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 |
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_mpeg_mmx.asm,v 1.5 2004-08-29 10:02:38 edgomez Exp $ |
; * $Id: quantize_mpeg_mmx.asm,v 1.15 2008-12-04 18:30:36 Isibaar Exp $ |
25 |
; * |
; * |
26 |
; *************************************************************************/ |
; *************************************************************************/ |
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 |
|
|
%else |
|
|
global _%1 |
|
|
%define %1 _%1 |
|
|
%endif |
|
|
%else |
|
|
%ifdef MARK_FUNCS |
|
|
global %1:function %1.endfunc-%1 |
|
|
%else |
|
|
global %1 |
|
|
%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) |
98 |
times 4 dw 0 |
times 4 dw 0 |
99 |
|
|
100 |
;============================================================================= |
;============================================================================= |
101 |
|
; rounding |
102 |
|
;============================================================================= |
103 |
|
|
104 |
|
ALIGN SECTION_ALIGN |
105 |
|
|
106 |
|
mmx_rounding: |
107 |
|
dd (1<<13) |
108 |
|
dd (1<<13) |
109 |
|
|
110 |
|
;============================================================================= |
111 |
; Code |
; Code |
112 |
;============================================================================= |
;============================================================================= |
113 |
|
|
114 |
SECTION .text |
TEXT |
115 |
|
|
116 |
cglobal quant_mpeg_intra_mmx |
cglobal quant_mpeg_intra_mmx |
117 |
cglobal quant_mpeg_inter_mmx |
cglobal quant_mpeg_inter_mmx |
118 |
cglobal dequant_mpeg_intra_mmx |
cglobal dequant_mpeg_intra_mmx |
119 |
cglobal dequant_mpeg_inter_mmx |
cglobal dequant_mpeg_inter_mmx |
120 |
|
|
121 |
|
|
122 |
|
%macro QUANT_MMX 1 |
123 |
|
movq mm0, [_EAX + 16*(%1)] ; data |
124 |
|
movq mm2, [TMP0 + 16*(%1) + 128] ; intra_matrix_rec |
125 |
|
movq mm4, [_EAX + 16*(%1) + 8] ; data |
126 |
|
movq mm6, [TMP0 + 16*(%1) + 128 + 8] ; intra_matrix_rec |
127 |
|
|
128 |
|
movq mm1, mm0 |
129 |
|
movq mm5, mm4 |
130 |
|
|
131 |
|
pmullw mm0, mm2 ; low results |
132 |
|
pmulhw mm1, mm2 ; high results |
133 |
|
pmullw mm4, mm6 ; low results |
134 |
|
pmulhw mm5, mm6 ; high results |
135 |
|
|
136 |
|
movq mm2, mm0 |
137 |
|
movq mm6, mm4 |
138 |
|
|
139 |
|
punpckhwd mm0, mm1 |
140 |
|
punpcklwd mm2, mm1 |
141 |
|
punpckhwd mm4, mm5 |
142 |
|
punpcklwd mm6, mm5 |
143 |
|
|
144 |
|
paddd mm2, mm7 |
145 |
|
paddd mm0, mm7 |
146 |
|
paddd mm6, mm7 |
147 |
|
paddd mm4, mm7 |
148 |
|
|
149 |
|
psrad mm2, 14 |
150 |
|
psrad mm0, 14 |
151 |
|
psrad mm6, 14 |
152 |
|
psrad mm4, 14 |
153 |
|
|
154 |
|
packssdw mm2, mm0 |
155 |
|
packssdw mm6, mm4 |
156 |
|
|
157 |
|
movq [TMP1 + 16*(%1)], mm2 |
158 |
|
movq [TMP1 + 16*(%1)+8], mm6 |
159 |
|
%endmacro |
160 |
|
|
161 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
162 |
; |
; |
163 |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
168 |
; |
; |
169 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
170 |
|
|
171 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
172 |
quant_mpeg_intra_mmx: |
quant_mpeg_intra_mmx: |
173 |
|
|
174 |
push ecx |
mov _EAX, prm2 ; data |
175 |
push esi |
mov TMP0, prm5 ; mpeg_quant_matrices |
176 |
push edi |
mov TMP1, prm1 ; coeff |
177 |
push ebx |
|
178 |
|
movq mm7, [mmx_rounding] |
179 |
mov edi, [esp + 16 + 4] ; coeff |
|
180 |
mov esi, [esp + 16 + 8] ; data |
QUANT_MMX(0) |
181 |
mov eax, [esp + 16 + 12] ; quant |
QUANT_MMX(1) |
182 |
mov ebx, [esp + 16 + 20] ; mpeg_quant_matrices |
QUANT_MMX(2) |
183 |
|
QUANT_MMX(3) |
184 |
|
QUANT_MMX(4) |
185 |
|
QUANT_MMX(5) |
186 |
|
QUANT_MMX(6) |
187 |
|
QUANT_MMX(7) |
188 |
|
|
189 |
|
; calculate DC |
190 |
|
movsx _EAX, word [_EAX] ; data[0] |
191 |
|
mov TMP0, prm4 ; dcscalar |
192 |
|
mov _EDX, _EAX |
193 |
|
shr TMP0, 1 ; TMP0 = dcscalar/2 |
194 |
|
sar _EDX, 31 ; TMP1 = sign extend of _EAX (ready for division too) |
195 |
|
|
196 |
|
xor TMP0, _EDX ; adjust TMP0 according to the sign of data[0] |
197 |
|
sub TMP0, _EDX |
198 |
|
add _EAX, TMP0 |
199 |
|
|
200 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
mov TMP0, prm4 ; dcscalar |
201 |
|
idiv TMP0 ; _EAX = _EDX:_EAX / dcscalar |
202 |
|
|
203 |
xor ecx, ecx |
mov _EDX, prm1 ; coeff again |
204 |
cmp al, 1 |
mov word [_EDX], ax ; coeff[0] = ax |
|
jz near .q1loop |
|
205 |
|
|
206 |
cmp al, 2 |
xor _EAX, _EAX ; return(0); |
|
jz near .q2loop |
|
|
|
|
|
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
|
|
|
|
|
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 ; |
|
|
psllw mm0, 4 ; level << 4 |
|
|
psllw mm3, 4 |
|
|
movq mm2, [ebx + 8*ecx] |
|
|
psrlw mm2, 1 ; intra_matrix[i]>>1 |
|
|
paddw mm0, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
movq mm2, [ebx + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
|
pmulhw mm3, mm2 |
|
|
paddw mm0, mm5 ; + quantd |
|
|
paddw mm3, mm5 |
|
|
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
|
|
pmulhw mm3, mm7 ; |
|
|
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
|
|
psrlw mm3, 1 |
|
|
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 near .loop |
|
|
|
|
|
.done |
|
|
; caclulate data[0] // (int32_t)dcscalar) |
|
|
mov ecx, [esp + 16 + 16] ; dcscalar |
|
|
mov edx, ecx |
|
|
movsx eax, word [esi] ; data[0] |
|
|
shr edx, 1 ; edx = dcscalar /2 |
|
|
cmp eax, 0 |
|
|
jg .gtzero |
|
|
|
|
|
sub eax, edx |
|
|
jmp short .mul |
|
|
.gtzero |
|
|
add eax, edx |
|
|
.mul |
|
|
cdq ; expand eax -> edx:eax |
|
|
idiv ecx ; eax = edx:eax / dcscalar |
|
|
|
|
|
mov [edi], ax ; coeff[0] = ax |
|
|
|
|
|
pop ebx |
|
|
pop edi |
|
|
pop esi |
|
|
pop ecx |
|
|
|
|
|
xor eax, eax ; return(0); |
|
207 |
ret |
ret |
208 |
|
ENDFUNC |
209 |
|
|
|
ALIGN 16 |
|
|
.q1loop |
|
|
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 ; |
|
|
psllw mm0, 4 |
|
|
psllw mm3, 4 |
|
|
movq mm2, [ebx + 8*ecx] |
|
|
psrlw mm2, 1 |
|
|
paddw mm0, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
movq mm2, [ebx + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
|
pmulhw mm3, mm2 |
|
|
paddw mm0, mm5 |
|
|
paddw mm3, mm5 |
|
|
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
|
|
psrlw mm3, 1 ; |
|
|
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 near .q1loop |
|
|
jmp near .done |
|
|
|
|
|
|
|
|
ALIGN 16 |
|
|
.q2loop |
|
|
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 ; |
|
|
psllw mm0, 4 |
|
|
psllw mm3, 4 |
|
|
movq mm2, [ebx + 8*ecx] |
|
|
psrlw mm2, 1 |
|
|
paddw mm0, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
movq mm2, [ebx + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
|
pmulhw mm3, mm2 |
|
|
paddw mm0, mm5 |
|
|
paddw mm3, mm5 |
|
|
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
|
|
psrlw mm3, 2 ; |
|
|
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 near .q2loop |
|
|
jmp near .done |
|
|
.endfunc |
|
210 |
|
|
211 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
212 |
; |
; |
217 |
; |
; |
218 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
219 |
|
|
220 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
221 |
quant_mpeg_inter_mmx: |
quant_mpeg_inter_mmx: |
222 |
|
|
223 |
push ecx |
mov TMP1, prm1 ; coeff |
224 |
push esi |
mov _EAX, prm3 ; quant |
225 |
push edi |
mov TMP0, prm4 ; mpeg_quant_matrices |
226 |
push ebx |
|
227 |
|
push _ESI |
228 |
mov edi, [esp + 16 + 4] ; coeff |
%ifdef ARCH_IS_X86_64 |
229 |
mov esi, [esp + 16 + 8] ; data |
mov _ESI, prm2 ; data |
230 |
mov eax, [esp + 16 + 12] ; quant |
%else |
231 |
mov ebx, [esp + 16 + 16] ; mpeg_quant_matrices |
mov _ESI, [_ESP + 4 + 8] ; data |
232 |
|
%endif |
233 |
|
|
234 |
xor ecx, ecx |
push _EBX |
235 |
|
xor _EBX, _EBX |
236 |
|
|
237 |
pxor mm5, mm5 ; sum |
pxor mm5, mm5 ; sum |
238 |
|
|
242 |
cmp al, 2 |
cmp al, 2 |
243 |
jz near .q2loop |
jz near .q2loop |
244 |
|
|
245 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
%ifdef ARCH_IS_X86_64 |
246 |
|
lea r9, [mmx_div] |
247 |
|
movq mm7, [r9 + _EAX * 8 - 8] |
248 |
|
%else |
249 |
|
movq mm7, [mmx_div + _EAX * 8 - 8] ; divider |
250 |
|
%endif |
251 |
|
|
252 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
253 |
.loop |
.loop: |
254 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
255 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [_ESI + 8*_EBX + 8] ; |
256 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
257 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
258 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
263 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
264 |
psllw mm0, 4 |
psllw mm0, 4 |
265 |
psllw mm3, 4 |
psllw mm3, 4 |
266 |
movq mm2, [ebx + 512 + 8*ecx] |
movq mm2, [TMP0 + 512 + 8*_EBX] |
267 |
psrlw mm2, 1 |
psrlw mm2, 1 |
268 |
paddw mm0, mm2 |
paddw mm0, mm2 |
269 |
movq mm2, [ebx + 768 + ecx*8] |
movq mm2, [TMP0 + 768 + _EBX*8] |
270 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
271 |
movq mm2, [ebx + 512 + 8*ecx + 8] |
movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
272 |
psrlw mm2, 1 |
psrlw mm2, 1 |
273 |
paddw mm3, mm2 |
paddw mm3, mm2 |
274 |
movq mm2, [ebx + 768 + ecx*8 + 8] |
movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
275 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
276 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
277 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
283 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
284 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
285 |
psubw mm3, mm4 |
psubw mm3, mm4 |
286 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*_EBX], mm0 |
287 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*_EBX + 8], mm3 |
288 |
|
|
289 |
add ecx, 2 |
add _EBX, 2 |
290 |
cmp ecx, 16 |
cmp _EBX, 16 |
291 |
jnz near .loop |
jnz near .loop |
292 |
|
|
293 |
.done |
.done: |
294 |
pmaddwd mm5, [mmx_one] |
pmaddwd mm5, [mmx_one] |
295 |
movq mm0, mm5 |
movq mm0, mm5 |
296 |
psrlq mm5, 32 |
psrlq mm5, 32 |
297 |
paddd mm0, mm5 |
paddd mm0, mm5 |
298 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
299 |
|
|
300 |
pop ebx |
pop _EBX |
301 |
pop edi |
pop _ESI |
|
pop esi |
|
|
pop ecx |
|
302 |
|
|
303 |
ret |
ret |
304 |
|
|
305 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
306 |
.q1loop |
.q1loop: |
307 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
308 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [_ESI + 8*_EBX+ 8] |
309 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
310 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
311 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
316 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
317 |
psllw mm0, 4 |
psllw mm0, 4 |
318 |
psllw mm3, 4 |
psllw mm3, 4 |
319 |
movq mm2, [ebx + 512 + 8*ecx] |
movq mm2, [TMP0 + 512 + 8*_EBX] |
320 |
psrlw mm2, 1 |
psrlw mm2, 1 |
321 |
paddw mm0, mm2 |
paddw mm0, mm2 |
322 |
movq mm2, [ebx + 768 + ecx*8] |
movq mm2, [TMP0 + 768 + _EBX*8] |
323 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
324 |
movq mm2, [ebx + 512 + 8*ecx + 8] |
movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
325 |
psrlw mm2, 1 |
psrlw mm2, 1 |
326 |
paddw mm3, mm2 |
paddw mm3, mm2 |
327 |
movq mm2, [ebx + 768 + ecx*8 + 8] |
movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
328 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
329 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
330 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
334 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
335 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
336 |
psubw mm3, mm4 |
psubw mm3, mm4 |
337 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*_EBX], mm0 |
338 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*_EBX + 8], mm3 |
339 |
|
|
340 |
add ecx, 2 |
add _EBX, 2 |
341 |
cmp ecx, 16 |
cmp _EBX, 16 |
342 |
jnz near .q1loop |
jnz near .q1loop |
343 |
|
|
344 |
jmp .done |
jmp .done |
345 |
|
|
346 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
347 |
.q2loop |
.q2loop: |
348 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [_ESI + 8*_EBX] ; mm0 = [1st] |
349 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [_ESI + 8*_EBX+ 8] |
350 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
351 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
352 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
357 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
358 |
psllw mm0, 4 |
psllw mm0, 4 |
359 |
psllw mm3, 4 |
psllw mm3, 4 |
360 |
movq mm2, [ebx + 512 + 8*ecx] |
movq mm2, [TMP0 + 512 + 8*_EBX] |
361 |
psrlw mm2, 1 |
psrlw mm2, 1 |
362 |
paddw mm0, mm2 |
paddw mm0, mm2 |
363 |
movq mm2, [ebx + 768 + ecx*8] |
movq mm2, [TMP0 + 768 + _EBX*8] |
364 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
365 |
movq mm2, [ebx + 512 + 8*ecx + 8] |
movq mm2, [TMP0 + 512 + 8*_EBX + 8] |
366 |
psrlw mm2, 1 |
psrlw mm2, 1 |
367 |
paddw mm3, mm2 |
paddw mm3, mm2 |
368 |
movq mm2, [ebx + 768 + ecx*8 + 8] |
movq mm2, [TMP0 + 768 + _EBX*8 + 8] |
369 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
370 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
371 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
375 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
376 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
377 |
psubw mm3, mm4 |
psubw mm3, mm4 |
378 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*_EBX], mm0 |
379 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*_EBX + 8], mm3 |
380 |
|
|
381 |
add ecx, 2 |
add _EBX, 2 |
382 |
cmp ecx, 16 |
cmp _EBX, 16 |
383 |
jnz near .q2loop |
jnz near .q2loop |
384 |
|
|
385 |
jmp .done |
jmp .done |
386 |
.endfunc |
ENDFUNC |
387 |
|
|
388 |
|
|
389 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
408 |
; The original loop is: |
; The original loop is: |
409 |
; |
; |
410 |
%if 0 |
%if 0 |
411 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = coeff[i] |
movq mm0, [TMP0+8*_EAX + 8*16] ; mm0 = coeff[i] |
412 |
pxor mm1, mm1 |
pxor mm1, mm1 |
413 |
pcmpgtw mm1, mm0 |
pcmpgtw mm1, mm0 |
414 |
pxor mm0, mm1 ; change sign if negative |
pxor mm0, mm1 ; change sign if negative |
415 |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
416 |
|
|
417 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
418 |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant. |
pmullw mm2, [_EBX + 8*_EAX + 8*16 ] ; matrix[i]*quant. |
419 |
|
|
420 |
movq mm6, mm2 |
movq mm6, mm2 |
421 |
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) |
429 |
por mm0, mm2 ; saturate to 2047 if needed |
por mm0, mm2 ; saturate to 2047 if needed |
430 |
pxor mm0, mm1 ; finish negating back |
pxor mm0, mm1 ; finish negating back |
431 |
|
|
432 |
movq [edx + 8*eax + 8*16], mm0 ; data[i] |
movq [TMP1 + 8*_EAX + 8*16], mm0 ; data[i] |
433 |
add eax, 1 |
add _EAX, 1 |
434 |
%endif |
%endif |
435 |
|
|
436 |
;******************************************************************** |
;******************************************************************** |
437 |
|
|
438 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
439 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
440 |
|
|
441 |
push ebx |
mov TMP1, prm1 ; data |
442 |
|
mov TMP0, prm2 ; coeff |
443 |
mov edx, [esp + 4 + 4] ; data |
mov _EAX, prm5 ; mpeg_quant_matrices |
444 |
mov ecx, [esp + 4 + 8] ; coeff |
|
445 |
mov eax, [esp + 4 + 12] ; quant |
push _EBX |
446 |
mov ebx, [esp + 4 + 20] ; mpeg_quant_matrices |
mov _EBX, _EAX |
447 |
|
%ifdef ARCH_IS_X86_64 |
448 |
|
mov _EAX, prm3 |
449 |
|
lea prm1, [mmx_mul_quant] |
450 |
|
movq mm7, [prm1 + _EAX*8 - 8] |
451 |
|
%else |
452 |
|
mov _EAX, [_ESP + 4 + 12] ; quant |
453 |
|
movq mm7, [mmx_mul_quant + _EAX*8 - 8] |
454 |
|
%endif |
455 |
|
|
456 |
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] |
|
457 |
psllw mm7, 2 ; << 2. See comment. |
psllw mm7, 2 ; << 2. See comment. |
458 |
pxor mm6, mm6 ; this is a NOP |
pxor mm6, mm6 ; this is a NOP |
459 |
|
|
460 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
461 |
.loop |
.loop: |
462 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
movq mm0, [TMP0+8*_EAX + 8*16] ; mm0 = c = coeff[i] |
463 |
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] |
464 |
pxor mm1, mm1 |
pxor mm1, mm1 |
465 |
pxor mm4, mm4 |
pxor mm4, mm4 |
466 |
pcmpgtw mm1, mm0 ; mm1 = sgn(c) |
pcmpgtw mm1, mm0 ; mm1 = sgn(c) |
467 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
468 |
|
|
469 |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
470 |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant |
pmullw mm2, [_EBX + 8*_EAX + 8*16 ] ; matrix[i]*quant |
471 |
|
|
472 |
pxor mm0, mm1 ; negate if negative |
pxor mm0, mm1 ; negate if negative |
473 |
pxor mm3, mm4 ; negate if negative |
pxor mm3, mm4 ; negate if negative |
483 |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
484 |
movq mm5, mm7 ; mm2 = quant |
movq mm5, mm7 ; mm2 = quant |
485 |
|
|
486 |
pmullw mm5, [ebx + 8*eax + 8*16 +8] ; matrix[i+1]*quant |
pmullw mm5, [_EBX + 8*_EAX + 8*16 +8] ; matrix[i+1]*quant |
487 |
|
|
488 |
movq mm6, mm5 |
movq mm6, mm5 |
489 |
add eax,2 ; z-flag will be tested later |
add _EAX,2 ; z-flag will be tested later |
490 |
|
|
491 |
pmullw mm6, mm3 ; low of coeff*(matrix*quant) |
pmullw mm6, mm3 ; low of coeff*(matrix*quant) |
492 |
pmulhw mm3, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm3, mm5 ; high of coeff*(matrix*quant) |
508 |
psubw mm2, mm1 ; finish negating back |
psubw mm2, mm1 ; finish negating back |
509 |
psubw mm6, mm4 ; finish negating back |
psubw mm6, mm4 ; finish negating back |
510 |
|
|
511 |
movq [edx + 8*eax + 8*16 -2*8 ], mm2 ; data[i] |
movq [TMP1 + 8*_EAX + 8*16 -2*8 ], mm2 ; data[i] |
512 |
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] |
513 |
|
|
514 |
jnz near .loop |
jnz near .loop |
515 |
|
|
516 |
|
pop _EBX |
517 |
|
|
518 |
; deal with DC |
; deal with DC |
519 |
movd mm0, [ecx] |
movd mm0, [TMP0] |
520 |
pmullw mm0, [esp + 4 + 16] ; dcscalar |
%ifdef ARCH_IS_X86_64 |
521 |
|
movq mm6, prm4 |
522 |
|
pmullw mm0, mm6 |
523 |
|
%else |
524 |
|
pmullw mm0, prm4 ; dcscalar |
525 |
|
%endif |
526 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
527 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
528 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
530 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
531 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
532 |
movd eax, mm0 |
movd eax, mm0 |
533 |
mov [edx], ax |
mov [TMP1], ax |
|
|
|
|
xor eax, eax |
|
534 |
|
|
535 |
pop ebx |
xor _EAX, _EAX |
536 |
|
|
537 |
ret |
ret |
538 |
.endfunc |
ENDFUNC |
539 |
|
|
540 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
541 |
; |
; |
551 |
; 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. |
552 |
; It's mixed with the extraction of the absolute value. |
; It's mixed with the extraction of the absolute value. |
553 |
|
|
554 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
555 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
556 |
|
|
|
push ebx |
|
557 |
|
|
558 |
mov edx, [esp + 4 + 4] ; data |
mov TMP1, prm1 ; data |
559 |
mov ecx, [esp + 4 + 8] ; coeff |
mov TMP0, prm2 ; coeff |
560 |
mov eax, [esp + 4 + 12] ; quant |
mov _EAX, prm3 ; quant |
561 |
mov ebx, [esp + 4 + 16] ; mpeg_quant_matrices |
|
562 |
|
push _EBX |
563 |
|
%ifdef ARCH_IS_X86_64 |
564 |
|
mov _EBX, prm4 |
565 |
|
lea r9, [mmx_mul_quant] |
566 |
|
movq mm7, [r9 + _EAX*8 - 8] |
567 |
|
%else |
568 |
|
mov _EBX, [_ESP + 4 + 16] ; mpeg_quant_matrices |
569 |
|
movq mm7, [mmx_mul_quant + _EAX*8 - 8] |
570 |
|
%endif |
571 |
|
|
572 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
mov _EAX, -16 |
|
mov eax, -16 |
|
573 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
574 |
pxor mm6, mm6 ; mismatch sum |
pxor mm6, mm6 ; mismatch sum |
575 |
|
|
576 |
ALIGN 16 |
ALIGN SECTION_ALIGN |
577 |
.loop |
.loop: |
578 |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
movq mm0, [TMP0+8*_EAX + 8*16 ] ; mm0 = coeff[i] |
579 |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
movq mm2, [TMP0+8*_EAX + 8*16 +8] ; mm2 = coeff[i+1] |
580 |
add eax, 2 |
add _EAX, 2 |
581 |
|
|
582 |
pxor mm1, mm1 |
pxor mm1, mm1 |
583 |
pxor mm3, mm3 |
pxor mm3, mm3 |
602 |
; we're short on register, here. Poor pairing... |
; we're short on register, here. Poor pairing... |
603 |
|
|
604 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
605 |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8] |
pmullw mm4, [_EBX + 512 + 8*_EAX + 8*16 -2*8] |
606 |
movq mm5, mm4 |
movq mm5, mm4 |
607 |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
608 |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
609 |
|
|
610 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
611 |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8 + 8] |
pmullw mm4, [_EBX + 512 + 8*_EAX + 8*16 -2*8 + 8] |
612 |
|
|
613 |
pcmpgtw mm5, [zero] |
pcmpgtw mm5, [zero] |
614 |
paddusw mm0, mm5 |
paddusw mm0, mm5 |
629 |
psubw mm2, mm3 ; finish restoring sign |
psubw mm2, mm3 ; finish restoring sign |
630 |
|
|
631 |
pxor mm6, mm0 ; mismatch control |
pxor mm6, mm0 ; mismatch control |
632 |
movq [edx + 8*eax + 8*16 -2*8 ], mm0 ; data[i] |
movq [TMP1 + 8*_EAX + 8*16 -2*8 ], mm0 ; data[i] |
633 |
pxor mm6, mm2 ; mismatch control |
pxor mm6, mm2 ; mismatch control |
634 |
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] |
635 |
|
|
636 |
jnz near .loop |
jnz near .loop |
637 |
|
|
647 |
pxor mm6, mm1 |
pxor mm6, mm1 |
648 |
pxor mm6, mm2 |
pxor mm6, mm2 |
649 |
movd eax, mm6 |
movd eax, mm6 |
650 |
and eax, 1 |
and _EAX, 1 |
651 |
xor eax, 1 |
xor _EAX, 1 |
652 |
xor word [edx + 2*63], ax |
xor word [TMP1 + 2*63], ax |
653 |
|
|
654 |
xor eax, eax |
xor _EAX, _EAX |
655 |
|
|
656 |
pop ebx |
pop _EBX |
657 |
|
|
658 |
ret |
ret |
659 |
.endfunc |
ENDFUNC |
660 |
|
|
661 |
|
|
662 |
|
%ifidn __OUTPUT_FORMAT__,elf |
663 |
|
section ".note.GNU-stack" noalloc noexec nowrite progbits |
664 |
|
%endif |
665 |
|
|