1 |
;/****************************************************************************** |
;/************************************************************************** |
2 |
; * * |
; * |
3 |
; * This file is part of XviD, a free MPEG-4 video encoder/decoder * |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * * |
; * - 3dne Quantization/Dequantization - |
5 |
; * XviD is an implementation of a part of one or more MPEG-4 Video tools * |
; * |
6 |
; * as specified in ISO/IEC 14496-2 standard. Those intending to use this * |
; * Copyright (C) 2002-2003 Peter Ross <pross@xvid.org> |
7 |
; * software module in hardware or software products are advised that its * |
; * 2002-2003 Michael Militzer <isibaar@xvid.org> |
8 |
; * use may infringe existing patents or copyrights, and any such use * |
; * 2002-2003 Pascal Massimino <skal@planet-d.net> |
9 |
; * would be at such party's own risk. The original developer of this * |
; * |
10 |
; * software module and his/her company, and subsequent editors and their * |
; * This program is free software ; you can redistribute it and/or modify |
11 |
; * companies, will have no liability for use of this software or * |
; * it under the terms of the GNU General Public License as published by |
12 |
; * modifications or derivatives thereof. * |
; * the Free Software Foundation ; either version 2 of the License, or |
13 |
; * * |
; * (at your option) any later version. |
14 |
; * XviD is free software; you can redistribute it and/or modify it * |
; * |
15 |
; * under the terms of the GNU General Public License as published by * |
; * This program is distributed in the hope that it will be useful, |
16 |
; * the Free Software Foundation; either version 2 of the License, or * |
; * but WITHOUT ANY WARRANTY ; without even the implied warranty of |
17 |
; * (at your option) any later version. * |
; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
18 |
; * * |
; * GNU General Public License for more details. |
19 |
; * XviD is distributed in the hope that it will be useful, but * |
; * |
20 |
; * WITHOUT ANY WARRANTY; without even the implied warranty of * |
; * You should have received a copy of the GNU General Public License |
21 |
; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
; * along with this program ; if not, write to the Free Software |
22 |
; * GNU General Public License for more details. * |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
; * * |
; * |
24 |
; * You should have received a copy of the GNU General Public License * |
; * $Id$ |
25 |
; * along with this program; if not, write to the Free Software * |
; * |
26 |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * |
; *************************************************************************/ |
|
; * * |
|
|
; ******************************************************************************/ |
|
|
; |
|
|
;/****************************************************************************** |
|
|
; * * |
|
|
; * quantize4.asm, MMX optimized MPEG quantization/dequantization * |
|
|
; * * |
|
|
; * Copyright (C) 2002 - Peter Ross <pross@cs.rmit.edu.au> * |
|
|
; * Copyright (C) 2002 - Michael Militzer <isibaar@xvid.org> * |
|
|
; * * |
|
|
; * For more information visit the XviD homepage: http://www.xvid.org * |
|
|
; * * |
|
|
; ******************************************************************************/ |
|
|
; |
|
|
;/****************************************************************************** |
|
|
; * * |
|
|
; * Revision history: * |
|
|
; * * |
|
|
; * 14.06.2002 mmx dequant4_* funcs revamped -Skal- * |
|
|
; * 22.01.2002 initial version * |
|
|
; * * |
|
|
; ******************************************************************************/ |
|
|
|
|
|
; data/text alignment |
|
|
%define ALIGN 8 |
|
27 |
|
|
28 |
%define SATURATE |
%define SATURATE |
29 |
|
|
30 |
bits 32 |
BITS 32 |
|
|
|
|
%ifdef FORMAT_COFF |
|
|
SECTION .data data |
|
|
%else |
|
|
SECTION .data data align=8 |
|
|
%endif |
|
|
|
|
31 |
|
|
32 |
%macro cglobal 1 |
%macro cglobal 1 |
33 |
%ifdef PREFIX |
%ifdef PREFIX |
34 |
|
%ifdef MARK_FUNCS |
35 |
|
global _%1:function %1.endfunc-%1 |
36 |
|
%define %1 _%1:function %1.endfunc-%1 |
37 |
|
%define ENDFUNC .endfunc |
38 |
|
%else |
39 |
global _%1 |
global _%1 |
40 |
%define %1 _%1 |
%define %1 _%1 |
41 |
|
%define ENDFUNC |
42 |
|
%endif |
43 |
|
%else |
44 |
|
%ifdef MARK_FUNCS |
45 |
|
global %1:function %1.endfunc-%1 |
46 |
|
%define ENDFUNC .endfunc |
47 |
%else |
%else |
48 |
global %1 |
global %1 |
49 |
|
%define ENDFUNC |
50 |
|
%endif |
51 |
%endif |
%endif |
52 |
%endmacro |
%endmacro |
53 |
|
|
60 |
%endif |
%endif |
61 |
%endmacro |
%endmacro |
62 |
|
|
63 |
mmx_one times 4 dw 1 |
;============================================================================= |
64 |
|
; Local data (Read Only) |
65 |
;=========================================================================== |
;============================================================================= |
|
; |
|
|
; divide by 2Q table |
|
|
; |
|
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_DIV 1 |
|
|
times 4 dw (1 << 17) / (%1 * 2) + 1 |
|
|
%endmacro |
|
|
|
|
|
align ALIGN |
|
|
mmx_div |
|
|
MMX_DIV 1 |
|
|
MMX_DIV 2 |
|
|
MMX_DIV 3 |
|
|
MMX_DIV 4 |
|
|
MMX_DIV 5 |
|
|
MMX_DIV 6 |
|
|
MMX_DIV 7 |
|
|
MMX_DIV 8 |
|
|
MMX_DIV 9 |
|
|
MMX_DIV 10 |
|
|
MMX_DIV 11 |
|
|
MMX_DIV 12 |
|
|
MMX_DIV 13 |
|
|
MMX_DIV 14 |
|
|
MMX_DIV 15 |
|
|
MMX_DIV 16 |
|
|
MMX_DIV 17 |
|
|
MMX_DIV 18 |
|
|
MMX_DIV 19 |
|
|
MMX_DIV 20 |
|
|
MMX_DIV 21 |
|
|
MMX_DIV 22 |
|
|
MMX_DIV 23 |
|
|
MMX_DIV 24 |
|
|
MMX_DIV 25 |
|
|
MMX_DIV 26 |
|
|
MMX_DIV 27 |
|
|
MMX_DIV 28 |
|
|
MMX_DIV 29 |
|
|
MMX_DIV 30 |
|
|
MMX_DIV 31 |
|
|
|
|
|
|
|
|
;=========================================================================== |
|
|
; |
|
|
; intra matrix |
|
|
; |
|
|
;=========================================================================== |
|
66 |
|
|
67 |
cextern intra_matrix |
%ifdef FORMAT_COFF |
68 |
cextern intra_matrix_fix |
SECTION .rodata |
69 |
|
%else |
70 |
|
SECTION .rodata align=16 |
71 |
|
%endif |
72 |
|
|
73 |
;=========================================================================== |
mmx_one: |
74 |
; |
times 4 dw 1 |
|
; inter matrix |
|
|
; |
|
|
;=========================================================================== |
|
75 |
|
|
76 |
cextern inter_matrix |
;----------------------------------------------------------------------------- |
77 |
cextern inter_matrix_fix |
; divide by 2Q table |
78 |
|
;----------------------------------------------------------------------------- |
79 |
|
|
80 |
|
ALIGN 16 |
81 |
|
mmx_div: |
82 |
|
times 4 dw 65535 ; the div by 2 formula will overflow for the case |
83 |
|
; quant=1 but we don't care much because quant=1 |
84 |
|
; is handled by a different piece of code that |
85 |
|
; doesn't use this table. |
86 |
|
%assign quant 2 |
87 |
|
%rep 30 |
88 |
|
times 4 dw (1<<17) / (quant*2) + 1 |
89 |
|
%assign quant quant+1 |
90 |
|
%endrep |
91 |
|
|
92 |
%define VM18P 3 |
%define VM18P 3 |
93 |
%define VM18Q 4 |
%define VM18Q 4 |
94 |
|
|
95 |
|
|
96 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
97 |
; quantd table |
; quantd table |
98 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_QUANTD 1 |
|
|
times 4 dw ((VM18P*%1) + (VM18Q/2)) / VM18Q |
|
|
%endmacro |
|
|
|
|
|
quantd |
|
|
MMX_QUANTD 1 |
|
|
MMX_QUANTD 2 |
|
|
MMX_QUANTD 3 |
|
|
MMX_QUANTD 4 |
|
|
MMX_QUANTD 5 |
|
|
MMX_QUANTD 6 |
|
|
MMX_QUANTD 7 |
|
|
MMX_QUANTD 8 |
|
|
MMX_QUANTD 9 |
|
|
MMX_QUANTD 10 |
|
|
MMX_QUANTD 11 |
|
|
MMX_QUANTD 12 |
|
|
MMX_QUANTD 13 |
|
|
MMX_QUANTD 14 |
|
|
MMX_QUANTD 15 |
|
|
MMX_QUANTD 16 |
|
|
MMX_QUANTD 17 |
|
|
MMX_QUANTD 18 |
|
|
MMX_QUANTD 19 |
|
|
MMX_QUANTD 20 |
|
|
MMX_QUANTD 21 |
|
|
MMX_QUANTD 22 |
|
|
MMX_QUANTD 23 |
|
|
MMX_QUANTD 24 |
|
|
MMX_QUANTD 25 |
|
|
MMX_QUANTD 26 |
|
|
MMX_QUANTD 27 |
|
|
MMX_QUANTD 28 |
|
|
MMX_QUANTD 29 |
|
|
MMX_QUANTD 30 |
|
|
MMX_QUANTD 31 |
|
99 |
|
|
100 |
|
quantd: |
101 |
|
%assign quant 1 |
102 |
|
%rep 31 |
103 |
|
times 4 dw ((VM18P*quant) + (VM18Q/2)) / VM18Q |
104 |
|
%assign quant quant+1 |
105 |
|
%endrep |
106 |
|
|
107 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
108 |
; multiple by 2Q table |
; multiple by 2Q table |
109 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_MUL_QUANT 1 |
|
|
times 4 dw %1 |
|
|
%endmacro |
|
110 |
|
|
111 |
mmx_mul_quant |
mmx_mul_quant: |
112 |
MMX_MUL_QUANT 1 |
%assign quant 1 |
113 |
MMX_MUL_QUANT 2 |
%rep 31 |
114 |
MMX_MUL_QUANT 3 |
times 4 dw quant |
115 |
MMX_MUL_QUANT 4 |
%assign quant quant+1 |
116 |
MMX_MUL_QUANT 5 |
%endrep |
|
MMX_MUL_QUANT 6 |
|
|
MMX_MUL_QUANT 7 |
|
|
MMX_MUL_QUANT 8 |
|
|
MMX_MUL_QUANT 9 |
|
|
MMX_MUL_QUANT 10 |
|
|
MMX_MUL_QUANT 11 |
|
|
MMX_MUL_QUANT 12 |
|
|
MMX_MUL_QUANT 13 |
|
|
MMX_MUL_QUANT 14 |
|
|
MMX_MUL_QUANT 15 |
|
|
MMX_MUL_QUANT 16 |
|
|
MMX_MUL_QUANT 17 |
|
|
MMX_MUL_QUANT 18 |
|
|
MMX_MUL_QUANT 19 |
|
|
MMX_MUL_QUANT 20 |
|
|
MMX_MUL_QUANT 21 |
|
|
MMX_MUL_QUANT 22 |
|
|
MMX_MUL_QUANT 23 |
|
|
MMX_MUL_QUANT 24 |
|
|
MMX_MUL_QUANT 25 |
|
|
MMX_MUL_QUANT 26 |
|
|
MMX_MUL_QUANT 27 |
|
|
MMX_MUL_QUANT 28 |
|
|
MMX_MUL_QUANT 29 |
|
|
MMX_MUL_QUANT 30 |
|
|
MMX_MUL_QUANT 31 |
|
117 |
|
|
118 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
119 |
; saturation limits |
; saturation limits |
120 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
121 |
|
|
122 |
align 16 |
ALIGN 16 |
123 |
|
|
124 |
mmx_32767_minus_2047 times 4 dw (32767-2047) |
mmx_32767_minus_2047: |
125 |
mmx_32768_minus_2048 times 4 dw (32768-2048) |
times 4 dw (32767-2047) |
126 |
mmx_2047 times 4 dw 2047 |
mmx_32768_minus_2048: |
127 |
mmx_minus_2048 times 4 dw (-2048) |
times 4 dw (32768-2048) |
128 |
zero times 4 dw 0 |
mmx_2047: |
129 |
|
times 4 dw 2047 |
130 |
|
mmx_minus_2048: |
131 |
|
times 4 dw (-2048) |
132 |
|
zero: |
133 |
|
times 4 dw 0 |
134 |
|
|
135 |
|
;============================================================================= |
136 |
|
; rounding |
137 |
|
;============================================================================= |
138 |
|
|
139 |
|
mmx_rounding: |
140 |
|
dw (1<<13) |
141 |
|
dw 0 |
142 |
|
dw (1<<13) |
143 |
|
dw 0 |
144 |
|
|
145 |
|
;============================================================================= |
146 |
|
; Code |
147 |
|
;============================================================================= |
148 |
|
|
149 |
section .text |
SECTION .text |
|
|
|
|
;=========================================================================== |
|
|
; |
|
|
; void quant_intra4_mmx(int16_t * coeff, |
|
|
; const int16_t const * data, |
|
|
; const uint32_t quant, |
|
|
; const uint32_t dcscalar); |
|
|
; |
|
|
;=========================================================================== |
|
150 |
|
|
|
align ALIGN |
|
151 |
cglobal quant_mpeg_intra_mmx |
cglobal quant_mpeg_intra_mmx |
152 |
quant_mpeg_intra_mmx: |
cglobal quant_mpeg_inter_mmx |
153 |
|
cglobal dequant_mpeg_intra_mmx |
154 |
push ecx |
cglobal dequant_mpeg_inter_mmx |
|
push esi |
|
|
push edi |
|
|
|
|
|
mov edi, [esp + 12 + 4] ; coeff |
|
|
mov esi, [esp + 12 + 8] ; data |
|
|
mov eax, [esp + 12 + 12] ; quant |
|
|
|
|
|
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
|
|
|
|
|
xor ecx, ecx |
|
|
cmp al, 1 |
|
|
jz near .q1loop |
|
|
|
|
|
cmp al, 2 |
|
|
jz near .q2loop |
|
|
|
|
|
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
|
|
|
|
|
align ALIGN |
|
|
.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, [intra_matrix + 8*ecx] |
|
|
psrlw mm2, 1 ; intra_matrix[i]>>1 |
|
|
paddw mm0, mm2 |
|
155 |
|
|
|
movq mm2, [intra_matrix_fix + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
156 |
|
|
157 |
movq mm2, [intra_matrix + 8*ecx + 8] |
%macro QUANT_MMX 1 |
158 |
psrlw mm2, 1 |
movq mm0, [eax + 16*(%1)] ; data |
159 |
paddw mm3, mm2 |
movq mm2, [ecx + 16*(%1) + 128] ; intra_matrix_rec |
160 |
|
movq mm4, [eax + 16*(%1) + 8] ; data |
161 |
|
movq mm6, [ecx + 16*(%1) + 128 + 8] ; intra_matrix_rec |
162 |
|
|
163 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm1, mm0 |
164 |
pmulhw mm3, mm2 |
movq mm5, mm4 |
165 |
|
|
166 |
paddw mm0, mm5 ; + quantd |
pmullw mm0, mm2 ; low results |
167 |
paddw mm3, mm5 |
pmulhw mm1, mm2 ; high results |
168 |
|
pmullw mm4, mm6 ; low results |
169 |
|
pmulhw mm5, mm6 ; high results |
170 |
|
|
171 |
|
movq mm2, mm0 |
172 |
|
movq mm6, mm4 |
173 |
|
|
174 |
|
punpckhwd mm0, mm1 |
175 |
|
punpcklwd mm2, mm1 |
176 |
|
punpckhwd mm4, mm5 |
177 |
|
punpcklwd mm6, mm5 |
178 |
|
|
179 |
|
paddd mm2, mm7 |
180 |
|
paddd mm0, mm7 |
181 |
|
paddd mm6, mm7 |
182 |
|
paddd mm4, mm7 |
183 |
|
|
184 |
|
psrad mm2, 14 |
185 |
|
psrad mm0, 14 |
186 |
|
psrad mm6, 14 |
187 |
|
psrad mm4, 14 |
188 |
|
|
189 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
packssdw mm2, mm0 |
190 |
pmulhw mm3, mm7 ; |
packssdw mm6, mm4 |
|
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
|
|
psrlw mm3, 1 |
|
191 |
|
|
192 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
movq [edi + 16*(%1)], mm2 |
193 |
pxor mm3, mm4 ; |
movq [edi + 16*(%1)+8], mm6 |
194 |
psubw mm0, mm1 ; undisplace |
%endmacro |
|
psubw mm3, mm4 ; |
|
|
|
|
|
movq [edi + 8*ecx], mm0 |
|
|
movq [edi + 8*ecx + 8], mm3 |
|
195 |
|
|
196 |
add ecx,2 |
;----------------------------------------------------------------------------- |
197 |
cmp ecx,16 |
; |
198 |
jnz near .loop |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
199 |
|
; const int16_t const * data, |
200 |
|
; const uint32_t quant, |
201 |
|
; const uint32_t dcscalar, |
202 |
|
; const uint16_t *mpeg_matrices); |
203 |
|
; |
204 |
|
;----------------------------------------------------------------------------- |
205 |
|
|
206 |
.done |
ALIGN 16 |
207 |
; caclulate data[0] // (int32_t)dcscalar) |
quant_mpeg_intra_mmx: |
208 |
|
|
209 |
mov ecx, [esp + 12 + 16] ; dcscalar |
push edi |
210 |
mov edx, ecx |
movq mm7, [mmx_rounding] |
|
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 |
|
211 |
|
|
212 |
mov [edi], ax ; coeff[0] = ax |
mov eax, [esp + 4 + 8] ; data |
213 |
|
mov ecx, [esp + 4 + 20] ; mpeg_quant_matrices |
214 |
|
mov edi, [esp + 4 + 4] ; coeff |
215 |
|
|
216 |
|
QUANT_MMX(0) |
217 |
|
QUANT_MMX(1) |
218 |
|
QUANT_MMX(2) |
219 |
|
QUANT_MMX(3) |
220 |
|
QUANT_MMX(4) |
221 |
|
QUANT_MMX(5) |
222 |
|
QUANT_MMX(6) |
223 |
|
QUANT_MMX(7) |
224 |
|
|
225 |
|
; calculate DC |
226 |
|
movsx eax, word [eax] ; data[0] |
227 |
|
mov ecx, [esp + 4 + 16] ; dcscalar |
228 |
|
mov edx, eax |
229 |
|
mov edi, ecx |
230 |
|
shr ecx, 1 ; ecx = dcscalar/2 |
231 |
|
sar edx, 31 ; edx = sign extend of eax (ready for division too) |
232 |
|
xor ecx, edx ; adjust ecx according to the sign of data[0] |
233 |
|
sub ecx, edx |
234 |
|
add eax, ecx |
235 |
|
|
236 |
|
mov ecx, [esp + 4 + 4] ; coeff again |
237 |
|
idiv edi ; eax = edx:eax / dcscalar |
238 |
|
mov [ecx], ax ; coeff[0] = ax |
239 |
|
|
240 |
pop edi |
pop edi |
|
pop esi |
|
|
pop ecx |
|
241 |
|
|
242 |
|
xor eax, eax ; return(0); |
243 |
ret |
ret |
244 |
|
ENDFUNC |
245 |
|
|
|
align ALIGN |
|
|
.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, [intra_matrix + 8*ecx] |
|
|
psrlw mm2, 1 |
|
|
paddw mm0, mm2 |
|
|
|
|
|
movq mm2, [intra_matrix_fix + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
|
|
movq mm2, [intra_matrix + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
|
|
|
movq mm2, [intra_matrix_fix + 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 ALIGN |
|
|
.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, [intra_matrix + 8*ecx] |
|
|
psrlw mm2, 1 |
|
|
paddw mm0, mm2 |
|
|
|
|
|
movq mm2, [intra_matrix_fix + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
|
|
movq mm2, [intra_matrix + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
|
|
|
movq mm2, [intra_matrix_fix + 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 |
|
246 |
|
|
247 |
add ecx,2 |
;----------------------------------------------------------------------------- |
|
cmp ecx,16 |
|
|
jnz near .q2loop |
|
|
jmp near .done |
|
|
|
|
|
|
|
|
;=========================================================================== |
|
248 |
; |
; |
249 |
; uint32_t quant4_inter_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
250 |
; const int16_t const * data, |
; const int16_t const * data, |
251 |
; const uint32_t quant); |
; const uint32_t quant, |
252 |
|
; const uint16_t *mpeg_matrices); |
253 |
; |
; |
254 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
255 |
|
|
256 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_mpeg_inter_mmx |
|
257 |
quant_mpeg_inter_mmx: |
quant_mpeg_inter_mmx: |
258 |
|
|
259 |
push ecx |
push ecx |
260 |
push esi |
push esi |
261 |
push edi |
push edi |
262 |
|
push ebx |
263 |
|
|
264 |
mov edi, [esp + 12 + 4] ; coeff |
mov edi, [esp + 16 + 4] ; coeff |
265 |
mov esi, [esp + 12 + 8] ; data |
mov esi, [esp + 16 + 8] ; data |
266 |
mov eax, [esp + 12 + 12] ; quant |
mov eax, [esp + 16 + 12] ; quant |
267 |
|
mov ebx, [esp + 16 + 16] ; mpeg_quant_matrices |
268 |
|
|
269 |
xor ecx, ecx |
xor ecx, ecx |
270 |
|
|
278 |
|
|
279 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
280 |
|
|
281 |
align ALIGN |
ALIGN 16 |
282 |
.loop |
.loop: |
283 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
284 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
285 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
290 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
291 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
292 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
293 |
psllw mm0, 4 |
psllw mm0, 4 |
294 |
psllw mm3, 4 |
psllw mm3, 4 |
295 |
|
movq mm2, [ebx + 512 + 8*ecx] |
|
movq mm2, [inter_matrix + 8*ecx] |
|
296 |
psrlw mm2, 1 |
psrlw mm2, 1 |
297 |
paddw mm0, mm2 |
paddw mm0, mm2 |
298 |
|
movq mm2, [ebx + 768 + ecx*8] |
|
movq mm2, [inter_matrix_fix + ecx*8] |
|
299 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
300 |
|
movq mm2, [ebx + 512 + 8*ecx + 8] |
|
movq mm2, [inter_matrix + 8*ecx + 8] |
|
301 |
psrlw mm2, 1 |
psrlw mm2, 1 |
302 |
paddw mm3, mm2 |
paddw mm3, mm2 |
303 |
|
movq mm2, [ebx + 768 + ecx*8 + 8] |
|
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
|
304 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
305 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
306 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
307 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
308 |
psrlw mm3, 1 |
psrlw mm3, 1 |
|
|
|
309 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
310 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
311 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
319 |
cmp ecx, 16 |
cmp ecx, 16 |
320 |
jnz near .loop |
jnz near .loop |
321 |
|
|
322 |
.done |
.done: |
323 |
pmaddwd mm5, [mmx_one] |
pmaddwd mm5, [mmx_one] |
324 |
movq mm0, mm5 |
movq mm0, mm5 |
325 |
psrlq mm5, 32 |
psrlq mm5, 32 |
326 |
paddd mm0, mm5 |
paddd mm0, mm5 |
327 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
328 |
|
|
329 |
|
pop ebx |
330 |
pop edi |
pop edi |
331 |
pop esi |
pop esi |
332 |
pop ecx |
pop ecx |
333 |
|
|
334 |
ret |
ret |
335 |
|
|
336 |
align ALIGN |
ALIGN 16 |
337 |
.q1loop |
.q1loop: |
338 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
339 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
340 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
341 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
342 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
343 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
344 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
345 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
346 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
347 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
348 |
psllw mm0, 4 |
psllw mm0, 4 |
349 |
psllw mm3, 4 |
psllw mm3, 4 |
350 |
|
movq mm2, [ebx + 512 + 8*ecx] |
|
movq mm2, [inter_matrix + 8*ecx] |
|
351 |
psrlw mm2, 1 |
psrlw mm2, 1 |
352 |
paddw mm0, mm2 |
paddw mm0, mm2 |
353 |
|
movq mm2, [ebx + 768 + ecx*8] |
|
movq mm2, [inter_matrix_fix + ecx*8] |
|
354 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
355 |
|
movq mm2, [ebx + 512 + 8*ecx + 8] |
|
movq mm2, [inter_matrix + 8*ecx + 8] |
|
356 |
psrlw mm2, 1 |
psrlw mm2, 1 |
357 |
paddw mm3, mm2 |
paddw mm3, mm2 |
358 |
|
movq mm2, [ebx + 768 + ecx*8 + 8] |
|
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
|
359 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
360 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
361 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
|
|
|
362 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
363 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
364 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
365 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
366 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
367 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
368 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
369 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
370 |
|
|
374 |
|
|
375 |
jmp .done |
jmp .done |
376 |
|
|
377 |
|
ALIGN 16 |
378 |
align ALIGN |
.q2loop: |
|
.q2loop |
|
379 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
380 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
381 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
382 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
383 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
384 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
385 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
386 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
387 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
388 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
389 |
psllw mm0, 4 |
psllw mm0, 4 |
390 |
psllw mm3, 4 |
psllw mm3, 4 |
391 |
|
movq mm2, [ebx + 512 + 8*ecx] |
|
movq mm2, [inter_matrix + 8*ecx] |
|
392 |
psrlw mm2, 1 |
psrlw mm2, 1 |
393 |
paddw mm0, mm2 |
paddw mm0, mm2 |
394 |
|
movq mm2, [ebx + 768 + ecx*8] |
|
movq mm2, [inter_matrix_fix + ecx*8] |
|
395 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
396 |
|
movq mm2, [ebx + 512 + 8*ecx + 8] |
|
movq mm2, [inter_matrix + 8*ecx + 8] |
|
397 |
psrlw mm2, 1 |
psrlw mm2, 1 |
398 |
paddw mm3, mm2 |
paddw mm3, mm2 |
399 |
|
movq mm2, [ebx + 768 + ecx*8 + 8] |
|
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
|
400 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
401 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
402 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
|
|
|
403 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
404 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
405 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
406 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
407 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
408 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
409 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
410 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
411 |
|
|
414 |
jnz near .q2loop |
jnz near .q2loop |
415 |
|
|
416 |
jmp .done |
jmp .done |
417 |
|
ENDFUNC |
418 |
|
|
419 |
|
|
420 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
421 |
; |
; |
422 |
; void dequant4_intra_mmx(int16_t *data, |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
423 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
424 |
; const uint32_t quant, |
; const uint32_t quant, |
425 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
426 |
|
; const uint16_t *mpeg_matrices); |
427 |
; |
; |
428 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
429 |
|
|
430 |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
431 |
; by 4. Then, the high-word of (coeff[]*matrix[i]*quant) are used to |
; by 4. Then, the high-word of (coeff[]*matrix[i]*quant) are used to |
446 |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
447 |
|
|
448 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
449 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant. |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant. |
450 |
|
|
451 |
movq mm6, mm2 |
movq mm6, mm2 |
452 |
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) |
466 |
|
|
467 |
;******************************************************************** |
;******************************************************************** |
468 |
|
|
469 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_intra_mmx |
|
470 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
471 |
|
|
472 |
mov edx, [esp+4] ; data |
push ebx |
473 |
mov ecx, [esp+8] ; coeff |
|
474 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
475 |
|
mov ecx, [esp + 4 + 8] ; coeff |
476 |
|
mov eax, [esp + 4 + 12] ; quant |
477 |
|
mov ebx, [esp + 4 + 20] ; mpeg_quant_matrices |
478 |
|
|
479 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
480 |
mov eax, -16 ; to keep aligned, we regularly process coeff[0] |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
481 |
psllw mm7, 2 ; << 2. See comment. |
psllw mm7, 2 ; << 2. See comment. |
482 |
pxor mm6, mm6 ; this is a NOP |
pxor mm6, mm6 ; this is a NOP |
483 |
|
|
484 |
align 16 |
ALIGN 16 |
485 |
.loop |
.loop: |
486 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
487 |
movq mm3, [ecx+8*eax + 8*16 +8]; mm3 = c' = coeff[i+1] |
movq mm3, [ecx+8*eax + 8*16 +8]; mm3 = c' = coeff[i+1] |
488 |
pxor mm1, mm1 |
pxor mm1, mm1 |
491 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
492 |
|
|
493 |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
494 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant |
495 |
|
|
496 |
pxor mm0, mm1 ; negate if negative |
pxor mm0, mm1 ; negate if negative |
497 |
pxor mm3, mm4 ; negate if negative |
pxor mm3, mm4 ; negate if negative |
507 |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
508 |
movq mm5, mm7 ; mm2 = quant |
movq mm5, mm7 ; mm2 = quant |
509 |
|
|
510 |
pmullw mm5, [intra_matrix + 8*eax + 8*16 +8] ; matrix[i+1]*quant |
pmullw mm5, [ebx + 8*eax + 8*16 +8] ; matrix[i+1]*quant |
511 |
|
|
512 |
movq mm6, mm5 |
movq mm6, mm5 |
513 |
add eax,2 ; z-flag will be tested later |
add eax,2 ; z-flag will be tested later |
538 |
jnz near .loop |
jnz near .loop |
539 |
|
|
540 |
; deal with DC |
; deal with DC |
|
|
|
541 |
movd mm0, [ecx] |
movd mm0, [ecx] |
542 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp + 4 + 16] ; dcscalar |
543 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
544 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
545 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
549 |
movd eax, mm0 |
movd eax, mm0 |
550 |
mov [edx], ax |
mov [edx], ax |
551 |
|
|
552 |
|
xor eax, eax |
553 |
|
|
554 |
|
pop ebx |
555 |
|
|
556 |
ret |
ret |
557 |
|
ENDFUNC |
558 |
|
|
559 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
560 |
; |
; |
561 |
; void dequant4_inter_mmx(int16_t * data, |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
562 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
563 |
; const uint32_t quant); |
; const uint32_t quant, |
564 |
|
; const uint16_t *mpeg_matrices); |
565 |
; |
; |
566 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
567 |
|
|
568 |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
569 |
; so we handle the 3 cases: c<0, c==0, and c>0 in one shot. |
; so we handle the 3 cases: c<0, c==0, and c>0 in one shot. |
570 |
; 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. |
571 |
; It's mixed with the extraction of the absolute value. |
; It's mixed with the extraction of the absolute value. |
572 |
|
|
573 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_inter_mmx |
|
574 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
575 |
|
|
576 |
mov edx, [esp+ 4] ; data |
push ebx |
577 |
mov ecx, [esp+ 8] ; coeff |
|
578 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
579 |
|
mov ecx, [esp + 4 + 8] ; coeff |
580 |
|
mov eax, [esp + 4 + 12] ; quant |
581 |
|
mov ebx, [esp + 4 + 16] ; mpeg_quant_matrices |
582 |
|
|
583 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
584 |
mov eax, -16 |
mov eax, -16 |
585 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
586 |
pxor mm6, mm6 ; mismatch sum |
pxor mm6, mm6 ; mismatch sum |
587 |
|
|
588 |
align 16 |
ALIGN 16 |
589 |
.loop |
.loop: |
590 |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
591 |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
592 |
add eax,2 |
add eax,2 |
614 |
; we're short on register, here. Poor pairing... |
; we're short on register, here. Poor pairing... |
615 |
|
|
616 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
617 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8] |
618 |
movq mm5, mm4 |
movq mm5, mm4 |
619 |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
620 |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
621 |
|
|
622 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
623 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8 + 8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8 + 8] |
624 |
|
|
625 |
pcmpgtw mm5, [zero] |
pcmpgtw mm5, [zero] |
626 |
paddusw mm0, mm5 |
paddusw mm0, mm5 |
663 |
xor eax, 1 |
xor eax, 1 |
664 |
xor word [edx + 2*63], ax |
xor word [edx + 2*63], ax |
665 |
|
|
666 |
|
xor eax, eax |
667 |
|
|
668 |
|
pop ebx |
669 |
|
|
670 |
ret |
ret |
671 |
|
ENDFUNC |
672 |
|
|
673 |
|
|
674 |
|
%ifidn __OUTPUT_FORMAT__,elf |
675 |
|
section ".note.GNU-stack" noalloc noexec nowrite progbits |
676 |
|
%endif |
677 |
|
|