1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libjpeg/simd/jiss2int-64.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,848 @@
1.4 +;
1.5 +; jiss2int-64.asm - accurate integer IDCT (64-bit SSE2)
1.6 +;
1.7 +; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
1.8 +; Copyright 2009 D. R. Commander
1.9 +;
1.10 +; Based on
1.11 +; x86 SIMD extension for IJG JPEG library
1.12 +; Copyright (C) 1999-2006, MIYASAKA Masaru.
1.13 +; For conditions of distribution and use, see copyright notice in jsimdext.inc
1.14 +;
1.15 +; This file should be assembled with NASM (Netwide Assembler),
1.16 +; can *not* be assembled with Microsoft's MASM or any compatible
1.17 +; assembler (including Borland's Turbo Assembler).
1.18 +; NASM is available from http://nasm.sourceforge.net/ or
1.19 +; http://sourceforge.net/project/showfiles.php?group_id=6208
1.20 +;
1.21 +; This file contains a slow-but-accurate integer implementation of the
1.22 +; inverse DCT (Discrete Cosine Transform). The following code is based
1.23 +; directly on the IJG's original jidctint.c; see the jidctint.c for
1.24 +; more details.
1.25 +;
1.26 +; [TAB8]
1.27 +
1.28 +%include "jsimdext.inc"
1.29 +%include "jdct.inc"
1.30 +
1.31 +; --------------------------------------------------------------------------
1.32 +
1.33 +%define CONST_BITS 13
1.34 +%define PASS1_BITS 2
1.35 +
1.36 +%define DESCALE_P1 (CONST_BITS-PASS1_BITS)
1.37 +%define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
1.38 +
1.39 +%if CONST_BITS == 13
1.40 +F_0_298 equ 2446 ; FIX(0.298631336)
1.41 +F_0_390 equ 3196 ; FIX(0.390180644)
1.42 +F_0_541 equ 4433 ; FIX(0.541196100)
1.43 +F_0_765 equ 6270 ; FIX(0.765366865)
1.44 +F_0_899 equ 7373 ; FIX(0.899976223)
1.45 +F_1_175 equ 9633 ; FIX(1.175875602)
1.46 +F_1_501 equ 12299 ; FIX(1.501321110)
1.47 +F_1_847 equ 15137 ; FIX(1.847759065)
1.48 +F_1_961 equ 16069 ; FIX(1.961570560)
1.49 +F_2_053 equ 16819 ; FIX(2.053119869)
1.50 +F_2_562 equ 20995 ; FIX(2.562915447)
1.51 +F_3_072 equ 25172 ; FIX(3.072711026)
1.52 +%else
1.53 +; NASM cannot do compile-time arithmetic on floating-point constants.
1.54 +%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
1.55 +F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336)
1.56 +F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644)
1.57 +F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
1.58 +F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
1.59 +F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
1.60 +F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602)
1.61 +F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110)
1.62 +F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
1.63 +F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560)
1.64 +F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869)
1.65 +F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
1.66 +F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026)
1.67 +%endif
1.68 +
1.69 +; --------------------------------------------------------------------------
1.70 + SECTION SEG_CONST
1.71 +
1.72 + alignz 16
1.73 + global EXTN(jconst_idct_islow_sse2)
1.74 +
1.75 +EXTN(jconst_idct_islow_sse2):
1.76 +
1.77 +PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
1.78 +PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
1.79 +PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
1.80 +PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
1.81 +PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
1.82 +PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
1.83 +PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
1.84 +PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
1.85 +PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
1.86 +PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
1.87 +PB_CENTERJSAMP times 16 db CENTERJSAMPLE
1.88 +
1.89 + alignz 16
1.90 +
1.91 +; --------------------------------------------------------------------------
1.92 + SECTION SEG_TEXT
1.93 + BITS 64
1.94 +;
1.95 +; Perform dequantization and inverse DCT on one block of coefficients.
1.96 +;
1.97 +; GLOBAL(void)
1.98 +; jsimd_idct_islow_sse2 (void * dct_table, JCOEFPTR coef_block,
1.99 +; JSAMPARRAY output_buf, JDIMENSION output_col)
1.100 +;
1.101 +
1.102 +; r10 = jpeg_component_info * compptr
1.103 +; r11 = JCOEFPTR coef_block
1.104 +; r12 = JSAMPARRAY output_buf
1.105 +; r13 = JDIMENSION output_col
1.106 +
1.107 +%define original_rbp rbp+0
1.108 +%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
1.109 +%define WK_NUM 12
1.110 +
1.111 + align 16
1.112 + global EXTN(jsimd_idct_islow_sse2)
1.113 +
1.114 +EXTN(jsimd_idct_islow_sse2):
1.115 + push rbp
1.116 + mov rax,rsp ; rax = original rbp
1.117 + sub rsp, byte 4
1.118 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
1.119 + mov [rsp],rax
1.120 + mov rbp,rsp ; rbp = aligned rbp
1.121 + lea rsp, [wk(0)]
1.122 + collect_args
1.123 +
1.124 + ; ---- Pass 1: process columns from input.
1.125 +
1.126 + mov rdx, r10 ; quantptr
1.127 + mov rsi, r11 ; inptr
1.128 +
1.129 +%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
1.130 + mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
1.131 + or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
1.132 + jnz near .columnDCT
1.133 +
1.134 + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
1.135 + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
1.136 + por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
1.137 + por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
1.138 + por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
1.139 + por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
1.140 + por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
1.141 + por xmm1,xmm0
1.142 + packsswb xmm1,xmm1
1.143 + packsswb xmm1,xmm1
1.144 + movd eax,xmm1
1.145 + test rax,rax
1.146 + jnz short .columnDCT
1.147 +
1.148 + ; -- AC terms all zero
1.149 +
1.150 + movdqa xmm5, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
1.151 + pmullw xmm5, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.152 +
1.153 + psllw xmm5,PASS1_BITS
1.154 +
1.155 + movdqa xmm4,xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07)
1.156 + punpcklwd xmm5,xmm5 ; xmm5=(00 00 01 01 02 02 03 03)
1.157 + punpckhwd xmm4,xmm4 ; xmm4=(04 04 05 05 06 06 07 07)
1.158 +
1.159 + pshufd xmm7,xmm5,0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00)
1.160 + pshufd xmm6,xmm5,0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01)
1.161 + pshufd xmm1,xmm5,0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02)
1.162 + pshufd xmm5,xmm5,0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03)
1.163 + pshufd xmm0,xmm4,0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04)
1.164 + pshufd xmm3,xmm4,0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05)
1.165 + pshufd xmm2,xmm4,0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06)
1.166 + pshufd xmm4,xmm4,0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07)
1.167 +
1.168 + movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1
1.169 + movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3
1.170 + movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
1.171 + movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
1.172 + jmp near .column_end
1.173 +%endif
1.174 +.columnDCT:
1.175 +
1.176 + ; -- Even part
1.177 +
1.178 + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
1.179 + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
1.180 + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.181 + pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.182 + movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
1.183 + movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
1.184 + pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.185 + pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.186 +
1.187 + ; (Original)
1.188 + ; z1 = (z2 + z3) * 0.541196100;
1.189 + ; tmp2 = z1 + z3 * -1.847759065;
1.190 + ; tmp3 = z1 + z2 * 0.765366865;
1.191 + ;
1.192 + ; (This implementation)
1.193 + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
1.194 + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
1.195 +
1.196 + movdqa xmm4,xmm1 ; xmm1=in2=z2
1.197 + movdqa xmm5,xmm1
1.198 + punpcklwd xmm4,xmm3 ; xmm3=in6=z3
1.199 + punpckhwd xmm5,xmm3
1.200 + movdqa xmm1,xmm4
1.201 + movdqa xmm3,xmm5
1.202 + pmaddwd xmm4,[rel PW_F130_F054] ; xmm4=tmp3L
1.203 + pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H
1.204 + pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L
1.205 + pmaddwd xmm3,[rel PW_F054_MF130] ; xmm3=tmp2H
1.206 +
1.207 + movdqa xmm6,xmm0
1.208 + paddw xmm0,xmm2 ; xmm0=in0+in4
1.209 + psubw xmm6,xmm2 ; xmm6=in0-in4
1.210 +
1.211 + pxor xmm7,xmm7
1.212 + pxor xmm2,xmm2
1.213 + punpcklwd xmm7,xmm0 ; xmm7=tmp0L
1.214 + punpckhwd xmm2,xmm0 ; xmm2=tmp0H
1.215 + psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
1.216 + psrad xmm2,(16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS
1.217 +
1.218 + movdqa xmm0,xmm7
1.219 + paddd xmm7,xmm4 ; xmm7=tmp10L
1.220 + psubd xmm0,xmm4 ; xmm0=tmp13L
1.221 + movdqa xmm4,xmm2
1.222 + paddd xmm2,xmm5 ; xmm2=tmp10H
1.223 + psubd xmm4,xmm5 ; xmm4=tmp13H
1.224 +
1.225 + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L
1.226 + movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H
1.227 + movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L
1.228 + movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H
1.229 +
1.230 + pxor xmm5,xmm5
1.231 + pxor xmm7,xmm7
1.232 + punpcklwd xmm5,xmm6 ; xmm5=tmp1L
1.233 + punpckhwd xmm7,xmm6 ; xmm7=tmp1H
1.234 + psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
1.235 + psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
1.236 +
1.237 + movdqa xmm2,xmm5
1.238 + paddd xmm5,xmm1 ; xmm5=tmp11L
1.239 + psubd xmm2,xmm1 ; xmm2=tmp12L
1.240 + movdqa xmm0,xmm7
1.241 + paddd xmm7,xmm3 ; xmm7=tmp11H
1.242 + psubd xmm0,xmm3 ; xmm0=tmp12H
1.243 +
1.244 + movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
1.245 + movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H
1.246 + movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L
1.247 + movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H
1.248 +
1.249 + ; -- Odd part
1.250 +
1.251 + movdqa xmm4, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
1.252 + movdqa xmm6, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
1.253 + pmullw xmm4, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.254 + pmullw xmm6, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.255 + movdqa xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
1.256 + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
1.257 + pmullw xmm1, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.258 + pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
1.259 +
1.260 + movdqa xmm5,xmm6
1.261 + movdqa xmm7,xmm4
1.262 + paddw xmm5,xmm3 ; xmm5=z3
1.263 + paddw xmm7,xmm1 ; xmm7=z4
1.264 +
1.265 + ; (Original)
1.266 + ; z5 = (z3 + z4) * 1.175875602;
1.267 + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
1.268 + ; z3 += z5; z4 += z5;
1.269 + ;
1.270 + ; (This implementation)
1.271 + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
1.272 + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
1.273 +
1.274 + movdqa xmm2,xmm5
1.275 + movdqa xmm0,xmm5
1.276 + punpcklwd xmm2,xmm7
1.277 + punpckhwd xmm0,xmm7
1.278 + movdqa xmm5,xmm2
1.279 + movdqa xmm7,xmm0
1.280 + pmaddwd xmm2,[rel PW_MF078_F117] ; xmm2=z3L
1.281 + pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3H
1.282 + pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L
1.283 + pmaddwd xmm7,[rel PW_F117_F078] ; xmm7=z4H
1.284 +
1.285 + movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L
1.286 + movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H
1.287 +
1.288 + ; (Original)
1.289 + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
1.290 + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
1.291 + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
1.292 + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
1.293 + ; tmp0 += z1 + z3; tmp1 += z2 + z4;
1.294 + ; tmp2 += z2 + z3; tmp3 += z1 + z4;
1.295 + ;
1.296 + ; (This implementation)
1.297 + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
1.298 + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
1.299 + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
1.300 + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
1.301 + ; tmp0 += z3; tmp1 += z4;
1.302 + ; tmp2 += z3; tmp3 += z4;
1.303 +
1.304 + movdqa xmm2,xmm3
1.305 + movdqa xmm0,xmm3
1.306 + punpcklwd xmm2,xmm4
1.307 + punpckhwd xmm0,xmm4
1.308 + movdqa xmm3,xmm2
1.309 + movdqa xmm4,xmm0
1.310 + pmaddwd xmm2,[rel PW_MF060_MF089] ; xmm2=tmp0L
1.311 + pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0H
1.312 + pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3L
1.313 + pmaddwd xmm4,[rel PW_MF089_F060] ; xmm4=tmp3H
1.314 +
1.315 + paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L
1.316 + paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H
1.317 + paddd xmm3,xmm5 ; xmm3=tmp3L
1.318 + paddd xmm4,xmm7 ; xmm4=tmp3H
1.319 +
1.320 + movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L
1.321 + movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H
1.322 +
1.323 + movdqa xmm2,xmm1
1.324 + movdqa xmm0,xmm1
1.325 + punpcklwd xmm2,xmm6
1.326 + punpckhwd xmm0,xmm6
1.327 + movdqa xmm1,xmm2
1.328 + movdqa xmm6,xmm0
1.329 + pmaddwd xmm2,[rel PW_MF050_MF256] ; xmm2=tmp1L
1.330 + pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1H
1.331 + pmaddwd xmm1,[rel PW_MF256_F050] ; xmm1=tmp2L
1.332 + pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H
1.333 +
1.334 + paddd xmm2,xmm5 ; xmm2=tmp1L
1.335 + paddd xmm0,xmm7 ; xmm0=tmp1H
1.336 + paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L
1.337 + paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
1.338 +
1.339 + movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L
1.340 + movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H
1.341 +
1.342 + ; -- Final output stage
1.343 +
1.344 + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
1.345 + movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H
1.346 +
1.347 + movdqa xmm2,xmm5
1.348 + movdqa xmm0,xmm7
1.349 + paddd xmm5,xmm3 ; xmm5=data0L
1.350 + paddd xmm7,xmm4 ; xmm7=data0H
1.351 + psubd xmm2,xmm3 ; xmm2=data7L
1.352 + psubd xmm0,xmm4 ; xmm0=data7H
1.353 +
1.354 + movdqa xmm3,[rel PD_DESCALE_P1] ; xmm3=[rel PD_DESCALE_P1]
1.355 +
1.356 + paddd xmm5,xmm3
1.357 + paddd xmm7,xmm3
1.358 + psrad xmm5,DESCALE_P1
1.359 + psrad xmm7,DESCALE_P1
1.360 + paddd xmm2,xmm3
1.361 + paddd xmm0,xmm3
1.362 + psrad xmm2,DESCALE_P1
1.363 + psrad xmm0,DESCALE_P1
1.364 +
1.365 + packssdw xmm5,xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07)
1.366 + packssdw xmm2,xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77)
1.367 +
1.368 + movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L
1.369 + movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H
1.370 +
1.371 + movdqa xmm7,xmm4
1.372 + movdqa xmm0,xmm3
1.373 + paddd xmm4,xmm1 ; xmm4=data1L
1.374 + paddd xmm3,xmm6 ; xmm3=data1H
1.375 + psubd xmm7,xmm1 ; xmm7=data6L
1.376 + psubd xmm0,xmm6 ; xmm0=data6H
1.377 +
1.378 + movdqa xmm1,[rel PD_DESCALE_P1] ; xmm1=[rel PD_DESCALE_P1]
1.379 +
1.380 + paddd xmm4,xmm1
1.381 + paddd xmm3,xmm1
1.382 + psrad xmm4,DESCALE_P1
1.383 + psrad xmm3,DESCALE_P1
1.384 + paddd xmm7,xmm1
1.385 + paddd xmm0,xmm1
1.386 + psrad xmm7,DESCALE_P1
1.387 + psrad xmm0,DESCALE_P1
1.388 +
1.389 + packssdw xmm4,xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17)
1.390 + packssdw xmm7,xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67)
1.391 +
1.392 + movdqa xmm6,xmm5 ; transpose coefficients(phase 1)
1.393 + punpcklwd xmm5,xmm4 ; xmm5=(00 10 01 11 02 12 03 13)
1.394 + punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
1.395 + movdqa xmm1,xmm7 ; transpose coefficients(phase 1)
1.396 + punpcklwd xmm7,xmm2 ; xmm7=(60 70 61 71 62 72 63 73)
1.397 + punpckhwd xmm1,xmm2 ; xmm1=(64 74 65 75 66 76 67 77)
1.398 +
1.399 + movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L
1.400 + movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H
1.401 + movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L
1.402 + movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H
1.403 +
1.404 + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13)
1.405 + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17)
1.406 + movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73)
1.407 + movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77)
1.408 +
1.409 + movdqa xmm5,xmm3
1.410 + movdqa xmm6,xmm0
1.411 + paddd xmm3,xmm4 ; xmm3=data2L
1.412 + paddd xmm0,xmm2 ; xmm0=data2H
1.413 + psubd xmm5,xmm4 ; xmm5=data5L
1.414 + psubd xmm6,xmm2 ; xmm6=data5H
1.415 +
1.416 + movdqa xmm7,[rel PD_DESCALE_P1] ; xmm7=[rel PD_DESCALE_P1]
1.417 +
1.418 + paddd xmm3,xmm7
1.419 + paddd xmm0,xmm7
1.420 + psrad xmm3,DESCALE_P1
1.421 + psrad xmm0,DESCALE_P1
1.422 + paddd xmm5,xmm7
1.423 + paddd xmm6,xmm7
1.424 + psrad xmm5,DESCALE_P1
1.425 + psrad xmm6,DESCALE_P1
1.426 +
1.427 + packssdw xmm3,xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27)
1.428 + packssdw xmm5,xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57)
1.429 +
1.430 + movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L
1.431 + movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H
1.432 + movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L
1.433 + movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H
1.434 +
1.435 + movdqa xmm0,xmm1
1.436 + movdqa xmm6,xmm4
1.437 + paddd xmm1,xmm2 ; xmm1=data3L
1.438 + paddd xmm4,xmm7 ; xmm4=data3H
1.439 + psubd xmm0,xmm2 ; xmm0=data4L
1.440 + psubd xmm6,xmm7 ; xmm6=data4H
1.441 +
1.442 + movdqa xmm2,[rel PD_DESCALE_P1] ; xmm2=[rel PD_DESCALE_P1]
1.443 +
1.444 + paddd xmm1,xmm2
1.445 + paddd xmm4,xmm2
1.446 + psrad xmm1,DESCALE_P1
1.447 + psrad xmm4,DESCALE_P1
1.448 + paddd xmm0,xmm2
1.449 + paddd xmm6,xmm2
1.450 + psrad xmm0,DESCALE_P1
1.451 + psrad xmm6,DESCALE_P1
1.452 +
1.453 + packssdw xmm1,xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37)
1.454 + packssdw xmm0,xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47)
1.455 +
1.456 + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13)
1.457 + movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17)
1.458 +
1.459 + movdqa xmm4,xmm3 ; transpose coefficients(phase 1)
1.460 + punpcklwd xmm3,xmm1 ; xmm3=(20 30 21 31 22 32 23 33)
1.461 + punpckhwd xmm4,xmm1 ; xmm4=(24 34 25 35 26 36 27 37)
1.462 + movdqa xmm6,xmm0 ; transpose coefficients(phase 1)
1.463 + punpcklwd xmm0,xmm5 ; xmm0=(40 50 41 51 42 52 43 53)
1.464 + punpckhwd xmm6,xmm5 ; xmm6=(44 54 45 55 46 56 47 57)
1.465 +
1.466 + movdqa xmm1,xmm7 ; transpose coefficients(phase 2)
1.467 + punpckldq xmm7,xmm3 ; xmm7=(00 10 20 30 01 11 21 31)
1.468 + punpckhdq xmm1,xmm3 ; xmm1=(02 12 22 32 03 13 23 33)
1.469 + movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
1.470 + punpckldq xmm2,xmm4 ; xmm2=(04 14 24 34 05 15 25 35)
1.471 + punpckhdq xmm5,xmm4 ; xmm5=(06 16 26 36 07 17 27 37)
1.472 +
1.473 + movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73)
1.474 + movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77)
1.475 +
1.476 + movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35)
1.477 + movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37)
1.478 +
1.479 + movdqa xmm2,xmm0 ; transpose coefficients(phase 2)
1.480 + punpckldq xmm0,xmm3 ; xmm0=(40 50 60 70 41 51 61 71)
1.481 + punpckhdq xmm2,xmm3 ; xmm2=(42 52 62 72 43 53 63 73)
1.482 + movdqa xmm5,xmm6 ; transpose coefficients(phase 2)
1.483 + punpckldq xmm6,xmm4 ; xmm6=(44 54 64 74 45 55 65 75)
1.484 + punpckhdq xmm5,xmm4 ; xmm5=(46 56 66 76 47 57 67 77)
1.485 +
1.486 + movdqa xmm3,xmm7 ; transpose coefficients(phase 3)
1.487 + punpcklqdq xmm7,xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70)
1.488 + punpckhqdq xmm3,xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71)
1.489 + movdqa xmm4,xmm1 ; transpose coefficients(phase 3)
1.490 + punpcklqdq xmm1,xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72)
1.491 + punpckhqdq xmm4,xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73)
1.492 +
1.493 + movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35)
1.494 + movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37)
1.495 +
1.496 + movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1
1.497 + movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3
1.498 +
1.499 + movdqa xmm3,xmm0 ; transpose coefficients(phase 3)
1.500 + punpcklqdq xmm0,xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74)
1.501 + punpckhqdq xmm3,xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75)
1.502 + movdqa xmm4,xmm2 ; transpose coefficients(phase 3)
1.503 + punpcklqdq xmm2,xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76)
1.504 + punpckhqdq xmm4,xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77)
1.505 +
1.506 + movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
1.507 + movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
1.508 +.column_end:
1.509 +
1.510 + ; -- Prefetch the next coefficient block
1.511 +
1.512 + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
1.513 + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
1.514 + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
1.515 + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
1.516 +
1.517 + ; ---- Pass 2: process rows from work array, store into output array.
1.518 +
1.519 + mov rax, [original_rbp]
1.520 + mov rdi, r12 ; (JSAMPROW *)
1.521 + mov rax, r13
1.522 +
1.523 + ; -- Even part
1.524 +
1.525 + ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6
1.526 +
1.527 + ; (Original)
1.528 + ; z1 = (z2 + z3) * 0.541196100;
1.529 + ; tmp2 = z1 + z3 * -1.847759065;
1.530 + ; tmp3 = z1 + z2 * 0.765366865;
1.531 + ;
1.532 + ; (This implementation)
1.533 + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
1.534 + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
1.535 +
1.536 + movdqa xmm6,xmm1 ; xmm1=in2=z2
1.537 + movdqa xmm5,xmm1
1.538 + punpcklwd xmm6,xmm2 ; xmm2=in6=z3
1.539 + punpckhwd xmm5,xmm2
1.540 + movdqa xmm1,xmm6
1.541 + movdqa xmm2,xmm5
1.542 + pmaddwd xmm6,[rel PW_F130_F054] ; xmm6=tmp3L
1.543 + pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H
1.544 + pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L
1.545 + pmaddwd xmm2,[rel PW_F054_MF130] ; xmm2=tmp2H
1.546 +
1.547 + movdqa xmm3,xmm7
1.548 + paddw xmm7,xmm0 ; xmm7=in0+in4
1.549 + psubw xmm3,xmm0 ; xmm3=in0-in4
1.550 +
1.551 + pxor xmm4,xmm4
1.552 + pxor xmm0,xmm0
1.553 + punpcklwd xmm4,xmm7 ; xmm4=tmp0L
1.554 + punpckhwd xmm0,xmm7 ; xmm0=tmp0H
1.555 + psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
1.556 + psrad xmm0,(16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS
1.557 +
1.558 + movdqa xmm7,xmm4
1.559 + paddd xmm4,xmm6 ; xmm4=tmp10L
1.560 + psubd xmm7,xmm6 ; xmm7=tmp13L
1.561 + movdqa xmm6,xmm0
1.562 + paddd xmm0,xmm5 ; xmm0=tmp10H
1.563 + psubd xmm6,xmm5 ; xmm6=tmp13H
1.564 +
1.565 + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L
1.566 + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H
1.567 + movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L
1.568 + movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H
1.569 +
1.570 + pxor xmm5,xmm5
1.571 + pxor xmm4,xmm4
1.572 + punpcklwd xmm5,xmm3 ; xmm5=tmp1L
1.573 + punpckhwd xmm4,xmm3 ; xmm4=tmp1H
1.574 + psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
1.575 + psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
1.576 +
1.577 + movdqa xmm0,xmm5
1.578 + paddd xmm5,xmm1 ; xmm5=tmp11L
1.579 + psubd xmm0,xmm1 ; xmm0=tmp12L
1.580 + movdqa xmm7,xmm4
1.581 + paddd xmm4,xmm2 ; xmm4=tmp11H
1.582 + psubd xmm7,xmm2 ; xmm7=tmp12H
1.583 +
1.584 + movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
1.585 + movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H
1.586 + movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L
1.587 + movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H
1.588 +
1.589 + ; -- Odd part
1.590 +
1.591 + movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3
1.592 + movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1
1.593 + movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7
1.594 + movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5
1.595 +
1.596 + movdqa xmm5,xmm6
1.597 + movdqa xmm4,xmm3
1.598 + paddw xmm5,xmm1 ; xmm5=z3
1.599 + paddw xmm4,xmm2 ; xmm4=z4
1.600 +
1.601 + ; (Original)
1.602 + ; z5 = (z3 + z4) * 1.175875602;
1.603 + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
1.604 + ; z3 += z5; z4 += z5;
1.605 + ;
1.606 + ; (This implementation)
1.607 + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
1.608 + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
1.609 +
1.610 + movdqa xmm0,xmm5
1.611 + movdqa xmm7,xmm5
1.612 + punpcklwd xmm0,xmm4
1.613 + punpckhwd xmm7,xmm4
1.614 + movdqa xmm5,xmm0
1.615 + movdqa xmm4,xmm7
1.616 + pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3L
1.617 + pmaddwd xmm7,[rel PW_MF078_F117] ; xmm7=z3H
1.618 + pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L
1.619 + pmaddwd xmm4,[rel PW_F117_F078] ; xmm4=z4H
1.620 +
1.621 + movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L
1.622 + movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H
1.623 +
1.624 + ; (Original)
1.625 + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
1.626 + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
1.627 + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
1.628 + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
1.629 + ; tmp0 += z1 + z3; tmp1 += z2 + z4;
1.630 + ; tmp2 += z2 + z3; tmp3 += z1 + z4;
1.631 + ;
1.632 + ; (This implementation)
1.633 + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
1.634 + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
1.635 + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
1.636 + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
1.637 + ; tmp0 += z3; tmp1 += z4;
1.638 + ; tmp2 += z3; tmp3 += z4;
1.639 +
1.640 + movdqa xmm0,xmm1
1.641 + movdqa xmm7,xmm1
1.642 + punpcklwd xmm0,xmm3
1.643 + punpckhwd xmm7,xmm3
1.644 + movdqa xmm1,xmm0
1.645 + movdqa xmm3,xmm7
1.646 + pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0L
1.647 + pmaddwd xmm7,[rel PW_MF060_MF089] ; xmm7=tmp0H
1.648 + pmaddwd xmm1,[rel PW_MF089_F060] ; xmm1=tmp3L
1.649 + pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3H
1.650 +
1.651 + paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L
1.652 + paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H
1.653 + paddd xmm1,xmm5 ; xmm1=tmp3L
1.654 + paddd xmm3,xmm4 ; xmm3=tmp3H
1.655 +
1.656 + movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L
1.657 + movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H
1.658 +
1.659 + movdqa xmm0,xmm2
1.660 + movdqa xmm7,xmm2
1.661 + punpcklwd xmm0,xmm6
1.662 + punpckhwd xmm7,xmm6
1.663 + movdqa xmm2,xmm0
1.664 + movdqa xmm6,xmm7
1.665 + pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1L
1.666 + pmaddwd xmm7,[rel PW_MF050_MF256] ; xmm7=tmp1H
1.667 + pmaddwd xmm2,[rel PW_MF256_F050] ; xmm2=tmp2L
1.668 + pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H
1.669 +
1.670 + paddd xmm0,xmm5 ; xmm0=tmp1L
1.671 + paddd xmm7,xmm4 ; xmm7=tmp1H
1.672 + paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L
1.673 + paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
1.674 +
1.675 + movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L
1.676 + movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H
1.677 +
1.678 + ; -- Final output stage
1.679 +
1.680 + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
1.681 + movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H
1.682 +
1.683 + movdqa xmm0,xmm5
1.684 + movdqa xmm7,xmm4
1.685 + paddd xmm5,xmm1 ; xmm5=data0L
1.686 + paddd xmm4,xmm3 ; xmm4=data0H
1.687 + psubd xmm0,xmm1 ; xmm0=data7L
1.688 + psubd xmm7,xmm3 ; xmm7=data7H
1.689 +
1.690 + movdqa xmm1,[rel PD_DESCALE_P2] ; xmm1=[rel PD_DESCALE_P2]
1.691 +
1.692 + paddd xmm5,xmm1
1.693 + paddd xmm4,xmm1
1.694 + psrad xmm5,DESCALE_P2
1.695 + psrad xmm4,DESCALE_P2
1.696 + paddd xmm0,xmm1
1.697 + paddd xmm7,xmm1
1.698 + psrad xmm0,DESCALE_P2
1.699 + psrad xmm7,DESCALE_P2
1.700 +
1.701 + packssdw xmm5,xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70)
1.702 + packssdw xmm0,xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77)
1.703 +
1.704 + movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L
1.705 + movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H
1.706 +
1.707 + movdqa xmm4,xmm3
1.708 + movdqa xmm7,xmm1
1.709 + paddd xmm3,xmm2 ; xmm3=data1L
1.710 + paddd xmm1,xmm6 ; xmm1=data1H
1.711 + psubd xmm4,xmm2 ; xmm4=data6L
1.712 + psubd xmm7,xmm6 ; xmm7=data6H
1.713 +
1.714 + movdqa xmm2,[rel PD_DESCALE_P2] ; xmm2=[rel PD_DESCALE_P2]
1.715 +
1.716 + paddd xmm3,xmm2
1.717 + paddd xmm1,xmm2
1.718 + psrad xmm3,DESCALE_P2
1.719 + psrad xmm1,DESCALE_P2
1.720 + paddd xmm4,xmm2
1.721 + paddd xmm7,xmm2
1.722 + psrad xmm4,DESCALE_P2
1.723 + psrad xmm7,DESCALE_P2
1.724 +
1.725 + packssdw xmm3,xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71)
1.726 + packssdw xmm4,xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76)
1.727 +
1.728 + packsswb xmm5,xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
1.729 + packsswb xmm3,xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
1.730 +
1.731 + movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L
1.732 + movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H
1.733 + movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L
1.734 + movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H
1.735 +
1.736 + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
1.737 + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
1.738 +
1.739 + movdqa xmm4,xmm6
1.740 + movdqa xmm0,xmm2
1.741 + paddd xmm6,xmm1 ; xmm6=data2L
1.742 + paddd xmm2,xmm7 ; xmm2=data2H
1.743 + psubd xmm4,xmm1 ; xmm4=data5L
1.744 + psubd xmm0,xmm7 ; xmm0=data5H
1.745 +
1.746 + movdqa xmm5,[rel PD_DESCALE_P2] ; xmm5=[rel PD_DESCALE_P2]
1.747 +
1.748 + paddd xmm6,xmm5
1.749 + paddd xmm2,xmm5
1.750 + psrad xmm6,DESCALE_P2
1.751 + psrad xmm2,DESCALE_P2
1.752 + paddd xmm4,xmm5
1.753 + paddd xmm0,xmm5
1.754 + psrad xmm4,DESCALE_P2
1.755 + psrad xmm0,DESCALE_P2
1.756 +
1.757 + packssdw xmm6,xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72)
1.758 + packssdw xmm4,xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75)
1.759 +
1.760 + movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L
1.761 + movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H
1.762 + movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L
1.763 + movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H
1.764 +
1.765 + movdqa xmm2,xmm3
1.766 + movdqa xmm0,xmm1
1.767 + paddd xmm3,xmm7 ; xmm3=data3L
1.768 + paddd xmm1,xmm5 ; xmm1=data3H
1.769 + psubd xmm2,xmm7 ; xmm2=data4L
1.770 + psubd xmm0,xmm5 ; xmm0=data4H
1.771 +
1.772 + movdqa xmm7,[rel PD_DESCALE_P2] ; xmm7=[rel PD_DESCALE_P2]
1.773 +
1.774 + paddd xmm3,xmm7
1.775 + paddd xmm1,xmm7
1.776 + psrad xmm3,DESCALE_P2
1.777 + psrad xmm1,DESCALE_P2
1.778 + paddd xmm2,xmm7
1.779 + paddd xmm0,xmm7
1.780 + psrad xmm2,DESCALE_P2
1.781 + psrad xmm0,DESCALE_P2
1.782 +
1.783 + movdqa xmm5,[rel PB_CENTERJSAMP] ; xmm5=[rel PB_CENTERJSAMP]
1.784 +
1.785 + packssdw xmm3,xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73)
1.786 + packssdw xmm2,xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74)
1.787 +
1.788 + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
1.789 + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
1.790 +
1.791 + packsswb xmm6,xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
1.792 + packsswb xmm3,xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
1.793 +
1.794 + paddb xmm7,xmm5
1.795 + paddb xmm1,xmm5
1.796 + paddb xmm6,xmm5
1.797 + paddb xmm3,xmm5
1.798 +
1.799 + movdqa xmm0,xmm7 ; transpose coefficients(phase 1)
1.800 + punpcklbw xmm7,xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
1.801 + punpckhbw xmm0,xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
1.802 + movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
1.803 + punpcklbw xmm6,xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
1.804 + punpckhbw xmm2,xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
1.805 +
1.806 + movdqa xmm4,xmm7 ; transpose coefficients(phase 2)
1.807 + punpcklwd xmm7,xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
1.808 + punpckhwd xmm4,xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
1.809 + movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
1.810 + punpcklwd xmm2,xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
1.811 + punpckhwd xmm5,xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
1.812 +
1.813 + movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
1.814 + punpckldq xmm7,xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
1.815 + punpckhdq xmm1,xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
1.816 + movdqa xmm3,xmm4 ; transpose coefficients(phase 3)
1.817 + punpckldq xmm4,xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
1.818 + punpckhdq xmm3,xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
1.819 +
1.820 + pshufd xmm6,xmm7,0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
1.821 + pshufd xmm0,xmm1,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
1.822 + pshufd xmm2,xmm4,0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
1.823 + pshufd xmm5,xmm3,0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
1.824 +
1.825 + mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
1.826 + mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
1.827 + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm7
1.828 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm1
1.829 + mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW]
1.830 + mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW]
1.831 + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
1.832 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
1.833 +
1.834 + mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
1.835 + mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
1.836 + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
1.837 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0
1.838 + mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW]
1.839 + mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW]
1.840 + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm2
1.841 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm5
1.842 +
1.843 + uncollect_args
1.844 + mov rsp,rbp ; rsp <- aligned rbp
1.845 + pop rsp ; rsp <- original rbp
1.846 + pop rbp
1.847 + ret
1.848 +
1.849 +; For some reason, the OS X linker does not honor the request to align the
1.850 +; segment unless we do this.
1.851 + align 16
