| 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> |
| 8 |
; * to use this software module in hardware or software products are |
; * 2004 Jean-Marc Bastide <jmtest@voila.fr> |
|
; * 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. |
|
| 9 |
; * |
; * |
| 10 |
; * This program is free software; you can redistribute it and/or modify |
; * This program is free software; you can redistribute it and/or modify |
| 11 |
; * 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 |
| 19 |
; * |
; * |
| 20 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
| 21 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
| 22 |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 23 |
; * |
; * |
| 24 |
; *************************************************************************/ |
; * $Id$ |
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
| 25 |
; * |
; * |
| 26 |
; * 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> |
|
|
; * |
|
|
; *************************************************************************/ |
|
| 27 |
|
|
| 28 |
; enable dequant saturate [-2048,2047], test purposes only. |
; enable dequant saturate [-2048,2047], test purposes only. |
| 29 |
%define SATURATE |
%define SATURATE |
| 30 |
|
|
| 31 |
; data/text alignment |
%include "nasm.inc" |
|
%define ALIGN 8 |
|
| 32 |
|
|
| 33 |
bits 32 |
;============================================================================= |
| 34 |
|
; Read only Local data |
| 35 |
|
;============================================================================= |
| 36 |
|
|
| 37 |
section .data |
DATA |
| 38 |
|
|
| 39 |
|
ALIGN SECTION_ALIGN |
| 40 |
|
plus_one: |
| 41 |
|
times 8 dw 1 |
| 42 |
|
|
| 43 |
%macro cglobal 1 |
;----------------------------------------------------------------------------- |
| 44 |
%ifdef PREFIX |
; |
| 45 |
global _%1 |
; quant table |
| 46 |
%define %1 _%1 |
; |
| 47 |
%else |
;----------------------------------------------------------------------------- |
|
global %1 |
|
|
%endif |
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
| 48 |
|
|
| 49 |
plus_one times 8 dw 1 |
ALIGN SECTION_ALIGN |
| 50 |
|
mmx_quant: |
| 51 |
|
%assign quant 0 |
| 52 |
|
%rep 32 |
| 53 |
|
times 4 dw quant |
| 54 |
|
%assign quant quant+1 |
| 55 |
|
%endrep |
| 56 |
|
|
| 57 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 58 |
; |
; |
| 59 |
; subtract by Q/2 table |
; subtract by Q/2 table |
| 60 |
; |
; |
| 61 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 62 |
|
|
| 63 |
%macro MMX_SUB 1 |
ALIGN SECTION_ALIGN |
| 64 |
times 4 dw %1 / 2 |
mmx_sub: |
| 65 |
%endmacro |
%assign quant 1 |
| 66 |
|
%rep 31 |
| 67 |
align 16 |
times 4 dw quant / 2 |
| 68 |
mmx_sub |
%assign quant quant+1 |
| 69 |
MMX_SUB 1 |
%endrep |
|
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 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
| 72 |
; |
; |
| 73 |
; divide by 2Q table |
; divide by 2Q table |
| 74 |
; |
; |
| 76 |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
; for q=1, _pmulhw_ will overflow so it is treated seperately |
| 77 |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
; (3dnow2 provides _pmulhuw_ which wont cause overflow) |
| 78 |
; |
; |
| 79 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
%macro MMX_DIV 1 |
|
|
times 4 dw (1 << 16) / (%1 * 2) + 1 |
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
|
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 |
|
|
|
|
|
|
|
|
|
|
|
;=========================================================================== |
|
|
; |
|
|
; add by (odd(Q) ? Q : Q - 1) table |
|
|
; |
|
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_ADD 1 |
|
|
%if %1 % 2 != 0 |
|
|
times 4 dw %1 |
|
|
%else |
|
|
times 4 dw %1 - 1 |
|
|
%endif |
|
|
%endmacro |
|
| 80 |
|
|
| 81 |
align 16 |
ALIGN SECTION_ALIGN |
| 82 |
mmx_add |
mmx_div: |
| 83 |
MMX_ADD 1 |
%assign quant 1 |
| 84 |
MMX_ADD 2 |
%rep 31 |
| 85 |
MMX_ADD 3 |
times 4 dw (1<<16) / (quant*2) + 1 |
| 86 |
MMX_ADD 4 |
%assign quant quant+1 |
| 87 |
MMX_ADD 5 |
%endrep |
| 88 |
MMX_ADD 6 |
|
| 89 |
MMX_ADD 7 |
;============================================================================= |
| 90 |
MMX_ADD 8 |
; Code |
| 91 |
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 |
|
|
|
|
|
|
|
|
;=========================================================================== |
|
|
; |
|
|
; multiple by 2Q table |
|
|
; |
|
|
;=========================================================================== |
|
|
|
|
|
%macro MMX_MUL 1 |
|
|
times 4 dw %1 * 2 |
|
|
%endmacro |
|
|
|
|
|
align 16 |
|
|
mmx_mul |
|
|
MMX_MUL 1 |
|
|
MMX_MUL 2 |
|
|
MMX_MUL 3 |
|
|
MMX_MUL 4 |
|
|
MMX_MUL 5 |
|
|
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 |
|
|
|
|
|
|
|
|
;=========================================================================== |
|
|
; |
|
|
; saturation limits |
|
|
; |
|
|
;=========================================================================== |
|
|
|
|
|
align 16 |
|
|
sse2_2047 times 8 dw 2047 |
|
|
|
|
|
align 16 |
|
|
mmx_2047 times 4 dw 2047 |
|
|
|
|
|
align 8 |
|
|
mmx_32768_minus_2048 times 4 dw (32768-2048) |
|
|
mmx_32767_minus_2047 times 4 dw (32767-2047) |
|
| 92 |
|
|
| 93 |
|
SECTION .rotext align=SECTION_ALIGN |
| 94 |
|
|
| 95 |
section .text |
cglobal quant_h263_intra_mmx |
| 96 |
|
cglobal quant_h263_intra_sse2 |
| 97 |
|
cglobal quant_h263_inter_mmx |
| 98 |
|
cglobal quant_h263_inter_sse2 |
| 99 |
|
cglobal dequant_h263_intra_mmx |
| 100 |
|
cglobal dequant_h263_intra_xmm |
| 101 |
|
cglobal dequant_h263_intra_sse2 |
| 102 |
|
cglobal dequant_h263_inter_mmx |
| 103 |
|
cglobal dequant_h263_inter_xmm |
| 104 |
|
cglobal dequant_h263_inter_sse2 |
| 105 |
|
|
| 106 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 107 |
; |
; |
| 108 |
; void quant_intra_mmx(int16_t * coeff, |
; uint32_t quant_h263_intra_mmx(int16_t * coeff, |
| 109 |
; const int16_t const * data, |
; const int16_t const * data, |
| 110 |
; const uint32_t quant, |
; const uint32_t quant, |
| 111 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
| 112 |
|
; const uint16_t *mpeg_matrices); |
| 113 |
; |
; |
| 114 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 115 |
|
|
| 116 |
align ALIGN |
ALIGN SECTION_ALIGN |
|
cglobal quant_h263_intra_mmx |
|
| 117 |
quant_h263_intra_mmx: |
quant_h263_intra_mmx: |
| 118 |
|
|
| 119 |
push ecx |
mov _EAX, prm2 ; data |
| 120 |
push esi |
mov TMP0, prm4 ; dcscalar |
| 121 |
push edi |
movsx _EAX, word [_EAX] ; data[0] |
| 122 |
|
|
| 123 |
mov edi, [esp + 12 + 4] ; coeff |
sar TMP0, 1 ; dcscalar /2 |
| 124 |
mov esi, [esp + 12 + 8] ; data |
mov TMP1, _EAX |
| 125 |
mov eax, [esp + 12 + 12] ; quant |
sar TMP1, 31 ; sgn(data[0]) |
| 126 |
|
xor TMP0,TMP1 ; *sgn(data[0]) |
| 127 |
xor ecx, ecx |
sub _EAX,TMP1 |
| 128 |
cmp al, 1 |
add _EAX,TMP0 ; + (dcscalar/2)*sgn(data[0]) |
| 129 |
jz .q1loop |
|
| 130 |
|
mov TMP0, prm3 ; quant |
| 131 |
movq mm7, [mmx_div + eax * 8 - 8] |
lea TMP1, [mmx_div] |
| 132 |
align ALIGN |
movq mm7, [TMP1+TMP0 * 8 - 8] |
| 133 |
.loop |
%ifdef ARCH_IS_X86_64 |
| 134 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
%ifdef WINDOWS |
| 135 |
movq mm3, [esi + 8*ecx + 8] ; |
mov TMP1, prm2 |
| 136 |
pxor mm1, mm1 ; mm1 = 0 |
%endif |
| 137 |
pxor mm4, mm4 ; |
%endif |
| 138 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
cdq |
| 139 |
pcmpgtw mm4, mm3 ; |
idiv prm4d ; dcscalar |
| 140 |
pxor mm0, mm1 ; mm0 = |mm0| |
%ifdef ARCH_IS_X86_64 |
| 141 |
pxor mm3, mm4 ; |
%ifdef WINDOWS |
| 142 |
psubw mm0, mm1 ; displace |
mov prm2, TMP1 |
| 143 |
psubw mm3, mm4 ; |
%endif |
| 144 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
%endif |
| 145 |
pmulhw mm3, mm7 ; |
cmp TMP0, 1 |
| 146 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
mov TMP1, prm1 ; coeff |
| 147 |
pxor mm3, mm4 ; |
je .low |
| 148 |
psubw mm0, mm1 ; undisplace |
|
| 149 |
psubw mm3, mm4 ; |
mov TMP0, prm2 ; data |
| 150 |
movq [edi + 8*ecx], mm0 |
push _EAX ; DC |
| 151 |
movq [edi + 8*ecx + 8], mm3 |
mov _EAX, TMP0 |
|
|
|
|
add ecx,2 |
|
|
cmp ecx,16 |
|
|
jnz .loop |
|
| 152 |
|
|
| 153 |
.done |
mov TMP0,4 |
|
; caclulate data[0] // (int32_t)dcscalar) |
|
| 154 |
|
|
| 155 |
mov ecx, [esp + 12 + 16] ; dcscalar |
.loop: |
| 156 |
mov edx, ecx |
movq mm0, [_EAX] ; data |
| 157 |
movsx eax, word [esi] ; data[0] |
pxor mm4,mm4 |
| 158 |
shr edx, 1 ; edx = dcscalar /2 |
movq mm1, [_EAX + 8] |
| 159 |
cmp eax, 0 |
pcmpgtw mm4,mm0 ; (data<0) |
| 160 |
jg .gtzero |
pxor mm5,mm5 |
| 161 |
|
pmulhw mm0,mm7 ; /(2*quant) |
| 162 |
sub eax, edx |
pcmpgtw mm5,mm1 |
| 163 |
jmp short .mul |
movq mm2, [_EAX+16] |
| 164 |
.gtzero |
psubw mm0,mm4 ; +(data<0) |
| 165 |
add eax, edx |
pmulhw mm1,mm7 |
| 166 |
.mul |
pxor mm4,mm4 |
| 167 |
cdq ; expand eax -> edx:eax |
movq mm3,[_EAX+24] |
| 168 |
idiv ecx ; eax = edx:eax / dcscalar |
pcmpgtw mm4,mm2 |
| 169 |
|
psubw mm1,mm5 |
| 170 |
mov [edi], ax ; coeff[0] = ax |
pmulhw mm2,mm7 |
| 171 |
|
pxor mm5,mm5 |
| 172 |
pop edi |
pcmpgtw mm5,mm3 |
| 173 |
pop esi |
pmulhw mm3,mm7 |
| 174 |
pop ecx |
psubw mm2,mm4 |
| 175 |
|
psubw mm3,mm5 |
| 176 |
|
movq [TMP1], mm0 |
| 177 |
|
lea _EAX, [_EAX+32] |
| 178 |
|
movq [TMP1 + 8], mm1 |
| 179 |
|
movq [TMP1 + 16], mm2 |
| 180 |
|
movq [TMP1 + 24], mm3 |
| 181 |
|
|
| 182 |
|
dec TMP0 |
| 183 |
|
lea TMP1, [TMP1+32] |
| 184 |
|
jne .loop |
| 185 |
|
jmp .end |
| 186 |
|
|
| 187 |
|
.low: |
| 188 |
|
movd mm7,TMP0d |
| 189 |
|
|
| 190 |
|
mov TMP0, prm2 |
| 191 |
|
push _EAX |
| 192 |
|
mov _EAX, TMP0 |
| 193 |
|
|
| 194 |
|
mov TMP0,4 |
| 195 |
|
.loop_low: |
| 196 |
|
movq mm0, [_EAX] |
| 197 |
|
pxor mm4,mm4 |
| 198 |
|
movq mm1, [_EAX + 8] |
| 199 |
|
pcmpgtw mm4,mm0 |
| 200 |
|
pxor mm5,mm5 |
| 201 |
|
psubw mm0,mm4 |
| 202 |
|
pcmpgtw mm5,mm1 |
| 203 |
|
psraw mm0,mm7 |
| 204 |
|
psubw mm1,mm5 |
| 205 |
|
movq mm2,[_EAX+16] |
| 206 |
|
pxor mm4,mm4 |
| 207 |
|
psraw mm1,mm7 |
| 208 |
|
pcmpgtw mm4,mm2 |
| 209 |
|
pxor mm5,mm5 |
| 210 |
|
psubw mm2,mm4 |
| 211 |
|
movq mm3,[_EAX+24] |
| 212 |
|
pcmpgtw mm5,mm3 |
| 213 |
|
psraw mm2,mm7 |
| 214 |
|
psubw mm3,mm5 |
| 215 |
|
movq [TMP1], mm0 |
| 216 |
|
psraw mm3,mm7 |
| 217 |
|
movq [TMP1 + 8], mm1 |
| 218 |
|
movq [TMP1+16],mm2 |
| 219 |
|
lea _EAX, [_EAX+32] |
| 220 |
|
movq [TMP1+24],mm3 |
| 221 |
|
|
| 222 |
|
dec TMP0 |
| 223 |
|
lea TMP1, [TMP1+32] |
| 224 |
|
jne .loop_low |
| 225 |
|
|
| 226 |
|
.end: |
| 227 |
|
|
| 228 |
|
pop _EAX |
| 229 |
|
|
| 230 |
|
mov TMP1, prm1 ; coeff |
| 231 |
|
mov [TMP1],ax |
| 232 |
|
xor _EAX,_EAX ; return 0 |
| 233 |
|
|
| 234 |
ret |
ret |
| 235 |
|
ENDFUNC |
|
align ALIGN |
|
|
.q1loop |
|
|
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
|
|
movq mm3, [esi + 8*ecx + 8] ; |
|
|
pxor mm1, mm1 ; mm1 = 0 |
|
|
pxor mm4, mm4 ; |
|
|
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
|
|
pcmpgtw mm4, mm3 ; |
|
|
pxor mm0, mm1 ; mm0 = |mm0| |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; displace |
|
|
psubw mm3, mm4 ; |
|
|
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
|
|
psrlw mm3, 1 ; |
|
|
pxor mm0, mm1 ; mm0 *= sign(mm0) |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; undisplace |
|
|
psubw mm3, mm4 ; |
|
|
movq [edi + 8*ecx], mm0 |
|
|
movq [edi + 8*ecx + 8], mm3 |
|
|
|
|
|
add ecx,2 |
|
|
cmp ecx,16 |
|
|
jnz .q1loop |
|
|
jmp short .done |
|
| 236 |
|
|
| 237 |
|
|
| 238 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
| 239 |
; |
; |
| 240 |
; void quant_intra_sse2(int16_t * coeff, |
; uint32_t quant_h263_intra_sse2(int16_t * coeff, |
| 241 |
; const int16_t const * data, |
; const int16_t const * data, |
| 242 |
; const uint32_t quant, |
; const uint32_t quant, |
| 243 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
| 244 |
|
; const uint16_t *mpeg_matrices); |
| 245 |
; |
; |
| 246 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 247 |
|
|
| 248 |
align ALIGN |
ALIGN SECTION_ALIGN |
|
cglobal quant_h263_intra_sse2 |
|
| 249 |
quant_h263_intra_sse2: |
quant_h263_intra_sse2: |
| 250 |
|
PUSH_XMM6_XMM7 |
| 251 |
|
mov _EAX, prm2 ; data |
| 252 |
|
|
| 253 |
push esi |
movsx _EAX, word [_EAX] ; data[0] |
|
push edi |
|
|
|
|
|
mov edi, [esp + 8 + 4] ; coeff |
|
|
mov esi, [esp + 8 + 8] ; data |
|
|
mov eax, [esp + 8 + 12] ; quant |
|
| 254 |
|
|
| 255 |
xor ecx, ecx |
mov TMP0,prm4 ; dcscalar |
| 256 |
cmp al, 1 |
mov TMP1,_EAX |
| 257 |
jz near .qas2_q1loop |
sar TMP0,1 |
| 258 |
|
add _EAX,TMP0 |
| 259 |
|
sub TMP1,TMP0 |
| 260 |
|
cmovl _EAX,TMP1 ; +/- dcscalar/2 |
| 261 |
|
mov TMP0, prm3 ; quant |
| 262 |
|
lea TMP1, [mmx_div] |
| 263 |
|
movq xmm7, [TMP1+TMP0 * 8 - 8] |
| 264 |
|
|
| 265 |
|
%ifdef ARCH_IS_X86_64 |
| 266 |
|
%ifdef WINDOWS |
| 267 |
|
mov TMP1, prm2 |
| 268 |
|
%endif |
| 269 |
|
%endif |
| 270 |
|
cdq |
| 271 |
|
idiv prm4d ; dcscalar |
| 272 |
|
%ifdef ARCH_IS_X86_64 |
| 273 |
|
%ifdef WINDOWS |
| 274 |
|
mov prm2, TMP1 |
| 275 |
|
%endif |
| 276 |
|
%endif |
| 277 |
|
cmp TMP0, 1 |
| 278 |
|
mov TMP1, prm1 ; coeff |
| 279 |
|
je near .low |
| 280 |
|
|
| 281 |
|
mov TMP0, prm2 |
| 282 |
|
push _EAX ; DC |
| 283 |
|
mov _EAX, TMP0 |
| 284 |
|
|
| 285 |
.qas2_not1 |
mov TMP0,2 |
|
movq mm7, [mmx_div + eax*8 - 8] |
|
|
movq2dq xmm7, mm7 |
|
| 286 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
| 287 |
|
|
| 288 |
align 16 |
.loop: |
| 289 |
.qas2_loop |
movdqa xmm0, [_EAX] |
|
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
|
|
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
|
|
pxor xmm1, xmm1 |
|
| 290 |
pxor xmm4, xmm4 |
pxor xmm4, xmm4 |
| 291 |
pcmpgtw xmm1, xmm0 |
movdqa xmm1, [_EAX + 16] |
| 292 |
pcmpgtw xmm4, xmm3 |
pcmpgtw xmm4,xmm0 |
| 293 |
pxor xmm0, xmm1 |
pxor xmm5,xmm5 |
|
pxor xmm3, xmm4 |
|
|
psubw xmm0, xmm1 |
|
|
psubw xmm3, xmm4 |
|
| 294 |
pmulhw xmm0, xmm7 |
pmulhw xmm0, xmm7 |
| 295 |
|
pcmpgtw xmm5,xmm1 |
| 296 |
|
movdqa xmm2, [_EAX+32] |
| 297 |
|
psubw xmm0,xmm4 |
| 298 |
|
pmulhw xmm1,xmm7 |
| 299 |
|
pxor xmm4,xmm4 |
| 300 |
|
movdqa xmm3,[_EAX+48] |
| 301 |
|
pcmpgtw xmm4,xmm2 |
| 302 |
|
psubw xmm1,xmm5 |
| 303 |
|
pmulhw xmm2,xmm7 |
| 304 |
|
pxor xmm5,xmm5 |
| 305 |
|
pcmpgtw xmm5,xmm3 |
| 306 |
pmulhw xmm3, xmm7 |
pmulhw xmm3, xmm7 |
| 307 |
pxor xmm0, xmm1 |
psubw xmm2,xmm4 |
| 308 |
pxor xmm3, xmm4 |
psubw xmm3,xmm5 |
| 309 |
psubw xmm0, xmm1 |
movdqa [TMP1], xmm0 |
| 310 |
psubw xmm3, xmm4 |
lea _EAX, [_EAX+64] |
| 311 |
movdqa [edi + ecx*8], xmm0 |
movdqa [TMP1 + 16], xmm1 |
| 312 |
movdqa [edi + ecx*8 + 16], xmm3 |
movdqa [TMP1 + 32], xmm2 |
| 313 |
|
movdqa [TMP1 + 48], xmm3 |
| 314 |
add ecx, 4 |
|
| 315 |
cmp ecx, 16 |
dec TMP0 |
| 316 |
jnz .qas2_loop |
lea TMP1, [TMP1+64] |
| 317 |
|
jne .loop |
| 318 |
.qas2_done |
jmp .end |
| 319 |
mov ecx, [esp + 8 + 16] ; dcscalar |
|
| 320 |
mov edx, ecx |
.low: |
| 321 |
movsx eax, word [esi] |
movd xmm7,TMP0d |
| 322 |
shr edx, 1 |
|
| 323 |
cmp eax, 0 |
mov TMP0, prm2 |
| 324 |
jg .qas2_gtzero |
push _EAX ; DC |
| 325 |
|
mov _EAX, TMP0 |
| 326 |
sub eax, edx |
|
| 327 |
jmp short .qas2_mul |
mov TMP0,2 |
| 328 |
.qas2_gtzero |
.loop_low: |
| 329 |
add eax, edx |
movdqa xmm0, [_EAX] |
|
.qas2_mul |
|
|
cdq |
|
|
idiv ecx |
|
|
|
|
|
mov [edi], ax |
|
|
|
|
|
pop edi |
|
|
pop esi |
|
|
|
|
|
ret |
|
|
|
|
|
align 16 |
|
|
.qas2_q1loop |
|
|
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
|
|
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
|
|
pxor xmm1, xmm1 |
|
| 330 |
pxor xmm4, xmm4 |
pxor xmm4, xmm4 |
| 331 |
pcmpgtw xmm1, xmm0 |
movdqa xmm1, [_EAX + 16] |
| 332 |
pcmpgtw xmm4, xmm3 |
pcmpgtw xmm4,xmm0 |
| 333 |
pxor xmm0, xmm1 |
pxor xmm5,xmm5 |
| 334 |
pxor xmm3, xmm4 |
psubw xmm0,xmm4 |
| 335 |
psubw xmm0, xmm1 |
pcmpgtw xmm5,xmm1 |
| 336 |
psubw xmm3, xmm4 |
psraw xmm0,xmm7 |
| 337 |
psrlw xmm0, 1 |
psubw xmm1,xmm5 |
| 338 |
psrlw xmm3, 1 |
movdqa xmm2,[_EAX+32] |
| 339 |
pxor xmm0, xmm1 |
pxor xmm4,xmm4 |
| 340 |
pxor xmm3, xmm4 |
psraw xmm1,xmm7 |
| 341 |
psubw xmm0, xmm1 |
pcmpgtw xmm4,xmm2 |
| 342 |
psubw xmm3, xmm4 |
pxor xmm5,xmm5 |
| 343 |
movdqa [edi + ecx*8], xmm0 |
psubw xmm2,xmm4 |
| 344 |
movdqa [edi + ecx*8 + 16], xmm3 |
movdqa xmm3,[_EAX+48] |
| 345 |
|
pcmpgtw xmm5,xmm3 |
| 346 |
add ecx, 4 |
psraw xmm2,xmm7 |
| 347 |
cmp ecx, 16 |
psubw xmm3,xmm5 |
| 348 |
jnz .qas2_q1loop |
movdqa [TMP1], xmm0 |
| 349 |
jmp near .qas2_done |
psraw xmm3,xmm7 |
| 350 |
|
movdqa [TMP1+16], xmm1 |
| 351 |
|
movdqa [TMP1+32],xmm2 |
| 352 |
|
lea _EAX, [_EAX+64] |
| 353 |
|
movdqa [TMP1+48],xmm3 |
| 354 |
|
|
| 355 |
|
dec TMP0 |
| 356 |
|
lea TMP1, [TMP1+64] |
| 357 |
|
jne .loop_low |
| 358 |
|
|
| 359 |
|
.end: |
| 360 |
|
|
| 361 |
|
pop _EAX |
| 362 |
|
|
| 363 |
|
mov TMP1, prm1 ; coeff |
| 364 |
|
mov [TMP1],ax |
| 365 |
|
xor _EAX,_EAX ; return 0 |
| 366 |
|
POP_XMM6_XMM7 |
| 367 |
|
ret |
| 368 |
|
ENDFUNC |
| 369 |
|
|
| 370 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 371 |
; |
; |
| 372 |
; uint32_t quant_inter_mmx(int16_t * coeff, |
; uint32_t quant_h263_inter_mmx(int16_t * coeff, |
| 373 |
; const int16_t const * data, |
; const int16_t const * data, |
| 374 |
; const uint32_t quant); |
; const uint32_t quant, |
| 375 |
|
; const uint16_t *mpeg_matrices); |
| 376 |
; |
; |
| 377 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 378 |
|
|
| 379 |
align ALIGN |
ALIGN SECTION_ALIGN |
|
cglobal quant_h263_inter_mmx |
|
| 380 |
quant_h263_inter_mmx: |
quant_h263_inter_mmx: |
| 381 |
|
|
| 382 |
push ecx |
mov TMP1, prm1 ; coeff |
| 383 |
push esi |
mov _EAX, prm3 ; quant |
|
push edi |
|
|
|
|
|
mov edi, [esp + 12 + 4] ; coeff |
|
|
mov esi, [esp + 12 + 8] ; data |
|
|
mov eax, [esp + 12 + 12] ; quant |
|
|
|
|
|
xor ecx, ecx |
|
| 384 |
|
|
| 385 |
pxor mm5, mm5 ; sum |
pxor mm5, mm5 ; sum |
| 386 |
movq mm6, [mmx_sub + eax * 8 - 8] ; sub |
lea TMP0, [mmx_sub] |
| 387 |
|
movq mm6, [TMP0 + _EAX * 8 - 8] ; sub |
| 388 |
|
|
| 389 |
cmp al, 1 |
cmp al, 1 |
| 390 |
jz .q1loop |
jz near .q1routine |
| 391 |
|
|
| 392 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
lea TMP0, [mmx_div] |
| 393 |
|
movq mm7, [TMP0 + _EAX * 8 - 8] ; divider |
| 394 |
|
|
| 395 |
align ALIGN |
xor TMP0, TMP0 |
| 396 |
.loop |
mov _EAX, prm2 ; data |
| 397 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
|
| 398 |
movq mm3, [esi + 8*ecx + 8] ; |
ALIGN SECTION_ALIGN |
| 399 |
|
.loop: |
| 400 |
|
movq mm0, [_EAX + 8*TMP0] ; mm0 = [1st] |
| 401 |
|
movq mm3, [_EAX + 8*TMP0 + 8] |
| 402 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
| 403 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
| 404 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
| 417 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
| 418 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
| 419 |
psubw mm3, mm4 |
psubw mm3, mm4 |
| 420 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*TMP0], mm0 |
| 421 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*TMP0 + 8], mm3 |
| 422 |
|
|
| 423 |
add ecx, 2 |
add TMP0, 2 |
| 424 |
cmp ecx, 16 |
cmp TMP0, 16 |
| 425 |
jnz .loop |
jnz .loop |
| 426 |
|
|
| 427 |
.done |
.done: |
| 428 |
pmaddwd mm5, [plus_one] |
pmaddwd mm5, [plus_one] |
| 429 |
movq mm0, mm5 |
movq mm0, mm5 |
| 430 |
psrlq mm5, 32 |
psrlq mm5, 32 |
| 431 |
paddd mm0, mm5 |
paddd mm0, mm5 |
|
movd eax, mm0 ; return sum |
|
| 432 |
|
|
| 433 |
pop edi |
movd eax, mm0 ; return sum |
|
pop esi |
|
|
pop ecx |
|
| 434 |
|
|
| 435 |
ret |
ret |
| 436 |
|
|
| 437 |
align ALIGN |
.q1routine: |
| 438 |
.q1loop |
xor TMP0, TMP0 |
| 439 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
mov _EAX, prm2 ; data |
| 440 |
movq mm3, [esi + 8*ecx+ 8] ; |
|
| 441 |
|
ALIGN SECTION_ALIGN |
| 442 |
|
.q1loop: |
| 443 |
|
movq mm0, [_EAX + 8*TMP0] ; mm0 = [1st] |
| 444 |
|
movq mm3, [_EAX + 8*TMP0+ 8] ; |
| 445 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
| 446 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
| 447 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
| 460 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
| 461 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
| 462 |
psubw mm3, mm4 |
psubw mm3, mm4 |
| 463 |
movq [edi + 8*ecx], mm0 |
movq [TMP1 + 8*TMP0], mm0 |
| 464 |
movq [edi + 8*ecx + 8], mm3 |
movq [TMP1 + 8*TMP0 + 8], mm3 |
| 465 |
|
|
| 466 |
add ecx,2 |
add TMP0, 2 |
| 467 |
cmp ecx,16 |
cmp TMP0, 16 |
| 468 |
jnz .q1loop |
jnz .q1loop |
| 469 |
|
|
| 470 |
jmp .done |
jmp .done |
| 471 |
|
ENDFUNC |
| 472 |
|
|
| 473 |
|
|
| 474 |
|
|
| 475 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 476 |
; |
; |
| 477 |
; uint32_t quant_inter_sse2(int16_t * coeff, |
; uint32_t quant_h263_inter_sse2(int16_t * coeff, |
| 478 |
; const int16_t const * data, |
; const int16_t const * data, |
| 479 |
; const uint32_t quant); |
; const uint32_t quant, |
| 480 |
|
; const uint16_t *mpeg_matrices); |
| 481 |
; |
; |
| 482 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 483 |
|
|
| 484 |
align 16 |
ALIGN SECTION_ALIGN |
|
cglobal quant_h263_inter_sse2 |
|
| 485 |
quant_h263_inter_sse2: |
quant_h263_inter_sse2: |
| 486 |
|
PUSH_XMM6_XMM7 |
| 487 |
|
|
| 488 |
push esi |
mov TMP1, prm1 ; coeff |
| 489 |
push edi |
mov _EAX, prm3 ; quant |
|
|
|
|
mov edi, [esp + 8 + 4] ; coeff |
|
|
mov esi, [esp + 8 + 8] ; data |
|
|
mov eax, [esp + 8 + 12] ; quant |
|
|
|
|
|
xor ecx, ecx |
|
| 490 |
|
|
| 491 |
pxor xmm5, xmm5 ; sum |
pxor xmm5, xmm5 ; sum |
| 492 |
|
|
| 493 |
movq mm0, [mmx_sub + eax*8 - 8] ; sub |
lea TMP0, [mmx_sub] |
| 494 |
|
movq mm0, [TMP0 + _EAX*8 - 8] ; sub |
| 495 |
movq2dq xmm6, mm0 ; load into low 8 bytes |
movq2dq xmm6, mm0 ; load into low 8 bytes |
| 496 |
movlhps xmm6, xmm6 ; duplicate into high 8 bytes |
movlhps xmm6, xmm6 ; duplicate into high 8 bytes |
| 497 |
|
|
| 498 |
cmp al, 1 |
cmp al, 1 |
| 499 |
jz near .qes2_q1loop |
jz near .qes2_q1_routine |
| 500 |
|
|
| 501 |
|
.qes2_not1: |
| 502 |
|
lea TMP0, [mmx_div] |
| 503 |
|
movq mm0, [TMP0 + _EAX*8 - 8] ; divider |
| 504 |
|
|
| 505 |
|
xor TMP0, TMP0 |
| 506 |
|
mov _EAX, prm2 ; data |
| 507 |
|
|
|
.qes2_not1 |
|
|
movq mm0, [mmx_div + eax*8 - 8] ; divider |
|
| 508 |
movq2dq xmm7, mm0 |
movq2dq xmm7, mm0 |
| 509 |
movlhps xmm7, xmm7 |
movlhps xmm7, xmm7 |
| 510 |
|
|
| 511 |
align 16 |
ALIGN SECTION_ALIGN |
| 512 |
.qes2_loop |
.qes2_loop: |
| 513 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
movdqa xmm0, [_EAX + TMP0*8] ; xmm0 = [1st] |
| 514 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
movdqa xmm3, [_EAX + TMP0*8 + 16] ; xmm3 = [2nd] |
| 515 |
pxor xmm1, xmm1 |
pxor xmm1, xmm1 |
| 516 |
pxor xmm4, xmm4 |
pxor xmm4, xmm4 |
| 517 |
pcmpgtw xmm1, xmm0 |
pcmpgtw xmm1, xmm0 |
| 530 |
pxor xmm3, xmm4 |
pxor xmm3, xmm4 |
| 531 |
psubw xmm0, xmm1 |
psubw xmm0, xmm1 |
| 532 |
psubw xmm3, xmm4 |
psubw xmm3, xmm4 |
| 533 |
movdqa [edi + ecx*8], xmm0 |
movdqa [TMP1 + TMP0*8], xmm0 |
| 534 |
movdqa [edi + ecx*8 + 16], xmm3 |
movdqa [TMP1 + TMP0*8 + 16], xmm3 |
| 535 |
|
|
| 536 |
add ecx, 4 |
add TMP0, 4 |
| 537 |
cmp ecx, 16 |
cmp TMP0, 16 |
| 538 |
jnz .qes2_loop |
jnz .qes2_loop |
| 539 |
|
|
| 540 |
.qes2_done |
.qes2_done: |
| 541 |
movdqu xmm6, [plus_one] |
movdqu xmm6, [plus_one] |
| 542 |
pmaddwd xmm5, xmm6 |
pmaddwd xmm5, xmm6 |
| 543 |
movhlps xmm6, xmm5 |
movhlps xmm6, xmm5 |
| 547 |
movq mm5, mm0 |
movq mm5, mm0 |
| 548 |
psrlq mm5, 32 |
psrlq mm5, 32 |
| 549 |
paddd mm0, mm5 |
paddd mm0, mm5 |
|
movd eax, mm0 ; return sum |
|
| 550 |
|
|
| 551 |
pop edi |
movd eax, mm0 ; return sum |
|
pop esi |
|
| 552 |
|
|
| 553 |
|
POP_XMM6_XMM7 |
| 554 |
ret |
ret |
| 555 |
|
|
| 556 |
align 16 |
.qes2_q1_routine: |
| 557 |
.qes2_q1loop |
xor TMP0, TMP0 |
| 558 |
movdqa xmm0, [esi + ecx*8] ; xmm0 = [1st] |
mov _EAX, prm2 ; data |
| 559 |
movdqa xmm3, [esi + ecx*8 + 16] ; xmm3 = [2nd] |
|
| 560 |
|
ALIGN SECTION_ALIGN |
| 561 |
|
.qes2_q1loop: |
| 562 |
|
movdqa xmm0, [_EAX + TMP0*8] ; xmm0 = [1st] |
| 563 |
|
movdqa xmm3, [_EAX + TMP0*8 + 16] ; xmm3 = [2nd] |
| 564 |
pxor xmm1, xmm1 |
pxor xmm1, xmm1 |
| 565 |
pxor xmm4, xmm4 |
pxor xmm4, xmm4 |
| 566 |
pcmpgtw xmm1, xmm0 |
pcmpgtw xmm1, xmm0 |
| 579 |
pxor xmm3, xmm4 |
pxor xmm3, xmm4 |
| 580 |
psubw xmm0, xmm1 |
psubw xmm0, xmm1 |
| 581 |
psubw xmm3, xmm4 |
psubw xmm3, xmm4 |
| 582 |
movdqa [edi + ecx*8], xmm0 |
movdqa [TMP1 + TMP0*8], xmm0 |
| 583 |
movdqa [edi + ecx*8 + 16], xmm3 |
movdqa [TMP1 + TMP0*8 + 16], xmm3 |
| 584 |
|
|
| 585 |
add ecx,4 |
add TMP0, 4 |
| 586 |
cmp ecx,16 |
cmp TMP0, 16 |
| 587 |
jnz .qes2_q1loop |
jnz .qes2_q1loop |
| 588 |
jmp .qes2_done |
jmp .qes2_done |
| 589 |
|
ENDFUNC |
| 590 |
|
|
| 591 |
|
|
| 592 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 593 |
; |
; |
| 594 |
; void dequant_intra_mmx(int16_t *data, |
; uint32_t dequant_h263_intra_mmx(int16_t *data, |
| 595 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
| 596 |
; const uint32_t quant, |
; const uint32_t quant, |
| 597 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
| 598 |
|
; const uint16_t *mpeg_matrices); |
| 599 |
; |
; |
| 600 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
; note: we only saturate to +2047 *before* restoring the sign. |
|
|
; Hence, final clamp really is [-2048,2047] |
|
| 601 |
|
|
| 602 |
align ALIGN |
ALIGN SECTION_ALIGN |
|
cglobal dequant_h263_intra_mmx |
|
| 603 |
dequant_h263_intra_mmx: |
dequant_h263_intra_mmx: |
| 604 |
|
|
| 605 |
mov edx, [esp+ 4] ; data |
mov TMP0, prm3 ; quant |
| 606 |
mov ecx, [esp+ 8] ; coeff |
mov _EAX, prm2 ; coeff |
| 607 |
mov eax, [esp+12] ; quant |
pcmpeqw mm0,mm0 |
| 608 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
lea TMP1, [mmx_quant] |
| 609 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm6, [TMP1 + TMP0*8] ; quant |
| 610 |
mov eax, -16 |
shl TMP0,31 ; quant & 1 ? 0 : - 1 |
| 611 |
|
movq mm7,mm6 |
| 612 |
align ALIGN |
movq mm5,mm0 |
| 613 |
.loop |
movd mm1,TMP0d |
| 614 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
mov TMP1, prm1 ; data |
| 615 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
psllw mm0,mm1 |
| 616 |
pxor mm1, mm1 |
paddw mm7,mm7 ; 2*quant |
| 617 |
|
paddw mm6,mm0 ; quant-1 |
| 618 |
|
psllw mm5,12 |
| 619 |
|
mov TMP0,8 |
| 620 |
|
psrlw mm5,1 |
| 621 |
|
|
| 622 |
|
.loop: |
| 623 |
|
movq mm0,[_EAX] |
| 624 |
|
pxor mm2,mm2 |
| 625 |
pxor mm4, mm4 |
pxor mm4, mm4 |
| 626 |
pcmpgtw mm1, mm0 ; sign(c) |
pcmpgtw mm2,mm0 |
| 627 |
pcmpgtw mm4, mm3 ; sign(c') |
pcmpeqw mm4,mm0 |
| 628 |
|
pmullw mm0,mm7 ; * 2 * quant |
| 629 |
|
movq mm1,[_EAX+8] |
| 630 |
|
psubw mm0,mm2 |
| 631 |
|
pxor mm2,mm6 |
| 632 |
|
pxor mm3,mm3 |
| 633 |
|
pandn mm4,mm2 |
| 634 |
pxor mm2, mm2 |
pxor mm2, mm2 |
| 635 |
pxor mm5, mm5 |
pcmpgtw mm3,mm1 |
| 636 |
pcmpeqw mm2, mm0 ; c is zero |
pcmpeqw mm2,mm1 |
| 637 |
pcmpeqw mm5, mm3 ; c' is zero |
pmullw mm1,mm7 |
| 638 |
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
paddw mm0,mm4 |
| 639 |
pandn mm5, mm6 |
psubw mm1,mm3 |
| 640 |
pxor mm0, mm1 ; negate if negative |
pxor mm3,mm6 |
| 641 |
pxor mm3, mm4 ; negate if negative |
pandn mm2,mm3 |
| 642 |
psubw mm0, mm1 |
paddsw mm0, mm5 ; saturate |
| 643 |
psubw mm3, mm4 |
paddw mm1,mm2 |
| 644 |
pmullw mm0, mm7 ; *= 2Q |
|
| 645 |
pmullw mm3, mm7 ; *= 2Q |
paddsw mm1, mm5 |
| 646 |
paddw mm0, mm2 ; + offset |
psubsw mm0, mm5 |
| 647 |
paddw mm3, mm5 ; + offset |
psubsw mm1, mm5 |
| 648 |
paddw mm0, mm1 ; negate back |
psubsw mm0, mm5 |
| 649 |
paddw mm3, mm4 ; negate back |
psubsw mm1, mm5 |
| 650 |
|
paddsw mm0, mm5 |
| 651 |
; saturates to +2047 |
paddsw mm1, mm5 |
| 652 |
movq mm2, [mmx_32767_minus_2047] |
|
| 653 |
add eax, 2 |
movq [TMP1],mm0 |
| 654 |
paddsw mm0, mm2 |
lea _EAX,[_EAX+16] |
| 655 |
paddsw mm3, mm2 |
movq [TMP1+8],mm1 |
| 656 |
psubsw mm0, mm2 |
|
| 657 |
psubsw mm3, mm2 |
dec TMP0 |
| 658 |
|
lea TMP1,[TMP1+16] |
| 659 |
pxor mm0, mm1 |
jne .loop |
|
pxor mm3, mm4 |
|
|
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
|
|
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
|
|
jnz near .loop |
|
| 660 |
|
|
| 661 |
; deal with DC |
; deal with DC |
| 662 |
|
mov _EAX, prm2 ; coeff |
| 663 |
movd mm0, [ecx] |
movd mm1,prm4d ; dcscalar |
| 664 |
pmullw mm0, [esp+16] ; dcscalar |
movd mm0,[_EAX] ; coeff[0] |
| 665 |
movq mm2, [mmx_32767_minus_2047] |
pmullw mm0,mm1 ; * dcscalar |
| 666 |
paddsw mm0, mm2 |
mov TMP1, prm1 ; data |
| 667 |
psubsw mm0, mm2 |
paddsw mm0, mm5 ; saturate + |
| 668 |
movq mm3, [mmx_32768_minus_2048] |
psubsw mm0, mm5 |
| 669 |
psubsw mm0, mm3 |
psubsw mm0, mm5 ; saturate - |
| 670 |
paddsw mm0, mm3 |
paddsw mm0, mm5 |
| 671 |
movd eax, mm0 |
movd eax, mm0 |
| 672 |
mov [edx], ax |
mov [TMP1], ax |
| 673 |
|
|
| 674 |
|
xor _EAX, _EAX ; return 0 |
| 675 |
ret |
ret |
| 676 |
|
ENDFUNC |
| 677 |
|
|
| 678 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 679 |
; |
; |
| 680 |
; void dequant_intra_xmm(int16_t *data, |
; uint32_t dequant_h263_intra_xmm(int16_t *data, |
| 681 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
| 682 |
; const uint32_t quant, |
; const uint32_t quant, |
| 683 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
| 684 |
|
; const uint16_t *mpeg_matrices); |
| 685 |
; |
; |
| 686 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 687 |
|
|
|
; this is the same as dequant_inter_mmx, except that we're |
|
|
; saturating using 'pminsw' (saves 2 cycles/loop => ~5% faster) |
|
| 688 |
|
|
| 689 |
align ALIGN |
ALIGN SECTION_ALIGN |
|
cglobal dequant_h263_intra_xmm |
|
| 690 |
dequant_h263_intra_xmm: |
dequant_h263_intra_xmm: |
| 691 |
|
|
| 692 |
mov edx, [esp+ 4] ; data |
mov TMP0, prm3 ; quant |
| 693 |
mov ecx, [esp+ 8] ; coeff |
mov _EAX, prm2 ; coeff |
| 694 |
mov eax, [esp+12] ; quant |
|
| 695 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
movd mm6,TMP0d ; quant |
| 696 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
pcmpeqw mm0,mm0 |
| 697 |
mov eax, -16 |
pshufw mm6,mm6,0 ; all quant |
| 698 |
|
shl TMP0,31 |
| 699 |
align ALIGN |
movq mm5,mm0 |
| 700 |
.loop |
movq mm7,mm6 |
| 701 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movd mm1,TMP0d |
| 702 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
mov TMP1, prm1 ; data |
| 703 |
pxor mm1, mm1 |
psllw mm0,mm1 ; quant & 1 ? 0 : - 1 |
| 704 |
pxor mm4, mm4 |
movq mm4,mm5 |
| 705 |
pcmpgtw mm1, mm0 ; sign(c) |
paddw mm7,mm7 ; quant*2 |
| 706 |
pcmpgtw mm4, mm3 ; sign(c') |
paddw mm6,mm0 ; quant-1 |
| 707 |
|
psrlw mm4,5 ; mm4=2047 |
| 708 |
|
mov TMP0,8 |
| 709 |
|
pxor mm5,mm4 ; mm5=-2048 |
| 710 |
|
|
| 711 |
|
.loop: |
| 712 |
|
movq mm0,[_EAX] |
| 713 |
pxor mm2, mm2 |
pxor mm2, mm2 |
| 714 |
pxor mm5, mm5 |
pxor mm3,mm3 |
|
pcmpeqw mm2, mm0 ; c is zero |
|
|
pcmpeqw mm5, mm3 ; c' is zero |
|
|
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
|
|
pandn mm5, mm6 |
|
|
pxor mm0, mm1 ; negate if negative |
|
|
pxor mm3, mm4 ; negate if negative |
|
|
psubw mm0, mm1 |
|
|
psubw mm3, mm4 |
|
|
pmullw mm0, mm7 ; *= 2Q |
|
|
pmullw mm3, mm7 ; *= 2Q |
|
|
paddw mm0, mm2 ; + offset |
|
|
paddw mm3, mm5 ; + offset |
|
|
paddw mm0, mm1 ; negate back |
|
|
paddw mm3, mm4 ; negate back |
|
|
|
|
|
; saturates to +2047 |
|
|
movq mm2, [mmx_2047] |
|
|
pminsw mm0, mm2 |
|
|
add eax, 2 |
|
|
pminsw mm3, mm2 |
|
|
|
|
|
pxor mm0, mm1 |
|
|
pxor mm3, mm4 |
|
|
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
|
|
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
|
|
jnz near .loop |
|
| 715 |
|
|
| 716 |
; deal with DC |
pcmpgtw mm2,mm0 |
| 717 |
|
pcmpeqw mm3,mm0 ; if coeff==0... |
| 718 |
|
pmullw mm0,mm7 ; * 2 * quant |
| 719 |
|
movq mm1,[_EAX+8] |
| 720 |
|
|
| 721 |
|
psubw mm0,mm2 |
| 722 |
|
pxor mm2,mm6 |
| 723 |
|
pandn mm3,mm2 ; ...then data=0 |
| 724 |
|
pxor mm2,mm2 |
| 725 |
|
paddw mm0,mm3 |
| 726 |
|
pxor mm3,mm3 |
| 727 |
|
pcmpeqw mm2,mm1 |
| 728 |
|
pcmpgtw mm3,mm1 |
| 729 |
|
pmullw mm1,mm7 |
| 730 |
|
|
| 731 |
|
pminsw mm0,mm4 |
| 732 |
|
psubw mm1,mm3 |
| 733 |
|
pxor mm3,mm6 |
| 734 |
|
pandn mm2,mm3 |
| 735 |
|
paddw mm1,mm2 |
| 736 |
|
|
| 737 |
|
pmaxsw mm0,mm5 |
| 738 |
|
pminsw mm1,mm4 |
| 739 |
|
movq [TMP1],mm0 |
| 740 |
|
pmaxsw mm1,mm5 |
| 741 |
|
lea _EAX,[_EAX+16] |
| 742 |
|
movq [TMP1+8],mm1 |
| 743 |
|
|
| 744 |
|
dec TMP0 |
| 745 |
|
lea TMP1,[TMP1+16] |
| 746 |
|
jne .loop |
| 747 |
|
|
| 748 |
movd mm0, [ecx] |
; deal with DC |
| 749 |
pmullw mm0, [esp+16] ; dcscalar |
mov _EAX, prm2 ; coeff |
| 750 |
movq mm2, [mmx_32767_minus_2047] |
movd mm1,prm4d ; dcscalar |
| 751 |
paddsw mm0, mm2 |
movd mm0, [_EAX] |
| 752 |
psubsw mm0, mm2 |
pmullw mm0, mm1 |
| 753 |
movq mm2, [mmx_32768_minus_2048] |
mov TMP1, prm1 ; data |
| 754 |
psubsw mm0, mm2 |
pminsw mm0,mm4 |
| 755 |
paddsw mm0, mm2 |
pmaxsw mm0,mm5 |
| 756 |
movd eax, mm0 |
movd eax, mm0 |
| 757 |
mov [edx], ax |
mov [TMP1], ax |
| 758 |
|
|
| 759 |
|
xor _EAX, _EAX ; return 0 |
| 760 |
ret |
ret |
| 761 |
|
ENDFUNC |
| 762 |
|
|
| 763 |
|
|
| 764 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 765 |
; |
; |
| 766 |
; void dequant_intra_sse2(int16_t *data, |
; uint32_t dequant_h263_intra_sse2(int16_t *data, |
| 767 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
| 768 |
; const uint32_t quant, |
; const uint32_t quant, |
| 769 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar, |
| 770 |
|
; const uint16_t *mpeg_matrices); |
| 771 |
; |
; |
| 772 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 773 |
align ALIGN |
|
| 774 |
cglobal dequant_h263_intra_sse2 |
ALIGN SECTION_ALIGN |
| 775 |
dequant_h263_intra_sse2: |
dequant_h263_intra_sse2: |
| 776 |
mov edx, [esp+ 4] ; data |
PUSH_XMM6_XMM7 |
|
mov ecx, [esp+ 8] ; coeff |
|
|
mov eax, [esp+12] ; quant |
|
|
movq mm6, [mmx_add + eax * 8 - 8] |
|
|
movq mm7, [mmx_mul + eax * 8 - 8] |
|
|
movq2dq xmm6, mm6 |
|
|
movq2dq xmm7, mm7 |
|
|
movlhps xmm6, xmm6 |
|
|
movlhps xmm7, xmm7 |
|
|
mov eax, -16 |
|
| 777 |
|
|
| 778 |
align ALIGN |
mov TMP0, prm3 ; quant |
| 779 |
.loop |
mov _EAX, prm2 ; coeff |
| 780 |
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
|
| 781 |
movdqa xmm3, [ecx + 8*16 + 8*eax+ 16] |
movd xmm6,TMP0d ; quant |
| 782 |
pxor xmm1, xmm1 |
|
| 783 |
pxor xmm4, xmm4 |
shl TMP0,31 |
| 784 |
pcmpgtw xmm1, xmm0 ; sign(c) |
pshuflw xmm6,xmm6,0 |
| 785 |
pcmpgtw xmm4, xmm3 |
pcmpeqw xmm0,xmm0 |
| 786 |
|
movlhps xmm6,xmm6 ; all quant |
| 787 |
|
movd xmm1,TMP0d |
| 788 |
|
movdqa xmm5,xmm0 |
| 789 |
|
movdqa xmm7,xmm6 |
| 790 |
|
mov TMP1, prm1 ; data |
| 791 |
|
paddw xmm7,xmm7 ; quant *2 |
| 792 |
|
psllw xmm0,xmm1 ; quant & 1 ? 0 : - 1 |
| 793 |
|
movdqa xmm4,xmm5 |
| 794 |
|
paddw xmm6,xmm0 ; quant-1 |
| 795 |
|
psrlw xmm4,5 ; 2047 |
| 796 |
|
mov TMP0,4 |
| 797 |
|
pxor xmm5,xmm4 ; mm5=-2048 |
| 798 |
|
|
| 799 |
|
.loop: |
| 800 |
|
movdqa xmm0,[_EAX] |
| 801 |
pxor xmm2, xmm2 |
pxor xmm2, xmm2 |
| 802 |
pxor xmm5, xmm5 |
pxor xmm3,xmm3 |
|
pcmpeqw xmm2, xmm0 ; c is zero |
|
|
pcmpeqw xmm5, xmm3 |
|
|
pandn xmm2, xmm6 ; offset = isZero ? 0 : quant_add |
|
|
pandn xmm5, xmm6 |
|
|
pxor xmm0, xmm1 ; negate if negative |
|
|
pxor xmm3, xmm4 |
|
|
psubw xmm0, xmm1 |
|
|
psubw xmm3, xmm4 |
|
|
pmullw xmm0, xmm7 ; *= 2Q |
|
|
pmullw xmm3, xmm7 |
|
|
paddw xmm0, xmm2 ; + offset |
|
|
paddw xmm3, xmm5 |
|
|
paddw xmm0, xmm1 ; negate back |
|
|
paddw xmm3, xmm4 |
|
|
|
|
|
; saturates to +2047 |
|
|
movdqa xmm2, [sse2_2047] |
|
|
pminsw xmm0, xmm2 |
|
|
add eax, 4 |
|
|
pminsw xmm3, xmm2 |
|
| 803 |
|
|
| 804 |
pxor xmm0, xmm1 |
pcmpgtw xmm2,xmm0 |
| 805 |
pxor xmm3, xmm4 |
pcmpeqw xmm3,xmm0 |
| 806 |
movdqa [edx + 8*16 - 8*4 + 8*eax], xmm0 |
pmullw xmm0,xmm7 ; * 2 * quant |
| 807 |
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
movdqa xmm1,[_EAX+16] |
| 808 |
jnz near .loop |
|
| 809 |
|
psubw xmm0,xmm2 |
| 810 |
|
pxor xmm2,xmm6 |
| 811 |
|
pandn xmm3,xmm2 |
| 812 |
|
pxor xmm2,xmm2 |
| 813 |
|
paddw xmm0,xmm3 |
| 814 |
|
pxor xmm3,xmm3 |
| 815 |
|
pcmpeqw xmm2,xmm1 |
| 816 |
|
pcmpgtw xmm3,xmm1 |
| 817 |
|
pmullw xmm1,xmm7 |
| 818 |
|
|
| 819 |
|
pminsw xmm0,xmm4 |
| 820 |
|
psubw xmm1,xmm3 |
| 821 |
|
pxor xmm3,xmm6 |
| 822 |
|
pandn xmm2,xmm3 |
| 823 |
|
paddw xmm1,xmm2 |
| 824 |
|
|
| 825 |
|
pmaxsw xmm0,xmm5 |
| 826 |
|
pminsw xmm1,xmm4 |
| 827 |
|
movdqa [TMP1],xmm0 |
| 828 |
|
pmaxsw xmm1,xmm5 |
| 829 |
|
lea _EAX,[_EAX+32] |
| 830 |
|
movdqa [TMP1+16],xmm1 |
| 831 |
|
|
| 832 |
|
dec TMP0 |
| 833 |
|
lea TMP1,[TMP1+32] |
| 834 |
|
jne .loop |
| 835 |
|
|
| 836 |
; deal with DC |
; deal with DC |
|
movd mm0, [ecx] |
|
|
pmullw mm0, [esp+16] ; dcscalar |
|
|
movq mm2, [mmx_32767_minus_2047] |
|
|
paddsw mm0, mm2 |
|
|
psubsw mm0, mm2 |
|
|
movq mm2, [mmx_32768_minus_2048] |
|
|
psubsw mm0, mm2 |
|
|
paddsw mm0, mm2 |
|
|
movd eax, mm0 |
|
|
mov [edx], ax |
|
| 837 |
|
|
| 838 |
ret |
mov _EAX, prm2 ; coeff |
| 839 |
|
movsx _EAX,word [_EAX] |
| 840 |
|
imul prm4d ; dcscalar |
| 841 |
|
mov TMP1, prm1 ; data |
| 842 |
|
movd xmm0,eax |
| 843 |
|
pminsw xmm0,xmm4 |
| 844 |
|
pmaxsw xmm0,xmm5 |
| 845 |
|
movd eax,xmm0 |
| 846 |
|
|
| 847 |
|
mov [TMP1], ax |
| 848 |
|
|
| 849 |
|
xor _EAX, _EAX ; return 0 |
| 850 |
|
|
| 851 |
|
POP_XMM6_XMM7 |
| 852 |
|
ret |
| 853 |
|
ENDFUNC |
| 854 |
|
|
| 855 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 856 |
; |
; |
| 857 |
; void dequant_inter_mmx(int16_t * data, |
; uint32t dequant_h263_inter_mmx(int16_t * data, |
| 858 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
| 859 |
; const uint32_t quant); |
; const uint32_t quant, |
| 860 |
|
; const uint16_t *mpeg_matrices); |
| 861 |
; |
; |
| 862 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 863 |
|
|
| 864 |
align ALIGN |
ALIGN SECTION_ALIGN |
|
cglobal dequant_h263_inter_mmx |
|
| 865 |
dequant_h263_inter_mmx: |
dequant_h263_inter_mmx: |
| 866 |
|
|
| 867 |
mov edx, [esp+ 4] ; data |
mov TMP0, prm3 ; quant |
| 868 |
mov ecx, [esp+ 8] ; coeff |
mov _EAX, prm2 ; coeff |
| 869 |
mov eax, [esp+12] ; quant |
pcmpeqw mm0,mm0 |
| 870 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
lea TMP1, [mmx_quant] |
| 871 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm6, [TMP1 + TMP0*8] ; quant |
| 872 |
mov eax, -16 |
shl TMP0,31 ; odd/even |
| 873 |
|
movq mm7,mm6 |
| 874 |
align ALIGN |
movd mm1,TMP0d |
| 875 |
.loop |
mov TMP1, prm1 ; data |
| 876 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
movq mm5,mm0 |
| 877 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
psllw mm0,mm1 ; quant & 1 ? 0 : - 1 |
| 878 |
pxor mm1, mm1 |
paddw mm7,mm7 ; quant*2 |
| 879 |
|
paddw mm6,mm0 ; quant & 1 ? quant : quant - 1 |
| 880 |
|
psllw mm5,12 |
| 881 |
|
mov TMP0,8 |
| 882 |
|
psrlw mm5,1 ; 32767-2047 (32768-2048) |
| 883 |
|
|
| 884 |
|
.loop: |
| 885 |
|
movq mm0,[_EAX] |
| 886 |
pxor mm4, mm4 |
pxor mm4, mm4 |
|
pcmpgtw mm1, mm0 ; sign(c) |
|
|
pcmpgtw mm4, mm3 ; sign(c') |
|
| 887 |
pxor mm2, mm2 |
pxor mm2, mm2 |
| 888 |
pxor mm5, mm5 |
pcmpeqw mm4,mm0 ; if coeff==0... |
| 889 |
pcmpeqw mm2, mm0 ; c is zero |
pcmpgtw mm2,mm0 |
| 890 |
pcmpeqw mm5, mm3 ; c' is zero |
pmullw mm0,mm7 ; * 2 * quant |
| 891 |
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
pxor mm3,mm3 |
| 892 |
pandn mm5, mm6 |
psubw mm0,mm2 |
| 893 |
pxor mm0, mm1 ; negate if negative |
movq mm1,[_EAX+8] |
| 894 |
pxor mm3, mm4 ; negate if negative |
pxor mm2,mm6 |
| 895 |
psubw mm0, mm1 |
pcmpgtw mm3,mm1 |
| 896 |
psubw mm3, mm4 |
pandn mm4,mm2 ; ... then data==0 |
| 897 |
pmullw mm0, mm7 ; *= 2Q |
pmullw mm1,mm7 |
| 898 |
pmullw mm3, mm7 ; *= 2Q |
pxor mm2,mm2 |
| 899 |
paddw mm0, mm2 ; + offset |
pcmpeqw mm2,mm1 |
| 900 |
paddw mm3, mm5 ; + offset |
psubw mm1,mm3 |
| 901 |
paddw mm0, mm1 ; negate back |
pxor mm3,mm6 |
| 902 |
paddw mm3, mm4 ; negate back |
pandn mm2,mm3 |
| 903 |
|
paddw mm0,mm4 |
| 904 |
; saturates to +2047 |
paddw mm1,mm2 |
| 905 |
movq mm2, [mmx_32767_minus_2047] |
|
| 906 |
add eax, 2 |
paddsw mm0, mm5 ; saturate |
| 907 |
paddsw mm0, mm2 |
paddsw mm1, mm5 |
| 908 |
paddsw mm3, mm2 |
psubsw mm0, mm5 |
| 909 |
psubsw mm0, mm2 |
psubsw mm1, mm5 |
| 910 |
psubsw mm3, mm2 |
psubsw mm0, mm5 |
| 911 |
|
psubsw mm1, mm5 |
| 912 |
pxor mm0, mm1 |
paddsw mm0, mm5 |
| 913 |
pxor mm3, mm4 |
paddsw mm1, mm5 |
| 914 |
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
|
| 915 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
movq [TMP1],mm0 |
| 916 |
jnz near .loop |
lea _EAX,[_EAX+16] |
| 917 |
|
movq [TMP1+8],mm1 |
| 918 |
|
|
| 919 |
|
dec TMP0 |
| 920 |
|
lea TMP1,[TMP1+16] |
| 921 |
|
jne .loop |
| 922 |
|
|
| 923 |
|
xor _EAX, _EAX ; return 0 |
| 924 |
ret |
ret |
| 925 |
|
ENDFUNC |
| 926 |
|
|
| 927 |
|
|
| 928 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 929 |
; |
; |
| 930 |
; void dequant_inter_xmm(int16_t * data, |
; uint32_t dequant_h263_inter_xmm(int16_t * data, |
| 931 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
| 932 |
; const uint32_t quant); |
; const uint32_t quant, |
| 933 |
|
; const uint16_t *mpeg_matrices); |
| 934 |
; |
; |
| 935 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 936 |
|
ALIGN SECTION_ALIGN |
|
; this is the same as dequant_inter_mmx, |
|
|
; except that we're saturating using 'pminsw' (saves 2 cycles/loop) |
|
|
|
|
|
align ALIGN |
|
|
cglobal dequant_h263_inter_xmm |
|
| 937 |
dequant_h263_inter_xmm: |
dequant_h263_inter_xmm: |
| 938 |
|
|
| 939 |
mov edx, [esp+ 4] ; data |
mov TMP0, prm3 ; quant |
| 940 |
mov ecx, [esp+ 8] ; coeff |
mov _EAX, prm2 ; coeff |
| 941 |
mov eax, [esp+12] ; quant |
pcmpeqw mm0,mm0 |
| 942 |
movq mm6, [mmx_add + eax*8 - 8] ; quant or quant-1 |
lea TMP1, [mmx_quant] |
| 943 |
movq mm7, [mmx_mul + eax*8 - 8] ; 2*quant |
movq mm6, [TMP1 + TMP0*8] ; quant |
| 944 |
mov eax, -16 |
shl TMP0,31 |
| 945 |
|
movq mm5,mm0 |
| 946 |
align ALIGN |
movd mm1,TMP0d |
| 947 |
.loop |
movq mm7,mm6 |
| 948 |
movq mm0, [ecx+8*eax+8*16] ; c = coeff[i] |
psllw mm0,mm1 |
| 949 |
movq mm3, [ecx+8*eax+8*16 + 8] ; c' = coeff[i+1] |
mov TMP1, prm1 ; data |
| 950 |
pxor mm1, mm1 |
movq mm4,mm5 |
| 951 |
pxor mm4, mm4 |
paddw mm7,mm7 |
| 952 |
pcmpgtw mm1, mm0 ; sign(c) |
paddw mm6,mm0 ; quant-1 |
| 953 |
pcmpgtw mm4, mm3 ; sign(c') |
|
| 954 |
|
psrlw mm4,5 |
| 955 |
|
mov TMP0,8 |
| 956 |
|
pxor mm5,mm4 ; mm5=-2048 |
| 957 |
|
|
| 958 |
|
.loop: |
| 959 |
|
movq mm0,[_EAX] |
| 960 |
|
pxor mm3,mm3 |
| 961 |
pxor mm2, mm2 |
pxor mm2, mm2 |
| 962 |
pxor mm5, mm5 |
pcmpeqw mm3,mm0 |
| 963 |
pcmpeqw mm2, mm0 ; c is zero |
pcmpgtw mm2,mm0 |
| 964 |
pcmpeqw mm5, mm3 ; c' is zero |
pmullw mm0,mm7 ; * 2 * quant |
| 965 |
pandn mm2, mm6 ; offset = isZero ? 0 : quant_add |
pandn mm3,mm6 |
| 966 |
pandn mm5, mm6 |
movq mm1,[_EAX+8] |
| 967 |
pxor mm0, mm1 ; negate if negative |
psubw mm0,mm2 |
| 968 |
pxor mm3, mm4 ; negate if negative |
pxor mm2,mm3 |
| 969 |
psubw mm0, mm1 |
pxor mm3,mm3 |
| 970 |
psubw mm3, mm4 |
paddw mm0,mm2 |
| 971 |
pmullw mm0, mm7 ; *= 2Q |
pxor mm2,mm2 |
| 972 |
pmullw mm3, mm7 ; *= 2Q |
pcmpgtw mm3,mm1 |
| 973 |
paddw mm0, mm2 ; + offset |
pcmpeqw mm2,mm1 |
| 974 |
paddw mm3, mm5 ; + offset |
pmullw mm1,mm7 |
| 975 |
paddw mm0, mm1 ; start restoring sign |
pandn mm2,mm6 |
| 976 |
paddw mm3, mm4 ; start restoring sign |
psubw mm1,mm3 |
| 977 |
|
pxor mm3,mm2 |
| 978 |
; saturates to +2047 |
paddw mm1,mm3 |
| 979 |
movq mm2, [mmx_2047] |
|
| 980 |
pminsw mm0, mm2 |
pminsw mm0,mm4 |
| 981 |
add eax, 2 |
pminsw mm1,mm4 |
| 982 |
pminsw mm3, mm2 |
pmaxsw mm0,mm5 |
| 983 |
|
pmaxsw mm1,mm5 |
| 984 |
pxor mm0, mm1 ; finish restoring sign |
|
| 985 |
pxor mm3, mm4 ; finish restoring sign |
movq [TMP1],mm0 |
| 986 |
movq [edx + 8*eax + 8*16 - 2*8], mm0 |
lea _EAX,[_EAX+16] |
| 987 |
movq [edx + 8*eax + 8*16+8 - 2*8], mm3 |
movq [TMP1+8],mm1 |
| 988 |
jnz near .loop |
|
| 989 |
|
dec TMP0 |
| 990 |
|
lea TMP1,[TMP1+16] |
| 991 |
|
jne .loop |
| 992 |
|
|
| 993 |
|
xor _EAX, _EAX ; return 0 |
| 994 |
ret |
ret |
| 995 |
|
ENDFUNC |
| 996 |
|
|
| 997 |
|
|
| 998 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 999 |
; |
; |
| 1000 |
; void dequant_inter_sse2(int16_t * data, |
; uint32_t dequant_h263_inter_sse2(int16_t * data, |
| 1001 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
| 1002 |
; const uint32_t quant); |
; const uint32_t quant, |
| 1003 |
|
; const uint16_t *mpeg_matrices); |
| 1004 |
; |
; |
| 1005 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 1006 |
align ALIGN |
|
| 1007 |
cglobal dequant_h263_inter_sse2 |
ALIGN SECTION_ALIGN |
| 1008 |
dequant_h263_inter_sse2: |
dequant_h263_inter_sse2: |
| 1009 |
mov edx, [esp + 4] ; data |
PUSH_XMM6_XMM7 |
|
mov ecx, [esp + 8] ; coeff |
|
|
mov eax, [esp + 12] ; quant |
|
|
movq mm6, [mmx_add + eax * 8 - 8] |
|
|
movq mm7, [mmx_mul + eax * 8 - 8] |
|
|
movq2dq xmm6, mm6 |
|
|
movq2dq xmm7, mm7 |
|
|
movlhps xmm6, xmm6 |
|
|
movlhps xmm7, xmm7 |
|
|
mov eax, -16 |
|
| 1010 |
|
|
| 1011 |
align ALIGN |
mov TMP0, prm3 ; quant |
| 1012 |
.loop |
mov _EAX, prm2 ; coeff |
|
movdqa xmm0, [ecx + 8*16 + 8*eax] ; c = coeff[i] |
|
|
movdqa xmm3, [ecx + 8*16 + 8*eax + 16] |
|
| 1013 |
|
|
| 1014 |
pxor xmm1, xmm1 |
lea TMP1, [mmx_quant] |
| 1015 |
pxor xmm4, xmm4 |
movq xmm6, [TMP1 + TMP0*8] ; quant |
| 1016 |
pcmpgtw xmm1, xmm0 ; sign(c) |
inc TMP0 |
| 1017 |
pcmpgtw xmm4, xmm3 |
pcmpeqw xmm5,xmm5 |
| 1018 |
|
and TMP0,1 |
| 1019 |
|
movlhps xmm6,xmm6 |
| 1020 |
|
movd xmm0,TMP0d |
| 1021 |
|
movdqa xmm7,xmm6 |
| 1022 |
|
pshuflw xmm0,xmm0,0 |
| 1023 |
|
movdqa xmm4,xmm5 |
| 1024 |
|
mov TMP1, prm1 ; data |
| 1025 |
|
movlhps xmm0,xmm0 |
| 1026 |
|
paddw xmm7,xmm7 |
| 1027 |
|
psubw xmm6,xmm0 |
| 1028 |
|
psrlw xmm4,5 ; 2047 |
| 1029 |
|
mov TMP0,4 |
| 1030 |
|
pxor xmm5,xmm4 ; mm5=-2048 |
| 1031 |
|
|
| 1032 |
|
.loop: |
| 1033 |
|
movdqa xmm0,[_EAX] |
| 1034 |
|
pxor xmm3,xmm3 |
| 1035 |
pxor xmm2, xmm2 |
pxor xmm2, xmm2 |
| 1036 |
pxor xmm5, xmm5 |
pcmpeqw xmm3,xmm0 |
| 1037 |
pcmpeqw xmm2, xmm0 ; c is zero |
pcmpgtw xmm2,xmm0 |
| 1038 |
pcmpeqw xmm5, xmm3 |
pmullw xmm0,xmm7 ; * 2 * quant |
| 1039 |
|
pandn xmm3,xmm6 |
| 1040 |
|
movdqa xmm1,[_EAX+16] |
| 1041 |
|
psubw xmm0,xmm2 |
| 1042 |
|
pxor xmm2,xmm3 |
| 1043 |
|
pxor xmm3,xmm3 |
| 1044 |
|
paddw xmm0,xmm2 |
| 1045 |
|
pxor xmm2,xmm2 |
| 1046 |
|
pcmpgtw xmm3,xmm1 |
| 1047 |
|
pcmpeqw xmm2,xmm1 |
| 1048 |
|
pmullw xmm1,xmm7 |
| 1049 |
pandn xmm2, xmm6 |
pandn xmm2, xmm6 |
| 1050 |
pandn xmm5, xmm6 |
psubw xmm1,xmm3 |
| 1051 |
pxor xmm0, xmm1 ; negate if negative |
pxor xmm3,xmm2 |
| 1052 |
pxor xmm3, xmm4 |
paddw xmm1,xmm3 |
| 1053 |
psubw xmm0, xmm1 |
|
| 1054 |
psubw xmm3, xmm4 |
pminsw xmm0,xmm4 |
| 1055 |
pmullw xmm0, xmm7 ; *= 2Q |
pminsw xmm1,xmm4 |
| 1056 |
pmullw xmm3, xmm7 |
pmaxsw xmm0,xmm5 |
| 1057 |
paddw xmm0, xmm2 ; + offset |
pmaxsw xmm1,xmm5 |
| 1058 |
paddw xmm3, xmm5 |
|
| 1059 |
|
movdqa [TMP1],xmm0 |
| 1060 |
paddw xmm0, xmm1 ; start restoring sign |
lea _EAX,[_EAX+32] |
| 1061 |
paddw xmm3, xmm4 |
movdqa [TMP1+16],xmm1 |
| 1062 |
|
|
| 1063 |
; saturates to +2047 |
dec TMP0 |
| 1064 |
movdqa xmm2, [sse2_2047] |
lea TMP1,[TMP1+32] |
| 1065 |
pminsw xmm0, xmm2 |
jne .loop |
|
add eax, 4 |
|
|
pminsw xmm3, xmm2 |
|
| 1066 |
|
|
| 1067 |
pxor xmm0, xmm1 ; finish restoring sign |
xor _EAX, _EAX ; return 0 |
|
pxor xmm3, xmm4 |
|
|
movdqa [edx + 8*16 - 8*4 + 8*eax], xmm0 |
|
|
movdqa [edx + 8*16 - 8*4 + 8*eax + 16], xmm3 |
|
|
jnz near .loop |
|
| 1068 |
|
|
| 1069 |
|
POP_XMM6_XMM7 |
| 1070 |
ret |
ret |
| 1071 |
|
ENDFUNC |
| 1072 |
|
|
| 1073 |
|
|
| 1074 |
|
%ifidn __OUTPUT_FORMAT__,elf |
| 1075 |
|
section ".note.GNU-stack" noalloc noexec nowrite progbits |
| 1076 |
|
%endif |
| 1077 |
|
|
Parent Directory
| 
Revision Log
| 
Patch