2 ; jfdctint.asm - accurate integer FDCT (64-bit SSE2)
4 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
5 ; Copyright (C) 2009, 2016, 2020, D. R. Commander.
6 ; Copyright (C) 2023, Aliaksiej Kandracienka.
8 ; Based on the x86 SIMD extension for IJG JPEG library
9 ; Copyright (C) 1999-2006, MIYASAKA Masaru.
10 ; For conditions of distribution and use, see copyright notice in jsimdext.inc
12 ; This file should be assembled with NASM (Netwide Assembler),
13 ; can *not* be assembled with Microsoft's MASM or any compatible
14 ; assembler (including Borland's Turbo Assembler).
15 ; NASM is available from http://nasm.sourceforge.net/ or
16 ; http://sourceforge.net/project/showfiles.php?group_id=6208
18 ; This file contains a slower but more accurate integer implementation of the
19 ; forward DCT (Discrete Cosine Transform). The following code is based
20 ; directly on the IJG's original jfdctint.c; see the jfdctint.c for
23 %include "jsimdext.inc"
26 ; --------------------------------------------------------------------------
31 %define DESCALE_P1 (CONST_BITS - PASS1_BITS)
32 %define DESCALE_P2 (CONST_BITS + PASS1_BITS)
35 F_0_298 equ 2446 ; FIX(0.298631336)
36 F_0_390 equ 3196 ; FIX(0.390180644)
37 F_0_541 equ 4433 ; FIX(0.541196100)
38 F_0_765 equ 6270 ; FIX(0.765366865)
39 F_0_899 equ 7373 ; FIX(0.899976223)
40 F_1_175 equ 9633 ; FIX(1.175875602)
41 F_1_501 equ 12299 ; FIX(1.501321110)
42 F_1_847 equ 15137 ; FIX(1.847759065)
43 F_1_961 equ 16069 ; FIX(1.961570560)
44 F_2_053 equ 16819 ; FIX(2.053119869)
45 F_2_562 equ 20995 ; FIX(2.562915447)
46 F_3_072 equ 25172 ; FIX(3.072711026)
48 ; NASM cannot do compile-time arithmetic on floating-point constants.
49 %define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n))
50 F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336)
51 F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644)
52 F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100)
53 F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865)
54 F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223)
55 F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602)
56 F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110)
57 F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065)
58 F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560)
59 F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869)
60 F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447)
61 F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026)
64 ; --------------------------------------------------------------------------
68 GLOBAL_DATA(jconst_fdct_islow_sse2)
70 EXTN(jconst_fdct_islow_sse2):
72 PW_F130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541
73 PW_F054_MF130 times 4 dw F_0_541, (F_0_541 - F_1_847)
74 PW_MF078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175
75 PW_F117_F078 times 4 dw F_1_175, (F_1_175 - F_0_390)
76 PW_MF060_MF089 times 4 dw (F_0_298 - F_0_899), -F_0_899
77 PW_MF089_F060 times 4 dw -F_0_899, (F_1_501 - F_0_899)
78 PW_MF050_MF256 times 4 dw (F_2_053 - F_2_562), -F_2_562
79 PW_MF256_F050 times 4 dw -F_2_562, (F_3_072 - F_2_562)
80 PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1 - 1)
81 PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2 - 1)
82 PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS - 1)
86 ; --------------------------------------------------------------------------
90 ; Perform the forward DCT on one block of samples.
93 ; jsimd_fdct_islow_sse2(DCTELEM *data)
98 %define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
102 GLOBAL_FUNCTION(jsimd_fdct_islow_sse2)
104 EXTN(jsimd_fdct_islow_sse2):
108 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
109 ; Allocate stack space for wk array. r15 is used to access it.
111 sub rsp, byte (SIZEOF_XMMWORD * WK_NUM)
114 ; ---- Pass 1: process rows.
116 mov rdx, r10 ; (DCTELEM *)
118 movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)]
119 movdqa xmm1, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)]
120 movdqa xmm2, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)]
121 movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)]
123 ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27)
124 ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37)
126 movdqa xmm4, xmm0 ; transpose coefficients(phase 1)
127 punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13)
128 punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17)
129 movdqa xmm5, xmm2 ; transpose coefficients(phase 1)
130 punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33)
131 punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37)
133 movdqa xmm6, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)]
134 movdqa xmm7, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)]
135 movdqa xmm1, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)]
136 movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)]
138 ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62)
139 ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63)
141 movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33)
142 movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37)
144 movdqa xmm2, xmm6 ; transpose coefficients(phase 1)
145 punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53)
146 punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57)
147 movdqa xmm5, xmm1 ; transpose coefficients(phase 1)
148 punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73)
149 punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77)
151 movdqa xmm7, xmm6 ; transpose coefficients(phase 2)
152 punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71)
153 punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73)
154 movdqa xmm3, xmm2 ; transpose coefficients(phase 2)
155 punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75)
156 punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77)
158 movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33)
159 movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37)
160 movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73)
161 movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75)
163 movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
164 punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31)
165 punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33)
166 movdqa xmm2, xmm4 ; transpose coefficients(phase 2)
167 punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35)
168 punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37)
170 movdqa xmm1, xmm0 ; transpose coefficients(phase 3)
171 punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0
172 punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1
173 movdqa xmm5, xmm2 ; transpose coefficients(phase 3)
174 punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6
175 punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7
179 psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6
180 psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7
181 paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1
182 paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0
184 movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73)
185 movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75)
186 movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6
187 movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7
189 movdqa xmm1, xmm7 ; transpose coefficients(phase 3)
190 punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2
191 punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3
192 movdqa xmm0, xmm4 ; transpose coefficients(phase 3)
193 punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4
194 punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5
198 paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3
199 paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2
200 psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4
201 psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5
207 paddw xmm3, xmm1 ; xmm3=tmp10
208 paddw xmm6, xmm7 ; xmm6=tmp11
209 psubw xmm4, xmm1 ; xmm4=tmp13
210 psubw xmm0, xmm7 ; xmm0=tmp12
213 paddw xmm3, xmm6 ; xmm3=tmp10+tmp11
214 psubw xmm1, xmm6 ; xmm1=tmp10-tmp11
216 psllw xmm3, PASS1_BITS ; xmm3=data0
217 psllw xmm1, PASS1_BITS ; xmm1=data4
219 movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0
220 movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4
223 ; z1 = (tmp12 + tmp13) * 0.541196100;
224 ; data2 = z1 + tmp13 * 0.765366865;
225 ; data6 = z1 + tmp12 * -1.847759065;
227 ; (This implementation)
228 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
229 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
231 movdqa xmm7, xmm4 ; xmm4=tmp13
233 punpcklwd xmm7, xmm0 ; xmm0=tmp12
237 pmaddwd xmm7, [rel PW_F130_F054] ; xmm7=data2L
238 pmaddwd xmm6, [rel PW_F130_F054] ; xmm6=data2H
239 pmaddwd xmm4, [rel PW_F054_MF130] ; xmm4=data6L
240 pmaddwd xmm0, [rel PW_F054_MF130] ; xmm0=data6H
242 paddd xmm7, [rel PD_DESCALE_P1]
243 paddd xmm6, [rel PD_DESCALE_P1]
244 psrad xmm7, DESCALE_P1
245 psrad xmm6, DESCALE_P1
246 paddd xmm4, [rel PD_DESCALE_P1]
247 paddd xmm0, [rel PD_DESCALE_P1]
248 psrad xmm4, DESCALE_P1
249 psrad xmm0, DESCALE_P1
251 packssdw xmm7, xmm6 ; xmm7=data2
252 packssdw xmm4, xmm0 ; xmm4=data6
254 movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2
255 movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6
259 movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6
260 movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7
262 movdqa xmm6, xmm2 ; xmm2=tmp4
263 movdqa xmm0, xmm5 ; xmm5=tmp5
264 paddw xmm6, xmm3 ; xmm6=z3
265 paddw xmm0, xmm1 ; xmm0=z4
268 ; z5 = (z3 + z4) * 1.175875602;
269 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
270 ; z3 += z5; z4 += z5;
272 ; (This implementation)
273 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
274 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
282 pmaddwd xmm7, [rel PW_MF078_F117] ; xmm7=z3L
283 pmaddwd xmm4, [rel PW_MF078_F117] ; xmm4=z3H
284 pmaddwd xmm6, [rel PW_F117_F078] ; xmm6=z4L
285 pmaddwd xmm0, [rel PW_F117_F078] ; xmm0=z4H
287 movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L
288 movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H
291 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
292 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
293 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
294 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
295 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
296 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
298 ; (This implementation)
299 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
300 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
301 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
302 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
303 ; data7 = tmp4 + z3; data5 = tmp5 + z4;
304 ; data3 = tmp6 + z3; data1 = tmp7 + z4;
312 pmaddwd xmm7, [rel PW_MF060_MF089] ; xmm7=tmp4L
313 pmaddwd xmm4, [rel PW_MF060_MF089] ; xmm4=tmp4H
314 pmaddwd xmm2, [rel PW_MF089_F060] ; xmm2=tmp7L
315 pmaddwd xmm1, [rel PW_MF089_F060] ; xmm1=tmp7H
317 paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L
318 paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H
319 paddd xmm2, xmm6 ; xmm2=data1L
320 paddd xmm1, xmm0 ; xmm1=data1H
322 paddd xmm7, [rel PD_DESCALE_P1]
323 paddd xmm4, [rel PD_DESCALE_P1]
324 psrad xmm7, DESCALE_P1
325 psrad xmm4, DESCALE_P1
326 paddd xmm2, [rel PD_DESCALE_P1]
327 paddd xmm1, [rel PD_DESCALE_P1]
328 psrad xmm2, DESCALE_P1
329 psrad xmm1, DESCALE_P1
331 packssdw xmm7, xmm4 ; xmm7=data7
332 packssdw xmm2, xmm1 ; xmm2=data1
340 pmaddwd xmm4, [rel PW_MF050_MF256] ; xmm4=tmp5L
341 pmaddwd xmm1, [rel PW_MF050_MF256] ; xmm1=tmp5H
342 pmaddwd xmm5, [rel PW_MF256_F050] ; xmm5=tmp6L
343 pmaddwd xmm3, [rel PW_MF256_F050] ; xmm3=tmp6H
345 paddd xmm4, xmm6 ; xmm4=data5L
346 paddd xmm1, xmm0 ; xmm1=data5H
347 paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L
348 paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H
350 paddd xmm4, [rel PD_DESCALE_P1]
351 paddd xmm1, [rel PD_DESCALE_P1]
352 psrad xmm4, DESCALE_P1
353 psrad xmm1, DESCALE_P1
354 paddd xmm5, [rel PD_DESCALE_P1]
355 paddd xmm3, [rel PD_DESCALE_P1]
356 psrad xmm5, DESCALE_P1
357 psrad xmm3, DESCALE_P1
359 packssdw xmm4, xmm1 ; xmm4=data5
360 packssdw xmm5, xmm3 ; xmm5=data3
362 ; ---- Pass 2: process columns.
364 movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0
365 movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2
367 ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72)
368 ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73)
370 movdqa xmm1, xmm6 ; transpose coefficients(phase 1)
371 punpcklwd xmm6, xmm2 ; xmm6=(00 01 10 11 20 21 30 31)
372 punpckhwd xmm1, xmm2 ; xmm1=(40 41 50 51 60 61 70 71)
373 movdqa xmm3, xmm0 ; transpose coefficients(phase 1)
374 punpcklwd xmm0, xmm5 ; xmm0=(02 03 12 13 22 23 32 33)
375 punpckhwd xmm3, xmm5 ; xmm3=(42 43 52 53 62 63 72 73)
377 movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4
378 movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6
380 ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76)
381 ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77)
383 movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33)
384 movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73)
386 movdqa xmm0, xmm2 ; transpose coefficients(phase 1)
387 punpcklwd xmm2, xmm4 ; xmm2=(04 05 14 15 24 25 34 35)
388 punpckhwd xmm0, xmm4 ; xmm0=(44 45 54 55 64 65 74 75)
389 movdqa xmm3, xmm5 ; transpose coefficients(phase 1)
390 punpcklwd xmm5, xmm7 ; xmm5=(06 07 16 17 26 27 36 37)
391 punpckhwd xmm3, xmm7 ; xmm3=(46 47 56 57 66 67 76 77)
393 movdqa xmm4, xmm2 ; transpose coefficients(phase 2)
394 punpckldq xmm2, xmm5 ; xmm2=(04 05 06 07 14 15 16 17)
395 punpckhdq xmm4, xmm5 ; xmm4=(24 25 26 27 34 35 36 37)
396 movdqa xmm7, xmm0 ; transpose coefficients(phase 2)
397 punpckldq xmm0, xmm3 ; xmm0=(44 45 46 47 54 55 56 57)
398 punpckhdq xmm7, xmm3 ; xmm7=(64 65 66 67 74 75 76 77)
400 movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33)
401 movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73)
402 movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37)
403 movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57)
405 movdqa xmm4, xmm6 ; transpose coefficients(phase 2)
406 punpckldq xmm6, xmm5 ; xmm6=(00 01 02 03 10 11 12 13)
407 punpckhdq xmm4, xmm5 ; xmm4=(20 21 22 23 30 31 32 33)
408 movdqa xmm0, xmm1 ; transpose coefficients(phase 2)
409 punpckldq xmm1, xmm3 ; xmm1=(40 41 42 43 50 51 52 53)
410 punpckhdq xmm0, xmm3 ; xmm0=(60 61 62 63 70 71 72 73)
412 movdqa xmm5, xmm6 ; transpose coefficients(phase 3)
413 punpcklqdq xmm6, xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0
414 punpckhqdq xmm5, xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1
415 movdqa xmm3, xmm0 ; transpose coefficients(phase 3)
416 punpcklqdq xmm0, xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6
417 punpckhqdq xmm3, xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7
421 psubw xmm5, xmm0 ; xmm5=data1-data6=tmp6
422 psubw xmm6, xmm3 ; xmm6=data0-data7=tmp7
423 paddw xmm2, xmm0 ; xmm2=data1+data6=tmp1
424 paddw xmm7, xmm3 ; xmm7=data0+data7=tmp0
426 movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37)
427 movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57)
428 movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6
429 movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7
431 movdqa xmm5, xmm4 ; transpose coefficients(phase 3)
432 punpcklqdq xmm4, xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2
433 punpckhqdq xmm5, xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3
434 movdqa xmm6, xmm1 ; transpose coefficients(phase 3)
435 punpcklqdq xmm1, xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4
436 punpckhqdq xmm6, xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5
440 paddw xmm5, xmm1 ; xmm5=data3+data4=tmp3
441 paddw xmm4, xmm6 ; xmm4=data2+data5=tmp2
442 psubw xmm0, xmm1 ; xmm0=data3-data4=tmp4
443 psubw xmm3, xmm6 ; xmm3=data2-data5=tmp5
449 paddw xmm7, xmm5 ; xmm7=tmp10
450 paddw xmm2, xmm4 ; xmm2=tmp11
451 psubw xmm1, xmm5 ; xmm1=tmp13
452 psubw xmm6, xmm4 ; xmm6=tmp12
455 paddw xmm7, xmm2 ; xmm7=tmp10+tmp11
456 psubw xmm5, xmm2 ; xmm5=tmp10-tmp11
458 paddw xmm7, [rel PW_DESCALE_P2X]
459 paddw xmm5, [rel PW_DESCALE_P2X]
460 psraw xmm7, PASS1_BITS ; xmm7=data0
461 psraw xmm5, PASS1_BITS ; xmm5=data4
463 movdqa XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)], xmm7
464 movdqa XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)], xmm5
467 ; z1 = (tmp12 + tmp13) * 0.541196100;
468 ; data2 = z1 + tmp13 * 0.765366865;
469 ; data6 = z1 + tmp12 * -1.847759065;
471 ; (This implementation)
472 ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100;
473 ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065);
475 movdqa xmm4, xmm1 ; xmm1=tmp13
477 punpcklwd xmm4, xmm6 ; xmm6=tmp12
481 pmaddwd xmm4, [rel PW_F130_F054] ; xmm4=data2L
482 pmaddwd xmm2, [rel PW_F130_F054] ; xmm2=data2H
483 pmaddwd xmm1, [rel PW_F054_MF130] ; xmm1=data6L
484 pmaddwd xmm6, [rel PW_F054_MF130] ; xmm6=data6H
486 paddd xmm4, [rel PD_DESCALE_P2]
487 paddd xmm2, [rel PD_DESCALE_P2]
488 psrad xmm4, DESCALE_P2
489 psrad xmm2, DESCALE_P2
490 paddd xmm1, [rel PD_DESCALE_P2]
491 paddd xmm6, [rel PD_DESCALE_P2]
492 psrad xmm1, DESCALE_P2
493 psrad xmm6, DESCALE_P2
495 packssdw xmm4, xmm2 ; xmm4=data2
496 packssdw xmm1, xmm6 ; xmm1=data6
498 movdqa XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)], xmm4
499 movdqa XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)], xmm1
503 movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6
504 movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7
506 movdqa xmm2, xmm0 ; xmm0=tmp4
507 movdqa xmm6, xmm3 ; xmm3=tmp5
508 paddw xmm2, xmm7 ; xmm2=z3
509 paddw xmm6, xmm5 ; xmm6=z4
512 ; z5 = (z3 + z4) * 1.175875602;
513 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
514 ; z3 += z5; z4 += z5;
516 ; (This implementation)
517 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
518 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
526 pmaddwd xmm4, [rel PW_MF078_F117] ; xmm4=z3L
527 pmaddwd xmm1, [rel PW_MF078_F117] ; xmm1=z3H
528 pmaddwd xmm2, [rel PW_F117_F078] ; xmm2=z4L
529 pmaddwd xmm6, [rel PW_F117_F078] ; xmm6=z4H
531 movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L
532 movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H
535 ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6;
536 ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869;
537 ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110;
538 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
539 ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4;
540 ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4;
542 ; (This implementation)
543 ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223;
544 ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447;
545 ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447);
546 ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223);
547 ; data7 = tmp4 + z3; data5 = tmp5 + z4;
548 ; data3 = tmp6 + z3; data1 = tmp7 + z4;
556 pmaddwd xmm4, [rel PW_MF060_MF089] ; xmm4=tmp4L
557 pmaddwd xmm1, [rel PW_MF060_MF089] ; xmm1=tmp4H
558 pmaddwd xmm0, [rel PW_MF089_F060] ; xmm0=tmp7L
559 pmaddwd xmm5, [rel PW_MF089_F060] ; xmm5=tmp7H
561 paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L
562 paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H
563 paddd xmm0, xmm2 ; xmm0=data1L
564 paddd xmm5, xmm6 ; xmm5=data1H
566 paddd xmm4, [rel PD_DESCALE_P2]
567 paddd xmm1, [rel PD_DESCALE_P2]
568 psrad xmm4, DESCALE_P2
569 psrad xmm1, DESCALE_P2
570 paddd xmm0, [rel PD_DESCALE_P2]
571 paddd xmm5, [rel PD_DESCALE_P2]
572 psrad xmm0, DESCALE_P2
573 psrad xmm5, DESCALE_P2
575 packssdw xmm4, xmm1 ; xmm4=data7
576 packssdw xmm0, xmm5 ; xmm0=data1
578 movdqa XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)], xmm4
579 movdqa XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)], xmm0
587 pmaddwd xmm1, [rel PW_MF050_MF256] ; xmm1=tmp5L
588 pmaddwd xmm5, [rel PW_MF050_MF256] ; xmm5=tmp5H
589 pmaddwd xmm3, [rel PW_MF256_F050] ; xmm3=tmp6L
590 pmaddwd xmm7, [rel PW_MF256_F050] ; xmm7=tmp6H
592 paddd xmm1, xmm2 ; xmm1=data5L
593 paddd xmm5, xmm6 ; xmm5=data5H
594 paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L
595 paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H
597 paddd xmm1, [rel PD_DESCALE_P2]
598 paddd xmm5, [rel PD_DESCALE_P2]
599 psrad xmm1, DESCALE_P2
600 psrad xmm5, DESCALE_P2
601 paddd xmm3, [rel PD_DESCALE_P2]
602 paddd xmm7, [rel PD_DESCALE_P2]
603 psrad xmm3, DESCALE_P2
604 psrad xmm7, DESCALE_P2
606 packssdw xmm1, xmm5 ; xmm1=data5
607 packssdw xmm3, xmm7 ; xmm3=data3
609 movdqa XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)], xmm1
610 movdqa XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)], xmm3
618 ; For some reason, the OS X linker does not honor the request to align the
619 ; segment unless we do this.