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: quantize_h263_mmx.asm,v 1.1.2.3 2003-10-28 22:23:03 edgomez Exp $ |
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * 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 |
|
SECTION .rodata |
46 |
|
|
47 |
plus_one times 8 dw 1 |
ALIGN 16 |
48 |
|
plus_one: |
49 |
|
times 8 dw 1 |
50 |
|
|
51 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
52 |
; |
; |
53 |
; subtract by Q/2 table |
; subtract by Q/2 table |
54 |
; |
; |
55 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro MMX_SUB 1 |
|
|
times 4 dw %1 / 2 |
|
|
%endmacro |
|
56 |
|
|
57 |
align 16 |
ALIGN 16 |
58 |
mmx_sub |
mmx_sub: |
59 |
MMX_SUB 1 |
%assign quant 1 |
60 |
MMX_SUB 2 |
%rep 31 |
61 |
MMX_SUB 3 |
times 4 dw quant / 2 |
62 |
MMX_SUB 4 |
%assign quant quant+1 |
63 |
MMX_SUB 5 |
%endrep |
|
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 |
|
64 |
|
|
65 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
66 |
; |
; |
67 |
; divide by 2Q table |
; divide by 2Q table |
68 |
; |
; |
70 |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
71 |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
72 |
; |
; |
73 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro MMX_DIV 1 |
|
|
times 4 dw (1 << 16) / (%1 * 2) + 1 |
|
|
%endmacro |
|
74 |
|
|
75 |
align 16 |
ALIGN 16 |
76 |
mmx_div |
mmx_div: |
77 |
MMX_DIV 1 |
%assign quant 1 |
78 |
MMX_DIV 2 |
%rep 31 |
79 |
MMX_DIV 3 |
times 4 dw (1<<16) / (quant*2) + 1 |
80 |
MMX_DIV 4 |
%assign quant quant+1 |
81 |
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 |
|
82 |
|
|
83 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
84 |
; |
; |
85 |
; add by (odd(Q) ? Q : Q - 1) table |
; add by (odd(Q) ? Q : Q - 1) table |
86 |
; |
; |
87 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
88 |
|
|
89 |
%macro MMX_ADD 1 |
ALIGN 16 |
90 |
%if %1 % 2 != 0 |
mmx_add: |
91 |
times 4 dw %1 |
%assign quant 1 |
92 |
|
%rep 31 |
93 |
|
%if quant % 2 != 0 |
94 |
|
times 4 dw quant |
95 |
%else |
%else |
96 |
times 4 dw %1 - 1 |
times 4 dw quant - 1 |
97 |
%endif |
%endif |
98 |
%endmacro |
%assign quant quant+1 |
99 |
|
%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 |
|
100 |
|
|
101 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
102 |
; |
; |
103 |
; multiple by 2Q table |
; multiple by 2Q table |
104 |
; |
; |
105 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro MMX_MUL 1 |
|
|
times 4 dw %1 * 2 |
|
|
%endmacro |
|
106 |
|
|
107 |
align 16 |
ALIGN 16 |
108 |
mmx_mul |
mmx_mul: |
109 |
MMX_MUL 1 |
%assign quant 1 |
110 |
MMX_MUL 2 |
%rep 31 |
111 |
MMX_MUL 3 |
times 4 dw quant*2 |
112 |
MMX_MUL 4 |
%assign quant quant+1 |
113 |
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 |
|
114 |
|
|
115 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
116 |
; |
; |
117 |
; saturation limits |
; saturation limits |
118 |
; |
; |
119 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
120 |
|
|
121 |
|
ALIGN 16 |
122 |
|
sse2_2047: |
123 |
|
times 8 dw 2047 |
124 |
|
|
125 |
align 16 |
ALIGN 16 |
126 |
sse2_2047 times 8 dw 2047 |
mmx_2047: |
127 |
|
times 4 dw 2047 |
128 |
|
|
129 |
align 16 |
ALIGN 8 |
130 |
mmx_2047 times 4 dw 2047 |
mmx_32768_minus_2048: |
131 |
|
times 4 dw (32768-2048) |
132 |
|
|
133 |
align 8 |
mmx_32767_minus_2047: |
134 |
mmx_32768_minus_2048 times 4 dw (32768-2048) |
times 4 dw (32767-2047) |
|
mmx_32767_minus_2047 times 4 dw (32767-2047) |
|
135 |
|
|
136 |
|
|
137 |
section .text |
;============================================================================= |
138 |
|
; Code |
139 |
|
;============================================================================= |
140 |
|
|
141 |
|
SECTION .text |
142 |
|
|
143 |
;=========================================================================== |
cglobal quant_h263_intra_mmx |
144 |
|
cglobal quant_h263_intra_sse2 |
145 |
|
cglobal quant_h263_inter_mmx |
146 |
|
cglobal quant_h263_inter_sse2 |
147 |
|
cglobal dequant_h263_intra_mmx |
148 |
|
cglobal dequant_h263_intra_xmm |
149 |
|
cglobal dequant_h263_intra_sse2 |
150 |
|
cglobal dequant_h263_inter_mmx |
151 |
|
cglobal dequant_h263_inter_xmm |
152 |
|
cglobal dequant_h263_inter_sse2 |
153 |
|
|
154 |
|
;----------------------------------------------------------------------------- |
155 |
; |
; |
156 |
; void quant_intra_mmx(int16_t * coeff, |
; uint32_t quant_h263_intra_mmx(int16_t * coeff, |
157 |
; const int16_t const * data, |
; const int16_t const * data, |
158 |
; const uint32_t quant, |
; const uint32_t quant, |
159 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
160 |
; |
; |
161 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
162 |
|
|
163 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_intra_mmx |
|
164 |
quant_h263_intra_mmx: |
quant_h263_intra_mmx: |
165 |
|
|
166 |
push ecx |
push ecx |
176 |
jz .q1loop |
jz .q1loop |
177 |
|
|
178 |
movq mm7, [mmx_div + eax * 8 - 8] |
movq mm7, [mmx_div + eax * 8 - 8] |
179 |
align ALIGN |
|
180 |
|
ALIGN 16 |
181 |
.loop |
.loop |
182 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
183 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] |
184 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
185 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
186 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
203 |
jnz .loop |
jnz .loop |
204 |
|
|
205 |
.done |
.done |
|
; caclulate data[0] // (int32_t)dcscalar) |
|
206 |
|
|
207 |
|
; caclulate data[0] // (int32_t)dcscalar) |
208 |
mov ecx, [esp + 12 + 16] ; dcscalar |
mov ecx, [esp + 12 + 16] ; dcscalar |
209 |
mov edx, ecx |
mov edx, ecx |
210 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
214 |
|
|
215 |
sub eax, edx |
sub eax, edx |
216 |
jmp short .mul |
jmp short .mul |
217 |
|
|
218 |
.gtzero |
.gtzero |
219 |
add eax, edx |
add eax, edx |
220 |
.mul |
.mul |
221 |
cdq ; expand eax -> edx:eax |
cdq ; expand eax -> edx:eax |
222 |
idiv ecx ; eax = edx:eax / dcscalar |
idiv ecx ; eax = edx:eax / dcscalar |
|
|
|
223 |
mov [edi], ax ; coeff[0] = ax |
mov [edi], ax ; coeff[0] = ax |
224 |
|
|
225 |
|
xor eax, eax ; return(0); |
226 |
pop edi |
pop edi |
227 |
pop esi |
pop esi |
228 |
pop ecx |
pop ecx |
229 |
|
|
230 |
ret |
ret |
231 |
|
|
232 |
align ALIGN |
ALIGN 16 |
233 |
.q1loop |
.q1loop |
234 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
235 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] |
236 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
237 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
238 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
244 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
245 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
246 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
247 |
pxor mm3, mm4 ; |
pxor mm3, mm4 |
248 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
249 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
250 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
257 |
|
|
258 |
|
|
259 |
|
|
260 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
261 |
; |
; |
262 |
; void quant_intra_sse2(int16_t * coeff, |
; uint32_t quant_h263_intra_sse2(int16_t * coeff, |
263 |
; const int16_t const * data, |
; const int16_t const * data, |
264 |
; const uint32_t quant, |
; const uint32_t quant, |
265 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
266 |
; |
; |
267 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
268 |
|
|
269 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_intra_sse2 |
|
270 |
quant_h263_intra_sse2: |
quant_h263_intra_sse2: |
271 |
|
|
272 |
push esi |
push esi |
285 |
movq2dq xmm7, mm7 |
movq2dq xmm7, mm7 |
286 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
287 |
|
|
288 |
align 16 |
ALIGN 16 |
289 |
.qas2_loop |
.qas2_loop |
290 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
291 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
320 |
|
|
321 |
sub eax, edx |
sub eax, edx |
322 |
jmp short .qas2_mul |
jmp short .qas2_mul |
323 |
|
|
324 |
.qas2_gtzero |
.qas2_gtzero |
325 |
add eax, edx |
add eax, edx |
326 |
|
|
327 |
.qas2_mul |
.qas2_mul |
328 |
cdq |
cdq |
329 |
idiv ecx |
idiv ecx |
330 |
|
|
331 |
mov [edi], ax |
mov [edi], ax |
332 |
|
|
333 |
|
xor eax, eax ; return(0); |
334 |
pop edi |
pop edi |
335 |
pop esi |
pop esi |
336 |
|
|
337 |
ret |
ret |
338 |
|
|
339 |
align 16 |
ALIGN 16 |
340 |
.qas2_q1loop |
.qas2_q1loop |
341 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
342 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
364 |
|
|
365 |
|
|
366 |
|
|
367 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
368 |
; |
; |
369 |
; uint32_t quant_inter_mmx(int16_t * coeff, |
; uint32_t quant_h263_inter_mmx(int16_t * coeff, |
370 |
; const int16_t const * data, |
; const int16_t const * data, |
371 |
; const uint32_t quant); |
; const uint32_t quant); |
372 |
; |
; |
373 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
374 |
|
|
375 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_h263_inter_mmx |
|
376 |
quant_h263_inter_mmx: |
quant_h263_inter_mmx: |
377 |
|
|
378 |
push ecx |
push ecx |
393 |
|
|
394 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
395 |
|
|
396 |
align ALIGN |
ALIGN 8 |
397 |
.loop |
.loop |
398 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
399 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] |
400 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
401 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
402 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
427 |
movq mm0, mm5 |
movq mm0, mm5 |
428 |
psrlq mm5, 32 |
psrlq mm5, 32 |
429 |
paddd mm0, mm5 |
paddd mm0, mm5 |
|
movd eax, mm0 ; return sum |
|
430 |
|
|
431 |
|
movd eax, mm0 ; return sum |
432 |
pop edi |
pop edi |
433 |
pop esi |
pop esi |
434 |
pop ecx |
pop ecx |
435 |
|
|
436 |
ret |
ret |
437 |
|
|
438 |
align ALIGN |
ALIGN 8 |
439 |
.q1loop |
.q1loop |
440 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
441 |
movq mm3, [esi + 8*ecx+ 8] ; |
movq mm3, [esi + 8*ecx+ 8] ; |
468 |
|
|
469 |
|
|
470 |
|
|
471 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
472 |
; |
; |
473 |
; uint32_t quant_inter_sse2(int16_t * coeff, |
; uint32_t quant_h263_inter_sse2(int16_t * coeff, |
474 |
; const int16_t const * data, |
; const int16_t const * data, |
475 |
; const uint32_t quant); |
; const uint32_t quant); |
476 |
; |
; |
477 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
478 |
|
|
479 |
align 16 |
ALIGN 16 |
|
cglobal quant_h263_inter_sse2 |
|
480 |
quant_h263_inter_sse2: |
quant_h263_inter_sse2: |
481 |
|
|
482 |
push esi |
push esi |
502 |
movq2dq xmm7, mm0 |
movq2dq xmm7, mm0 |
503 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
504 |
|
|
505 |
align 16 |
ALIGN 16 |
506 |
.qes2_loop |
.qes2_loop |
507 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
508 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
541 |
movq mm5, mm0 |
movq mm5, mm0 |
542 |
psrlq mm5, 32 |
psrlq mm5, 32 |
543 |
paddd mm0, mm5 |
paddd mm0, mm5 |
544 |
|
|
545 |
movd eax, mm0 ; return sum |
movd eax, mm0 ; return sum |
546 |
|
|
547 |
pop edi |
pop edi |
549 |
|
|
550 |
ret |
ret |
551 |
|
|
552 |
align 16 |
ALIGN 16 |
553 |
.qes2_q1loop |
.qes2_q1loop |
554 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
555 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
580 |
jmp .qes2_done |
jmp .qes2_done |
581 |
|
|
582 |
|
|
583 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
584 |
; |
; |
585 |
; void dequant_intra_mmx(int16_t *data, |
; uint32_t dequant_h263_intra_mmx(int16_t *data, |
586 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
587 |
; const uint32_t quant, |
; const uint32_t quant, |
588 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
589 |
; |
; |
590 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
591 |
|
|
592 |
; note: we only saturate to +2047 *before* restoring the sign. |
; note: we only saturate to +2047 *before* restoring the sign. |
593 |
; Hence, final clamp really is [-2048,2047] |
; Hence, final clamp really is [-2048,2047] |
594 |
|
|
595 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_intra_mmx |
|
596 |
dequant_h263_intra_mmx: |
dequant_h263_intra_mmx: |
597 |
|
|
598 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
602 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
603 |
mov eax, -16 |
mov eax, -16 |
604 |
|
|
605 |
align ALIGN |
ALIGN 16 |
606 |
.loop |
.loop |
607 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
608 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
642 |
jnz near .loop |
jnz near .loop |
643 |
|
|
644 |
; deal with DC |
; deal with DC |
|
|
|
645 |
movd mm0, [ecx] |
movd mm0, [ecx] |
646 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp+16] ; dcscalar |
647 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
653 |
movd eax, mm0 |
movd eax, mm0 |
654 |
mov [edx], ax |
mov [edx], ax |
655 |
|
|
656 |
|
xor eax, eax ; return(0); |
657 |
ret |
ret |
658 |
|
|
659 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
660 |
; |
; |
661 |
; void dequant_intra_xmm(int16_t *data, |
; uint32_t dequant_h263_intra_xmm(int16_t *data, |
662 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
663 |
; const uint32_t quant, |
; const uint32_t quant, |
664 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
665 |
; |
; |
666 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
667 |
|
|
668 |
; this is the same as dequant_inter_mmx, except that we're |
; this is the same as dequant_inter_mmx, except that we're |
669 |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
670 |
|
|
671 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_intra_xmm |
|
672 |
dequant_h263_intra_xmm: |
dequant_h263_intra_xmm: |
673 |
|
|
674 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
678 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
679 |
mov eax, -16 |
mov eax, -16 |
680 |
|
|
681 |
align ALIGN |
ALIGN 16 |
682 |
.loop |
.loop |
683 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
684 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
716 |
jnz near .loop |
jnz near .loop |
717 |
|
|
718 |
; deal with DC |
; deal with DC |
|
|
|
719 |
movd mm0, [ecx] |
movd mm0, [ecx] |
720 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp+16] ; dcscalar |
721 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
727 |
movd eax, mm0 |
movd eax, mm0 |
728 |
mov [edx], ax |
mov [edx], ax |
729 |
|
|
730 |
|
xor eax, eax |
731 |
ret |
ret |
732 |
|
|
733 |
|
|
734 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
735 |
; |
; |
736 |
; void dequant_intra_sse2(int16_t *data, |
; uint32_t dequant_h263_intra_sse2(int16_t *data, |
737 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
738 |
; const uint32_t quant, |
; const uint32_t quant, |
739 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
740 |
; |
; |
741 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
742 |
align ALIGN |
|
743 |
cglobal dequant_h263_intra_sse2 |
ALIGN 16 |
744 |
dequant_h263_intra_sse2: |
dequant_h263_intra_sse2: |
745 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
746 |
mov ecx, [esp+ 8] ; coeff |
mov ecx, [esp+ 8] ; coeff |
753 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
754 |
mov eax, -16 |
mov eax, -16 |
755 |
|
|
756 |
align ALIGN |
ALIGN 16 |
757 |
.loop |
.loop |
758 |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
759 |
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
802 |
movd eax, mm0 |
movd eax, mm0 |
803 |
mov [edx], ax |
mov [edx], ax |
804 |
|
|
805 |
|
xor eax, eax |
806 |
ret |
ret |
807 |
|
|
808 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
|
|
|
;=========================================================================== |
|
809 |
; |
; |
810 |
; void dequant_inter_mmx(int16_t * data, |
; uint32t dequant_h263_inter_mmx(int16_t * data, |
811 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
812 |
; const uint32_t quant); |
; const uint32_t quant); |
813 |
; |
; |
814 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
815 |
|
|
816 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_inter_mmx |
|
817 |
dequant_h263_inter_mmx: |
dequant_h263_inter_mmx: |
818 |
|
|
819 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
823 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
824 |
mov eax, -16 |
mov eax, -16 |
825 |
|
|
826 |
align ALIGN |
ALIGN 16 |
827 |
.loop |
.loop |
828 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
829 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
847 |
paddw mm3, mm5 ; + offset |
paddw mm3, mm5 ; + offset |
848 |
paddw mm0, mm1 ; negate back |
paddw mm0, mm1 ; negate back |
849 |
paddw mm3, mm4 ; negate back |
paddw mm3, mm4 ; negate back |
|
|
|
850 |
; saturates to +2047 |
; saturates to +2047 |
851 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
852 |
add eax, 2 |
add eax, 2 |
861 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
862 |
jnz near .loop |
jnz near .loop |
863 |
|
|
864 |
|
xor eax, eax |
865 |
ret |
ret |
866 |
|
|
867 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
868 |
; |
; |
869 |
; void dequant_inter_xmm(int16_t * data, |
; uint32_t dequant_h263_inter_xmm(int16_t * data, |
870 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
871 |
; const uint32_t quant); |
; const uint32_t quant); |
872 |
; |
; |
873 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
874 |
|
|
875 |
; this is the same as dequant_inter_mmx, |
; this is the same as dequant_inter_mmx, |
876 |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
877 |
|
|
878 |
align ALIGN |
ALIGN 16 |
|
cglobal dequant_h263_inter_xmm |
|
879 |
dequant_h263_inter_xmm: |
dequant_h263_inter_xmm: |
880 |
|
|
881 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
885 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
886 |
mov eax, -16 |
mov eax, -16 |
887 |
|
|
888 |
align ALIGN |
ALIGN 16 |
889 |
.loop |
.loop |
890 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
891 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
909 |
paddw mm3, mm5 ; + offset |
paddw mm3, mm5 ; + offset |
910 |
paddw mm0, mm1 ; start restoring sign |
paddw mm0, mm1 ; start restoring sign |
911 |
paddw mm3, mm4 ; start restoring sign |
paddw mm3, mm4 ; start restoring sign |
|
|
|
912 |
; saturates to +2047 |
; saturates to +2047 |
913 |
movq mm2, [mmx_2047] |
movq mm2, [mmx_2047] |
914 |
pminsw mm0, mm2 |
pminsw mm0, mm2 |
921 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
922 |
jnz near .loop |
jnz near .loop |
923 |
|
|
924 |
|
xor eax, eax |
925 |
ret |
ret |
926 |
|
|
927 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
928 |
; |
; |
929 |
; void dequant_inter_sse2(int16_t * data, |
; uint32_t dequant_h263_inter_sse2(int16_t * data, |
930 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
931 |
; const uint32_t quant); |
; const uint32_t quant); |
932 |
; |
; |
933 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
934 |
align ALIGN |
|
935 |
cglobal dequant_h263_inter_sse2 |
ALIGN 16 |
936 |
dequant_h263_inter_sse2: |
dequant_h263_inter_sse2: |
937 |
mov edx, [esp + 4] ; data |
mov edx, [esp + 4] ; data |
938 |
mov ecx, [esp + 8] ; coeff |
mov ecx, [esp + 8] ; coeff |
945 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
946 |
mov eax, -16 |
mov eax, -16 |
947 |
|
|
948 |
align ALIGN |
ALIGN 16 |
949 |
.loop |
.loop |
950 |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
951 |
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |
984 |
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
985 |
jnz near .loop |
jnz near .loop |
986 |
|
|
987 |
|
xor eax, eax |
988 |
ret |
ret |