |
|
%endmacro |
859 |
|
|
860 |
|
ALIGN SECTION_ALIGN |
861 |
|
interpolate8x8_6tap_lowpass_h_mmx: |
862 |
|
|
863 |
|
mov _EAX, prm4 ; rounding |
864 |
|
|
865 |
|
lea TMP0, [rounding_lowpass_mmx] |
866 |
|
movq mm6, [TMP0 + _EAX * 8] |
867 |
|
|
868 |
|
mov TMP0, prm1 ; dst -> edi |
869 |
|
mov _EAX, prm2 ; src -> esi |
870 |
|
mov TMP1, prm3 ; stride -> edx |
871 |
|
|
872 |
|
pxor mm7, mm7 |
873 |
|
|
874 |
|
LOWPASS_6TAP_H_MMX |
875 |
|
lea TMP0, [TMP0+TMP1] |
876 |
|
LOWPASS_6TAP_H_MMX |
877 |
|
lea TMP0, [TMP0+TMP1] |
878 |
|
LOWPASS_6TAP_H_MMX |
879 |
|
lea TMP0, [TMP0+TMP1] |
880 |
|
LOWPASS_6TAP_H_MMX |
881 |
|
lea TMP0, [TMP0+TMP1] |
882 |
|
LOWPASS_6TAP_H_MMX |
883 |
|
lea TMP0, [TMP0+TMP1] |
884 |
|
LOWPASS_6TAP_H_MMX |
885 |
|
lea TMP0, [TMP0+TMP1] |
886 |
|
LOWPASS_6TAP_H_MMX |
887 |
|
lea TMP0, [TMP0+TMP1] |
888 |
|
LOWPASS_6TAP_H_MMX |
889 |
|
|
890 |
|
ret |
891 |
|
ENDFUNC |
892 |
|
|
893 |
|
;----------------------------------------------------------------------------- |
894 |
|
; |
895 |
|
; void interpolate8x8_6tap_lowpass_v_mmx(uint8_t const *dst, |
896 |
|
; const uint8_t * const src, |
897 |
|
; const uint32_t stride, |
898 |
|
; const uint32_t rounding); |
899 |
|
; |
900 |
|
;----------------------------------------------------------------------------- |
901 |
|
|
902 |
|
%macro LOWPASS_6TAP_V_MMX 0 |
903 |
|
movq mm0, [_EAX] |
904 |
|
movq mm2, [_EAX+TMP1] |
905 |
|
|
906 |
|
movq mm1, mm0 |
907 |
|
movq mm3, mm2 |
908 |
|
|
909 |
|
punpcklbw mm0, mm7 |
910 |
|
punpcklbw mm2, mm7 |
911 |
|
|
912 |
|
punpckhbw mm1, mm7 |
913 |
|
punpckhbw mm3, mm7 |
914 |
|
|
915 |
|
paddw mm0, mm2 |
916 |
|
paddw mm1, mm3 |
917 |
|
|
918 |
|
psllw mm0, 2 |
919 |
|
psllw mm1, 2 |
920 |
|
|
921 |
|
movq mm4, [_EAX+2*TMP1] |
922 |
|
sub _EAX, _EBX |
923 |
|
movq mm2, [_EAX+2*TMP1] |
924 |
|
|
925 |
|
movq mm3, mm2 |
926 |
|
movq mm5, mm4 |
927 |
|
|
928 |
|
punpcklbw mm2, mm7 |
929 |
|
punpcklbw mm4, mm7 |
930 |
|
|
931 |
|
punpckhbw mm3, mm7 |
932 |
|
punpckhbw mm5, mm7 |
933 |
|
|
934 |
|
paddw mm2, mm4 |
935 |
|
paddw mm3, mm5 |
936 |
|
|
937 |
|
psubsw mm0, mm2 |
938 |
|
psubsw mm1, mm3 |
939 |
|
|
940 |
|
pmullw mm0, [mmx_five] |
941 |
|
pmullw mm1, [mmx_five] |
942 |
|
|
943 |
|
movq mm2, [_EAX+TMP1] |
944 |
|
movq mm4, [_EAX+2*_EBX] |
945 |
|
|
946 |
|
movq mm3, mm2 |
947 |
|
movq mm5, mm4 |
948 |
|
|
949 |
|
punpcklbw mm2, mm7 |
950 |
|
punpcklbw mm4, mm7 |
951 |
|
|
952 |
|
punpckhbw mm3, mm7 |
953 |
|
punpckhbw mm5, mm7 |
954 |
|
|
955 |
|
paddw mm2, mm4 |
956 |
|
paddw mm3, mm5 |
957 |
|
|
958 |
|
paddsw mm0, mm2 |
959 |
|
paddsw mm1, mm3 |
960 |
|
|
961 |
|
paddsw mm0, mm6 |
962 |
|
paddsw mm1, mm6 |
963 |
|
|
964 |
|
psraw mm0, 5 |
965 |
|
psraw mm1, 5 |
966 |
|
|
967 |
|
lea _EAX, [_EAX+4*TMP1] |
968 |
|
packuswb mm0, mm1 |
969 |
|
movq [TMP0], mm0 |
970 |
|
%endmacro |
971 |
|
|
972 |
|
ALIGN SECTION_ALIGN |
973 |
|
interpolate8x8_6tap_lowpass_v_mmx: |
974 |
|
|
975 |
|
mov _EAX, prm4 ; rounding |
976 |
|
|
977 |
|
lea TMP0, [rounding_lowpass_mmx] |
978 |
|
movq mm6, [TMP0 + _EAX * 8] |
979 |
|
|
980 |
|
mov TMP0, prm1 ; dst -> edi |
981 |
|
mov _EAX, prm2 ; src -> esi |
982 |
|
mov TMP1, prm3 ; stride -> edx |
983 |
|
|
984 |
|
push _EBX |
985 |
|
lea _EBX, [TMP1+TMP1*2] |
986 |
|
|
987 |
|
pxor mm7, mm7 |
988 |
|
|
989 |
|
LOWPASS_6TAP_V_MMX |
990 |
|
lea TMP0, [TMP0+TMP1] |
991 |
|
LOWPASS_6TAP_V_MMX |
992 |
|
lea TMP0, [TMP0+TMP1] |
993 |
|
LOWPASS_6TAP_V_MMX |
994 |
|
lea TMP0, [TMP0+TMP1] |
995 |
|
LOWPASS_6TAP_V_MMX |
996 |
|
lea TMP0, [TMP0+TMP1] |
997 |
|
LOWPASS_6TAP_V_MMX |
998 |
|
lea TMP0, [TMP0+TMP1] |
999 |
|
LOWPASS_6TAP_V_MMX |
1000 |
|
lea TMP0, [TMP0+TMP1] |
1001 |
|
LOWPASS_6TAP_V_MMX |
1002 |
|
lea TMP0, [TMP0+TMP1] |
1003 |
|
LOWPASS_6TAP_V_MMX |
1004 |
|
|
1005 |
|
pop _EBX |
1006 |
|
ret |
1007 |
|
ENDFUNC |
1008 |
|
|
1009 |
|
;=========================================================================== |
1010 |
|
; |
1011 |
|
; The next functions combine both source halfpel interpolation step and the |
1012 |
|
; averaging (with rouding) step to avoid wasting memory bandwidth computing |
1013 |
|
; intermediate halfpel images and then averaging them. |
1014 |
|
; |
1015 |
|
;=========================================================================== |
1016 |
|
|
1017 |
|
%macro PROLOG0 0 |
1018 |
|
mov TMP0, prm1 ; Dst |
1019 |
|
mov _EAX, prm2 ; Src |
1020 |
|
mov TMP1, prm3 ; BpS |
1021 |
|
%endmacro |
1022 |
|
|
1023 |
|
%macro PROLOG 2 ; %1: Rounder, %2 load Dst-Rounder |
1024 |
|
pxor mm6, mm6 |
1025 |
|
movq mm7, [%1] ; TODO: dangerous! (eax isn't checked) |
1026 |
|
%if %2 |
1027 |
|
movq mm5, [rounding1_mmx] |
1028 |
|
%endif |
1029 |
|
|
1030 |
|
PROLOG0 |
1031 |
|
%endmacro |
1032 |
|
|
1033 |
|
; performs: mm0 == (mm0+mm2) mm1 == (mm1+mm3) |
1034 |
|
%macro MIX 0 |
1035 |
|
punpcklbw mm0, mm6 |
1036 |
|
punpcklbw mm2, mm6 |
1037 |
|
punpckhbw mm1, mm6 |
1038 |
|
punpckhbw mm3, mm6 |
1039 |
|
paddusw mm0, mm2 |
1040 |
|
paddusw mm1, mm3 |
1041 |
|
%endmacro |
1042 |
|
|
1043 |
|
%macro MIX_DST 0 |
1044 |
|
movq mm3, mm2 |
1045 |
|
paddusw mm0, mm7 ; rounder |
1046 |
|
paddusw mm1, mm7 ; rounder |
1047 |
|
punpcklbw mm2, mm6 |
1048 |
|
punpckhbw mm3, mm6 |
1049 |
|
psrlw mm0, 1 |
1050 |
|
psrlw mm1, 1 |
1051 |
|
|
1052 |
|
paddusw mm0, mm2 ; mix Src(mm0/mm1) with Dst(mm2/mm3) |
1053 |
|
paddusw mm1, mm3 |
1054 |
|
paddusw mm0, mm5 |
1055 |
|
paddusw mm1, mm5 |
1056 |
|
psrlw mm0, 1 |
1057 |
|
psrlw mm1, 1 |
1058 |
|
|
1059 |
|
packuswb mm0, mm1 |
1060 |
|
%endmacro |
1061 |
|
|
1062 |
|
%macro MIX2 0 |
1063 |
|
punpcklbw mm0, mm6 |
1064 |
|
punpcklbw mm2, mm6 |
1065 |
|
paddusw mm0, mm2 |
1066 |
|
paddusw mm0, mm7 |
1067 |
|
punpckhbw mm1, mm6 |
1068 |
|
punpckhbw mm3, mm6 |
1069 |
|
paddusw mm1, mm7 |
1070 |
|
paddusw mm1, mm3 |
1071 |
|
psrlw mm0, 1 |
1072 |
|
psrlw mm1, 1 |
1073 |
|
|
1074 |
|
packuswb mm0, mm1 |
1075 |
|
%endmacro |
1076 |
|
|
1077 |
|
;=========================================================================== |
1078 |
|
; |
1079 |
|
; void interpolate8x8_halfpel_add_mmx(uint8_t * const dst, |
1080 |
|
; const uint8_t * const src, |
1081 |
|
; const uint32_t stride, |
1082 |
|
; const uint32_t rounding); |
1083 |
|
; |
1084 |
|
; |
1085 |
|
;=========================================================================== |
1086 |
|
|
1087 |
|
%macro ADD_FF_MMX 1 |
1088 |
|
movq mm0, [_EAX] |
1089 |
|
movq mm2, [TMP0] |
1090 |
|
movq mm1, mm0 |
1091 |
|
movq mm3, mm2 |
1092 |
|
%if (%1!=0) |
1093 |
|
lea _EAX,[_EAX+%1*TMP1] |
1094 |
|
%endif |
1095 |
|
MIX |
1096 |
|
paddusw mm0, mm5 ; rounder |
1097 |
|
paddusw mm1, mm5 ; rounder |
1098 |
|
psrlw mm0, 1 |
1099 |
|
psrlw mm1, 1 |
1100 |
|
|
1101 |
|
packuswb mm0, mm1 |
1102 |
|
movq [TMP0], mm0 |
1103 |
|
%if (%1!=0) |
1104 |
|
lea TMP0,[TMP0+%1*TMP1] |
1105 |
|
%endif |
1106 |
|
%endmacro |
1107 |
|
|
1108 |
|
ALIGN SECTION_ALIGN |
1109 |
|
interpolate8x8_halfpel_add_mmx: |
1110 |
|
PROLOG rounding1_mmx, 1 |
1111 |
|
ADD_FF_MMX 1 |
1112 |
|
ADD_FF_MMX 1 |
1113 |
|
ADD_FF_MMX 1 |
1114 |
|
ADD_FF_MMX 1 |
1115 |
|
ADD_FF_MMX 1 |
1116 |
|
ADD_FF_MMX 1 |
1117 |
|
ADD_FF_MMX 1 |
1118 |
|
ADD_FF_MMX 0 |
1119 |
|
ret |
1120 |
|
ENDFUNC |
1121 |
|
|
1122 |
|
;=========================================================================== |
1123 |
|
; |
1124 |
|
; void interpolate8x8_halfpel_h_add_mmx(uint8_t * const dst, |
1125 |
|
; const uint8_t * const src, |
1126 |
|
; const uint32_t stride, |
1127 |
|
; const uint32_t rounding); |
1128 |
|
; |
1129 |
|
; |
1130 |
|
;=========================================================================== |
1131 |
|
|
1132 |
|
%macro ADD_FH_MMX 0 |
1133 |
|
movq mm0, [_EAX] |
1134 |
|
movq mm2, [_EAX+1] |
1135 |
|
movq mm1, mm0 |
1136 |
|
movq mm3, mm2 |
1137 |
|
|
1138 |
|
lea _EAX,[_EAX+TMP1] |
1139 |
|
|
1140 |
|
MIX |
1141 |
|
movq mm2, [TMP0] ; prepare mix with Dst[0] |
1142 |
|
MIX_DST |
1143 |
|
movq [TMP0], mm0 |
1144 |
|
%endmacro |
1145 |
|
|
1146 |
|
ALIGN SECTION_ALIGN |
1147 |
|
interpolate8x8_halfpel_h_add_mmx: |
1148 |
|
PROLOG rounding1_mmx, 1 |
1149 |
|
|
1150 |
|
ADD_FH_MMX |
1151 |
|
lea TMP0,[TMP0+TMP1] |
1152 |
|
ADD_FH_MMX |
1153 |
|
lea TMP0,[TMP0+TMP1] |
1154 |
|
ADD_FH_MMX |
1155 |
|
lea TMP0,[TMP0+TMP1] |
1156 |
|
ADD_FH_MMX |
1157 |
|
lea TMP0,[TMP0+TMP1] |
1158 |
|
ADD_FH_MMX |
1159 |
|
lea TMP0,[TMP0+TMP1] |
1160 |
|
ADD_FH_MMX |
1161 |
|
lea TMP0,[TMP0+TMP1] |
1162 |
|
ADD_FH_MMX |
1163 |
|
lea TMP0,[TMP0+TMP1] |
1164 |
|
ADD_FH_MMX |
1165 |
|
ret |
1166 |
|
ENDFUNC |
1167 |
|
|
1168 |
|
;=========================================================================== |
1169 |
|
; |
1170 |
|
; void interpolate8x8_halfpel_v_add_mmx(uint8_t * const dst, |
1171 |
|
; const uint8_t * const src, |
1172 |
|
; const uint32_t stride, |
1173 |
|
; const uint32_t rounding); |
1174 |
|
; |
1175 |
|
; |
1176 |
|
;=========================================================================== |
1177 |
|
|
1178 |
|
%macro ADD_HF_MMX 0 |
1179 |
|
movq mm0, [_EAX] |
1180 |
|
movq mm2, [_EAX+TMP1] |
1181 |
|
movq mm1, mm0 |
1182 |
|
movq mm3, mm2 |
1183 |
|
|
1184 |
|
lea _EAX,[_EAX+TMP1] |
1185 |
|
|
1186 |
|
MIX |
1187 |
|
movq mm2, [TMP0] ; prepare mix with Dst[0] |
1188 |
|
MIX_DST |
1189 |
|
movq [TMP0], mm0 |
1190 |
|
|
1191 |
|
%endmacro |
1192 |
|
|
1193 |
|
ALIGN SECTION_ALIGN |
1194 |
|
interpolate8x8_halfpel_v_add_mmx: |
1195 |
|
PROLOG rounding1_mmx, 1 |
1196 |
|
|
1197 |
|
ADD_HF_MMX |
1198 |
|
lea TMP0,[TMP0+TMP1] |
1199 |
|
ADD_HF_MMX |
1200 |
|
lea TMP0,[TMP0+TMP1] |
1201 |
|
ADD_HF_MMX |
1202 |
|
lea TMP0,[TMP0+TMP1] |
1203 |
|
ADD_HF_MMX |
1204 |
|
lea TMP0,[TMP0+TMP1] |
1205 |
|
ADD_HF_MMX |
1206 |
|
lea TMP0,[TMP0+TMP1] |
1207 |
|
ADD_HF_MMX |
1208 |
|
lea TMP0,[TMP0+TMP1] |
1209 |
|
ADD_HF_MMX |
1210 |
|
lea TMP0,[TMP0+TMP1] |
1211 |
|
ADD_HF_MMX |
1212 |
|
ret |
1213 |
|
ENDFUNC |
1214 |
|
|
1215 |
|
; The trick is to correct the result of 'pavgb' with some combination of the |
1216 |
|
; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgb' (s and t). |
1217 |
|
; The boolean relations are: |
1218 |
|
; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st |
1219 |
|
; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st |
1220 |
|
; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st |
1221 |
|
; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st |
1222 |
|
; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. |
1223 |
|
|
1224 |
|
; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). |
1225 |
|
|
1226 |
|
;=========================================================================== |
1227 |
|
; |
1228 |
|
; void interpolate8x8_halfpel_hv_add_mmx(uint8_t * const dst, |
1229 |
|
; const uint8_t * const src, |
1230 |
|
; const uint32_t stride, |
1231 |
|
; const uint32_t rounding); |
1232 |
|
; |
1233 |
|
; |
1234 |
|
;=========================================================================== |
1235 |
|
|
1236 |
|
%macro ADD_HH_MMX 0 |
1237 |
|
lea _EAX,[_EAX+TMP1] |
1238 |
|
|
1239 |
|
; transfert prev line to mm0/mm1 |
1240 |
|
movq mm0, mm2 |
1241 |
|
movq mm1, mm3 |
1242 |
|
|
1243 |
|
; load new line in mm2/mm3 |
1244 |
|
movq mm2, [_EAX] |
1245 |
|
movq mm4, [_EAX+1] |
1246 |
|
movq mm3, mm2 |
1247 |
|
movq mm5, mm4 |
1248 |
|
|
1249 |
|
punpcklbw mm2, mm6 |
1250 |
|
punpcklbw mm4, mm6 |
1251 |
|
paddusw mm2, mm4 |
1252 |
|
punpckhbw mm3, mm6 |
1253 |
|
punpckhbw mm5, mm6 |
1254 |
|
paddusw mm3, mm5 |
1255 |
|
|
1256 |
|
; mix current line (mm2/mm3) with previous (mm0,mm1); |
1257 |
|
; we'll preserve mm2/mm3 for next line... |
1258 |
|
|
1259 |
|
paddusw mm0, mm2 |
1260 |
|
paddusw mm1, mm3 |
1261 |
|
|
1262 |
|
movq mm4, [TMP0] ; prepare mix with Dst[0] |
1263 |
|
movq mm5, mm4 |
1264 |
|
|
1265 |
|
paddusw mm0, mm7 ; finish mixing current line |
1266 |
|
paddusw mm1, mm7 |
1267 |
|
|
1268 |
|
punpcklbw mm4, mm6 |
1269 |
|
punpckhbw mm5, mm6 |
1270 |
|
|
1271 |
|
psrlw mm0, 2 |
1272 |
|
psrlw mm1, 2 |
1273 |
|
|
1274 |
|
paddusw mm0, mm4 ; mix Src(mm0/mm1) with Dst(mm2/mm3) |
1275 |
|
paddusw mm1, mm5 |
1276 |
|
|
1277 |
|
paddusw mm0, [rounding1_mmx] |
1278 |
|
paddusw mm1, [rounding1_mmx] |
1279 |
|
|
1280 |
|
psrlw mm0, 1 |
1281 |
|
psrlw mm1, 1 |
1282 |
|
|
1283 |
|
packuswb mm0, mm1 |
1284 |
|
|
1285 |
|
movq [TMP0], mm0 |
1286 |
|
%endmacro |
1287 |
|
|
1288 |
|
ALIGN SECTION_ALIGN |
1289 |
|
interpolate8x8_halfpel_hv_add_mmx: |
1290 |
|
PROLOG rounding2_mmx, 0 ; mm5 is busy. Don't load dst-rounder |
1291 |
|
|
1292 |
|
; preprocess first line |
1293 |
|
movq mm0, [_EAX] |
1294 |
|
movq mm2, [_EAX+1] |
1295 |
|
movq mm1, mm0 |
1296 |
|
movq mm3, mm2 |
1297 |
|
|
1298 |
|
punpcklbw mm0, mm6 |
1299 |
|
punpcklbw mm2, mm6 |
1300 |
|
punpckhbw mm1, mm6 |
1301 |
|
punpckhbw mm3, mm6 |
1302 |
|
paddusw mm2, mm0 |
1303 |
|
paddusw mm3, mm1 |
1304 |
|
|
1305 |
|
; Input: mm2/mm3 contains the value (Src[0]+Src[1]) of previous line |
1306 |
|
|
1307 |
|
ADD_HH_MMX |
1308 |
|
lea TMP0,[TMP0+TMP1] |
1309 |
|
ADD_HH_MMX |
1310 |
|
lea TMP0,[TMP0+TMP1] |
1311 |
|
ADD_HH_MMX |
1312 |
|
lea TMP0,[TMP0+TMP1] |
1313 |
|
ADD_HH_MMX |
1314 |
|
lea TMP0,[TMP0+TMP1] |
1315 |
|
ADD_HH_MMX |
1316 |
|
lea TMP0,[TMP0+TMP1] |
1317 |
|
ADD_HH_MMX |
1318 |
|
lea TMP0,[TMP0+TMP1] |
1319 |
|
ADD_HH_MMX |
1320 |
|
lea TMP0,[TMP0+TMP1] |
1321 |
|
ADD_HH_MMX |
1322 |
|
|
1323 |
|
ret |
1324 |
|
ENDFUNC |
1325 |
|
|
1326 |
|
|
1327 |
|
%ifidn __OUTPUT_FORMAT__,elf |
1328 |
|
section ".note.GNU-stack" noalloc noexec nowrite progbits |
1329 |
|
%endif |
1330 |
|
|