4 |
; * - 3dne Quantization/Dequantization - |
; * - 3dne Quantization/Dequantization - |
5 |
; * |
; * |
6 |
; * Copyright (C) 2002-2003 Peter Ross <pross@xvid.org> |
; * Copyright (C) 2002-2003 Peter Ross <pross@xvid.org> |
7 |
; * 2002-2003 Michael Militzer <isibaar@xvid.org> |
; * 2002-2008 Michael Militzer <michael@xvid.org> |
8 |
; * 2002-2003 Pascal Massimino <skal@planet-d.net> |
; * 2002-2003 Pascal Massimino <skal@planet-d.net> |
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 |
27 |
|
|
28 |
%define SATURATE |
%define SATURATE |
29 |
|
|
30 |
BITS 32 |
%include "nasm.inc" |
|
|
|
|
%macro cglobal 1 |
|
|
%ifdef PREFIX |
|
|
%ifdef MARK_FUNCS |
|
|
global _%1:function %1.endfunc-%1 |
|
|
%define %1 _%1:function %1.endfunc-%1 |
|
|
%define ENDFUNC .endfunc |
|
|
%else |
|
|
global _%1 |
|
|
%define %1 _%1 |
|
|
%define ENDFUNC |
|
|
%endif |
|
|
%else |
|
|
%ifdef MARK_FUNCS |
|
|
global %1:function %1.endfunc-%1 |
|
|
%define ENDFUNC .endfunc |
|
|
%else |
|
|
global %1 |
|
|
%define ENDFUNC |
|
|
%endif |
|
|
%endif |
|
|
%endmacro |
|
|
|
|
|
%macro cextern 1 |
|
|
%ifdef PREFIX |
|
|
extern _%1 |
|
|
%define %1 _%1 |
|
|
%else |
|
|
extern %1 |
|
|
%endif |
|
|
%endmacro |
|
31 |
|
|
32 |
;============================================================================= |
;============================================================================= |
33 |
; Local data (Read Only) |
; Local data (Read Only) |
34 |
;============================================================================= |
;============================================================================= |
35 |
|
|
36 |
%ifdef FORMAT_COFF |
DATA |
|
SECTION .rodata |
|
|
%else |
|
|
SECTION .rodata align=16 |
|
|
%endif |
|
37 |
|
|
38 |
mmx_one: |
mmx_one: |
39 |
times 4 dw 1 |
times 4 dw 1 |
42 |
; divide by 2Q table |
; divide by 2Q table |
43 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
44 |
|
|
45 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
46 |
mmx_div: |
mmx_div: |
47 |
times 4 dw 65535 ; the div by 2 formula will overflow for the case |
times 4 dw 65535 ; the div by 2 formula will overflow for the case |
48 |
; quant=1 but we don't care much because quant=1 |
; quant=1 but we don't care much because quant=1 |
84 |
; saturation limits |
; saturation limits |
85 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
86 |
|
|
87 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
88 |
|
|
89 |
mmx_32767_minus_2047: |
mmx_32767_minus_2047: |
90 |
times 4 dw (32767-2047) |
times 4 dw (32767-2047) |
101 |
; rounding |
; rounding |
102 |
;============================================================================= |
;============================================================================= |
103 |
|
|
104 |
|
ALIGN SECTION_ALIGN |
105 |
|
|
106 |
mmx_rounding: |
mmx_rounding: |
107 |
dw (1<<13) |
dd (1<<13) |
108 |
dw 0 |
dd (1<<13) |
|
dw (1<<13) |
|
|
dw 0 |
|
109 |
|
|
110 |
;============================================================================= |
;============================================================================= |
111 |
; Code |
; Code |
112 |
;============================================================================= |
;============================================================================= |
113 |
|
|
114 |
SECTION .text |
SECTION .rotext align=SECTION_ALIGN |
115 |
|
|
116 |
cglobal quant_mpeg_intra_mmx |
cglobal quant_mpeg_intra_mmx |
117 |
cglobal quant_mpeg_inter_mmx |
cglobal quant_mpeg_inter_mmx |
120 |
|
|
121 |
|
|
122 |
%macro QUANT_MMX 1 |
%macro QUANT_MMX 1 |
123 |
movq mm0, [eax + 16*(%1)] ; data |
movq mm0, [_EAX + 16*(%1)] ; data |
124 |
movq mm2, [ecx + 16*(%1) + 128] ; intra_matrix_rec |
movq mm2, [TMP0 + 16*(%1) + 128] ; intra_matrix_rec |
125 |
movq mm4, [eax + 16*(%1) + 8] ; data |
movq mm4, [_EAX + 16*(%1) + 8] ; data |
126 |
movq mm6, [ecx + 16*(%1) + 128 + 8] ; intra_matrix_rec |
movq mm6, [TMP0 + 16*(%1) + 128 + 8] ; intra_matrix_rec |
127 |
|
|
128 |
movq mm1, mm0 |
movq mm1, mm0 |
129 |
movq mm5, mm4 |
movq mm5, mm4 |
154 |
packssdw mm2, mm0 |
packssdw mm2, mm0 |
155 |
packssdw mm6, mm4 |
packssdw mm6, mm4 |
156 |
|
|
157 |
movq [edi + 16*(%1)], mm2 |
movq [TMP1 + 16*(%1)], mm2 |
158 |
movq [edi + 16*(%1)+8], mm6 |
movq [TMP1 + 16*(%1)+8], mm6 |
159 |
%endmacro |
%endmacro |
160 |
|
|
161 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
168 |
; |
; |
169 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
170 |
|
|
171 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
172 |
quant_mpeg_intra_mmx: |
quant_mpeg_intra_mmx: |
173 |
|
|
174 |
push edi |
mov _EAX, prm2 ; data |
175 |
|
mov TMP0, prm5 ; mpeg_quant_matrices |
176 |
|
mov TMP1, prm1 ; coeff |
177 |
|
|
178 |
movq mm7, [mmx_rounding] |
movq mm7, [mmx_rounding] |
179 |
|
|
180 |
mov eax, [esp + 4 + 8] ; data |
; QUANT_MMX(0) |
181 |
mov ecx, [esp + 4 + 20] ; mpeg_quant_matrices |
; QUANT_MMX(1) |
182 |
mov edi, [esp + 4 + 4] ; coeff |
; QUANT_MMX(2) |
183 |
|
; QUANT_MMX(3) |
184 |
QUANT_MMX(0) |
; QUANT_MMX(4) |
185 |
QUANT_MMX(1) |
; QUANT_MMX(5) |
186 |
QUANT_MMX(2) |
; QUANT_MMX(6) |
187 |
QUANT_MMX(3) |
; QUANT_MMX(7) |
|
QUANT_MMX(4) |
|
|
QUANT_MMX(5) |
|
|
QUANT_MMX(6) |
|
|
QUANT_MMX(7) |
|
188 |
|
|
189 |
; calculate DC |
; calculate DC |
190 |
movsx eax, word [eax] ; data[0] |
; movsx _EAX, word [_EAX] ; data[0] |
191 |
mov ecx, [esp + 4 + 16] ; dcscalar |
; mov TMP0, prm4 ; dcscalar |
192 |
mov edx, eax |
; mov _EDX, _EAX |
193 |
mov edi, ecx |
; shr TMP0, 1 ; TMP0 = dcscalar/2 |
194 |
shr ecx, 1 ; ecx = dcscalar/2 |
; sar _EDX, 31 ; TMP1 = sign extend of _EAX (ready for division too) |
195 |
sar edx, 31 ; edx = sign extend of eax (ready for division too) |
; xor TMP0, _EDX ; adjust TMP0 according to the sign of data[0] |
196 |
xor ecx, edx ; adjust ecx according to the sign of data[0] |
; sub TMP0, _EDX |
197 |
sub ecx, edx |
; add _EAX, TMP0 |
198 |
add eax, ecx |
|
199 |
|
; mov TMP0, prm4 ; dcscalar |
200 |
mov ecx, [esp + 4 + 4] ; coeff again |
; idiv TMP0 ; _EAX = _EDX:_EAX / dcscalar |
|
idiv edi ; eax = edx:eax / dcscalar |
|
|
mov [ecx], ax ; coeff[0] = ax |
|
201 |
|
|
202 |
pop edi |
; mov _EDX, prm1 ; coeff again |
203 |
|
; mov word [_EDX], ax ; coeff[0] = ax |
204 |
|
|
205 |
xor eax, eax ; return(0); |
xor _EAX, _EAX ; return(0); |
206 |
ret |
ret |
207 |
ENDFUNC |
ENDFUNC |
208 |
|
|
216 |
; |
; |
217 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
218 |
|
|
219 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
220 |
quant_mpeg_inter_mmx: |
quant_mpeg_inter_mmx: |
221 |
|
|
222 |
push ecx |
mov TMP1, prm1 ; coeff |
223 |
push esi |
mov _EAX, prm3 ; quant |
224 |
push edi |
mov TMP0, prm4 ; mpeg_quant_matrices |
225 |
push ebx |
|
226 |
|
push _ESI |
227 |
mov edi, [esp + 16 + 4] ; coeff |
%ifdef ARCH_IS_X86_64 |
228 |
mov esi, [esp + 16 + 8] ; data |
mov _ESI, prm2 ; data |
229 |
mov eax, [esp + 16 + 12] ; quant |
%else |
230 |
mov ebx, [esp + 16 + 16] ; mpeg_quant_matrices |
mov _ESI, [_ESP + 4 + 8] ; data |
231 |
|
%endif |
232 |
|
|
233 |
xor ecx, ecx |
push _EBX |
234 |
|
xor _EBX, _EBX |
235 |
|
|
236 |
pxor mm5, mm5 ; sum |
pxor mm5, mm5 ; sum |
237 |
|
|
241 |
cmp al, 2 |
cmp al, 2 |
242 |
jz near .q2loop |
jz near .q2loop |
243 |
|
|
244 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
%ifdef ARCH_IS_X86_64 |
245 |
|
lea r9, [mmx_div] |
246 |
|
movq mm7, [r9 + _EAX * 8 - 8] |
247 |
|
%else |
248 |
|
movq mm7, [mmx_div + _EAX * 8 - 8] ; divider |
249 |
|
%endif |
250 |
|
|
251 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
252 |
.loop: |
.loop: |
253 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
254 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [_ESI + 8*_EBX + 8] ; |
255 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
256 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
257 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
262 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
263 |
psllw mm0, 4 |
psllw mm0, 4 |
264 |
psllw mm3, 4 |
psllw mm3, 4 |
265 |
movq mm2, [ebx + 512 + 8*ecx] |
movq mm2, [TMP0 + 512 + 8*_EBX] |
266 |
psrlw mm2, 1 |
psrlw mm2, 1 |
267 |
paddw mm0, mm2 |
paddw mm0, mm2 |
268 |
movq mm2, [ebx + 768 + ecx*8] |
movq mm2, [TMP0 + 768 + _EBX*8] |
269 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
270 |
movq mm2, [ebx + 512 + 8*ecx + 8] |
movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
271 |
psrlw mm2, 1 |
psrlw mm2, 1 |
272 |
paddw mm3, mm2 |
paddw mm3, mm2 |
273 |
movq mm2, [ebx + 768 + ecx*8 + 8] |
movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
274 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
275 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
276 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
282 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
283 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
284 |
psubw mm3, mm4 |
psubw mm3, mm4 |
285 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*_EBX], mm0 |
286 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*_EBX + 8], mm3 |
287 |
|
|
288 |
add ecx, 2 |
add _EBX, 2 |
289 |
cmp ecx, 16 |
cmp _EBX, 16 |
290 |
jnz near .loop |
jnz near .loop |
291 |
|
|
292 |
.done: |
.done: |
296 |
paddd mm0, mm5 |
paddd mm0, mm5 |
297 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
298 |
|
|
299 |
pop ebx |
pop _EBX |
300 |
pop edi |
pop _ESI |
|
pop esi |
|
|
pop ecx |
|
301 |
|
|
302 |
ret |
ret |
303 |
|
|
304 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
305 |
.q1loop: |
.q1loop: |
306 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
307 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [_ESI + 8*_EBX+ 8] |
308 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
309 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
310 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
315 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
316 |
psllw mm0, 4 |
psllw mm0, 4 |
317 |
psllw mm3, 4 |
psllw mm3, 4 |
318 |
movq mm2, [ebx + 512 + 8*ecx] |
movq mm2, [TMP0 + 512 + 8*_EBX] |
319 |
psrlw mm2, 1 |
psrlw mm2, 1 |
320 |
paddw mm0, mm2 |
paddw mm0, mm2 |
321 |
movq mm2, [ebx + 768 + ecx*8] |
movq mm2, [TMP0 + 768 + _EBX*8] |
322 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
323 |
movq mm2, [ebx + 512 + 8*ecx + 8] |
movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
324 |
psrlw mm2, 1 |
psrlw mm2, 1 |
325 |
paddw mm3, mm2 |
paddw mm3, mm2 |
326 |
movq mm2, [ebx + 768 + ecx*8 + 8] |
movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
327 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
328 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
329 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
333 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
334 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
335 |
psubw mm3, mm4 |
psubw mm3, mm4 |
336 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*_EBX], mm0 |
337 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*_EBX + 8], mm3 |
338 |
|
|
339 |
add ecx, 2 |
add _EBX, 2 |
340 |
cmp ecx, 16 |
cmp _EBX, 16 |
341 |
jnz near .q1loop |
jnz near .q1loop |
342 |
|
|
343 |
jmp .done |
jmp .done |
344 |
|
|
345 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
346 |
.q2loop: |
.q2loop: |
347 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
348 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [_ESI + 8*_EBX+ 8] |
349 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
350 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
351 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
356 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
357 |
psllw mm0, 4 |
psllw mm0, 4 |
358 |
psllw mm3, 4 |
psllw mm3, 4 |
359 |
movq mm2, [ebx + 512 + 8*ecx] |
movq mm2, [TMP0 + 512 + 8*_EBX] |
360 |
psrlw mm2, 1 |
psrlw mm2, 1 |
361 |
paddw mm0, mm2 |
paddw mm0, mm2 |
362 |
movq mm2, [ebx + 768 + ecx*8] |
movq mm2, [TMP0 + 768 + _EBX*8] |
363 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
364 |
movq mm2, [ebx + 512 + 8*ecx + 8] |
movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
365 |
psrlw mm2, 1 |
psrlw mm2, 1 |
366 |
paddw mm3, mm2 |
paddw mm3, mm2 |
367 |
movq mm2, [ebx + 768 + ecx*8 + 8] |
movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
368 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
369 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
370 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
374 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
375 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
376 |
psubw mm3, mm4 |
psubw mm3, mm4 |
377 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*_EBX], mm0 |
378 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*_EBX + 8], mm3 |
379 |
|
|
380 |
add ecx, 2 |
add _EBX, 2 |
381 |
cmp ecx, 16 |
cmp _EBX, 16 |
382 |
jnz near .q2loop |
jnz near .q2loop |
383 |
|
|
384 |
jmp .done |
jmp .done |
407 |
; The original loop is: |
; The original loop is: |
408 |
; |
; |
409 |
%if 0 |
%if 0 |
410 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = coeff[i] |
movq mm0, [TMP0+8*_EAX + 8*16] ; mm0 = coeff[i] |
411 |
pxor mm1, mm1 |
pxor mm1, mm1 |
412 |
pcmpgtw mm1, mm0 |
pcmpgtw mm1, mm0 |
413 |
pxor mm0, mm1 ; change sign if negative |
pxor mm0, mm1 ; change sign if negative |
414 |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
415 |
|
|
416 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
417 |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant. |
pmullw mm2, [_EBX + 8*_EAX + 8*16 ] ; matrix[i]*quant. |
418 |
|
|
419 |
movq mm6, mm2 |
movq mm6, mm2 |
420 |
pmulhw mm2, mm0 ; high of coeff*(matrix*quant) (should be 0 if no overflow) |
pmulhw mm2, mm0 ; high of coeff*(matrix*quant) (should be 0 if no overflow) |
428 |
por mm0, mm2 ; saturate to 2047 if needed |
por mm0, mm2 ; saturate to 2047 if needed |
429 |
pxor mm0, mm1 ; finish negating back |
pxor mm0, mm1 ; finish negating back |
430 |
|
|
431 |
movq [edx + 8*eax + 8*16], mm0 ; data[i] |
movq [TMP1 + 8*_EAX + 8*16], mm0 ; data[i] |
432 |
add eax, 1 |
add _EAX, 1 |
433 |
%endif |
%endif |
434 |
|
|
435 |
;******************************************************************** |
;******************************************************************** |
436 |
|
|
437 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
438 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
439 |
|
|
440 |
push ebx |
mov TMP1, prm1 ; data |
441 |
|
mov TMP0, prm2 ; coeff |
442 |
mov edx, [esp + 4 + 4] ; data |
mov _EAX, prm5 ; mpeg_quant_matrices |
443 |
mov ecx, [esp + 4 + 8] ; coeff |
|
444 |
mov eax, [esp + 4 + 12] ; quant |
push _EBX |
445 |
mov ebx, [esp + 4 + 20] ; mpeg_quant_matrices |
mov _EBX, _EAX |
446 |
|
%ifdef ARCH_IS_X86_64 |
447 |
|
mov _EAX, prm3 |
448 |
|
lea r9, [mmx_mul_quant] |
449 |
|
movq mm7, [r9 + _EAX*8 - 8] |
450 |
|
%else |
451 |
|
mov _EAX, [_ESP + 4 + 12] ; quant |
452 |
|
movq mm7, [mmx_mul_quant + _EAX*8 - 8] |
453 |
|
%endif |
454 |
|
|
455 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
mov _EAX, -16 ; to keep ALIGNed, we regularly process coeff[0] |
|
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
|
456 |
psllw mm7, 2 ; << 2. See comment. |
psllw mm7, 2 ; << 2. See comment. |
457 |
pxor mm6, mm6 ; this is a NOP |
pxor mm6, mm6 ; this is a NOP |
458 |
|
|
459 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
460 |
.loop: |
.loop: |
461 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
movq mm0, [TMP0+8*_EAX + 8*16] ; mm0 = c = coeff[i] |
462 |
movq mm3, [ecx+8*eax + 8*16 +8]; mm3 = c' = coeff[i+1] |
movq mm3, [TMP0+8*_EAX + 8*16 +8]; mm3 = c' = coeff[i+1] |
463 |
pxor mm1, mm1 |
pxor mm1, mm1 |
464 |
pxor mm4, mm4 |
pxor mm4, mm4 |
465 |
pcmpgtw mm1, mm0 ; mm1 = sgn(c) |
pcmpgtw mm1, mm0 ; mm1 = sgn(c) |
466 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
467 |
|
|
468 |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
469 |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant |
pmullw mm2, [_EBX + 8*_EAX + 8*16 ] ; matrix[i]*quant |
470 |
|
|
471 |
pxor mm0, mm1 ; negate if negative |
pxor mm0, mm1 ; negate if negative |
472 |
pxor mm3, mm4 ; negate if negative |
pxor mm3, mm4 ; negate if negative |
482 |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
483 |
movq mm5, mm7 ; mm2 = quant |
movq mm5, mm7 ; mm2 = quant |
484 |
|
|
485 |
pmullw mm5, [ebx + 8*eax + 8*16 +8] ; matrix[i+1]*quant |
pmullw mm5, [_EBX + 8*_EAX + 8*16 +8] ; matrix[i+1]*quant |
486 |
|
|
487 |
movq mm6, mm5 |
movq mm6, mm5 |
488 |
add eax,2 ; z-flag will be tested later |
add _EAX,2 ; z-flag will be tested later |
489 |
|
|
490 |
pmullw mm6, mm3 ; low of coeff*(matrix*quant) |
pmullw mm6, mm3 ; low of coeff*(matrix*quant) |
491 |
pmulhw mm3, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm3, mm5 ; high of coeff*(matrix*quant) |
507 |
psubw mm2, mm1 ; finish negating back |
psubw mm2, mm1 ; finish negating back |
508 |
psubw mm6, mm4 ; finish negating back |
psubw mm6, mm4 ; finish negating back |
509 |
|
|
510 |
movq [edx + 8*eax + 8*16 -2*8 ], mm2 ; data[i] |
movq [TMP1 + 8*_EAX + 8*16 -2*8 ], mm2 ; data[i] |
511 |
movq [edx + 8*eax + 8*16 -2*8 +8], mm6 ; data[i+1] |
movq [TMP1 + 8*_EAX + 8*16 -2*8 +8], mm6 ; data[i+1] |
512 |
|
|
513 |
jnz near .loop |
jnz near .loop |
514 |
|
|
515 |
|
pop _EBX |
516 |
|
|
517 |
; deal with DC |
; deal with DC |
518 |
movd mm0, [ecx] |
movd mm0, [TMP0] |
519 |
pmullw mm0, [esp + 4 + 16] ; dcscalar |
%ifdef ARCH_IS_X86_64 |
520 |
|
movq mm6, prm4 |
521 |
|
pmullw mm0, mm6 |
522 |
|
%else |
523 |
|
pmullw mm0, prm4 ; dcscalar |
524 |
|
%endif |
525 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
526 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
527 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
529 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
530 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
531 |
movd eax, mm0 |
movd eax, mm0 |
532 |
mov [edx], ax |
mov [TMP1], ax |
533 |
|
|
534 |
xor eax, eax |
xor _EAX, _EAX |
|
|
|
|
pop ebx |
|
535 |
|
|
536 |
ret |
ret |
537 |
ENDFUNC |
ENDFUNC |
550 |
; sgn(x) is the result of 'pcmpgtw 0,x': 0 if x>=0, -1 if x<0. |
; sgn(x) is the result of 'pcmpgtw 0,x': 0 if x>=0, -1 if x<0. |
551 |
; It's mixed with the extraction of the absolute value. |
; It's mixed with the extraction of the absolute value. |
552 |
|
|
553 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
554 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
555 |
|
|
|
push ebx |
|
556 |
|
|
557 |
mov edx, [esp + 4 + 4] ; data |
mov TMP1, prm1 ; data |
558 |
mov ecx, [esp + 4 + 8] ; coeff |
mov TMP0, prm2 ; coeff |
559 |
mov eax, [esp + 4 + 12] ; quant |
mov _EAX, prm3 ; quant |
560 |
mov ebx, [esp + 4 + 16] ; mpeg_quant_matrices |
|
561 |
|
push _EBX |
562 |
|
%ifdef ARCH_IS_X86_64 |
563 |
|
mov _EBX, prm4 |
564 |
|
lea r9, [mmx_mul_quant] |
565 |
|
movq mm7, [r9 + _EAX*8 - 8] |
566 |
|
%else |
567 |
|
mov _EBX, [_ESP + 4 + 16] ; mpeg_quant_matrices |
568 |
|
movq mm7, [mmx_mul_quant + _EAX*8 - 8] |
569 |
|
%endif |
570 |
|
|
571 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
mov _EAX, -16 |
|
mov eax, -16 |
|
572 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
573 |
pxor mm6, mm6 ; mismatch sum |
pxor mm6, mm6 ; mismatch sum |
574 |
|
|
575 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
576 |
.loop: |
.loop: |
577 |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
movq mm0, [TMP0+8*_EAX + 8*16 ] ; mm0 = coeff[i] |
578 |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
movq mm2, [TMP0+8*_EAX + 8*16 +8] ; mm2 = coeff[i+1] |
579 |
add eax, 2 |
add _EAX, 2 |
580 |
|
|
581 |
pxor mm1, mm1 |
pxor mm1, mm1 |
582 |
pxor mm3, mm3 |
pxor mm3, mm3 |
601 |
; we're short on register, here. Poor pairing... |
; we're short on register, here. Poor pairing... |
602 |
|
|
603 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
604 |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8] |
pmullw mm4, [_EBX + 512 + 8*_EAX + 8*16 -2*8] |
605 |
movq mm5, mm4 |
movq mm5, mm4 |
606 |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
607 |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
608 |
|
|
609 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
610 |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8 + 8] |
pmullw mm4, [_EBX + 512 + 8*_EAX + 8*16 -2*8 + 8] |
611 |
|
|
612 |
pcmpgtw mm5, [zero] |
pcmpgtw mm5, [zero] |
613 |
paddusw mm0, mm5 |
paddusw mm0, mm5 |
628 |
psubw mm2, mm3 ; finish restoring sign |
psubw mm2, mm3 ; finish restoring sign |
629 |
|
|
630 |
pxor mm6, mm0 ; mismatch control |
pxor mm6, mm0 ; mismatch control |
631 |
movq [edx + 8*eax + 8*16 -2*8 ], mm0 ; data[i] |
movq [TMP1 + 8*_EAX + 8*16 -2*8 ], mm0 ; data[i] |
632 |
pxor mm6, mm2 ; mismatch control |
pxor mm6, mm2 ; mismatch control |
633 |
movq [edx + 8*eax + 8*16 -2*8 +8], mm2 ; data[i+1] |
movq [TMP1 + 8*_EAX + 8*16 -2*8 +8], mm2 ; data[i+1] |
634 |
|
|
635 |
jnz near .loop |
jnz near .loop |
636 |
|
|
646 |
pxor mm6, mm1 |
pxor mm6, mm1 |
647 |
pxor mm6, mm2 |
pxor mm6, mm2 |
648 |
movd eax, mm6 |
movd eax, mm6 |
649 |
and eax, 1 |
and _EAX, 1 |
650 |
xor eax, 1 |
xor _EAX, 1 |
651 |
xor word [edx + 2*63], ax |
xor word [TMP1 + 2*63], ax |
652 |
|
|
653 |
xor eax, eax |
xor _EAX, _EAX |
654 |
|
|
655 |
pop ebx |
pop _EBX |
656 |
|
|
657 |
ret |
ret |
658 |
ENDFUNC |
ENDFUNC |