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 |
47 |
%endif |
%endif |
48 |
%endmacro |
%endmacro |
49 |
|
|
50 |
mmx_one times 4 dw 1 |
;============================================================================= |
51 |
|
; Local data (Read Only) |
52 |
;=========================================================================== |
;============================================================================= |
|
; |
|
|
; 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 |
|
|
|
|
53 |
|
|
54 |
;=========================================================================== |
%ifdef FORMAT_COFF |
55 |
; |
SECTION .rodata |
56 |
; intra matrix |
%else |
57 |
; |
SECTION .rodata align=16 |
58 |
;=========================================================================== |
%endif |
|
|
|
|
cextern intra_matrix |
|
|
cextern intra_matrix_fix |
|
59 |
|
|
60 |
;=========================================================================== |
mmx_one: |
61 |
; |
times 4 dw 1 |
|
; inter matrix |
|
|
; |
|
|
;=========================================================================== |
|
62 |
|
|
63 |
cextern inter_matrix |
;----------------------------------------------------------------------------- |
64 |
cextern inter_matrix_fix |
; divide by 2Q table |
65 |
|
;----------------------------------------------------------------------------- |
66 |
|
|
67 |
|
ALIGN 16 |
68 |
|
mmx_div: |
69 |
|
times 4 dw 65535 ; the div by 2 formula will overflow for the case |
70 |
|
; quant=1 but we don't care much because quant=1 |
71 |
|
; is handled by a different piece of code that |
72 |
|
; doesn't use this table. |
73 |
|
%assign quant 2 |
74 |
|
%rep 30 |
75 |
|
times 4 dw (1<<17) / (quant*2) + 1 |
76 |
|
%assign quant quant+1 |
77 |
|
%endrep |
78 |
|
|
79 |
%define VM18P 3 |
%define VM18P 3 |
80 |
%define VM18Q 4 |
%define VM18Q 4 |
81 |
|
|
82 |
|
|
83 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
84 |
; quantd table |
; quantd table |
85 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
86 |
|
|
87 |
%macro MMX_QUANTD 1 |
quantd: |
88 |
times 4 dw ((VM18P*%1) + (VM18Q/2)) / VM18Q |
%assign quant 1 |
89 |
%endmacro |
%rep 31 |
90 |
|
times 4 dw ((VM18P*quant) + (VM18Q/2)) / VM18Q |
91 |
|
%assign quant quant+1 |
92 |
|
%endrep |
93 |
|
|
94 |
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 |
|
|
|
|
|
|
|
|
;=========================================================================== |
|
|
; |
|
95 |
; multiple by 2Q table |
; multiple by 2Q table |
96 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_MUL_QUANT 1 |
|
|
times 4 dw %1 |
|
|
%endmacro |
|
97 |
|
|
98 |
mmx_mul_quant |
mmx_mul_quant: |
99 |
MMX_MUL_QUANT 1 |
%assign quant 1 |
100 |
MMX_MUL_QUANT 2 |
%rep 31 |
101 |
MMX_MUL_QUANT 3 |
times 4 dw quant |
102 |
MMX_MUL_QUANT 4 |
%assign quant quant+1 |
103 |
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 |
|
104 |
|
|
105 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
106 |
; saturation limits |
; saturation limits |
107 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
108 |
|
|
109 |
align 16 |
ALIGN 16 |
110 |
|
|
111 |
mmx_32767_minus_2047 times 4 dw (32767-2047) |
mmx_32767_minus_2047: |
112 |
mmx_32768_minus_2048 times 4 dw (32768-2048) |
times 4 dw (32767-2047) |
113 |
mmx_2047 times 4 dw 2047 |
mmx_32768_minus_2048: |
114 |
mmx_minus_2048 times 4 dw (-2048) |
times 4 dw (32768-2048) |
115 |
zero times 4 dw 0 |
mmx_2047: |
116 |
|
times 4 dw 2047 |
117 |
|
mmx_minus_2048: |
118 |
|
times 4 dw (-2048) |
119 |
|
zero: |
120 |
|
times 4 dw 0 |
121 |
|
|
122 |
|
;============================================================================= |
123 |
|
; Code |
124 |
|
;============================================================================= |
125 |
|
|
126 |
section .text |
SECTION .text |
127 |
|
|
128 |
;=========================================================================== |
cglobal quant_mpeg_intra_mmx |
129 |
|
cglobal quant_mpeg_inter_mmx |
130 |
|
cglobal dequant_mpeg_intra_mmx |
131 |
|
cglobal dequant_mpeg_inter_mmx |
132 |
|
|
133 |
|
;----------------------------------------------------------------------------- |
134 |
; |
; |
135 |
; void quant_intra4_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
136 |
; const int16_t const * data, |
; const int16_t const * data, |
137 |
; const uint32_t quant, |
; const uint32_t quant, |
138 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
139 |
|
; const uint16_t *mpeg_matrices); |
140 |
; |
; |
141 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
142 |
|
|
143 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_mpeg_intra_mmx |
|
144 |
quant_mpeg_intra_mmx: |
quant_mpeg_intra_mmx: |
145 |
|
|
146 |
push ecx |
push ecx |
147 |
push esi |
push esi |
148 |
push edi |
push edi |
149 |
|
push ebx |
150 |
|
|
151 |
mov edi, [esp + 12 + 4] ; coeff |
mov edi, [esp + 16 + 4] ; coeff |
152 |
mov esi, [esp + 12 + 8] ; data |
mov esi, [esp + 16 + 8] ; data |
153 |
mov eax, [esp + 12 + 12] ; quant |
mov eax, [esp + 16 + 12] ; quant |
154 |
|
mov ebx, [esp + 16 + 20] ; mpeg_quant_matrices |
155 |
|
|
156 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
157 |
|
|
164 |
|
|
165 |
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
166 |
|
|
167 |
align ALIGN |
ALIGN 16 |
168 |
.loop |
.loop |
169 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
170 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
171 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
172 |
pxor mm4, mm4 |
pxor mm4, mm4 |
|
|
|
173 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
174 |
pcmpgtw mm4, mm3 |
pcmpgtw mm4, mm3 |
|
|
|
175 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
176 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
177 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
178 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
179 |
psllw mm0, 4 ; level << 4 |
psllw mm0, 4 ; level << 4 |
180 |
psllw mm3, 4 ; |
psllw mm3, 4 |
181 |
|
movq mm2, [ebx + 8*ecx] |
|
movq mm2, [intra_matrix + 8*ecx] |
|
182 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
183 |
paddw mm0, mm2 |
paddw mm0, mm2 |
184 |
|
movq mm2, [ebx + 256 + ecx*8] |
|
movq mm2, [intra_matrix_fix + ecx*8] |
|
185 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
186 |
|
movq mm2, [ebx + 8*ecx + 8] |
|
movq mm2, [intra_matrix + 8*ecx + 8] |
|
187 |
psrlw mm2, 1 |
psrlw mm2, 1 |
188 |
paddw mm3, mm2 |
paddw mm3, mm2 |
189 |
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
|
190 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
191 |
paddw mm0, mm5 ; + quantd |
paddw mm0, mm5 ; + quantd |
192 |
paddw mm3, mm5 |
paddw mm3, mm5 |
|
|
|
193 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
194 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
195 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
196 |
psrlw mm3, 1 |
psrlw mm3, 1 |
|
|
|
197 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
198 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
199 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
208 |
|
|
209 |
.done |
.done |
210 |
; caclulate data[0] // (int32_t)dcscalar) |
; caclulate data[0] // (int32_t)dcscalar) |
211 |
|
mov ecx, [esp + 16 + 16] ; dcscalar |
|
mov ecx, [esp + 12 + 16] ; dcscalar |
|
212 |
mov edx, ecx |
mov edx, ecx |
213 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
214 |
shr edx, 1 ; edx = dcscalar /2 |
shr edx, 1 ; edx = dcscalar /2 |
225 |
|
|
226 |
mov [edi], ax ; coeff[0] = ax |
mov [edi], ax ; coeff[0] = ax |
227 |
|
|
228 |
|
pop ebx |
229 |
pop edi |
pop edi |
230 |
pop esi |
pop esi |
231 |
pop ecx |
pop ecx |
232 |
|
|
233 |
|
xor eax, eax ; return(0); |
234 |
ret |
ret |
235 |
|
|
236 |
align ALIGN |
ALIGN 16 |
237 |
.q1loop |
.q1loop |
238 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
239 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
240 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
241 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
242 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
243 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
244 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
245 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
246 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
247 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
248 |
psllw mm0, 4 |
psllw mm0, 4 |
249 |
psllw mm3, 4 |
psllw mm3, 4 |
250 |
|
movq mm2, [ebx + 8*ecx] |
|
movq mm2, [intra_matrix + 8*ecx] |
|
251 |
psrlw mm2, 1 |
psrlw mm2, 1 |
252 |
paddw mm0, mm2 |
paddw mm0, mm2 |
253 |
|
movq mm2, [ebx + 256 + ecx*8] |
|
movq mm2, [intra_matrix_fix + ecx*8] |
|
254 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
255 |
|
movq mm2, [ebx + 8*ecx + 8] |
|
movq mm2, [intra_matrix + 8*ecx + 8] |
|
256 |
psrlw mm2, 1 |
psrlw mm2, 1 |
257 |
paddw mm3, mm2 |
paddw mm3, mm2 |
258 |
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
|
259 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
260 |
paddw mm0, mm5 |
paddw mm0, mm5 |
261 |
paddw mm3, mm5 |
paddw mm3, mm5 |
|
|
|
262 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
263 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
|
|
|
264 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
265 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
266 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
267 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
268 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
269 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
270 |
|
|
274 |
jmp near .done |
jmp near .done |
275 |
|
|
276 |
|
|
277 |
align ALIGN |
ALIGN 16 |
278 |
.q2loop |
.q2loop |
279 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
280 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
281 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
282 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
283 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
284 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
285 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
286 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
287 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
288 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
289 |
psllw mm0, 4 |
psllw mm0, 4 |
290 |
psllw mm3, 4 |
psllw mm3, 4 |
291 |
|
movq mm2, [ebx + 8*ecx] |
|
movq mm2, [intra_matrix + 8*ecx] |
|
292 |
psrlw mm2, 1 |
psrlw mm2, 1 |
293 |
paddw mm0, mm2 |
paddw mm0, mm2 |
294 |
|
movq mm2, [ebx + 256 + ecx*8] |
|
movq mm2, [intra_matrix_fix + ecx*8] |
|
295 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
296 |
|
movq mm2, [ebx + 8*ecx + 8] |
|
movq mm2, [intra_matrix + 8*ecx + 8] |
|
297 |
psrlw mm2, 1 |
psrlw mm2, 1 |
298 |
paddw mm3, mm2 |
paddw mm3, mm2 |
299 |
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
|
300 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
301 |
paddw mm0, mm5 |
paddw mm0, mm5 |
302 |
paddw mm3, mm5 |
paddw mm3, mm5 |
|
|
|
303 |
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
304 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
|
|
|
305 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
306 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
307 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
308 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
309 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
310 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
311 |
|
|
315 |
jmp near .done |
jmp near .done |
316 |
|
|
317 |
|
|
318 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
319 |
; |
; |
320 |
; uint32_t quant4_inter_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
321 |
; const int16_t const * data, |
; const int16_t const * data, |
322 |
; const uint32_t quant); |
; const uint32_t quant, |
323 |
|
; const uint16_t *mpeg_matrices); |
324 |
; |
; |
325 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
326 |
|
|
327 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_mpeg_inter_mmx |
|
328 |
quant_mpeg_inter_mmx: |
quant_mpeg_inter_mmx: |
329 |
|
|
330 |
push ecx |
push ecx |
331 |
push esi |
push esi |
332 |
push edi |
push edi |
333 |
|
push ebx |
334 |
|
|
335 |
mov edi, [esp + 12 + 4] ; coeff |
mov edi, [esp + 16 + 4] ; coeff |
336 |
mov esi, [esp + 12 + 8] ; data |
mov esi, [esp + 16 + 8] ; data |
337 |
mov eax, [esp + 12 + 12] ; quant |
mov eax, [esp + 16 + 12] ; quant |
338 |
|
mov ebx, [esp + 16 + 16] ; mpeg_quant_matrices |
339 |
|
|
340 |
xor ecx, ecx |
xor ecx, ecx |
341 |
|
|
349 |
|
|
350 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
351 |
|
|
352 |
align ALIGN |
ALIGN 16 |
353 |
.loop |
.loop |
354 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
355 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
361 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
362 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
363 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
364 |
psllw mm0, 4 |
psllw mm0, 4 |
365 |
psllw mm3, 4 |
psllw mm3, 4 |
366 |
|
movq mm2, [ebx + 512 + 8*ecx] |
|
movq mm2, [inter_matrix + 8*ecx] |
|
367 |
psrlw mm2, 1 |
psrlw mm2, 1 |
368 |
paddw mm0, mm2 |
paddw mm0, mm2 |
369 |
|
movq mm2, [ebx + 768 + ecx*8] |
|
movq mm2, [inter_matrix_fix + ecx*8] |
|
370 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
371 |
|
movq mm2, [ebx + 512 + 8*ecx + 8] |
|
movq mm2, [inter_matrix + 8*ecx + 8] |
|
372 |
psrlw mm2, 1 |
psrlw mm2, 1 |
373 |
paddw mm3, mm2 |
paddw mm3, mm2 |
374 |
|
movq mm2, [ebx + 768 + ecx*8 + 8] |
|
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
|
375 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
376 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
377 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
378 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
379 |
psrlw mm3, 1 |
psrlw mm3, 1 |
|
|
|
380 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
381 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
382 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
397 |
paddd mm0, mm5 |
paddd mm0, mm5 |
398 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
399 |
|
|
400 |
|
pop ebx |
401 |
pop edi |
pop edi |
402 |
pop esi |
pop esi |
403 |
pop ecx |
pop ecx |
404 |
|
|
405 |
ret |
ret |
406 |
|
|
407 |
align ALIGN |
ALIGN 16 |
408 |
.q1loop |
.q1loop |
409 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
410 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
411 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
412 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
413 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
414 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
415 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
416 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
417 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
418 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
419 |
psllw mm0, 4 |
psllw mm0, 4 |
420 |
psllw mm3, 4 |
psllw mm3, 4 |
421 |
|
movq mm2, [ebx + 512 + 8*ecx] |
|
movq mm2, [inter_matrix + 8*ecx] |
|
422 |
psrlw mm2, 1 |
psrlw mm2, 1 |
423 |
paddw mm0, mm2 |
paddw mm0, mm2 |
424 |
|
movq mm2, [ebx + 768 + ecx*8] |
|
movq mm2, [inter_matrix_fix + ecx*8] |
|
425 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
426 |
|
movq mm2, [ebx + 512 + 8*ecx + 8] |
|
movq mm2, [inter_matrix + 8*ecx + 8] |
|
427 |
psrlw mm2, 1 |
psrlw mm2, 1 |
428 |
paddw mm3, mm2 |
paddw mm3, mm2 |
429 |
|
movq mm2, [ebx + 768 + ecx*8 + 8] |
|
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
|
430 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
431 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
432 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
|
|
|
433 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
434 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
435 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
436 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
437 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
438 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
439 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
440 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
441 |
|
|
446 |
jmp .done |
jmp .done |
447 |
|
|
448 |
|
|
449 |
align ALIGN |
ALIGN 16 |
450 |
.q2loop |
.q2loop |
451 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
452 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
453 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
454 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
455 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
456 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
457 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
458 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
459 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
460 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
461 |
psllw mm0, 4 |
psllw mm0, 4 |
462 |
psllw mm3, 4 |
psllw mm3, 4 |
463 |
|
movq mm2, [ebx + 512 + 8*ecx] |
|
movq mm2, [inter_matrix + 8*ecx] |
|
464 |
psrlw mm2, 1 |
psrlw mm2, 1 |
465 |
paddw mm0, mm2 |
paddw mm0, mm2 |
466 |
|
movq mm2, [ebx + 768 + ecx*8] |
|
movq mm2, [inter_matrix_fix + ecx*8] |
|
467 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
468 |
|
movq mm2, [ebx + 512 + 8*ecx + 8] |
|
movq mm2, [inter_matrix + 8*ecx + 8] |
|
469 |
psrlw mm2, 1 |
psrlw mm2, 1 |
470 |
paddw mm3, mm2 |
paddw mm3, mm2 |
471 |
|
movq mm2, [ebx + 768 + ecx*8 + 8] |
|
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
|
472 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
473 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
474 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
|
|
|
475 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
476 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
477 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
478 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
479 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
480 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
481 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
482 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
483 |
|
|
488 |
jmp .done |
jmp .done |
489 |
|
|
490 |
|
|
491 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
492 |
; |
; |
493 |
; void dequant4_intra_mmx(int16_t *data, |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
494 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
495 |
; const uint32_t quant, |
; const uint32_t quant, |
496 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
497 |
|
; const uint16_t *mpeg_matrices); |
498 |
; |
; |
499 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
500 |
|
|
501 |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
502 |
; 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 |
517 |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
psubw mm0, mm1 ; -> mm0 = abs(coeff[i]), mm1 = sign of coeff[i] |
518 |
|
|
519 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
520 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant. |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant. |
521 |
|
|
522 |
movq mm6, mm2 |
movq mm6, mm2 |
523 |
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) |
537 |
|
|
538 |
;******************************************************************** |
;******************************************************************** |
539 |
|
|
540 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_intra_mmx |
|
541 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
542 |
|
|
543 |
mov edx, [esp+4] ; data |
push ebx |
544 |
mov ecx, [esp+8] ; coeff |
|
545 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
546 |
|
mov ecx, [esp + 4 + 8] ; coeff |
547 |
|
mov eax, [esp + 4 + 12] ; quant |
548 |
|
mov ebx, [esp + 4 + 20] ; mpeg_quant_matrices |
549 |
|
|
550 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
551 |
mov eax, -16 ; to keep aligned, we regularly process coeff[0] |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
552 |
psllw mm7, 2 ; << 2. See comment. |
psllw mm7, 2 ; << 2. See comment. |
553 |
pxor mm6, mm6 ; this is a NOP |
pxor mm6, mm6 ; this is a NOP |
554 |
|
|
555 |
align 16 |
ALIGN 16 |
556 |
.loop |
.loop |
557 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
558 |
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] |
562 |
movq mm2, mm7 ; mm2 = quant |
movq mm2, mm7 ; mm2 = quant |
563 |
|
|
564 |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
pcmpgtw mm4, mm3 ; mm4 = sgn(c') |
565 |
pmullw mm2, [intra_matrix + 8*eax + 8*16 ] ; matrix[i]*quant |
pmullw mm2, [ebx + 8*eax + 8*16 ] ; matrix[i]*quant |
566 |
|
|
567 |
pxor mm0, mm1 ; negate if negative |
pxor mm0, mm1 ; negate if negative |
568 |
pxor mm3, mm4 ; negate if negative |
pxor mm3, mm4 ; negate if negative |
578 |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
pmulhw mm0, mm5 ; high of coeff*(matrix*quant) |
579 |
movq mm5, mm7 ; mm2 = quant |
movq mm5, mm7 ; mm2 = quant |
580 |
|
|
581 |
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 |
582 |
|
|
583 |
movq mm6, mm5 |
movq mm6, mm5 |
584 |
add eax,2 ; z-flag will be tested later |
add eax,2 ; z-flag will be tested later |
609 |
jnz near .loop |
jnz near .loop |
610 |
|
|
611 |
; deal with DC |
; deal with DC |
|
|
|
612 |
movd mm0, [ecx] |
movd mm0, [ecx] |
613 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp + 4 + 16] ; dcscalar |
614 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
615 |
paddsw mm0, mm2 |
paddsw mm0, mm2 |
616 |
psubsw mm0, mm2 |
psubsw mm0, mm2 |
620 |
movd eax, mm0 |
movd eax, mm0 |
621 |
mov [edx], ax |
mov [edx], ax |
622 |
|
|
623 |
|
xor eax, eax |
624 |
|
|
625 |
|
pop ebx |
626 |
|
|
627 |
ret |
ret |
628 |
|
|
629 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
630 |
; |
; |
631 |
; void dequant4_inter_mmx(int16_t * data, |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
632 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
633 |
; const uint32_t quant); |
; const uint32_t quant, |
634 |
|
; const uint16_t *mpeg_matrices); |
635 |
; |
; |
636 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
637 |
|
|
638 |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
639 |
; 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. |
640 |
; 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. |
641 |
; It's mixed with the extraction of the absolute value. |
; It's mixed with the extraction of the absolute value. |
642 |
|
|
643 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_inter_mmx |
|
644 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
645 |
|
|
646 |
mov edx, [esp+ 4] ; data |
push ebx |
647 |
mov ecx, [esp+ 8] ; coeff |
|
648 |
mov eax, [esp+12] ; quant |
mov edx, [esp + 4 + 4] ; data |
649 |
|
mov ecx, [esp + 4 + 8] ; coeff |
650 |
|
mov eax, [esp + 4 + 12] ; quant |
651 |
|
mov ebx, [esp + 4 + 16] ; mpeg_quant_matrices |
652 |
|
|
653 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
654 |
mov eax, -16 |
mov eax, -16 |
655 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
656 |
pxor mm6, mm6 ; mismatch sum |
pxor mm6, mm6 ; mismatch sum |
657 |
|
|
658 |
align 16 |
ALIGN 16 |
659 |
.loop |
.loop |
660 |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
661 |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
684 |
; we're short on register, here. Poor pairing... |
; we're short on register, here. Poor pairing... |
685 |
|
|
686 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
687 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8] |
688 |
movq mm5, mm4 |
movq mm5, mm4 |
689 |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
pmulhw mm5, mm0 ; high of c*(matrix*quant) |
690 |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
pmullw mm0, mm4 ; low of c*(matrix*quant) |
691 |
|
|
692 |
movq mm4, mm7 ; (matrix*quant) |
movq mm4, mm7 ; (matrix*quant) |
693 |
pmullw mm4, [inter_matrix + 8*eax + 8*16 -2*8 + 8] |
pmullw mm4, [ebx + 512 + 8*eax + 8*16 -2*8 + 8] |
694 |
|
|
695 |
pcmpgtw mm5, [zero] |
pcmpgtw mm5, [zero] |
696 |
paddusw mm0, mm5 |
paddusw mm0, mm5 |
733 |
xor eax, 1 |
xor eax, 1 |
734 |
xor word [edx + 2*63], ax |
xor word [edx + 2*63], ax |
735 |
|
|
736 |
ret |
xor eax, eax |
737 |
|
|
738 |
|
pop ebx |
739 |
|
|
740 |
|
ret |