1 |
;/************************************************************************** |
;/***************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * mmx quantization/dequantization |
; * - MPEG4 Quantization H263 implementation / MMX optimized - |
5 |
; * |
; * |
6 |
; * This program is an implementation of a part of one or more MPEG-4 |
; * Copyright(C) 2001-2003 Peter Ross <pross@xvid.org> |
7 |
; * Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
; * 2002-2003 Pascal Massimino <skal@planet-d.net> |
|
; * to use this software module in hardware or software products are |
|
|
; * advised that its use may infringe existing patents or copyrights, and |
|
|
; * any such use would be at such party's own risk. The original |
|
|
; * developer of this software module and his/her company, and subsequent |
|
|
; * editors and their companies, will have no liability for use of this |
|
|
; * software or modifications or derivatives thereof. |
|
8 |
; * |
; * |
9 |
; * This program is free software; you can redistribute it and/or modify |
; * This program is free software; you can redistribute it and/or modify |
10 |
; * it under the terms of the GNU General Public License as published by |
; * it under the terms of the GNU General Public License as published by |
18 |
; * |
; * |
19 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
20 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
21 |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
22 |
; * |
; * |
23 |
; *************************************************************************/ |
; * $Id$ |
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
|
; * |
|
|
; * 09.08.2002 sse2 dequant funcs revamped |
|
|
; * 14.06.2002 mmx+xmm dequant_* funcs revamped -Skal- |
|
|
; * 24.02.2002 sse2 quant_intra / dequant_intra (have to use movdqu ???) |
|
|
; * 17.04.2002 sse2 quant_inter / dequant_inter |
|
|
; * 26.12.2001 minor bug fixes, dequant saturate, further optimization |
|
|
; * 19.11.2001 quant_inter_mmx now returns sum of abs. coefficient values |
|
|
; * 04.11.2001 nasm version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
|
24 |
; * |
; * |
25 |
; *************************************************************************/ |
; ****************************************************************************/ |
26 |
|
|
27 |
; enable dequant saturate [-2048,2047], test purposes only. |
; enable dequant saturate [-2048,2047], test purposes only. |
28 |
%define SATURATE |
%define SATURATE |
29 |
|
|
30 |
; data/text alignment |
BITS 32 |
|
%define ALIGN 8 |
|
|
|
|
|
bits 32 |
|
|
|
|
|
section .data |
|
|
|
|
31 |
|
|
32 |
%macro cglobal 1 |
%macro cglobal 1 |
33 |
%ifdef PREFIX |
%ifdef PREFIX |
38 |
%endif |
%endif |
39 |
%endmacro |
%endmacro |
40 |
|
|
41 |
align 16 |
;============================================================================= |
42 |
|
; Read only Local data |
43 |
|
;============================================================================= |
44 |
|
|
45 |
plus_one times 8 dw 1 |
%ifdef FORMAT_COFF |
46 |
|
SECTION .rodata data |
47 |
|
%else |
48 |
|
SECTION .rodata data align=16 |
49 |
|
%endif |
50 |
|
|
51 |
;=========================================================================== |
ALIGN 16 |
52 |
|
plus_one: |
53 |
|
times 8 dw 1 |
54 |
|
|
55 |
|
;----------------------------------------------------------------------------- |
56 |
; |
; |
57 |
; subtract by Q/2 table |
; subtract by Q/2 table |
58 |
; |
; |
59 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
60 |
|
|
61 |
%macro MMX_SUB 1 |
ALIGN 16 |
62 |
times 4 dw %1 / 2 |
mmx_sub: |
63 |
%endmacro |
%assign quant 1 |
64 |
|
%rep 31 |
65 |
|
times 4 dw quant / 2 |
66 |
|
%assign quant quant+1 |
67 |
|
%endrep |
68 |
|
|
69 |
align 16 |
;----------------------------------------------------------------------------- |
|
mmx_sub |
|
|
MMX_SUB 1 |
|
|
MMX_SUB 2 |
|
|
MMX_SUB 3 |
|
|
MMX_SUB 4 |
|
|
MMX_SUB 5 |
|
|
MMX_SUB 6 |
|
|
MMX_SUB 7 |
|
|
MMX_SUB 8 |
|
|
MMX_SUB 9 |
|
|
MMX_SUB 10 |
|
|
MMX_SUB 11 |
|
|
MMX_SUB 12 |
|
|
MMX_SUB 13 |
|
|
MMX_SUB 14 |
|
|
MMX_SUB 15 |
|
|
MMX_SUB 16 |
|
|
MMX_SUB 17 |
|
|
MMX_SUB 18 |
|
|
MMX_SUB 19 |
|
|
MMX_SUB 20 |
|
|
MMX_SUB 21 |
|
|
MMX_SUB 22 |
|
|
MMX_SUB 23 |
|
|
MMX_SUB 24 |
|
|
MMX_SUB 25 |
|
|
MMX_SUB 26 |
|
|
MMX_SUB 27 |
|
|
MMX_SUB 28 |
|
|
MMX_SUB 29 |
|
|
MMX_SUB 30 |
|
|
MMX_SUB 31 |
|
|
|
|
|
|
|
|
|
|
|
;=========================================================================== |
|
70 |
; |
; |
71 |
; divide by 2Q table |
; divide by 2Q table |
72 |
; |
; |
74 |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
75 |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
76 |
; |
; |
77 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro MMX_DIV 1 |
|
|
times 4 dw (1 << 16) / (%1 * 2) + 1 |
|
|
%endmacro |
|
78 |
|
|
79 |
align 16 |
ALIGN 16 |
80 |
mmx_div |
mmx_div: |
81 |
MMX_DIV 1 |
%assign quant 1 |
82 |
MMX_DIV 2 |
%rep 31 |
83 |
MMX_DIV 3 |
times 4 dw (1<<16) / (quant*2) + 1 |
84 |
MMX_DIV 4 |
%assign quant quant+1 |
85 |
MMX_DIV 5 |
%endrep |
|
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 |
|
86 |
|
|
87 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
88 |
; |
; |
89 |
; add by (odd(Q) ? Q : Q - 1) table |
; add by (odd(Q) ? Q : Q - 1) table |
90 |
; |
; |
91 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
92 |
|
|
93 |
%macro MMX_ADD 1 |
ALIGN 16 |
94 |
%if %1 % 2 != 0 |
mmx_add: |
95 |
times 4 dw %1 |
%assign quant 1 |
96 |
|
%rep 31 |
97 |
|
%if quant % 2 != 0 |
98 |
|
times 4 dw quant |
99 |
%else |
%else |
100 |
times 4 dw %1 - 1 |
times 4 dw quant - 1 |
101 |
%endif |
%endif |
102 |
%endmacro |
%assign quant quant+1 |
103 |
|
%endrep |
|
align 16 |
|
|
mmx_add |
|
|
MMX_ADD 1 |
|
|
MMX_ADD 2 |
|
|
MMX_ADD 3 |
|
|
MMX_ADD 4 |
|
|
MMX_ADD 5 |
|
|
MMX_ADD 6 |
|
|
MMX_ADD 7 |
|
|
MMX_ADD 8 |
|
|
MMX_ADD 9 |
|
|
MMX_ADD 10 |
|
|
MMX_ADD 11 |
|
|
MMX_ADD 12 |
|
|
MMX_ADD 13 |
|
|
MMX_ADD 14 |
|
|
MMX_ADD 15 |
|
|
MMX_ADD 16 |
|
|
MMX_ADD 17 |
|
|
MMX_ADD 18 |
|
|
MMX_ADD 19 |
|
|
MMX_ADD 20 |
|
|
MMX_ADD 21 |
|
|
MMX_ADD 22 |
|
|
MMX_ADD 23 |
|
|
MMX_ADD 24 |
|
|
MMX_ADD 25 |
|
|
MMX_ADD 26 |
|
|
MMX_ADD 27 |
|
|
MMX_ADD 28 |
|
|
MMX_ADD 29 |
|
|
MMX_ADD 30 |
|
|
MMX_ADD 31 |
|
104 |
|
|
105 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
106 |
; |
; |
107 |
; multiple by 2Q table |
; multiple by 2Q table |
108 |
; |
; |
109 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro MMX_MUL 1 |
|
|
times 4 dw %1 * 2 |
|
|
%endmacro |
|
110 |
|
|
111 |
align 16 |
ALIGN 16 |
112 |
mmx_mul |
mmx_mul: |
113 |
MMX_MUL 1 |
%assign quant 1 |
114 |
MMX_MUL 2 |
%rep 31 |
115 |
MMX_MUL 3 |
times 4 dw quant*2 |
116 |
MMX_MUL 4 |
%assign quant quant+1 |
117 |
MMX_MUL 5 |
%endrep |
|
MMX_MUL 6 |
|
|
MMX_MUL 7 |
|
|
MMX_MUL 8 |
|
|
MMX_MUL 9 |
|
|
MMX_MUL 10 |
|
|
MMX_MUL 11 |
|
|
MMX_MUL 12 |
|
|
MMX_MUL 13 |
|
|
MMX_MUL 14 |
|
|
MMX_MUL 15 |
|
|
MMX_MUL 16 |
|
|
MMX_MUL 17 |
|
|
MMX_MUL 18 |
|
|
MMX_MUL 19 |
|
|
MMX_MUL 20 |
|
|
MMX_MUL 21 |
|
|
MMX_MUL 22 |
|
|
MMX_MUL 23 |
|
|
MMX_MUL 24 |
|
|
MMX_MUL 25 |
|
|
MMX_MUL 26 |
|
|
MMX_MUL 27 |
|
|
MMX_MUL 28 |
|
|
MMX_MUL 29 |
|
|
MMX_MUL 30 |
|
|
MMX_MUL 31 |
|
118 |
|
|
119 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
120 |
; |
; |
121 |
; saturation limits |
; saturation limits |
122 |
; |
; |
123 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
124 |
|
|
125 |
|
ALIGN 16 |
126 |
|
sse2_2047: |
127 |
|
times 8 dw 2047 |
128 |
|
|
129 |
align 16 |
ALIGN 16 |
130 |
sse2_2047 times 8 dw 2047 |
mmx_2047: |
131 |
|
times 4 dw 2047 |
132 |
|
|
133 |
align 16 |
ALIGN 8 |
134 |
mmx_2047 times 4 dw 2047 |
mmx_32768_minus_2048: |
135 |
|
times 4 dw (32768-2048) |
136 |
|
|
137 |
align 8 |
mmx_32767_minus_2047: |
138 |
mmx_32768_minus_2048 times 4 dw (32768-2048) |
times 4 dw (32767-2047) |
|
mmx_32767_minus_2047 times 4 dw (32767-2047) |
|
139 |
|
|
140 |
|
|
141 |
section .text |
;============================================================================= |
142 |
|
; Code |
143 |
|
;============================================================================= |
144 |
|
|
145 |
|
SECTION .text |
146 |
|
|
147 |
|
cglobal quant_h263_intra_mmx |
148 |
|
cglobal quant_h263_intra_sse2 |
149 |
|
cglobal quant_h263_inter_mmx |
150 |
|
cglobal quant_h263_inter_sse2 |
151 |
|
cglobal dequant_h263_intra_mmx |
152 |
|
cglobal dequant_h263_intra_xmm |
153 |
|
cglobal dequant_h263_intra_sse2 |
154 |
|
cglobal dequant_h263_inter_mmx |
155 |
|
cglobal dequant_h263_inter_xmm |
156 |
|
cglobal dequant_h263_inter_sse2 |
157 |
|
|
158 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
159 |
; |
; |
160 |
; void quant_intra_mmx(int16_t * coeff, |
; uint32_t quant_h263_intra_mmx(int16_t * coeff, |
161 |
; const int16_t const * data, |
; const int16_t const * data, |
162 |
; const uint32_t quant, |
; const uint32_t quant, |
163 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
164 |
|
; const uint16_t *mpeg_matrices); |
165 |
; |
; |
166 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
167 |
|
|
168 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_intra_mmx |
|
169 |
quant_h263_intra_mmx: |
quant_h263_intra_mmx: |
170 |
|
|
171 |
push ecx |
push ecx |
181 |
jz .q1loop |
jz .q1loop |
182 |
|
|
183 |
movq mm7, [mmx_div + eax * 8 - 8] |
movq mm7, [mmx_div + eax * 8 - 8] |
184 |
align ALIGN |
|
185 |
|
ALIGN 16 |
186 |
.loop |
.loop |
187 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
188 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] |
189 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
190 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
191 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
208 |
jnz .loop |
jnz .loop |
209 |
|
|
210 |
.done |
.done |
|
; caclulate data[0] // (int32_t)dcscalar) |
|
211 |
|
|
212 |
|
; caclulate data[0] // (int32_t)dcscalar) |
213 |
mov ecx, [esp + 12 + 16] ; dcscalar |
mov ecx, [esp + 12 + 16] ; dcscalar |
214 |
mov edx, ecx |
mov edx, ecx |
215 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
219 |
|
|
220 |
sub eax, edx |
sub eax, edx |
221 |
jmp short .mul |
jmp short .mul |
222 |
|
|
223 |
.gtzero |
.gtzero |
224 |
add eax, edx |
add eax, edx |
225 |
.mul |
.mul |
226 |
cdq ; expand eax -> edx:eax |
cdq ; expand eax -> edx:eax |
227 |
idiv ecx ; eax = edx:eax / dcscalar |
idiv ecx ; eax = edx:eax / dcscalar |
|
|
|
228 |
mov [edi], ax ; coeff[0] = ax |
mov [edi], ax ; coeff[0] = ax |
229 |
|
|
230 |
|
xor eax, eax ; return(0); |
231 |
pop edi |
pop edi |
232 |
pop esi |
pop esi |
233 |
pop ecx |
pop ecx |
234 |
|
|
235 |
ret |
ret |
236 |
|
|
237 |
align ALIGN |
ALIGN 16 |
238 |
.q1loop |
.q1loop |
239 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
240 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] |
241 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
242 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
243 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
249 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
250 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
251 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
252 |
pxor mm3, mm4 ; |
pxor mm3, mm4 |
253 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
254 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
255 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
262 |
|
|
263 |
|
|
264 |
|
|
265 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
266 |
; |
; |
267 |
; void quant_intra_sse2(int16_t * coeff, |
; uint32_t quant_h263_intra_sse2(int16_t * coeff, |
268 |
; const int16_t const * data, |
; const int16_t const * data, |
269 |
; const uint32_t quant, |
; const uint32_t quant, |
270 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
271 |
|
; const uint16_t *mpeg_matrices); |
272 |
; |
; |
273 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
274 |
|
|
275 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_intra_sse2 |
|
276 |
quant_h263_intra_sse2: |
quant_h263_intra_sse2: |
277 |
|
|
278 |
push esi |
push esi |
291 |
movq2dq xmm7, mm7 |
movq2dq xmm7, mm7 |
292 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
293 |
|
|
294 |
align 16 |
ALIGN 16 |
295 |
.qas2_loop |
.qas2_loop |
296 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
297 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
326 |
|
|
327 |
sub eax, edx |
sub eax, edx |
328 |
jmp short .qas2_mul |
jmp short .qas2_mul |
329 |
|
|
330 |
.qas2_gtzero |
.qas2_gtzero |
331 |
add eax, edx |
add eax, edx |
332 |
|
|
333 |
.qas2_mul |
.qas2_mul |
334 |
cdq |
cdq |
335 |
idiv ecx |
idiv ecx |
336 |
|
|
337 |
mov [edi], ax |
mov [edi], ax |
338 |
|
|
339 |
|
xor eax, eax ; return(0); |
340 |
pop edi |
pop edi |
341 |
pop esi |
pop esi |
342 |
|
|
343 |
ret |
ret |
344 |
|
|
345 |
align 16 |
ALIGN 16 |
346 |
.qas2_q1loop |
.qas2_q1loop |
347 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
348 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
370 |
|
|
371 |
|
|
372 |
|
|
373 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
374 |
; |
; |
375 |
; uint32_t quant_inter_mmx(int16_t * coeff, |
; uint32_t quant_h263_inter_mmx(int16_t * coeff, |
376 |
; const int16_t const * data, |
; const int16_t const * data, |
377 |
; const uint32_t quant); |
; const uint32_t quant, |
378 |
|
; const uint16_t *mpeg_matrices); |
379 |
; |
; |
380 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
381 |
|
|
382 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_inter_mmx |
|
383 |
quant_h263_inter_mmx: |
quant_h263_inter_mmx: |
384 |
|
|
385 |
push ecx |
push ecx |
400 |
|
|
401 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
402 |
|
|
403 |
align ALIGN |
ALIGN 8 |
404 |
.loop |
.loop |
405 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
406 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] |
407 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
408 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
409 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
434 |
movq mm0, mm5 |
movq mm0, mm5 |
435 |
psrlq mm5, 32 |
psrlq mm5, 32 |
436 |
paddd mm0, mm5 |
paddd mm0, mm5 |
|
movd eax, mm0 ; return sum |
|
437 |
|
|
438 |
|
movd eax, mm0 ; return sum |
439 |
pop edi |
pop edi |
440 |
pop esi |
pop esi |
441 |
pop ecx |
pop ecx |
442 |
|
|
443 |
ret |
ret |
444 |
|
|
445 |
align ALIGN |
ALIGN 8 |
446 |
.q1loop |
.q1loop |
447 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
448 |
movq mm3, [esi + 8*ecx+ 8] ; |
movq mm3, [esi + 8*ecx+ 8] ; |
475 |
|
|
476 |
|
|
477 |
|
|
478 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
479 |
; |
; |
480 |
; uint32_t quant_inter_sse2(int16_t * coeff, |
; uint32_t quant_h263_inter_sse2(int16_t * coeff, |
481 |
; const int16_t const * data, |
; const int16_t const * data, |
482 |
; const uint32_t quant); |
; const uint32_t quant, |
483 |
|
; const uint16_t *mpeg_matrices); |
484 |
; |
; |
485 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
486 |
|
|
487 |
align 16 |
ALIGN 16 |
|
cglobal quant_h263_inter_sse2 |
|
488 |
quant_h263_inter_sse2: |
quant_h263_inter_sse2: |
489 |
|
|
490 |
push esi |
push esi |
510 |
movq2dq xmm7, mm0 |
movq2dq xmm7, mm0 |
511 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
512 |
|
|
513 |
align 16 |
ALIGN 16 |
514 |
.qes2_loop |
.qes2_loop |
515 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
516 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
549 |
movq mm5, mm0 |
movq mm5, mm0 |
550 |
psrlq mm5, 32 |
psrlq mm5, 32 |
551 |
paddd mm0, mm5 |
paddd mm0, mm5 |
552 |
|
|
553 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
554 |
|
|
555 |
pop edi |
pop edi |
557 |
|
|
558 |
ret |
ret |
559 |
|
|
560 |
align 16 |
ALIGN 16 |
561 |
.qes2_q1loop |
.qes2_q1loop |
562 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
563 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
588 |
jmp .qes2_done |
jmp .qes2_done |
589 |
|
|
590 |
|
|
591 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
592 |
; |
; |
593 |
; void dequant_intra_mmx(int16_t *data, |
; uint32_t dequant_h263_intra_mmx(int16_t *data, |
594 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
595 |
; const uint32_t quant, |
; const uint32_t quant, |
596 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
597 |
|
; const uint16_t *mpeg_matrices); |
598 |
; |
; |
599 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
600 |
|
|
601 |
; note: we only saturate to +2047 *before* restoring the sign. |
; note: we only saturate to +2047 *before* restoring the sign. |
602 |
; Hence, final clamp really is [-2048,2047] |
; Hence, final clamp really is [-2048,2047] |
603 |
|
|
604 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_intra_mmx |
|
605 |
dequant_h263_intra_mmx: |
dequant_h263_intra_mmx: |
606 |
|
|
607 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
611 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
612 |
mov eax, -16 |
mov eax, -16 |
613 |
|
|
614 |
align ALIGN |
ALIGN 16 |
615 |
.loop |
.loop |
616 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
617 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
651 |
jnz near .loop |
jnz near .loop |
652 |
|
|
653 |
; deal with DC |
; deal with DC |
|
|
|
654 |
movd mm0, [ecx] |
movd mm0, [ecx] |
655 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp+16] ; dcscalar |
656 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
662 |
movd eax, mm0 |
movd eax, mm0 |
663 |
mov [edx], ax |
mov [edx], ax |
664 |
|
|
665 |
|
xor eax, eax ; return(0); |
666 |
ret |
ret |
667 |
|
|
668 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
669 |
; |
; |
670 |
; void dequant_intra_xmm(int16_t *data, |
; uint32_t dequant_h263_intra_xmm(int16_t *data, |
671 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
672 |
; const uint32_t quant, |
; const uint32_t quant, |
673 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
674 |
|
; const uint16_t *mpeg_matrices); |
675 |
; |
; |
676 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
677 |
|
|
678 |
; this is the same as dequant_inter_mmx, except that we're |
; this is the same as dequant_inter_mmx, except that we're |
679 |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
680 |
|
|
681 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_intra_xmm |
|
682 |
dequant_h263_intra_xmm: |
dequant_h263_intra_xmm: |
683 |
|
|
684 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
688 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
689 |
mov eax, -16 |
mov eax, -16 |
690 |
|
|
691 |
align ALIGN |
ALIGN 16 |
692 |
.loop |
.loop |
693 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
694 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
726 |
jnz near .loop |
jnz near .loop |
727 |
|
|
728 |
; deal with DC |
; deal with DC |
|
|
|
729 |
movd mm0, [ecx] |
movd mm0, [ecx] |
730 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp+16] ; dcscalar |
731 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
737 |
movd eax, mm0 |
movd eax, mm0 |
738 |
mov [edx], ax |
mov [edx], ax |
739 |
|
|
740 |
|
xor eax, eax |
741 |
ret |
ret |
742 |
|
|
743 |
|
|
744 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
745 |
; |
; |
746 |
; void dequant_intra_sse2(int16_t *data, |
; uint32_t dequant_h263_intra_sse2(int16_t *data, |
747 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
748 |
; const uint32_t quant, |
; const uint32_t quant, |
749 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
750 |
|
; const uint16_t *mpeg_matrices); |
751 |
; |
; |
752 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
753 |
align ALIGN |
|
754 |
cglobal dequant_h263_intra_sse2 |
ALIGN 16 |
755 |
dequant_h263_intra_sse2: |
dequant_h263_intra_sse2: |
756 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
757 |
mov ecx, [esp+ 8] ; coeff |
mov ecx, [esp+ 8] ; coeff |
764 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
765 |
mov eax, -16 |
mov eax, -16 |
766 |
|
|
767 |
align ALIGN |
ALIGN 16 |
768 |
.loop |
.loop |
769 |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
770 |
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
813 |
movd eax, mm0 |
movd eax, mm0 |
814 |
mov [edx], ax |
mov [edx], ax |
815 |
|
|
816 |
|
xor eax, eax |
817 |
ret |
ret |
818 |
|
|
819 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
|
|
;=========================================================================== |
|
820 |
; |
; |
821 |
; void dequant_inter_mmx(int16_t * data, |
; uint32t dequant_h263_inter_mmx(int16_t * data, |
822 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
823 |
; const uint32_t quant); |
; const uint32_t quant, |
824 |
|
; const uint16_t *mpeg_matrices); |
825 |
; |
; |
826 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
827 |
|
|
828 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_inter_mmx |
|
829 |
dequant_h263_inter_mmx: |
dequant_h263_inter_mmx: |
830 |
|
|
831 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
835 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
836 |
mov eax, -16 |
mov eax, -16 |
837 |
|
|
838 |
align ALIGN |
ALIGN 16 |
839 |
.loop |
.loop |
840 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
841 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
859 |
paddw mm3, mm5 ; + offset |
paddw mm3, mm5 ; + offset |
860 |
paddw mm0, mm1 ; negate back |
paddw mm0, mm1 ; negate back |
861 |
paddw mm3, mm4 ; negate back |
paddw mm3, mm4 ; negate back |
|
|
|
862 |
; saturates to +2047 |
; saturates to +2047 |
863 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
864 |
add eax, 2 |
add eax, 2 |
873 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
874 |
jnz near .loop |
jnz near .loop |
875 |
|
|
876 |
|
xor eax, eax |
877 |
ret |
ret |
878 |
|
|
879 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
880 |
; |
; |
881 |
; void dequant_inter_xmm(int16_t * data, |
; uint32_t dequant_h263_inter_xmm(int16_t * data, |
882 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
883 |
; const uint32_t quant); |
; const uint32_t quant, |
884 |
|
; const uint16_t *mpeg_matrices); |
885 |
; |
; |
886 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
887 |
|
|
888 |
; this is the same as dequant_inter_mmx, |
; this is the same as dequant_inter_mmx, |
889 |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
890 |
|
|
891 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_inter_xmm |
|
892 |
dequant_h263_inter_xmm: |
dequant_h263_inter_xmm: |
893 |
|
|
894 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
898 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
899 |
mov eax, -16 |
mov eax, -16 |
900 |
|
|
901 |
align ALIGN |
ALIGN 16 |
902 |
.loop |
.loop |
903 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
904 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
922 |
paddw mm3, mm5 ; + offset |
paddw mm3, mm5 ; + offset |
923 |
paddw mm0, mm1 ; start restoring sign |
paddw mm0, mm1 ; start restoring sign |
924 |
paddw mm3, mm4 ; start restoring sign |
paddw mm3, mm4 ; start restoring sign |
|
|
|
925 |
; saturates to +2047 |
; saturates to +2047 |
926 |
movq mm2, [mmx_2047] |
movq mm2, [mmx_2047] |
927 |
pminsw mm0, mm2 |
pminsw mm0, mm2 |
934 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
935 |
jnz near .loop |
jnz near .loop |
936 |
|
|
937 |
|
xor eax, eax |
938 |
ret |
ret |
939 |
|
|
940 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
941 |
; |
; |
942 |
; void dequant_inter_sse2(int16_t * data, |
; uint32_t dequant_h263_inter_sse2(int16_t * data, |
943 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
944 |
; const uint32_t quant); |
; const uint32_t quant, |
945 |
|
; const uint16_t *mpeg_matrices); |
946 |
; |
; |
947 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
948 |
align ALIGN |
|
949 |
cglobal dequant_h263_inter_sse2 |
ALIGN 16 |
950 |
dequant_h263_inter_sse2: |
dequant_h263_inter_sse2: |
951 |
mov edx, [esp + 4] ; data |
mov edx, [esp + 4] ; data |
952 |
mov ecx, [esp + 8] ; coeff |
mov ecx, [esp + 8] ; coeff |
959 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
960 |
mov eax, -16 |
mov eax, -16 |
961 |
|
|
962 |
align ALIGN |
ALIGN 16 |
963 |
.loop |
.loop |
964 |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
965 |
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |
998 |
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
999 |
jnz near .loop |
jnz near .loop |
1000 |
|
|
1001 |
|
xor eax, eax |
1002 |
ret |
ret |