1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libjpeg/simd/jimmxint.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,852 @@
1.4 +;
1.5 +; jimmxint.asm - accurate integer IDCT (MMX)
1.6 +;
1.7 +; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
1.8 +;
1.9 +; Based on
1.10 +; x86 SIMD extension for IJG JPEG library
1.11 +; Copyright (C) 1999-2006, MIYASAKA Masaru.
1.12 +; For conditions of distribution and use, see copyright notice in jsimdext.inc
1.13 +;
1.14 +; This file should be assembled with NASM (Netwide Assembler),
1.15 +; can *not* be assembled with Microsoft's MASM or any compatible
1.16 +; assembler (including Borland's Turbo Assembler).
1.17 +; NASM is available from http://nasm.sourceforge.net/ or
1.18 +; http://sourceforge.net/project/showfiles.php?group_id=6208
1.19 +;
1.20 +; This file contains a slow-but-accurate integer implementation of the
1.21 +; inverse DCT (Discrete Cosine Transform). The following code is based
1.22 +; directly on the IJG's original jidctint.c; see the jidctint.c for
1.23 +; more details.
1.24 +;
1.25 +; [TAB8]
1.26 +
1.27 +%include "jsimdext.inc"
1.28 +%include "jdct.inc"
1.29 +
1.30 +; --------------------------------------------------------------------------
1.31 +
1.32 +%define CONST_BITS 13
1.33 +%define PASS1_BITS 2
1.34 +
1.35 +%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
1.36 +%define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
1.37 +
1.38 +%if CONST_BITS == 13
1.39 +F_0_298 equ 2446 ; FIX(0.298631336)
1.40 +F_0_390 equ 3196 ; FIX(0.390180644)
1.41 +F_0_541 equ 4433 ; FIX(0.541196100)
1.42 +F_0_765 equ 6270 ; FIX(0.765366865)
1.43 +F_0_899 equ 7373 ; FIX(0.899976223)
1.44 +F_1_175 equ 9633 ; FIX(1.175875602)
1.45 +F_1_501 equ 12299 ; FIX(1.501321110)
1.46 +F_1_847 equ 15137 ; FIX(1.847759065)
1.47 +F_1_961 equ 16069 ; FIX(1.961570560)
1.48 +F_2_053 equ 16819 ; FIX(2.053119869)
1.49 +F_2_562 equ 20995 ; FIX(2.562915447)
1.50 +F_3_072 equ 25172 ; FIX(3.072711026)
1.51 +%else
1.52 +; NASM cannot do compile-time arithmetic on floating-point constants.
1.53 +%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
1.54 +F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336)
1.55 +F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644)
1.56 +F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
1.57 +F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
1.58 +F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
1.59 +F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602)
1.60 +F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110)
1.61 +F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
1.62 +F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560)
1.63 +F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869)
1.64 +F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
1.65 +F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026)
1.66 +%endif
1.67 +
1.68 +; --------------------------------------------------------------------------
1.69 + SECTION SEG_CONST
1.70 +
1.71 + alignz 16
1.72 + global EXTN(jconst_idct_islow_mmx)
1.73 +
1.74 +EXTN(jconst_idct_islow_mmx):
1.75 +
1.76 +PW_F130_F054 times 2 dw (F_0_541+F_0_765), F_0_541
1.77 +PW_F054_MF130 times 2 dw F_0_541, (F_0_541-F_1_847)
1.78 +PW_MF078_F117 times 2 dw (F_1_175-F_1_961), F_1_175
1.79 +PW_F117_F078 times 2 dw F_1_175, (F_1_175-F_0_390)
1.80 +PW_MF060_MF089 times 2 dw (F_0_298-F_0_899),-F_0_899
1.81 +PW_MF089_F060 times 2 dw -F_0_899, (F_1_501-F_0_899)
1.82 +PW_MF050_MF256 times 2 dw (F_2_053-F_2_562),-F_2_562
1.83 +PW_MF256_F050 times 2 dw -F_2_562, (F_3_072-F_2_562)
1.84 +PD_DESCALE_P1 times 2 dd 1 << (DESCALE_P1-1)
1.85 +PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2-1)
1.86 +PB_CENTERJSAMP times 8 db CENTERJSAMPLE
1.87 +
1.88 + alignz 16
1.89 +
1.90 +; --------------------------------------------------------------------------
1.91 + SECTION SEG_TEXT
1.92 + BITS 32
1.93 +;
1.94 +; Perform dequantization and inverse DCT on one block of coefficients.
1.95 +;
1.96 +; GLOBAL(void)
1.97 +; jsimd_idct_islow_mmx (void * dct_table, JCOEFPTR coef_block,
1.98 +; JSAMPARRAY output_buf, JDIMENSION output_col)
1.99 +;
1.100 +
1.101 +%define dct_table(b) (b)+8 ; jpeg_component_info * compptr
1.102 +%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
1.103 +%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
1.104 +%define output_col(b) (b)+20 ; JDIMENSION output_col
1.105 +
1.106 +%define original_ebp ebp+0
1.107 +%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM]
1.108 +%define WK_NUM 12
1.109 +%define workspace wk(0)-DCTSIZE2*SIZEOF_JCOEF
1.110 + ; JCOEF workspace[DCTSIZE2]
1.111 +
1.112 + align 16
1.113 + global EXTN(jsimd_idct_islow_mmx)
1.114 +
1.115 +EXTN(jsimd_idct_islow_mmx):
1.116 + push ebp
1.117 + mov eax,esp ; eax = original ebp
1.118 + sub esp, byte 4
1.119 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits
1.120 + mov [esp],eax
1.121 + mov ebp,esp ; ebp = aligned ebp
1.122 + lea esp, [workspace]
1.123 + push ebx
1.124 +; push ecx ; need not be preserved
1.125 +; push edx ; need not be preserved
1.126 + push esi
1.127 + push edi
1.128 +
1.129 + get_GOT ebx ; get GOT address
1.130 +
1.131 + ; ---- Pass 1: process columns from input, store into work array.
1.132 +
1.133 +; mov eax, [original_ebp]
1.134 + mov edx, POINTER [dct_table(eax)] ; quantptr
1.135 + mov esi, JCOEFPTR [coef_block(eax)] ; inptr
1.136 + lea edi, [workspace] ; JCOEF * wsptr
1.137 + mov ecx, DCTSIZE/4 ; ctr
1.138 + alignx 16,7
1.139 +.columnloop:
1.140 +%ifndef NO_ZERO_COLUMN_TEST_ISLOW_MMX
1.141 + mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
1.142 + or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
1.143 + jnz short .columnDCT
1.144 +
1.145 + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
1.146 + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
1.147 + por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
1.148 + por mm1, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
1.149 + por mm0, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
1.150 + por mm1, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
1.151 + por mm0, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
1.152 + por mm1,mm0
1.153 + packsswb mm1,mm1
1.154 + movd eax,mm1
1.155 + test eax,eax
1.156 + jnz short .columnDCT
1.157 +
1.158 + ; -- AC terms all zero
1.159 +
1.160 + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
1.161 + pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.162 +
1.163 + psllw mm0,PASS1_BITS
1.164 +
1.165 + movq mm2,mm0 ; mm0=in0=(00 01 02 03)
1.166 + punpcklwd mm0,mm0 ; mm0=(00 00 01 01)
1.167 + punpckhwd mm2,mm2 ; mm2=(02 02 03 03)
1.168 +
1.169 + movq mm1,mm0
1.170 + punpckldq mm0,mm0 ; mm0=(00 00 00 00)
1.171 + punpckhdq mm1,mm1 ; mm1=(01 01 01 01)
1.172 + movq mm3,mm2
1.173 + punpckldq mm2,mm2 ; mm2=(02 02 02 02)
1.174 + punpckhdq mm3,mm3 ; mm3=(03 03 03 03)
1.175 +
1.176 + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0
1.177 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0
1.178 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1
1.179 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm1
1.180 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2
1.181 + movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm2
1.182 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm3
1.183 + movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm3
1.184 + jmp near .nextcolumn
1.185 + alignx 16,7
1.186 +%endif
1.187 +.columnDCT:
1.188 +
1.189 + ; -- Even part
1.190 +
1.191 + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
1.192 + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
1.193 + pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.194 + pmullw mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.195 + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
1.196 + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
1.197 + pmullw mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.198 + pmullw mm3, MMWORD [MMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.199 +
1.200 + ; (Original)
1.201 + ; z1 = (z2 + z3) * 0.541196100;
1.202 + ; tmp2 = z1 + z3 * -1.847759065;
1.203 + ; tmp3 = z1 + z2 * 0.765366865;
1.204 + ;
1.205 + ; (This implementation)
1.206 + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
1.207 + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
1.208 +
1.209 + movq mm4,mm1 ; mm1=in2=z2
1.210 + movq mm5,mm1
1.211 + punpcklwd mm4,mm3 ; mm3=in6=z3
1.212 + punpckhwd mm5,mm3
1.213 + movq mm1,mm4
1.214 + movq mm3,mm5
1.215 + pmaddwd mm4,[GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L
1.216 + pmaddwd mm5,[GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H
1.217 + pmaddwd mm1,[GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L
1.218 + pmaddwd mm3,[GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H
1.219 +
1.220 + movq mm6,mm0
1.221 + paddw mm0,mm2 ; mm0=in0+in4
1.222 + psubw mm6,mm2 ; mm6=in0-in4
1.223 +
1.224 + pxor mm7,mm7
1.225 + pxor mm2,mm2
1.226 + punpcklwd mm7,mm0 ; mm7=tmp0L
1.227 + punpckhwd mm2,mm0 ; mm2=tmp0H
1.228 + psrad mm7,(16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
1.229 + psrad mm2,(16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS
1.230 +
1.231 + movq mm0,mm7
1.232 + paddd mm7,mm4 ; mm7=tmp10L
1.233 + psubd mm0,mm4 ; mm0=tmp13L
1.234 + movq mm4,mm2
1.235 + paddd mm2,mm5 ; mm2=tmp10H
1.236 + psubd mm4,mm5 ; mm4=tmp13H
1.237 +
1.238 + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L
1.239 + movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H
1.240 + movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L
1.241 + movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H
1.242 +
1.243 + pxor mm5,mm5
1.244 + pxor mm7,mm7
1.245 + punpcklwd mm5,mm6 ; mm5=tmp1L
1.246 + punpckhwd mm7,mm6 ; mm7=tmp1H
1.247 + psrad mm5,(16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS
1.248 + psrad mm7,(16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
1.249 +
1.250 + movq mm2,mm5
1.251 + paddd mm5,mm1 ; mm5=tmp11L
1.252 + psubd mm2,mm1 ; mm2=tmp12L
1.253 + movq mm0,mm7
1.254 + paddd mm7,mm3 ; mm7=tmp11H
1.255 + psubd mm0,mm3 ; mm0=tmp12H
1.256 +
1.257 + movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L
1.258 + movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H
1.259 + movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L
1.260 + movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H
1.261 +
1.262 + ; -- Odd part
1.263 +
1.264 + movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
1.265 + movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
1.266 + pmullw mm4, MMWORD [MMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.267 + pmullw mm6, MMWORD [MMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.268 + movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
1.269 + movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
1.270 + pmullw mm1, MMWORD [MMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.271 + pmullw mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
1.272 +
1.273 + movq mm5,mm6
1.274 + movq mm7,mm4
1.275 + paddw mm5,mm3 ; mm5=z3
1.276 + paddw mm7,mm1 ; mm7=z4
1.277 +
1.278 + ; (Original)
1.279 + ; z5 = (z3 + z4) * 1.175875602;
1.280 + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
1.281 + ; z3 += z5; z4 += z5;
1.282 + ;
1.283 + ; (This implementation)
1.284 + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
1.285 + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
1.286 +
1.287 + movq mm2,mm5
1.288 + movq mm0,mm5
1.289 + punpcklwd mm2,mm7
1.290 + punpckhwd mm0,mm7
1.291 + movq mm5,mm2
1.292 + movq mm7,mm0
1.293 + pmaddwd mm2,[GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L
1.294 + pmaddwd mm0,[GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H
1.295 + pmaddwd mm5,[GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L
1.296 + pmaddwd mm7,[GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H
1.297 +
1.298 + movq MMWORD [wk(10)], mm2 ; wk(10)=z3L
1.299 + movq MMWORD [wk(11)], mm0 ; wk(11)=z3H
1.300 +
1.301 + ; (Original)
1.302 + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
1.303 + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
1.304 + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
1.305 + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
1.306 + ; tmp0 += z1 + z3; tmp1 += z2 + z4;
1.307 + ; tmp2 += z2 + z3; tmp3 += z1 + z4;
1.308 + ;
1.309 + ; (This implementation)
1.310 + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
1.311 + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
1.312 + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
1.313 + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
1.314 + ; tmp0 += z3; tmp1 += z4;
1.315 + ; tmp2 += z3; tmp3 += z4;
1.316 +
1.317 + movq mm2,mm3
1.318 + movq mm0,mm3
1.319 + punpcklwd mm2,mm4
1.320 + punpckhwd mm0,mm4
1.321 + movq mm3,mm2
1.322 + movq mm4,mm0
1.323 + pmaddwd mm2,[GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L
1.324 + pmaddwd mm0,[GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H
1.325 + pmaddwd mm3,[GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L
1.326 + pmaddwd mm4,[GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H
1.327 +
1.328 + paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L
1.329 + paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H
1.330 + paddd mm3,mm5 ; mm3=tmp3L
1.331 + paddd mm4,mm7 ; mm4=tmp3H
1.332 +
1.333 + movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L
1.334 + movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H
1.335 +
1.336 + movq mm2,mm1
1.337 + movq mm0,mm1
1.338 + punpcklwd mm2,mm6
1.339 + punpckhwd mm0,mm6
1.340 + movq mm1,mm2
1.341 + movq mm6,mm0
1.342 + pmaddwd mm2,[GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L
1.343 + pmaddwd mm0,[GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H
1.344 + pmaddwd mm1,[GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L
1.345 + pmaddwd mm6,[GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H
1.346 +
1.347 + paddd mm2,mm5 ; mm2=tmp1L
1.348 + paddd mm0,mm7 ; mm0=tmp1H
1.349 + paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L
1.350 + paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H
1.351 +
1.352 + movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L
1.353 + movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H
1.354 +
1.355 + ; -- Final output stage
1.356 +
1.357 + movq mm5, MMWORD [wk(0)] ; mm5=tmp10L
1.358 + movq mm7, MMWORD [wk(1)] ; mm7=tmp10H
1.359 +
1.360 + movq mm2,mm5
1.361 + movq mm0,mm7
1.362 + paddd mm5,mm3 ; mm5=data0L
1.363 + paddd mm7,mm4 ; mm7=data0H
1.364 + psubd mm2,mm3 ; mm2=data7L
1.365 + psubd mm0,mm4 ; mm0=data7H
1.366 +
1.367 + movq mm3,[GOTOFF(ebx,PD_DESCALE_P1)] ; mm3=[PD_DESCALE_P1]
1.368 +
1.369 + paddd mm5,mm3
1.370 + paddd mm7,mm3
1.371 + psrad mm5,DESCALE_P1
1.372 + psrad mm7,DESCALE_P1
1.373 + paddd mm2,mm3
1.374 + paddd mm0,mm3
1.375 + psrad mm2,DESCALE_P1
1.376 + psrad mm0,DESCALE_P1
1.377 +
1.378 + packssdw mm5,mm7 ; mm5=data0=(00 01 02 03)
1.379 + packssdw mm2,mm0 ; mm2=data7=(70 71 72 73)
1.380 +
1.381 + movq mm4, MMWORD [wk(4)] ; mm4=tmp11L
1.382 + movq mm3, MMWORD [wk(5)] ; mm3=tmp11H
1.383 +
1.384 + movq mm7,mm4
1.385 + movq mm0,mm3
1.386 + paddd mm4,mm1 ; mm4=data1L
1.387 + paddd mm3,mm6 ; mm3=data1H
1.388 + psubd mm7,mm1 ; mm7=data6L
1.389 + psubd mm0,mm6 ; mm0=data6H
1.390 +
1.391 + movq mm1,[GOTOFF(ebx,PD_DESCALE_P1)] ; mm1=[PD_DESCALE_P1]
1.392 +
1.393 + paddd mm4,mm1
1.394 + paddd mm3,mm1
1.395 + psrad mm4,DESCALE_P1
1.396 + psrad mm3,DESCALE_P1
1.397 + paddd mm7,mm1
1.398 + paddd mm0,mm1
1.399 + psrad mm7,DESCALE_P1
1.400 + psrad mm0,DESCALE_P1
1.401 +
1.402 + packssdw mm4,mm3 ; mm4=data1=(10 11 12 13)
1.403 + packssdw mm7,mm0 ; mm7=data6=(60 61 62 63)
1.404 +
1.405 + movq mm6,mm5 ; transpose coefficients(phase 1)
1.406 + punpcklwd mm5,mm4 ; mm5=(00 10 01 11)
1.407 + punpckhwd mm6,mm4 ; mm6=(02 12 03 13)
1.408 + movq mm1,mm7 ; transpose coefficients(phase 1)
1.409 + punpcklwd mm7,mm2 ; mm7=(60 70 61 71)
1.410 + punpckhwd mm1,mm2 ; mm1=(62 72 63 73)
1.411 +
1.412 + movq mm3, MMWORD [wk(6)] ; mm3=tmp12L
1.413 + movq mm0, MMWORD [wk(7)] ; mm0=tmp12H
1.414 + movq mm4, MMWORD [wk(10)] ; mm4=tmp1L
1.415 + movq mm2, MMWORD [wk(11)] ; mm2=tmp1H
1.416 +
1.417 + movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 01 11)
1.418 + movq MMWORD [wk(1)], mm6 ; wk(1)=(02 12 03 13)
1.419 + movq MMWORD [wk(4)], mm7 ; wk(4)=(60 70 61 71)
1.420 + movq MMWORD [wk(5)], mm1 ; wk(5)=(62 72 63 73)
1.421 +
1.422 + movq mm5,mm3
1.423 + movq mm6,mm0
1.424 + paddd mm3,mm4 ; mm3=data2L
1.425 + paddd mm0,mm2 ; mm0=data2H
1.426 + psubd mm5,mm4 ; mm5=data5L
1.427 + psubd mm6,mm2 ; mm6=data5H
1.428 +
1.429 + movq mm7,[GOTOFF(ebx,PD_DESCALE_P1)] ; mm7=[PD_DESCALE_P1]
1.430 +
1.431 + paddd mm3,mm7
1.432 + paddd mm0,mm7
1.433 + psrad mm3,DESCALE_P1
1.434 + psrad mm0,DESCALE_P1
1.435 + paddd mm5,mm7
1.436 + paddd mm6,mm7
1.437 + psrad mm5,DESCALE_P1
1.438 + psrad mm6,DESCALE_P1
1.439 +
1.440 + packssdw mm3,mm0 ; mm3=data2=(20 21 22 23)
1.441 + packssdw mm5,mm6 ; mm5=data5=(50 51 52 53)
1.442 +
1.443 + movq mm1, MMWORD [wk(2)] ; mm1=tmp13L
1.444 + movq mm4, MMWORD [wk(3)] ; mm4=tmp13H
1.445 + movq mm2, MMWORD [wk(8)] ; mm2=tmp0L
1.446 + movq mm7, MMWORD [wk(9)] ; mm7=tmp0H
1.447 +
1.448 + movq mm0,mm1
1.449 + movq mm6,mm4
1.450 + paddd mm1,mm2 ; mm1=data3L
1.451 + paddd mm4,mm7 ; mm4=data3H
1.452 + psubd mm0,mm2 ; mm0=data4L
1.453 + psubd mm6,mm7 ; mm6=data4H
1.454 +
1.455 + movq mm2,[GOTOFF(ebx,PD_DESCALE_P1)] ; mm2=[PD_DESCALE_P1]
1.456 +
1.457 + paddd mm1,mm2
1.458 + paddd mm4,mm2
1.459 + psrad mm1,DESCALE_P1
1.460 + psrad mm4,DESCALE_P1
1.461 + paddd mm0,mm2
1.462 + paddd mm6,mm2
1.463 + psrad mm0,DESCALE_P1
1.464 + psrad mm6,DESCALE_P1
1.465 +
1.466 + packssdw mm1,mm4 ; mm1=data3=(30 31 32 33)
1.467 + packssdw mm0,mm6 ; mm0=data4=(40 41 42 43)
1.468 +
1.469 + movq mm7, MMWORD [wk(0)] ; mm7=(00 10 01 11)
1.470 + movq mm2, MMWORD [wk(1)] ; mm2=(02 12 03 13)
1.471 +
1.472 + movq mm4,mm3 ; transpose coefficients(phase 1)
1.473 + punpcklwd mm3,mm1 ; mm3=(20 30 21 31)
1.474 + punpckhwd mm4,mm1 ; mm4=(22 32 23 33)
1.475 + movq mm6,mm0 ; transpose coefficients(phase 1)
1.476 + punpcklwd mm0,mm5 ; mm0=(40 50 41 51)
1.477 + punpckhwd mm6,mm5 ; mm6=(42 52 43 53)
1.478 +
1.479 + movq mm1,mm7 ; transpose coefficients(phase 2)
1.480 + punpckldq mm7,mm3 ; mm7=(00 10 20 30)
1.481 + punpckhdq mm1,mm3 ; mm1=(01 11 21 31)
1.482 + movq mm5,mm2 ; transpose coefficients(phase 2)
1.483 + punpckldq mm2,mm4 ; mm2=(02 12 22 32)
1.484 + punpckhdq mm5,mm4 ; mm5=(03 13 23 33)
1.485 +
1.486 + movq mm3, MMWORD [wk(4)] ; mm3=(60 70 61 71)
1.487 + movq mm4, MMWORD [wk(5)] ; mm4=(62 72 63 73)
1.488 +
1.489 + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm7
1.490 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1
1.491 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2
1.492 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm5
1.493 +
1.494 + movq mm7,mm0 ; transpose coefficients(phase 2)
1.495 + punpckldq mm0,mm3 ; mm0=(40 50 60 70)
1.496 + punpckhdq mm7,mm3 ; mm7=(41 51 61 71)
1.497 + movq mm1,mm6 ; transpose coefficients(phase 2)
1.498 + punpckldq mm6,mm4 ; mm6=(42 52 62 72)
1.499 + punpckhdq mm1,mm4 ; mm1=(43 53 63 73)
1.500 +
1.501 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0
1.502 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm7
1.503 + movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm6
1.504 + movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm1
1.505 +
1.506 +.nextcolumn:
1.507 + add esi, byte 4*SIZEOF_JCOEF ; coef_block
1.508 + add edx, byte 4*SIZEOF_ISLOW_MULT_TYPE ; quantptr
1.509 + add edi, byte 4*DCTSIZE*SIZEOF_JCOEF ; wsptr
1.510 + dec ecx ; ctr
1.511 + jnz near .columnloop
1.512 +
1.513 + ; ---- Pass 2: process rows from work array, store into output array.
1.514 +
1.515 + mov eax, [original_ebp]
1.516 + lea esi, [workspace] ; JCOEF * wsptr
1.517 + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
1.518 + mov eax, JDIMENSION [output_col(eax)]
1.519 + mov ecx, DCTSIZE/4 ; ctr
1.520 + alignx 16,7
1.521 +.rowloop:
1.522 +
1.523 + ; -- Even part
1.524 +
1.525 + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
1.526 + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
1.527 + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
1.528 + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
1.529 +
1.530 + ; (Original)
1.531 + ; z1 = (z2 + z3) * 0.541196100;
1.532 + ; tmp2 = z1 + z3 * -1.847759065;
1.533 + ; tmp3 = z1 + z2 * 0.765366865;
1.534 + ;
1.535 + ; (This implementation)
1.536 + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
1.537 + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
1.538 +
1.539 + movq mm4,mm1 ; mm1=in2=z2
1.540 + movq mm5,mm1
1.541 + punpcklwd mm4,mm3 ; mm3=in6=z3
1.542 + punpckhwd mm5,mm3
1.543 + movq mm1,mm4
1.544 + movq mm3,mm5
1.545 + pmaddwd mm4,[GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L
1.546 + pmaddwd mm5,[GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H
1.547 + pmaddwd mm1,[GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L
1.548 + pmaddwd mm3,[GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H
1.549 +
1.550 + movq mm6,mm0
1.551 + paddw mm0,mm2 ; mm0=in0+in4
1.552 + psubw mm6,mm2 ; mm6=in0-in4
1.553 +
1.554 + pxor mm7,mm7
1.555 + pxor mm2,mm2
1.556 + punpcklwd mm7,mm0 ; mm7=tmp0L
1.557 + punpckhwd mm2,mm0 ; mm2=tmp0H
1.558 + psrad mm7,(16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
1.559 + psrad mm2,(16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS
1.560 +
1.561 + movq mm0,mm7
1.562 + paddd mm7,mm4 ; mm7=tmp10L
1.563 + psubd mm0,mm4 ; mm0=tmp13L
1.564 + movq mm4,mm2
1.565 + paddd mm2,mm5 ; mm2=tmp10H
1.566 + psubd mm4,mm5 ; mm4=tmp13H
1.567 +
1.568 + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L
1.569 + movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H
1.570 + movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L
1.571 + movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H
1.572 +
1.573 + pxor mm5,mm5
1.574 + pxor mm7,mm7
1.575 + punpcklwd mm5,mm6 ; mm5=tmp1L
1.576 + punpckhwd mm7,mm6 ; mm7=tmp1H
1.577 + psrad mm5,(16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS
1.578 + psrad mm7,(16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
1.579 +
1.580 + movq mm2,mm5
1.581 + paddd mm5,mm1 ; mm5=tmp11L
1.582 + psubd mm2,mm1 ; mm2=tmp12L
1.583 + movq mm0,mm7
1.584 + paddd mm7,mm3 ; mm7=tmp11H
1.585 + psubd mm0,mm3 ; mm0=tmp12H
1.586 +
1.587 + movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L
1.588 + movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H
1.589 + movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L
1.590 + movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H
1.591 +
1.592 + ; -- Odd part
1.593 +
1.594 + movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
1.595 + movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
1.596 + movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
1.597 + movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
1.598 +
1.599 + movq mm5,mm6
1.600 + movq mm7,mm4
1.601 + paddw mm5,mm3 ; mm5=z3
1.602 + paddw mm7,mm1 ; mm7=z4
1.603 +
1.604 + ; (Original)
1.605 + ; z5 = (z3 + z4) * 1.175875602;
1.606 + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
1.607 + ; z3 += z5; z4 += z5;
1.608 + ;
1.609 + ; (This implementation)
1.610 + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
1.611 + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
1.612 +
1.613 + movq mm2,mm5
1.614 + movq mm0,mm5
1.615 + punpcklwd mm2,mm7
1.616 + punpckhwd mm0,mm7
1.617 + movq mm5,mm2
1.618 + movq mm7,mm0
1.619 + pmaddwd mm2,[GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L
1.620 + pmaddwd mm0,[GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H
1.621 + pmaddwd mm5,[GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L
1.622 + pmaddwd mm7,[GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H
1.623 +
1.624 + movq MMWORD [wk(10)], mm2 ; wk(10)=z3L
1.625 + movq MMWORD [wk(11)], mm0 ; wk(11)=z3H
1.626 +
1.627 + ; (Original)
1.628 + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
1.629 + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
1.630 + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
1.631 + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
1.632 + ; tmp0 += z1 + z3; tmp1 += z2 + z4;
1.633 + ; tmp2 += z2 + z3; tmp3 += z1 + z4;
1.634 + ;
1.635 + ; (This implementation)
1.636 + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
1.637 + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
1.638 + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
1.639 + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
1.640 + ; tmp0 += z3; tmp1 += z4;
1.641 + ; tmp2 += z3; tmp3 += z4;
1.642 +
1.643 + movq mm2,mm3
1.644 + movq mm0,mm3
1.645 + punpcklwd mm2,mm4
1.646 + punpckhwd mm0,mm4
1.647 + movq mm3,mm2
1.648 + movq mm4,mm0
1.649 + pmaddwd mm2,[GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L
1.650 + pmaddwd mm0,[GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H
1.651 + pmaddwd mm3,[GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L
1.652 + pmaddwd mm4,[GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H
1.653 +
1.654 + paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L
1.655 + paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H
1.656 + paddd mm3,mm5 ; mm3=tmp3L
1.657 + paddd mm4,mm7 ; mm4=tmp3H
1.658 +
1.659 + movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L
1.660 + movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H
1.661 +
1.662 + movq mm2,mm1
1.663 + movq mm0,mm1
1.664 + punpcklwd mm2,mm6
1.665 + punpckhwd mm0,mm6
1.666 + movq mm1,mm2
1.667 + movq mm6,mm0
1.668 + pmaddwd mm2,[GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L
1.669 + pmaddwd mm0,[GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H
1.670 + pmaddwd mm1,[GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L
1.671 + pmaddwd mm6,[GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H
1.672 +
1.673 + paddd mm2,mm5 ; mm2=tmp1L
1.674 + paddd mm0,mm7 ; mm0=tmp1H
1.675 + paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L
1.676 + paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H
1.677 +
1.678 + movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L
1.679 + movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H
1.680 +
1.681 + ; -- Final output stage
1.682 +
1.683 + movq mm5, MMWORD [wk(0)] ; mm5=tmp10L
1.684 + movq mm7, MMWORD [wk(1)] ; mm7=tmp10H
1.685 +
1.686 + movq mm2,mm5
1.687 + movq mm0,mm7
1.688 + paddd mm5,mm3 ; mm5=data0L
1.689 + paddd mm7,mm4 ; mm7=data0H
1.690 + psubd mm2,mm3 ; mm2=data7L
1.691 + psubd mm0,mm4 ; mm0=data7H
1.692 +
1.693 + movq mm3,[GOTOFF(ebx,PD_DESCALE_P2)] ; mm3=[PD_DESCALE_P2]
1.694 +
1.695 + paddd mm5,mm3
1.696 + paddd mm7,mm3
1.697 + psrad mm5,DESCALE_P2
1.698 + psrad mm7,DESCALE_P2
1.699 + paddd mm2,mm3
1.700 + paddd mm0,mm3
1.701 + psrad mm2,DESCALE_P2
1.702 + psrad mm0,DESCALE_P2
1.703 +
1.704 + packssdw mm5,mm7 ; mm5=data0=(00 10 20 30)
1.705 + packssdw mm2,mm0 ; mm2=data7=(07 17 27 37)
1.706 +
1.707 + movq mm4, MMWORD [wk(4)] ; mm4=tmp11L
1.708 + movq mm3, MMWORD [wk(5)] ; mm3=tmp11H
1.709 +
1.710 + movq mm7,mm4
1.711 + movq mm0,mm3
1.712 + paddd mm4,mm1 ; mm4=data1L
1.713 + paddd mm3,mm6 ; mm3=data1H
1.714 + psubd mm7,mm1 ; mm7=data6L
1.715 + psubd mm0,mm6 ; mm0=data6H
1.716 +
1.717 + movq mm1,[GOTOFF(ebx,PD_DESCALE_P2)] ; mm1=[PD_DESCALE_P2]
1.718 +
1.719 + paddd mm4,mm1
1.720 + paddd mm3,mm1
1.721 + psrad mm4,DESCALE_P2
1.722 + psrad mm3,DESCALE_P2
1.723 + paddd mm7,mm1
1.724 + paddd mm0,mm1
1.725 + psrad mm7,DESCALE_P2
1.726 + psrad mm0,DESCALE_P2
1.727 +
1.728 + packssdw mm4,mm3 ; mm4=data1=(01 11 21 31)
1.729 + packssdw mm7,mm0 ; mm7=data6=(06 16 26 36)
1.730 +
1.731 + packsswb mm5,mm7 ; mm5=(00 10 20 30 06 16 26 36)
1.732 + packsswb mm4,mm2 ; mm4=(01 11 21 31 07 17 27 37)
1.733 +
1.734 + movq mm6, MMWORD [wk(6)] ; mm6=tmp12L
1.735 + movq mm1, MMWORD [wk(7)] ; mm1=tmp12H
1.736 + movq mm3, MMWORD [wk(10)] ; mm3=tmp1L
1.737 + movq mm0, MMWORD [wk(11)] ; mm0=tmp1H
1.738 +
1.739 + movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 20 30 06 16 26 36)
1.740 + movq MMWORD [wk(1)], mm4 ; wk(1)=(01 11 21 31 07 17 27 37)
1.741 +
1.742 + movq mm7,mm6
1.743 + movq mm2,mm1
1.744 + paddd mm6,mm3 ; mm6=data2L
1.745 + paddd mm1,mm0 ; mm1=data2H
1.746 + psubd mm7,mm3 ; mm7=data5L
1.747 + psubd mm2,mm0 ; mm2=data5H
1.748 +
1.749 + movq mm5,[GOTOFF(ebx,PD_DESCALE_P2)] ; mm5=[PD_DESCALE_P2]
1.750 +
1.751 + paddd mm6,mm5
1.752 + paddd mm1,mm5
1.753 + psrad mm6,DESCALE_P2
1.754 + psrad mm1,DESCALE_P2
1.755 + paddd mm7,mm5
1.756 + paddd mm2,mm5
1.757 + psrad mm7,DESCALE_P2
1.758 + psrad mm2,DESCALE_P2
1.759 +
1.760 + packssdw mm6,mm1 ; mm6=data2=(02 12 22 32)
1.761 + packssdw mm7,mm2 ; mm7=data5=(05 15 25 35)
1.762 +
1.763 + movq mm4, MMWORD [wk(2)] ; mm4=tmp13L
1.764 + movq mm3, MMWORD [wk(3)] ; mm3=tmp13H
1.765 + movq mm0, MMWORD [wk(8)] ; mm0=tmp0L
1.766 + movq mm5, MMWORD [wk(9)] ; mm5=tmp0H
1.767 +
1.768 + movq mm1,mm4
1.769 + movq mm2,mm3
1.770 + paddd mm4,mm0 ; mm4=data3L
1.771 + paddd mm3,mm5 ; mm3=data3H
1.772 + psubd mm1,mm0 ; mm1=data4L
1.773 + psubd mm2,mm5 ; mm2=data4H
1.774 +
1.775 + movq mm0,[GOTOFF(ebx,PD_DESCALE_P2)] ; mm0=[PD_DESCALE_P2]
1.776 +
1.777 + paddd mm4,mm0
1.778 + paddd mm3,mm0
1.779 + psrad mm4,DESCALE_P2
1.780 + psrad mm3,DESCALE_P2
1.781 + paddd mm1,mm0
1.782 + paddd mm2,mm0
1.783 + psrad mm1,DESCALE_P2
1.784 + psrad mm2,DESCALE_P2
1.785 +
1.786 + movq mm5,[GOTOFF(ebx,PB_CENTERJSAMP)] ; mm5=[PB_CENTERJSAMP]
1.787 +
1.788 + packssdw mm4,mm3 ; mm4=data3=(03 13 23 33)
1.789 + packssdw mm1,mm2 ; mm1=data4=(04 14 24 34)
1.790 +
1.791 + movq mm0, MMWORD [wk(0)] ; mm0=(00 10 20 30 06 16 26 36)
1.792 + movq mm3, MMWORD [wk(1)] ; mm3=(01 11 21 31 07 17 27 37)
1.793 +
1.794 + packsswb mm6,mm1 ; mm6=(02 12 22 32 04 14 24 34)
1.795 + packsswb mm4,mm7 ; mm4=(03 13 23 33 05 15 25 35)
1.796 +
1.797 + paddb mm0,mm5
1.798 + paddb mm3,mm5
1.799 + paddb mm6,mm5
1.800 + paddb mm4,mm5
1.801 +
1.802 + movq mm2,mm0 ; transpose coefficients(phase 1)
1.803 + punpcklbw mm0,mm3 ; mm0=(00 01 10 11 20 21 30 31)
1.804 + punpckhbw mm2,mm3 ; mm2=(06 07 16 17 26 27 36 37)
1.805 + movq mm1,mm6 ; transpose coefficients(phase 1)
1.806 + punpcklbw mm6,mm4 ; mm6=(02 03 12 13 22 23 32 33)
1.807 + punpckhbw mm1,mm4 ; mm1=(04 05 14 15 24 25 34 35)
1.808 +
1.809 + movq mm7,mm0 ; transpose coefficients(phase 2)
1.810 + punpcklwd mm0,mm6 ; mm0=(00 01 02 03 10 11 12 13)
1.811 + punpckhwd mm7,mm6 ; mm7=(20 21 22 23 30 31 32 33)
1.812 + movq mm5,mm1 ; transpose coefficients(phase 2)
1.813 + punpcklwd mm1,mm2 ; mm1=(04 05 06 07 14 15 16 17)
1.814 + punpckhwd mm5,mm2 ; mm5=(24 25 26 27 34 35 36 37)
1.815 +
1.816 + movq mm3,mm0 ; transpose coefficients(phase 3)
1.817 + punpckldq mm0,mm1 ; mm0=(00 01 02 03 04 05 06 07)
1.818 + punpckhdq mm3,mm1 ; mm3=(10 11 12 13 14 15 16 17)
1.819 + movq mm4,mm7 ; transpose coefficients(phase 3)
1.820 + punpckldq mm7,mm5 ; mm7=(20 21 22 23 24 25 26 27)
1.821 + punpckhdq mm4,mm5 ; mm4=(30 31 32 33 34 35 36 37)
1.822 +
1.823 + pushpic ebx ; save GOT address
1.824 +
1.825 + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
1.826 + mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
1.827 + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm0
1.828 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm3
1.829 + mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
1.830 + mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
1.831 + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm7
1.832 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm4
1.833 +
1.834 + poppic ebx ; restore GOT address
1.835 +
1.836 + add esi, byte 4*SIZEOF_JCOEF ; wsptr
1.837 + add edi, byte 4*SIZEOF_JSAMPROW
1.838 + dec ecx ; ctr
1.839 + jnz near .rowloop
1.840 +
1.841 + emms ; empty MMX state
1.842 +
1.843 + pop edi
1.844 + pop esi
1.845 +; pop edx ; need not be preserved
1.846 +; pop ecx ; need not be preserved
1.847 + pop ebx
1.848 + mov esp,ebp ; esp <- aligned ebp
1.849 + pop esp ; esp <- original ebp
1.850 + pop ebp
1.851 + ret
1.852 +
1.853 +; For some reason, the OS X linker does not honor the request to align the
1.854 +; segment unless we do this.
1.855 + align 16
