5 |
; * |
; * |
6 |
; * Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
; * Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
7 |
; * 2002-2003 Pascal Massimino <skal@planet-d.net> |
; * 2002-2003 Pascal Massimino <skal@planet-d.net> |
8 |
|
; * 2004 Jean-Marc Bastide <jmtest@voila.fr> |
9 |
; * |
; * |
10 |
; * This program is free software ; you can redistribute it and/or modify |
; * This program is free software ; you can redistribute it and/or modify |
11 |
; * it under the terms of the GNU General Public License as published by |
; * it under the terms of the GNU General Public License as published by |
21 |
; * along with this program ; if not, write to the Free Software |
; * along with this program ; if not, write to the Free Software |
22 |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
; * |
; * |
24 |
; * $Id: quantize_h263_mmx.asm,v 1.2 2004-03-22 22:36:24 edgomez Exp $ |
; * $Id: quantize_h263_mmx.asm,v 1.9 2008-11-11 20:46:24 Isibaar Exp $ |
25 |
; * |
; * |
26 |
; ****************************************************************************/ |
; ****************************************************************************/ |
27 |
|
|
32 |
|
|
33 |
%macro cglobal 1 |
%macro cglobal 1 |
34 |
%ifdef PREFIX |
%ifdef PREFIX |
35 |
|
%ifdef MARK_FUNCS |
36 |
|
global _%1:function %1.endfunc-%1 |
37 |
|
%define %1 _%1:function %1.endfunc-%1 |
38 |
|
%define ENDFUNC .endfunc |
39 |
|
%else |
40 |
global _%1 |
global _%1 |
41 |
%define %1 _%1 |
%define %1 _%1 |
42 |
|
%define ENDFUNC |
43 |
|
%endif |
44 |
|
%else |
45 |
|
%ifdef MARK_FUNCS |
46 |
|
global %1:function %1.endfunc-%1 |
47 |
|
%define ENDFUNC .endfunc |
48 |
%else |
%else |
49 |
global %1 |
global %1 |
50 |
|
%define ENDFUNC |
51 |
|
%endif |
52 |
%endif |
%endif |
53 |
%endmacro |
%endmacro |
54 |
|
|
57 |
;============================================================================= |
;============================================================================= |
58 |
|
|
59 |
%ifdef FORMAT_COFF |
%ifdef FORMAT_COFF |
60 |
SECTION .rodata data |
SECTION .rodata |
61 |
%else |
%else |
62 |
SECTION .rodata data align=16 |
SECTION .rodata align=16 |
63 |
%endif |
%endif |
64 |
|
|
65 |
ALIGN 16 |
ALIGN 16 |
68 |
|
|
69 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
70 |
; |
; |
71 |
; subtract by Q/2 table |
; quant table |
72 |
; |
; |
73 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
74 |
|
|
75 |
ALIGN 16 |
ALIGN 16 |
76 |
mmx_sub: |
mmx_quant: |
77 |
%assign quant 1 |
%assign quant 0 |
78 |
%rep 31 |
%rep 32 |
79 |
times 4 dw quant / 2 |
times 4 dw quant |
80 |
%assign quant quant+1 |
%assign quant quant+1 |
81 |
%endrep |
%endrep |
82 |
|
|
83 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
84 |
; |
; |
85 |
; divide by 2Q table |
; subtract by Q/2 table |
|
; |
|
|
; use a shift of 16 to take full advantage of _pmulhw_ |
|
|
; for q=1, _pmulhw_ will overflow so it is treated seperately |
|
|
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
|
86 |
; |
; |
87 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
88 |
|
|
89 |
ALIGN 16 |
ALIGN 16 |
90 |
mmx_div: |
mmx_sub: |
91 |
%assign quant 1 |
%assign quant 1 |
92 |
%rep 31 |
%rep 31 |
93 |
times 4 dw (1<<16) / (quant*2) + 1 |
times 4 dw quant / 2 |
94 |
%assign quant quant+1 |
%assign quant quant+1 |
95 |
%endrep |
%endrep |
96 |
|
|
97 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
98 |
; |
; |
99 |
; add by (odd(Q) ? Q : Q - 1) table |
; divide by 2Q table |
|
; |
|
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
ALIGN 16 |
|
|
mmx_add: |
|
|
%assign quant 1 |
|
|
%rep 31 |
|
|
%if quant % 2 != 0 |
|
|
times 4 dw quant |
|
|
%else |
|
|
times 4 dw quant - 1 |
|
|
%endif |
|
|
%assign quant quant+1 |
|
|
%endrep |
|
|
|
|
|
;----------------------------------------------------------------------------- |
|
100 |
; |
; |
101 |
; multiple by 2Q table |
; use a shift of 16 to take full advantage of _pmulhw_ |
102 |
|
; for q=1, _pmulhw_ will overflow so it is treated seperately |
103 |
|
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
104 |
; |
; |
105 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
106 |
|
|
107 |
ALIGN 16 |
ALIGN 16 |
108 |
mmx_mul: |
mmx_div: |
109 |
%assign quant 1 |
%assign quant 1 |
110 |
%rep 31 |
%rep 31 |
111 |
times 4 dw quant*2 |
times 4 dw (1<<16) / (quant*2) + 1 |
112 |
%assign quant quant+1 |
%assign quant quant+1 |
113 |
%endrep |
%endrep |
114 |
|
|
|
;----------------------------------------------------------------------------- |
|
|
; |
|
|
; saturation limits |
|
|
; |
|
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
ALIGN 16 |
|
|
sse2_2047: |
|
|
times 8 dw 2047 |
|
|
|
|
|
ALIGN 16 |
|
|
mmx_2047: |
|
|
times 4 dw 2047 |
|
|
|
|
|
ALIGN 8 |
|
|
mmx_32768_minus_2048: |
|
|
times 4 dw (32768-2048) |
|
|
|
|
|
mmx_32767_minus_2047: |
|
|
times 4 dw (32767-2047) |
|
|
|
|
|
|
|
115 |
;============================================================================= |
;============================================================================= |
116 |
; Code |
; Code |
117 |
;============================================================================= |
;============================================================================= |
142 |
ALIGN 16 |
ALIGN 16 |
143 |
quant_h263_intra_mmx: |
quant_h263_intra_mmx: |
144 |
|
|
|
push ecx |
|
145 |
push esi |
push esi |
|
push edi |
|
|
|
|
|
mov edi, [esp + 12 + 4] ; coeff |
|
|
mov esi, [esp + 12 + 8] ; data |
|
|
mov eax, [esp + 12 + 12] ; quant |
|
|
|
|
|
xor ecx, ecx |
|
|
cmp al, 1 |
|
|
jz .q1loop |
|
|
|
|
|
movq mm7, [mmx_div + eax * 8 - 8] |
|
|
|
|
|
ALIGN 16 |
|
|
.loop |
|
|
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
|
|
movq mm3, [esi + 8*ecx + 8] |
|
|
pxor mm1, mm1 ; mm1 = 0 |
|
|
pxor mm4, mm4 ; |
|
|
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
|
|
pcmpgtw mm4, mm3 ; |
|
|
pxor mm0, mm1 ; mm0 = |mm0| |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; displace |
|
|
psubw mm3, mm4 ; |
|
|
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
|
|
pmulhw mm3, mm7 ; |
|
|
pxor mm0, mm1 ; mm0 *= sign(mm0) |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; undisplace |
|
|
psubw mm3, mm4 ; |
|
|
movq [edi + 8*ecx], mm0 |
|
|
movq [edi + 8*ecx + 8], mm3 |
|
|
|
|
|
add ecx, 2 |
|
|
cmp ecx, 16 |
|
|
jnz .loop |
|
146 |
|
|
147 |
.done |
mov esi, [esp + 4 + 8] ; data |
148 |
|
mov ecx,[esp + 4 + 16] ; dcscalar |
|
; caclulate data[0] // (int32_t)dcscalar) |
|
|
mov ecx, [esp + 12 + 16] ; dcscalar |
|
|
mov edx, ecx |
|
149 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
|
shr edx, 1 ; edx = dcscalar /2 |
|
|
cmp eax, 0 |
|
|
jg .gtzero |
|
150 |
|
|
151 |
|
sar ecx,1 ; dcscalar /2 |
152 |
|
mov edx,eax |
153 |
|
sar edx,31 ; sgn(data[0]) |
154 |
|
xor ecx,edx ; *sgn(data[0]) |
155 |
sub eax, edx |
sub eax, edx |
156 |
jmp short .mul |
add eax,ecx ; + (dcscalar/2)*sgn(data[0]) |
157 |
|
|
158 |
.gtzero |
mov ecx, [esp + 4 + 12] ; quant |
159 |
add eax, edx |
cdq |
160 |
.mul |
idiv dword [esp + 4 + 16] ; dcscalar |
161 |
cdq ; expand eax -> edx:eax |
cmp ecx, 1 |
162 |
idiv ecx ; eax = edx:eax / dcscalar |
mov edx, [esp + 4 + 4] ; coeff |
163 |
mov [edi], ax ; coeff[0] = ax |
je .low |
164 |
|
|
165 |
xor eax, eax ; return(0); |
movq mm7, [mmx_div+ecx * 8 - 8] |
166 |
pop edi |
mov ecx,4 |
|
pop esi |
|
|
pop ecx |
|
167 |
|
|
168 |
ret |
.loop: |
169 |
|
movq mm0, [esi] ; data |
170 |
ALIGN 16 |
pxor mm4,mm4 |
171 |
.q1loop |
movq mm1, [esi + 8] |
172 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
pcmpgtw mm4,mm0 ; (data<0) |
173 |
movq mm3, [esi + 8*ecx + 8] |
pxor mm5,mm5 |
174 |
pxor mm1, mm1 ; mm1 = 0 |
pmulhw mm0,mm7 ; /(2*quant) |
175 |
pxor mm4, mm4 ; |
pcmpgtw mm5,mm1 |
176 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
movq mm2, [esi+16] |
177 |
pcmpgtw mm4, mm3 ; |
psubw mm0,mm4 ; +(data<0) |
178 |
pxor mm0, mm1 ; mm0 = |mm0| |
pmulhw mm1,mm7 |
179 |
pxor mm3, mm4 ; |
pxor mm4,mm4 |
180 |
psubw mm0, mm1 ; displace |
movq mm3,[esi+24] |
181 |
psubw mm3, mm4 ; |
pcmpgtw mm4,mm2 |
182 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psubw mm1,mm5 |
183 |
psrlw mm3, 1 ; |
pmulhw mm2,mm7 |
184 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm5,mm5 |
185 |
pxor mm3, mm4 |
pcmpgtw mm5,mm3 |
186 |
psubw mm0, mm1 ; undisplace |
pmulhw mm3,mm7 |
187 |
psubw mm3, mm4 ; |
psubw mm2,mm4 |
188 |
movq [edi + 8*ecx], mm0 |
psubw mm3,mm5 |
189 |
movq [edi + 8*ecx + 8], mm3 |
movq [edx], mm0 |
190 |
|
lea esi, [esi+32] |
191 |
|
movq [edx + 8], mm1 |
192 |
|
movq [edx + 16], mm2 |
193 |
|
movq [edx + 24], mm3 |
194 |
|
|
195 |
|
dec ecx |
196 |
|
lea edx, [edx+32] |
197 |
|
jne .loop |
198 |
|
jmp .end |
199 |
|
|
200 |
|
.low: |
201 |
|
movd mm7,ecx |
202 |
|
mov ecx,4 |
203 |
|
.loop_low: |
204 |
|
movq mm0, [esi] |
205 |
|
pxor mm4,mm4 |
206 |
|
movq mm1, [esi + 8] |
207 |
|
pcmpgtw mm4,mm0 |
208 |
|
pxor mm5,mm5 |
209 |
|
psubw mm0,mm4 |
210 |
|
pcmpgtw mm5,mm1 |
211 |
|
psraw mm0,mm7 |
212 |
|
psubw mm1,mm5 |
213 |
|
movq mm2,[esi+16] |
214 |
|
pxor mm4,mm4 |
215 |
|
psraw mm1,mm7 |
216 |
|
pcmpgtw mm4,mm2 |
217 |
|
pxor mm5,mm5 |
218 |
|
psubw mm2,mm4 |
219 |
|
movq mm3,[esi+24] |
220 |
|
pcmpgtw mm5,mm3 |
221 |
|
psraw mm2,mm7 |
222 |
|
psubw mm3,mm5 |
223 |
|
movq [edx], mm0 |
224 |
|
psraw mm3,mm7 |
225 |
|
movq [edx + 8], mm1 |
226 |
|
movq [edx+16],mm2 |
227 |
|
lea esi, [esi+32] |
228 |
|
movq [edx+24],mm3 |
229 |
|
|
230 |
|
dec ecx |
231 |
|
lea edx, [edx+32] |
232 |
|
jne .loop_low |
233 |
|
|
234 |
add ecx, 2 |
.end: |
235 |
cmp ecx, 16 |
mov edx, [esp + 4 + 4] ; coeff |
236 |
jnz .q1loop |
mov [edx],ax |
237 |
jmp short .done |
xor eax,eax ; return 0 |
238 |
|
|
239 |
|
pop esi |
240 |
|
ret |
241 |
|
ENDFUNC |
242 |
|
|
243 |
|
|
244 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
255 |
quant_h263_intra_sse2: |
quant_h263_intra_sse2: |
256 |
|
|
257 |
push esi |
push esi |
|
push edi |
|
258 |
|
|
259 |
mov edi, [esp + 8 + 4] ; coeff |
mov esi, [esp + 4 + 8] ; data |
|
mov esi, [esp + 8 + 8] ; data |
|
|
mov eax, [esp + 8 + 12] ; quant |
|
260 |
|
|
261 |
xor ecx, ecx |
movsx eax, word [esi] ; data[0] |
|
cmp al, 1 |
|
|
jz near .qas2_q1loop |
|
262 |
|
|
263 |
.qas2_not1 |
mov ecx,[esp + 4 + 16] ; dcscalar |
264 |
movq mm7, [mmx_div + eax*8 - 8] |
mov edx,eax |
265 |
movq2dq xmm7, mm7 |
sar ecx,1 |
266 |
|
add eax,ecx |
267 |
|
sub edx,ecx |
268 |
|
cmovl eax,edx ; +/- dcscalar/2 |
269 |
|
mov ecx, [esp + 4 + 12] ; quant |
270 |
|
cdq |
271 |
|
idiv dword [esp + 4 + 16] ; dcscalar |
272 |
|
cmp ecx, 1 |
273 |
|
mov edx, [esp + 4 + 4] ; coeff |
274 |
|
movq xmm7, [mmx_div+ecx * 8 - 8] |
275 |
|
je .low |
276 |
|
|
277 |
|
mov ecx,2 |
278 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
279 |
|
|
280 |
ALIGN 16 |
.loop: |
281 |
.qas2_loop |
movdqa xmm0, [esi] |
|
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
|
|
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
|
|
pxor xmm1, xmm1 |
|
282 |
pxor xmm4, xmm4 |
pxor xmm4, xmm4 |
283 |
pcmpgtw xmm1, xmm0 |
movdqa xmm1, [esi + 16] |
284 |
pcmpgtw xmm4, xmm3 |
pcmpgtw xmm4,xmm0 |
285 |
pxor xmm0, xmm1 |
pxor xmm5,xmm5 |
|
pxor xmm3, xmm4 |
|
|
psubw xmm0, xmm1 |
|
|
psubw xmm3, xmm4 |
|
286 |
pmulhw xmm0, xmm7 |
pmulhw xmm0, xmm7 |
287 |
|
pcmpgtw xmm5,xmm1 |
288 |
|
movdqa xmm2, [esi+32] |
289 |
|
psubw xmm0,xmm4 |
290 |
|
pmulhw xmm1,xmm7 |
291 |
|
pxor xmm4,xmm4 |
292 |
|
movdqa xmm3,[esi+48] |
293 |
|
pcmpgtw xmm4,xmm2 |
294 |
|
psubw xmm1,xmm5 |
295 |
|
pmulhw xmm2,xmm7 |
296 |
|
pxor xmm5,xmm5 |
297 |
|
pcmpgtw xmm5,xmm3 |
298 |
pmulhw xmm3, xmm7 |
pmulhw xmm3, xmm7 |
299 |
pxor xmm0, xmm1 |
psubw xmm2,xmm4 |
300 |
pxor xmm3, xmm4 |
psubw xmm3,xmm5 |
301 |
psubw xmm0, xmm1 |
movdqa [edx], xmm0 |
302 |
psubw xmm3, xmm4 |
lea esi, [esi+64] |
303 |
movdqa [edi + ecx*8], xmm0 |
movdqa [edx + 16], xmm1 |
304 |
movdqa [edi + ecx*8 + 16], xmm3 |
movdqa [edx + 32], xmm2 |
305 |
|
movdqa [edx + 48], xmm3 |
306 |
add ecx, 4 |
|
307 |
cmp ecx, 16 |
dec ecx |
308 |
jnz .qas2_loop |
lea edx, [edx+64] |
309 |
|
jne .loop |
310 |
.qas2_done |
jmp .end |
311 |
mov ecx, [esp + 8 + 16] ; dcscalar |
|
312 |
mov edx, ecx |
.low: |
313 |
movsx eax, word [esi] |
movd xmm7,ecx |
314 |
shr edx, 1 |
mov ecx,2 |
315 |
cmp eax, 0 |
.loop_low: |
316 |
jg .qas2_gtzero |
movdqa xmm0, [esi] |
317 |
|
pxor xmm4,xmm4 |
318 |
sub eax, edx |
movdqa xmm1, [esi + 16] |
319 |
jmp short .qas2_mul |
pcmpgtw xmm4,xmm0 |
320 |
|
pxor xmm5,xmm5 |
321 |
.qas2_gtzero |
psubw xmm0,xmm4 |
322 |
add eax, edx |
pcmpgtw xmm5,xmm1 |
323 |
|
psraw xmm0,xmm7 |
324 |
.qas2_mul |
psubw xmm1,xmm5 |
325 |
cdq |
movdqa xmm2,[esi+32] |
326 |
idiv ecx |
pxor xmm4,xmm4 |
327 |
|
psraw xmm1,xmm7 |
328 |
|
pcmpgtw xmm4,xmm2 |
329 |
|
pxor xmm5,xmm5 |
330 |
|
psubw xmm2,xmm4 |
331 |
|
movdqa xmm3,[esi+48] |
332 |
|
pcmpgtw xmm5,xmm3 |
333 |
|
psraw xmm2,xmm7 |
334 |
|
psubw xmm3,xmm5 |
335 |
|
movdqa [edx], xmm0 |
336 |
|
psraw xmm3,xmm7 |
337 |
|
movdqa [edx+16], xmm1 |
338 |
|
movdqa [edx+32],xmm2 |
339 |
|
lea esi, [esi+64] |
340 |
|
movdqa [edx+48],xmm3 |
341 |
|
|
342 |
|
dec ecx |
343 |
|
lea edx, [edx+64] |
344 |
|
jne .loop_low |
345 |
|
|
346 |
mov [edi], ax |
.end: |
347 |
|
mov edx, [esp + 4 + 4] ; coeff |
348 |
|
mov [edx],ax |
349 |
|
xor eax,eax ; return 0 |
350 |
|
|
|
xor eax, eax ; return(0); |
|
|
pop edi |
|
351 |
pop esi |
pop esi |
|
|
|
352 |
ret |
ret |
353 |
|
ENDFUNC |
|
ALIGN 16 |
|
|
.qas2_q1loop |
|
|
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
|
|
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
|
|
pxor xmm1, xmm1 |
|
|
pxor xmm4, xmm4 |
|
|
pcmpgtw xmm1, xmm0 |
|
|
pcmpgtw xmm4, xmm3 |
|
|
pxor xmm0, xmm1 |
|
|
pxor xmm3, xmm4 |
|
|
psubw xmm0, xmm1 |
|
|
psubw xmm3, xmm4 |
|
|
psrlw xmm0, 1 |
|
|
psrlw xmm3, 1 |
|
|
pxor xmm0, xmm1 |
|
|
pxor xmm3, xmm4 |
|
|
psubw xmm0, xmm1 |
|
|
psubw xmm3, xmm4 |
|
|
movdqa [edi + ecx*8], xmm0 |
|
|
movdqa [edi + ecx*8 + 16], xmm3 |
|
|
|
|
|
add ecx, 4 |
|
|
cmp ecx, 16 |
|
|
jnz .qas2_q1loop |
|
|
jmp near .qas2_done |
|
|
|
|
|
|
|
354 |
|
|
355 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
356 |
; |
; |
383 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
384 |
|
|
385 |
ALIGN 8 |
ALIGN 8 |
386 |
.loop |
.loop: |
387 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
388 |
movq mm3, [esi + 8*ecx + 8] |
movq mm3, [esi + 8*ecx + 8] |
389 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
411 |
cmp ecx, 16 |
cmp ecx, 16 |
412 |
jnz .loop |
jnz .loop |
413 |
|
|
414 |
.done |
.done: |
415 |
pmaddwd mm5, [plus_one] |
pmaddwd mm5, [plus_one] |
416 |
movq mm0, mm5 |
movq mm0, mm5 |
417 |
psrlq mm5, 32 |
psrlq mm5, 32 |
425 |
ret |
ret |
426 |
|
|
427 |
ALIGN 8 |
ALIGN 8 |
428 |
.q1loop |
.q1loop: |
429 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
430 |
movq mm3, [esi + 8*ecx+ 8] ; |
movq mm3, [esi + 8*ecx+ 8] ; |
431 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
454 |
jnz .q1loop |
jnz .q1loop |
455 |
|
|
456 |
jmp .done |
jmp .done |
457 |
|
ENDFUNC |
458 |
|
|
459 |
|
|
460 |
|
|
488 |
cmp al, 1 |
cmp al, 1 |
489 |
jz near .qes2_q1loop |
jz near .qes2_q1loop |
490 |
|
|
491 |
.qes2_not1 |
.qes2_not1: |
492 |
movq mm0, [mmx_div + eax*8 - 8] ; divider |
movq mm0, [mmx_div + eax*8 - 8] ; divider |
493 |
movq2dq xmm7, mm0 |
movq2dq xmm7, mm0 |
494 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
495 |
|
|
496 |
ALIGN 16 |
ALIGN 16 |
497 |
.qes2_loop |
.qes2_loop: |
498 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
499 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
500 |
pxor xmm1, xmm1 |
pxor xmm1, xmm1 |
522 |
cmp ecx, 16 |
cmp ecx, 16 |
523 |
jnz .qes2_loop |
jnz .qes2_loop |
524 |
|
|
525 |
.qes2_done |
.qes2_done: |
526 |
movdqu xmm6, [plus_one] |
movdqu xmm6, [plus_one] |
527 |
pmaddwd xmm5, xmm6 |
pmaddwd xmm5, xmm6 |
528 |
movhlps xmm6, xmm5 |
movhlps xmm6, xmm5 |
541 |
ret |
ret |
542 |
|
|
543 |
ALIGN 16 |
ALIGN 16 |
544 |
.qes2_q1loop |
.qes2_q1loop: |
545 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
546 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
547 |
pxor xmm1, xmm1 |
pxor xmm1, xmm1 |
569 |
cmp ecx, 16 |
cmp ecx, 16 |
570 |
jnz .qes2_q1loop |
jnz .qes2_q1loop |
571 |
jmp .qes2_done |
jmp .qes2_done |
572 |
|
ENDFUNC |
573 |
|
|
574 |
|
|
575 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
582 |
; |
; |
583 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
584 |
|
|
|
; note: we only saturate to +2047 *before* restoring the sign. |
|
|
; Hence, final clamp really is [-2048,2047] |
|
|
|
|
585 |
ALIGN 16 |
ALIGN 16 |
586 |
dequant_h263_intra_mmx: |
dequant_h263_intra_mmx: |
587 |
|
|
588 |
|
mov ecx, [esp+12] ; quant |
589 |
|
mov eax, [esp+ 8] ; coeff |
590 |
|
pcmpeqw mm0,mm0 |
591 |
|
movq mm6, [mmx_quant + ecx*8] ; quant |
592 |
|
shl ecx,31 ; quant & 1 ? 0 : - 1 |
593 |
|
movq mm7,mm6 |
594 |
|
movq mm5,mm0 |
595 |
|
movd mm1,ecx |
596 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
597 |
mov ecx, [esp+ 8] ; coeff |
psllw mm0,mm1 |
598 |
mov eax, [esp+12] ; quant |
paddw mm7,mm7 ; 2*quant |
599 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
paddw mm6,mm0 ; quant-1 |
600 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
psllw mm5,12 |
601 |
mov eax, -16 |
mov ecx,8 |
602 |
|
psrlw mm5,1 |
603 |
|
|
604 |
ALIGN 16 |
.loop: |
605 |
.loop |
movq mm0,[eax] |
606 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
pxor mm2,mm2 |
|
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
|
|
pxor mm1, mm1 |
|
607 |
pxor mm4, mm4 |
pxor mm4, mm4 |
608 |
pcmpgtw mm1, mm0 ; sign(c) |
pcmpgtw mm2,mm0 |
609 |
pcmpgtw mm4, mm3 ; sign(c') |
pcmpeqw mm4,mm0 |
610 |
|
pmullw mm0,mm7 ; * 2 * quant |
611 |
|
movq mm1,[eax+8] |
612 |
|
psubw mm0,mm2 |
613 |
|
pxor mm2,mm6 |
614 |
|
pxor mm3,mm3 |
615 |
|
pandn mm4,mm2 |
616 |
pxor mm2, mm2 |
pxor mm2, mm2 |
617 |
pxor mm5, mm5 |
pcmpgtw mm3,mm1 |
618 |
pcmpeqw mm2, mm0 ; c is zero |
pcmpeqw mm2,mm1 |
619 |
pcmpeqw mm5, mm3 ; c' is zero |
pmullw mm1,mm7 |
620 |
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
paddw mm0,mm4 |
621 |
pandn mm5, mm6 |
psubw mm1,mm3 |
622 |
pxor mm0, mm1 ; negate if negative |
pxor mm3,mm6 |
623 |
pxor mm3, mm4 ; negate if negative |
pandn mm2,mm3 |
624 |
psubw mm0, mm1 |
paddsw mm0, mm5 ; saturate |
625 |
psubw mm3, mm4 |
paddw mm1,mm2 |
626 |
pmullw mm0, mm7 ; *= 2Q |
|
627 |
pmullw mm3, mm7 ; *= 2Q |
paddsw mm1, mm5 |
628 |
paddw mm0, mm2 ; + offset |
psubsw mm0, mm5 |
629 |
paddw mm3, mm5 ; + offset |
psubsw mm1, mm5 |
630 |
paddw mm0, mm1 ; negate back |
psubsw mm0, mm5 |
631 |
paddw mm3, mm4 ; negate back |
psubsw mm1, mm5 |
632 |
|
paddsw mm0, mm5 |
633 |
; saturates to +2047 |
paddsw mm1, mm5 |
634 |
movq mm2, [mmx_32767_minus_2047] |
|
635 |
add eax, 2 |
movq [edx],mm0 |
636 |
paddsw mm0, mm2 |
lea eax,[eax+16] |
637 |
paddsw mm3, mm2 |
movq [edx+8],mm1 |
638 |
psubsw mm0, mm2 |
|
639 |
psubsw mm3, mm2 |
dec ecx |
640 |
|
lea edx,[edx+16] |
641 |
pxor mm0, mm1 |
jne .loop |
|
pxor mm3, mm4 |
|
|
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
|
|
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
|
|
jnz near .loop |
|
642 |
|
|
643 |
; deal with DC |
; deal with DC |
644 |
movd mm0, [ecx] |
mov eax, [esp+ 8] ; coeff |
645 |
pmullw mm0, [esp+16] ; dcscalar |
movd mm1,[esp+16] ; dcscalar |
646 |
movq mm2, [mmx_32767_minus_2047] |
movd mm0,[eax] ; coeff[0] |
647 |
paddsw mm0, mm2 |
pmullw mm0,mm1 ; * dcscalar |
648 |
psubsw mm0, mm2 |
mov edx, [esp+ 4] ; data |
649 |
movq mm3, [mmx_32768_minus_2048] |
paddsw mm0, mm5 ; saturate + |
650 |
psubsw mm0, mm3 |
psubsw mm0, mm5 |
651 |
paddsw mm0, mm3 |
psubsw mm0, mm5 ; saturate - |
652 |
|
paddsw mm0, mm5 |
653 |
movd eax, mm0 |
movd eax, mm0 |
654 |
mov [edx], ax |
mov [edx], ax |
655 |
|
|
656 |
xor eax, eax ; return(0); |
xor eax, eax ; return 0 |
657 |
ret |
ret |
658 |
|
ENDFUNC |
659 |
|
|
660 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
661 |
; |
; |
667 |
; |
; |
668 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
669 |
|
|
|
; this is the same as dequant_inter_mmx, except that we're |
|
|
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
|
670 |
|
|
671 |
ALIGN 16 |
ALIGN 16 |
672 |
dequant_h263_intra_xmm: |
dequant_h263_intra_xmm: |
673 |
|
|
674 |
|
mov ecx, [esp+12] ; quant |
675 |
|
mov eax, [esp+ 8] ; coeff |
676 |
|
|
677 |
|
movd mm6,ecx ; quant |
678 |
|
pcmpeqw mm0,mm0 |
679 |
|
pshufw mm6,mm6,0 ; all quant |
680 |
|
shl ecx,31 |
681 |
|
movq mm5,mm0 |
682 |
|
movq mm7,mm6 |
683 |
|
movd mm1,ecx |
684 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
685 |
mov ecx, [esp+ 8] ; coeff |
psllw mm0,mm1 ; quant & 1 ? 0 : - 1 |
686 |
mov eax, [esp+12] ; quant |
movq mm4,mm5 |
687 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
paddw mm7,mm7 ; quant*2 |
688 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
paddw mm6,mm0 ; quant-1 |
689 |
mov eax, -16 |
psrlw mm4,5 ; mm4=2047 |
690 |
|
mov ecx,8 |
691 |
|
pxor mm5,mm4 ; mm5=-2048 |
692 |
|
|
693 |
ALIGN 16 |
.loop: |
694 |
.loop |
movq mm0,[eax] |
|
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
|
|
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
|
|
pxor mm1, mm1 |
|
|
pxor mm4, mm4 |
|
|
pcmpgtw mm1, mm0 ; sign(c) |
|
|
pcmpgtw mm4, mm3 ; sign(c') |
|
695 |
pxor mm2, mm2 |
pxor mm2, mm2 |
696 |
pxor mm5, mm5 |
pxor mm3,mm3 |
697 |
pcmpeqw mm2, mm0 ; c is zero |
|
698 |
pcmpeqw mm5, mm3 ; c' is zero |
pcmpgtw mm2,mm0 |
699 |
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
pcmpeqw mm3,mm0 ; if coeff==0... |
700 |
pandn mm5, mm6 |
pmullw mm0,mm7 ; * 2 * quant |
701 |
pxor mm0, mm1 ; negate if negative |
movq mm1,[eax+8] |
702 |
pxor mm3, mm4 ; negate if negative |
|
703 |
psubw mm0, mm1 |
psubw mm0,mm2 |
704 |
psubw mm3, mm4 |
pxor mm2,mm6 |
705 |
pmullw mm0, mm7 ; *= 2Q |
pandn mm3,mm2 ; ...then data=0 |
706 |
pmullw mm3, mm7 ; *= 2Q |
pxor mm2,mm2 |
707 |
paddw mm0, mm2 ; + offset |
paddw mm0,mm3 |
708 |
paddw mm3, mm5 ; + offset |
pxor mm3,mm3 |
709 |
paddw mm0, mm1 ; negate back |
pcmpeqw mm2,mm1 |
710 |
paddw mm3, mm4 ; negate back |
pcmpgtw mm3,mm1 |
711 |
|
pmullw mm1,mm7 |
712 |
; saturates to +2047 |
|
713 |
movq mm2, [mmx_2047] |
pminsw mm0,mm4 |
714 |
pminsw mm0, mm2 |
psubw mm1,mm3 |
715 |
add eax, 2 |
pxor mm3,mm6 |
716 |
pminsw mm3, mm2 |
pandn mm2,mm3 |
717 |
|
paddw mm1,mm2 |
718 |
pxor mm0, mm1 |
|
719 |
pxor mm3, mm4 |
pmaxsw mm0,mm5 |
720 |
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
pminsw mm1,mm4 |
721 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
movq [edx],mm0 |
722 |
jnz near .loop |
pmaxsw mm1,mm5 |
723 |
|
lea eax,[eax+16] |
724 |
|
movq [edx+8],mm1 |
725 |
|
|
726 |
|
dec ecx |
727 |
|
lea edx,[edx+16] |
728 |
|
jne .loop |
729 |
|
|
730 |
; deal with DC |
; deal with DC |
731 |
movd mm0, [ecx] |
mov eax, [esp+ 8] ; coeff |
732 |
pmullw mm0, [esp+16] ; dcscalar |
movd mm1,[esp+16] ; dcscalar |
733 |
movq mm2, [mmx_32767_minus_2047] |
movd mm0, [eax] |
734 |
paddsw mm0, mm2 |
pmullw mm0, mm1 |
735 |
psubsw mm0, mm2 |
mov edx, [esp+ 4] ; data |
736 |
movq mm2, [mmx_32768_minus_2048] |
pminsw mm0,mm4 |
737 |
psubsw mm0, mm2 |
pmaxsw mm0,mm5 |
|
paddsw mm0, mm2 |
|
738 |
movd eax, mm0 |
movd eax, mm0 |
739 |
mov [edx], ax |
mov [edx], ax |
740 |
|
|
741 |
xor eax, eax |
xor eax, eax ; return 0 |
742 |
ret |
ret |
743 |
|
ENDFUNC |
744 |
|
|
745 |
|
|
746 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
755 |
|
|
756 |
ALIGN 16 |
ALIGN 16 |
757 |
dequant_h263_intra_sse2: |
dequant_h263_intra_sse2: |
758 |
|
|
759 |
|
mov ecx, [esp+12] ; quant |
760 |
|
mov eax, [esp+ 8] ; coeff |
761 |
|
|
762 |
|
movd xmm6,ecx ; quant |
763 |
|
|
764 |
|
shl ecx,31 |
765 |
|
pshuflw xmm6,xmm6,0 |
766 |
|
pcmpeqw xmm0,xmm0 |
767 |
|
movlhps xmm6,xmm6 ; all quant |
768 |
|
movd xmm1,ecx |
769 |
|
movdqa xmm5,xmm0 |
770 |
|
movdqa xmm7,xmm6 |
771 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
772 |
mov ecx, [esp+ 8] ; coeff |
paddw xmm7,xmm7 ; quant *2 |
773 |
mov eax, [esp+12] ; quant |
psllw xmm0,xmm1 ; quant & 1 ? 0 : - 1 |
774 |
movq mm6, [mmx_add + eax * 8 - 8] |
movdqa xmm4,xmm5 |
775 |
movq mm7, [mmx_mul + eax * 8 - 8] |
paddw xmm6,xmm0 ; quant-1 |
776 |
movq2dq xmm6, mm6 |
psrlw xmm4,5 ; 2047 |
777 |
movq2dq xmm7, mm7 |
mov ecx,4 |
778 |
movlhps xmm6, xmm6 |
pxor xmm5,xmm4 ; mm5=-2048 |
|
movlhps xmm7, xmm7 |
|
|
mov eax, -16 |
|
779 |
|
|
780 |
ALIGN 16 |
.loop: |
781 |
.loop |
movdqa xmm0,[eax] |
|
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
|
|
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
|
|
pxor xmm1, xmm1 |
|
|
pxor xmm4, xmm4 |
|
|
pcmpgtw xmm1, xmm0 ; sign(c) |
|
|
pcmpgtw xmm4, xmm3 |
|
782 |
pxor xmm2, xmm2 |
pxor xmm2, xmm2 |
783 |
pxor xmm5, xmm5 |
pxor xmm3,xmm3 |
|
pcmpeqw xmm2, xmm0 ; c is zero |
|
|
pcmpeqw xmm5, xmm3 |
|
|
pandn xmm2, xmm6 ; offset = isZero ? 0 : quant_add |
|
|
pandn xmm5, xmm6 |
|
|
pxor xmm0, xmm1 ; negate if negative |
|
|
pxor xmm3, xmm4 |
|
|
psubw xmm0, xmm1 |
|
|
psubw xmm3, xmm4 |
|
|
pmullw xmm0, xmm7 ; *= 2Q |
|
|
pmullw xmm3, xmm7 |
|
|
paddw xmm0, xmm2 ; + offset |
|
|
paddw xmm3, xmm5 |
|
|
paddw xmm0, xmm1 ; negate back |
|
|
paddw xmm3, xmm4 |
|
|
|
|
|
; saturates to +2047 |
|
|
movdqa xmm2, [sse2_2047] |
|
|
pminsw xmm0, xmm2 |
|
|
add eax, 4 |
|
|
pminsw xmm3, xmm2 |
|
784 |
|
|
785 |
pxor xmm0, xmm1 |
pcmpgtw xmm2,xmm0 |
786 |
pxor xmm3, xmm4 |
pcmpeqw xmm3,xmm0 |
787 |
movdqa [edx + 8*16 - 8*4 + 8*eax], xmm0 |
pmullw xmm0,xmm7 ; * 2 * quant |
788 |
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
movdqa xmm1,[eax+16] |
789 |
jnz near .loop |
|
790 |
|
psubw xmm0,xmm2 |
791 |
|
pxor xmm2,xmm6 |
792 |
|
pandn xmm3,xmm2 |
793 |
|
pxor xmm2,xmm2 |
794 |
|
paddw xmm0,xmm3 |
795 |
|
pxor xmm3,xmm3 |
796 |
|
pcmpeqw xmm2,xmm1 |
797 |
|
pcmpgtw xmm3,xmm1 |
798 |
|
pmullw xmm1,xmm7 |
799 |
|
|
800 |
|
pminsw xmm0,xmm4 |
801 |
|
psubw xmm1,xmm3 |
802 |
|
pxor xmm3,xmm6 |
803 |
|
pandn xmm2,xmm3 |
804 |
|
paddw xmm1,xmm2 |
805 |
|
|
806 |
|
pmaxsw xmm0,xmm5 |
807 |
|
pminsw xmm1,xmm4 |
808 |
|
movdqa [edx],xmm0 |
809 |
|
pmaxsw xmm1,xmm5 |
810 |
|
lea eax,[eax+32] |
811 |
|
movdqa [edx+16],xmm1 |
812 |
|
|
813 |
|
dec ecx |
814 |
|
lea edx,[edx+32] |
815 |
|
jne .loop |
816 |
|
|
817 |
; deal with DC |
; deal with DC |
818 |
movd mm0, [ecx] |
|
819 |
pmullw mm0, [esp+16] ; dcscalar |
mov eax, [esp+ 8] ; coeff |
820 |
movq mm2, [mmx_32767_minus_2047] |
movsx eax,word [eax] |
821 |
paddsw mm0, mm2 |
imul dword [esp+16] ; dcscalar |
822 |
psubsw mm0, mm2 |
mov edx, [esp+ 4] ; data |
823 |
movq mm2, [mmx_32768_minus_2048] |
movd xmm0,eax |
824 |
psubsw mm0, mm2 |
pminsw xmm0,xmm4 |
825 |
paddsw mm0, mm2 |
pmaxsw xmm0,xmm5 |
826 |
movd eax, mm0 |
movd eax,xmm0 |
827 |
|
|
828 |
mov [edx], ax |
mov [edx], ax |
829 |
|
|
830 |
xor eax, eax |
xor eax, eax ; return 0 |
831 |
ret |
ret |
832 |
|
ENDFUNC |
833 |
|
|
834 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
835 |
; |
; |
843 |
ALIGN 16 |
ALIGN 16 |
844 |
dequant_h263_inter_mmx: |
dequant_h263_inter_mmx: |
845 |
|
|
846 |
|
mov ecx, [esp+12] ; quant |
847 |
|
mov eax, [esp+ 8] ; coeff |
848 |
|
pcmpeqw mm0,mm0 |
849 |
|
movq mm6, [mmx_quant + ecx*8] ; quant |
850 |
|
shl ecx,31 ; odd/even |
851 |
|
movq mm7,mm6 |
852 |
|
movd mm1,ecx |
853 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
854 |
mov ecx, [esp+ 8] ; coeff |
movq mm5,mm0 |
855 |
mov eax, [esp+12] ; quant |
psllw mm0,mm1 ; quant & 1 ? 0 : - 1 |
856 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
paddw mm7,mm7 ; quant*2 |
857 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
paddw mm6,mm0 ; quant & 1 ? quant : quant - 1 |
858 |
mov eax, -16 |
psllw mm5,12 |
859 |
|
mov ecx,8 |
860 |
|
psrlw mm5,1 ; 32767-2047 (32768-2048) |
861 |
|
|
862 |
ALIGN 16 |
.loop: |
863 |
.loop |
movq mm0,[eax] |
|
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
|
|
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
|
|
pxor mm1, mm1 |
|
864 |
pxor mm4, mm4 |
pxor mm4, mm4 |
|
pcmpgtw mm1, mm0 ; sign(c) |
|
|
pcmpgtw mm4, mm3 ; sign(c') |
|
865 |
pxor mm2, mm2 |
pxor mm2, mm2 |
866 |
pxor mm5, mm5 |
pcmpeqw mm4,mm0 ; if coeff==0... |
867 |
pcmpeqw mm2, mm0 ; c is zero |
pcmpgtw mm2,mm0 |
868 |
pcmpeqw mm5, mm3 ; c' is zero |
pmullw mm0,mm7 ; * 2 * quant |
869 |
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
pxor mm3,mm3 |
870 |
pandn mm5, mm6 |
psubw mm0,mm2 |
871 |
pxor mm0, mm1 ; negate if negative |
movq mm1,[eax+8] |
872 |
pxor mm3, mm4 ; negate if negative |
pxor mm2,mm6 |
873 |
psubw mm0, mm1 |
pcmpgtw mm3,mm1 |
874 |
psubw mm3, mm4 |
pandn mm4,mm2 ; ... then data==0 |
875 |
pmullw mm0, mm7 ; *= 2Q |
pmullw mm1,mm7 |
876 |
pmullw mm3, mm7 ; *= 2Q |
pxor mm2,mm2 |
877 |
paddw mm0, mm2 ; + offset |
pcmpeqw mm2,mm1 |
878 |
paddw mm3, mm5 ; + offset |
psubw mm1,mm3 |
879 |
paddw mm0, mm1 ; negate back |
pxor mm3,mm6 |
880 |
paddw mm3, mm4 ; negate back |
pandn mm2,mm3 |
881 |
; saturates to +2047 |
paddw mm0,mm4 |
882 |
movq mm2, [mmx_32767_minus_2047] |
paddw mm1,mm2 |
883 |
add eax, 2 |
|
884 |
paddsw mm0, mm2 |
paddsw mm0, mm5 ; saturate |
885 |
paddsw mm3, mm2 |
paddsw mm1, mm5 |
886 |
psubsw mm0, mm2 |
psubsw mm0, mm5 |
887 |
psubsw mm3, mm2 |
psubsw mm1, mm5 |
888 |
|
psubsw mm0, mm5 |
889 |
pxor mm0, mm1 |
psubsw mm1, mm5 |
890 |
pxor mm3, mm4 |
paddsw mm0, mm5 |
891 |
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
paddsw mm1, mm5 |
892 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
|
893 |
jnz near .loop |
movq [edx],mm0 |
894 |
|
lea eax,[eax+16] |
895 |
|
movq [edx+8],mm1 |
896 |
|
|
897 |
|
dec ecx |
898 |
|
lea edx,[edx+16] |
899 |
|
jne .loop |
900 |
|
|
901 |
xor eax, eax |
xor eax, eax ; return 0 |
902 |
ret |
ret |
903 |
|
ENDFUNC |
904 |
|
|
905 |
|
|
906 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
907 |
; |
; |
911 |
; const uint16_t *mpeg_matrices); |
; const uint16_t *mpeg_matrices); |
912 |
; |
; |
913 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
|
|
|
|
; this is the same as dequant_inter_mmx, |
|
|
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
|
|
|
|
914 |
ALIGN 16 |
ALIGN 16 |
915 |
dequant_h263_inter_xmm: |
dequant_h263_inter_xmm: |
916 |
|
|
917 |
|
mov ecx, [esp+12] ; quant |
918 |
|
mov eax, [esp+ 8] ; coeff |
919 |
|
pcmpeqw mm0,mm0 |
920 |
|
movq mm6, [mmx_quant + ecx*8] ; quant |
921 |
|
shl ecx,31 |
922 |
|
movq mm5,mm0 |
923 |
|
movd mm1,ecx |
924 |
|
movq mm7,mm6 |
925 |
|
psllw mm0,mm1 |
926 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
927 |
mov ecx, [esp+ 8] ; coeff |
movq mm4,mm5 |
928 |
mov eax, [esp+12] ; quant |
paddw mm7,mm7 |
929 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
paddw mm6,mm0 ; quant-1 |
930 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
|
931 |
mov eax, -16 |
psrlw mm4,5 |
932 |
|
mov ecx,8 |
933 |
ALIGN 16 |
pxor mm5,mm4 ; mm5=-2048 |
934 |
.loop |
|
935 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
.loop: |
936 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm0,[eax] |
937 |
pxor mm1, mm1 |
pxor mm3,mm3 |
|
pxor mm4, mm4 |
|
|
pcmpgtw mm1, mm0 ; sign(c) |
|
|
pcmpgtw mm4, mm3 ; sign(c') |
|
938 |
pxor mm2, mm2 |
pxor mm2, mm2 |
939 |
pxor mm5, mm5 |
pcmpeqw mm3,mm0 |
940 |
pcmpeqw mm2, mm0 ; c is zero |
pcmpgtw mm2,mm0 |
941 |
pcmpeqw mm5, mm3 ; c' is zero |
pmullw mm0,mm7 ; * 2 * quant |
942 |
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
pandn mm3,mm6 |
943 |
pandn mm5, mm6 |
movq mm1,[eax+8] |
944 |
pxor mm0, mm1 ; negate if negative |
psubw mm0,mm2 |
945 |
pxor mm3, mm4 ; negate if negative |
pxor mm2,mm3 |
946 |
psubw mm0, mm1 |
pxor mm3,mm3 |
947 |
psubw mm3, mm4 |
paddw mm0,mm2 |
948 |
pmullw mm0, mm7 ; *= 2Q |
pxor mm2,mm2 |
949 |
pmullw mm3, mm7 ; *= 2Q |
pcmpgtw mm3,mm1 |
950 |
paddw mm0, mm2 ; + offset |
pcmpeqw mm2,mm1 |
951 |
paddw mm3, mm5 ; + offset |
pmullw mm1,mm7 |
952 |
paddw mm0, mm1 ; start restoring sign |
pandn mm2,mm6 |
953 |
paddw mm3, mm4 ; start restoring sign |
psubw mm1,mm3 |
954 |
; saturates to +2047 |
pxor mm3,mm2 |
955 |
movq mm2, [mmx_2047] |
paddw mm1,mm3 |
956 |
pminsw mm0, mm2 |
|
957 |
add eax, 2 |
pminsw mm0,mm4 |
958 |
pminsw mm3, mm2 |
pminsw mm1,mm4 |
959 |
|
pmaxsw mm0,mm5 |
960 |
pxor mm0, mm1 ; finish restoring sign |
pmaxsw mm1,mm5 |
961 |
pxor mm3, mm4 ; finish restoring sign |
|
962 |
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
movq [edx],mm0 |
963 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
lea eax,[eax+16] |
964 |
jnz near .loop |
movq [edx+8],mm1 |
965 |
|
|
966 |
|
dec ecx |
967 |
|
lea edx,[edx+16] |
968 |
|
jne .loop |
969 |
|
|
970 |
xor eax, eax |
xor eax, eax ; return 0 |
971 |
ret |
ret |
972 |
|
ENDFUNC |
973 |
|
|
974 |
|
|
975 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
976 |
; |
; |
983 |
|
|
984 |
ALIGN 16 |
ALIGN 16 |
985 |
dequant_h263_inter_sse2: |
dequant_h263_inter_sse2: |
|
mov edx, [esp + 4] ; data |
|
|
mov ecx, [esp + 8] ; coeff |
|
|
mov eax, [esp + 12] ; quant |
|
|
movq mm6, [mmx_add + eax * 8 - 8] |
|
|
movq mm7, [mmx_mul + eax * 8 - 8] |
|
|
movq2dq xmm6, mm6 |
|
|
movq2dq xmm7, mm7 |
|
|
movlhps xmm6, xmm6 |
|
|
movlhps xmm7, xmm7 |
|
|
mov eax, -16 |
|
986 |
|
|
987 |
ALIGN 16 |
mov ecx, [esp+12] ; quant |
988 |
.loop |
mov eax, [esp+ 8] ; coeff |
|
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
|
|
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |
|
989 |
|
|
990 |
pxor xmm1, xmm1 |
movq xmm6, [mmx_quant + ecx*8] ; quant |
991 |
pxor xmm4, xmm4 |
inc ecx |
992 |
pcmpgtw xmm1, xmm0 ; sign(c) |
pcmpeqw xmm5,xmm5 |
993 |
pcmpgtw xmm4, xmm3 |
and ecx,1 |
994 |
|
movlhps xmm6,xmm6 |
995 |
|
movd xmm0,ecx |
996 |
|
movdqa xmm7,xmm6 |
997 |
|
pshuflw xmm0,xmm0,0 |
998 |
|
movdqa xmm4,xmm5 |
999 |
|
mov edx, [esp+ 4] ; data |
1000 |
|
movlhps xmm0,xmm0 |
1001 |
|
paddw xmm7,xmm7 |
1002 |
|
psubw xmm6,xmm0 |
1003 |
|
psrlw xmm4,5 ; 2047 |
1004 |
|
mov ecx,4 |
1005 |
|
pxor xmm5,xmm4 ; mm5=-2048 |
1006 |
|
|
1007 |
|
.loop: |
1008 |
|
movdqa xmm0,[eax] |
1009 |
|
pxor xmm3,xmm3 |
1010 |
pxor xmm2, xmm2 |
pxor xmm2, xmm2 |
1011 |
pxor xmm5, xmm5 |
pcmpeqw xmm3,xmm0 |
1012 |
pcmpeqw xmm2, xmm0 ; c is zero |
pcmpgtw xmm2,xmm0 |
1013 |
pcmpeqw xmm5, xmm3 |
pmullw xmm0,xmm7 ; * 2 * quant |
1014 |
|
pandn xmm3,xmm6 |
1015 |
|
movdqa xmm1,[eax+16] |
1016 |
|
psubw xmm0,xmm2 |
1017 |
|
pxor xmm2,xmm3 |
1018 |
|
pxor xmm3,xmm3 |
1019 |
|
paddw xmm0,xmm2 |
1020 |
|
pxor xmm2,xmm2 |
1021 |
|
pcmpgtw xmm3,xmm1 |
1022 |
|
pcmpeqw xmm2,xmm1 |
1023 |
|
pmullw xmm1,xmm7 |
1024 |
pandn xmm2, xmm6 |
pandn xmm2, xmm6 |
1025 |
pandn xmm5, xmm6 |
psubw xmm1,xmm3 |
1026 |
pxor xmm0, xmm1 ; negate if negative |
pxor xmm3,xmm2 |
1027 |
pxor xmm3, xmm4 |
paddw xmm1,xmm3 |
1028 |
psubw xmm0, xmm1 |
|
1029 |
psubw xmm3, xmm4 |
pminsw xmm0,xmm4 |
1030 |
pmullw xmm0, xmm7 ; *= 2Q |
pminsw xmm1,xmm4 |
1031 |
pmullw xmm3, xmm7 |
pmaxsw xmm0,xmm5 |
1032 |
paddw xmm0, xmm2 ; + offset |
pmaxsw xmm1,xmm5 |
1033 |
paddw xmm3, xmm5 |
|
1034 |
|
movdqa [edx],xmm0 |
1035 |
paddw xmm0, xmm1 ; start restoring sign |
lea eax,[eax+32] |
1036 |
paddw xmm3, xmm4 |
movdqa [edx+16],xmm1 |
1037 |
|
|
1038 |
; saturates to +2047 |
dec ecx |
1039 |
movdqa xmm2, [sse2_2047] |
lea edx,[edx+32] |
1040 |
pminsw xmm0, xmm2 |
jne .loop |
|
add eax, 4 |
|
|
pminsw xmm3, xmm2 |
|
|
|
|
|
pxor xmm0, xmm1 ; finish restoring sign |
|
|
pxor xmm3, xmm4 |
|
|
movdqa [edx + 8*16 - 8*4 + 8*eax], xmm0 |
|
|
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
|
|
jnz near .loop |
|
1041 |
|
|
1042 |
xor eax, eax |
xor eax, eax ; return 0 |
1043 |
ret |
ret |
1044 |
|
ENDFUNC |
1045 |
|
|
1046 |
|
|
1047 |
|
%ifidn __OUTPUT_FORMAT__,elf |
1048 |
|
section ".note.GNU-stack" noalloc noexec nowrite progbits |
1049 |
|
%endif |
1050 |
|
|