2 ; jiss2int-64.asm - accurate integer IDCT (64-bit SSE2)
4 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
5 ; Copyright 2009 D. R. Commander
8 ; 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 slow-but-accurate integer implementation of the
19 ; inverse DCT (Discrete Cosine Transform). The following code is based
20 ; directly on the IJG's original jidctint.c; see the jidctint.c for
25 %include "jsimdext.inc"
28 ; --------------------------------------------------------------------------
33 %define DESCALE_P1 (CONST_BITS-PASS1_BITS)
34 %define DESCALE_P2 (CONST_BITS+PASS1_BITS+3)
37 F_0_298 equ 2446 ; FIX(0.298631336)
38 F_0_390 equ 3196 ; FIX(0.390180644)
39 F_0_541 equ 4433 ; FIX(0.541196100)
40 F_0_765 equ 6270 ; FIX(0.765366865)
41 F_0_899 equ 7373 ; FIX(0.899976223)
42 F_1_175 equ 9633 ; FIX(1.175875602)
43 F_1_501 equ 12299 ; FIX(1.501321110)
44 F_1_847 equ 15137 ; FIX(1.847759065)
45 F_1_961 equ 16069 ; FIX(1.961570560)
46 F_2_053 equ 16819 ; FIX(2.053119869)
47 F_2_562 equ 20995 ; FIX(2.562915447)
48 F_3_072 equ 25172 ; FIX(3.072711026)
50 ; NASM cannot do compile-time arithmetic on floating-point constants.
51 %define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
52 F_0_298 equ DESCALE( 320652955,30-CONST_BITS) ; FIX(0.298631336)
53 F_0_390 equ DESCALE( 418953276,30-CONST_BITS) ; FIX(0.390180644)
54 F_0_541 equ DESCALE( 581104887,30-CONST_BITS) ; FIX(0.541196100)
55 F_0_765 equ DESCALE( 821806413,30-CONST_BITS) ; FIX(0.765366865)
56 F_0_899 equ DESCALE( 966342111,30-CONST_BITS) ; FIX(0.899976223)
57 F_1_175 equ DESCALE(1262586813,30-CONST_BITS) ; FIX(1.175875602)
58 F_1_501 equ DESCALE(1612031267,30-CONST_BITS) ; FIX(1.501321110)
59 F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
60 F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560)
61 F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869)
62 F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447)
63 F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026)
66 ; --------------------------------------------------------------------------
70 global EXTN(jconst_idct_islow_sse2)
72 EXTN(jconst_idct_islow_sse2):
74 PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541
75 PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847)
76 PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175
77 PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390)
78 PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899
79 PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899)
80 PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562
81 PW_MF256_F050 times 4 dw -F_2_562, (F_3_072-F_2_562)
82 PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1-1)
83 PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2-1)
84 PB_CENTERJSAMP times 16 db CENTERJSAMPLE
88 ; --------------------------------------------------------------------------
92 ; Perform dequantization and inverse DCT on one block of coefficients.
95 ; jsimd_idct_islow_sse2 (void * dct_table, JCOEFPTR coef_block,
96 ; JSAMPARRAY output_buf, JDIMENSION output_col)
99 ; r10 = jpeg_component_info * compptr
100 ; r11 = JCOEFPTR coef_block
101 ; r12 = JSAMPARRAY output_buf
102 ; r13 = JDIMENSION output_col
104 %define original_rbp rbp+0
105 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
109 global EXTN(jsimd_idct_islow_sse2)
111 EXTN(jsimd_idct_islow_sse2):
113 mov rax,rsp ; rax = original rbp
115 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
117 mov rbp,rsp ; rbp = aligned rbp
121 ; ---- Pass 1: process columns from input.
123 mov rdx, r10 ; quantptr
126 %ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2
127 mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
128 or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
131 movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
132 movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
133 por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
134 por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
135 por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
136 por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
137 por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
145 ; -- AC terms all zero
147 movdqa xmm5, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
148 pmullw xmm5, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
150 psllw xmm5,PASS1_BITS
152 movdqa xmm4,xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07)
153 punpcklwd xmm5,xmm5 ; xmm5=(00 00 01 01 02 02 03 03)
154 punpckhwd xmm4,xmm4 ; xmm4=(04 04 05 05 06 06 07 07)
156 pshufd xmm7,xmm5,0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00)
157 pshufd xmm6,xmm5,0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01)
158 pshufd xmm1,xmm5,0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02)
159 pshufd xmm5,xmm5,0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03)
160 pshufd xmm0,xmm4,0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04)
161 pshufd xmm3,xmm4,0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05)
162 pshufd xmm2,xmm4,0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06)
163 pshufd xmm4,xmm4,0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07)
165 movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1
166 movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3
167 movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
168 movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
175 movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
176 movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
177 pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
178 pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
179 movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
180 movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
181 pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
182 pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
185 ; z1 = (z2 + z3) * 0.541196100;
186 ; tmp2 = z1 + z3 * -1.847759065;
187 ; tmp3 = z1 + z2 * 0.765366865;
189 ; (This implementation)
190 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
191 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
193 movdqa xmm4,xmm1 ; xmm1=in2=z2
195 punpcklwd xmm4,xmm3 ; xmm3=in6=z3
199 pmaddwd xmm4,[rel PW_F130_F054] ; xmm4=tmp3L
200 pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H
201 pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L
202 pmaddwd xmm3,[rel PW_F054_MF130] ; xmm3=tmp2H
205 paddw xmm0,xmm2 ; xmm0=in0+in4
206 psubw xmm6,xmm2 ; xmm6=in0-in4
210 punpcklwd xmm7,xmm0 ; xmm7=tmp0L
211 punpckhwd xmm2,xmm0 ; xmm2=tmp0H
212 psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
213 psrad xmm2,(16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS
216 paddd xmm7,xmm4 ; xmm7=tmp10L
217 psubd xmm0,xmm4 ; xmm0=tmp13L
219 paddd xmm2,xmm5 ; xmm2=tmp10H
220 psubd xmm4,xmm5 ; xmm4=tmp13H
222 movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L
223 movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H
224 movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L
225 movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H
229 punpcklwd xmm5,xmm6 ; xmm5=tmp1L
230 punpckhwd xmm7,xmm6 ; xmm7=tmp1H
231 psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
232 psrad xmm7,(16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS
235 paddd xmm5,xmm1 ; xmm5=tmp11L
236 psubd xmm2,xmm1 ; xmm2=tmp12L
238 paddd xmm7,xmm3 ; xmm7=tmp11H
239 psubd xmm0,xmm3 ; xmm0=tmp12H
241 movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
242 movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H
243 movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L
244 movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H
248 movdqa xmm4, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
249 movdqa xmm6, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
250 pmullw xmm4, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
251 pmullw xmm6, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
252 movdqa xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
253 movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
254 pmullw xmm1, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
255 pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
259 paddw xmm5,xmm3 ; xmm5=z3
260 paddw xmm7,xmm1 ; xmm7=z4
263 ; z5 = (z3 + z4) * 1.175875602;
264 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
265 ; z3 += z5; z4 += z5;
267 ; (This implementation)
268 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
269 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
277 pmaddwd xmm2,[rel PW_MF078_F117] ; xmm2=z3L
278 pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3H
279 pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L
280 pmaddwd xmm7,[rel PW_F117_F078] ; xmm7=z4H
282 movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L
283 movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H
286 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
287 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
288 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
289 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
290 ; tmp0 += z1 + z3; tmp1 += z2 + z4;
291 ; tmp2 += z2 + z3; tmp3 += z1 + z4;
293 ; (This implementation)
294 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
295 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
296 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
297 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
298 ; tmp0 += z3; tmp1 += z4;
299 ; tmp2 += z3; tmp3 += z4;
307 pmaddwd xmm2,[rel PW_MF060_MF089] ; xmm2=tmp0L
308 pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0H
309 pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3L
310 pmaddwd xmm4,[rel PW_MF089_F060] ; xmm4=tmp3H
312 paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L
313 paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H
314 paddd xmm3,xmm5 ; xmm3=tmp3L
315 paddd xmm4,xmm7 ; xmm4=tmp3H
317 movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L
318 movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H
326 pmaddwd xmm2,[rel PW_MF050_MF256] ; xmm2=tmp1L
327 pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1H
328 pmaddwd xmm1,[rel PW_MF256_F050] ; xmm1=tmp2L
329 pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H
331 paddd xmm2,xmm5 ; xmm2=tmp1L
332 paddd xmm0,xmm7 ; xmm0=tmp1H
333 paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L
334 paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
336 movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L
337 movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H
339 ; -- Final output stage
341 movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
342 movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H
346 paddd xmm5,xmm3 ; xmm5=data0L
347 paddd xmm7,xmm4 ; xmm7=data0H
348 psubd xmm2,xmm3 ; xmm2=data7L
349 psubd xmm0,xmm4 ; xmm0=data7H
351 movdqa xmm3,[rel PD_DESCALE_P1] ; xmm3=[rel PD_DESCALE_P1]
355 psrad xmm5,DESCALE_P1
356 psrad xmm7,DESCALE_P1
359 psrad xmm2,DESCALE_P1
360 psrad xmm0,DESCALE_P1
362 packssdw xmm5,xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07)
363 packssdw xmm2,xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77)
365 movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L
366 movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H
370 paddd xmm4,xmm1 ; xmm4=data1L
371 paddd xmm3,xmm6 ; xmm3=data1H
372 psubd xmm7,xmm1 ; xmm7=data6L
373 psubd xmm0,xmm6 ; xmm0=data6H
375 movdqa xmm1,[rel PD_DESCALE_P1] ; xmm1=[rel PD_DESCALE_P1]
379 psrad xmm4,DESCALE_P1
380 psrad xmm3,DESCALE_P1
383 psrad xmm7,DESCALE_P1
384 psrad xmm0,DESCALE_P1
386 packssdw xmm4,xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17)
387 packssdw xmm7,xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67)
389 movdqa xmm6,xmm5 ; transpose coefficients(phase 1)
390 punpcklwd xmm5,xmm4 ; xmm5=(00 10 01 11 02 12 03 13)
391 punpckhwd xmm6,xmm4 ; xmm6=(04 14 05 15 06 16 07 17)
392 movdqa xmm1,xmm7 ; transpose coefficients(phase 1)
393 punpcklwd xmm7,xmm2 ; xmm7=(60 70 61 71 62 72 63 73)
394 punpckhwd xmm1,xmm2 ; xmm1=(64 74 65 75 66 76 67 77)
396 movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L
397 movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H
398 movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L
399 movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H
401 movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13)
402 movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17)
403 movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73)
404 movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77)
408 paddd xmm3,xmm4 ; xmm3=data2L
409 paddd xmm0,xmm2 ; xmm0=data2H
410 psubd xmm5,xmm4 ; xmm5=data5L
411 psubd xmm6,xmm2 ; xmm6=data5H
413 movdqa xmm7,[rel PD_DESCALE_P1] ; xmm7=[rel PD_DESCALE_P1]
417 psrad xmm3,DESCALE_P1
418 psrad xmm0,DESCALE_P1
421 psrad xmm5,DESCALE_P1
422 psrad xmm6,DESCALE_P1
424 packssdw xmm3,xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27)
425 packssdw xmm5,xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57)
427 movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L
428 movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H
429 movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L
430 movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H
434 paddd xmm1,xmm2 ; xmm1=data3L
435 paddd xmm4,xmm7 ; xmm4=data3H
436 psubd xmm0,xmm2 ; xmm0=data4L
437 psubd xmm6,xmm7 ; xmm6=data4H
439 movdqa xmm2,[rel PD_DESCALE_P1] ; xmm2=[rel PD_DESCALE_P1]
443 psrad xmm1,DESCALE_P1
444 psrad xmm4,DESCALE_P1
447 psrad xmm0,DESCALE_P1
448 psrad xmm6,DESCALE_P1
450 packssdw xmm1,xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37)
451 packssdw xmm0,xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47)
453 movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13)
454 movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17)
456 movdqa xmm4,xmm3 ; transpose coefficients(phase 1)
457 punpcklwd xmm3,xmm1 ; xmm3=(20 30 21 31 22 32 23 33)
458 punpckhwd xmm4,xmm1 ; xmm4=(24 34 25 35 26 36 27 37)
459 movdqa xmm6,xmm0 ; transpose coefficients(phase 1)
460 punpcklwd xmm0,xmm5 ; xmm0=(40 50 41 51 42 52 43 53)
461 punpckhwd xmm6,xmm5 ; xmm6=(44 54 45 55 46 56 47 57)
463 movdqa xmm1,xmm7 ; transpose coefficients(phase 2)
464 punpckldq xmm7,xmm3 ; xmm7=(00 10 20 30 01 11 21 31)
465 punpckhdq xmm1,xmm3 ; xmm1=(02 12 22 32 03 13 23 33)
466 movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
467 punpckldq xmm2,xmm4 ; xmm2=(04 14 24 34 05 15 25 35)
468 punpckhdq xmm5,xmm4 ; xmm5=(06 16 26 36 07 17 27 37)
470 movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73)
471 movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77)
473 movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35)
474 movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37)
476 movdqa xmm2,xmm0 ; transpose coefficients(phase 2)
477 punpckldq xmm0,xmm3 ; xmm0=(40 50 60 70 41 51 61 71)
478 punpckhdq xmm2,xmm3 ; xmm2=(42 52 62 72 43 53 63 73)
479 movdqa xmm5,xmm6 ; transpose coefficients(phase 2)
480 punpckldq xmm6,xmm4 ; xmm6=(44 54 64 74 45 55 65 75)
481 punpckhdq xmm5,xmm4 ; xmm5=(46 56 66 76 47 57 67 77)
483 movdqa xmm3,xmm7 ; transpose coefficients(phase 3)
484 punpcklqdq xmm7,xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70)
485 punpckhqdq xmm3,xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71)
486 movdqa xmm4,xmm1 ; transpose coefficients(phase 3)
487 punpcklqdq xmm1,xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72)
488 punpckhqdq xmm4,xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73)
490 movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35)
491 movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37)
493 movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1
494 movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3
496 movdqa xmm3,xmm0 ; transpose coefficients(phase 3)
497 punpcklqdq xmm0,xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74)
498 punpckhqdq xmm3,xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75)
499 movdqa xmm4,xmm2 ; transpose coefficients(phase 3)
500 punpcklqdq xmm2,xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76)
501 punpckhqdq xmm4,xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77)
503 movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5
504 movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7
507 ; -- Prefetch the next coefficient block
509 prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
510 prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
511 prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
512 prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
514 ; ---- Pass 2: process rows from work array, store into output array.
516 mov rax, [original_rbp]
517 mov rdi, r12 ; (JSAMPROW *)
522 ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6
525 ; z1 = (z2 + z3) * 0.541196100;
526 ; tmp2 = z1 + z3 * -1.847759065;
527 ; tmp3 = z1 + z2 * 0.765366865;
529 ; (This implementation)
530 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
531 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
533 movdqa xmm6,xmm1 ; xmm1=in2=z2
535 punpcklwd xmm6,xmm2 ; xmm2=in6=z3
539 pmaddwd xmm6,[rel PW_F130_F054] ; xmm6=tmp3L
540 pmaddwd xmm5,[rel PW_F130_F054] ; xmm5=tmp3H
541 pmaddwd xmm1,[rel PW_F054_MF130] ; xmm1=tmp2L
542 pmaddwd xmm2,[rel PW_F054_MF130] ; xmm2=tmp2H
545 paddw xmm7,xmm0 ; xmm7=in0+in4
546 psubw xmm3,xmm0 ; xmm3=in0-in4
550 punpcklwd xmm4,xmm7 ; xmm4=tmp0L
551 punpckhwd xmm0,xmm7 ; xmm0=tmp0H
552 psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
553 psrad xmm0,(16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS
556 paddd xmm4,xmm6 ; xmm4=tmp10L
557 psubd xmm7,xmm6 ; xmm7=tmp13L
559 paddd xmm0,xmm5 ; xmm0=tmp10H
560 psubd xmm6,xmm5 ; xmm6=tmp13H
562 movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L
563 movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H
564 movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L
565 movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H
569 punpcklwd xmm5,xmm3 ; xmm5=tmp1L
570 punpckhwd xmm4,xmm3 ; xmm4=tmp1H
571 psrad xmm5,(16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS
572 psrad xmm4,(16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS
575 paddd xmm5,xmm1 ; xmm5=tmp11L
576 psubd xmm0,xmm1 ; xmm0=tmp12L
578 paddd xmm4,xmm2 ; xmm4=tmp11H
579 psubd xmm7,xmm2 ; xmm7=tmp12H
581 movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L
582 movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H
583 movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L
584 movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H
588 movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3
589 movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1
590 movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7
591 movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5
595 paddw xmm5,xmm1 ; xmm5=z3
596 paddw xmm4,xmm2 ; xmm4=z4
599 ; z5 = (z3 + z4) * 1.175875602;
600 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
601 ; z3 += z5; z4 += z5;
603 ; (This implementation)
604 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
605 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
613 pmaddwd xmm0,[rel PW_MF078_F117] ; xmm0=z3L
614 pmaddwd xmm7,[rel PW_MF078_F117] ; xmm7=z3H
615 pmaddwd xmm5,[rel PW_F117_F078] ; xmm5=z4L
616 pmaddwd xmm4,[rel PW_F117_F078] ; xmm4=z4H
618 movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L
619 movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H
622 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
623 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
624 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
625 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
626 ; tmp0 += z1 + z3; tmp1 += z2 + z4;
627 ; tmp2 += z2 + z3; tmp3 += z1 + z4;
629 ; (This implementation)
630 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
631 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
632 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
633 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
634 ; tmp0 += z3; tmp1 += z4;
635 ; tmp2 += z3; tmp3 += z4;
643 pmaddwd xmm0,[rel PW_MF060_MF089] ; xmm0=tmp0L
644 pmaddwd xmm7,[rel PW_MF060_MF089] ; xmm7=tmp0H
645 pmaddwd xmm1,[rel PW_MF089_F060] ; xmm1=tmp3L
646 pmaddwd xmm3,[rel PW_MF089_F060] ; xmm3=tmp3H
648 paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L
649 paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H
650 paddd xmm1,xmm5 ; xmm1=tmp3L
651 paddd xmm3,xmm4 ; xmm3=tmp3H
653 movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L
654 movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H
662 pmaddwd xmm0,[rel PW_MF050_MF256] ; xmm0=tmp1L
663 pmaddwd xmm7,[rel PW_MF050_MF256] ; xmm7=tmp1H
664 pmaddwd xmm2,[rel PW_MF256_F050] ; xmm2=tmp2L
665 pmaddwd xmm6,[rel PW_MF256_F050] ; xmm6=tmp2H
667 paddd xmm0,xmm5 ; xmm0=tmp1L
668 paddd xmm7,xmm4 ; xmm7=tmp1H
669 paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L
670 paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H
672 movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L
673 movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H
675 ; -- Final output stage
677 movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L
678 movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H
682 paddd xmm5,xmm1 ; xmm5=data0L
683 paddd xmm4,xmm3 ; xmm4=data0H
684 psubd xmm0,xmm1 ; xmm0=data7L
685 psubd xmm7,xmm3 ; xmm7=data7H
687 movdqa xmm1,[rel PD_DESCALE_P2] ; xmm1=[rel PD_DESCALE_P2]
691 psrad xmm5,DESCALE_P2
692 psrad xmm4,DESCALE_P2
695 psrad xmm0,DESCALE_P2
696 psrad xmm7,DESCALE_P2
698 packssdw xmm5,xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70)
699 packssdw xmm0,xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77)
701 movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L
702 movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H
706 paddd xmm3,xmm2 ; xmm3=data1L
707 paddd xmm1,xmm6 ; xmm1=data1H
708 psubd xmm4,xmm2 ; xmm4=data6L
709 psubd xmm7,xmm6 ; xmm7=data6H
711 movdqa xmm2,[rel PD_DESCALE_P2] ; xmm2=[rel PD_DESCALE_P2]
715 psrad xmm3,DESCALE_P2
716 psrad xmm1,DESCALE_P2
719 psrad xmm4,DESCALE_P2
720 psrad xmm7,DESCALE_P2
722 packssdw xmm3,xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71)
723 packssdw xmm4,xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76)
725 packsswb xmm5,xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
726 packsswb xmm3,xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
728 movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L
729 movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H
730 movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L
731 movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H
733 movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
734 movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
738 paddd xmm6,xmm1 ; xmm6=data2L
739 paddd xmm2,xmm7 ; xmm2=data2H
740 psubd xmm4,xmm1 ; xmm4=data5L
741 psubd xmm0,xmm7 ; xmm0=data5H
743 movdqa xmm5,[rel PD_DESCALE_P2] ; xmm5=[rel PD_DESCALE_P2]
747 psrad xmm6,DESCALE_P2
748 psrad xmm2,DESCALE_P2
751 psrad xmm4,DESCALE_P2
752 psrad xmm0,DESCALE_P2
754 packssdw xmm6,xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72)
755 packssdw xmm4,xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75)
757 movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L
758 movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H
759 movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L
760 movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H
764 paddd xmm3,xmm7 ; xmm3=data3L
765 paddd xmm1,xmm5 ; xmm1=data3H
766 psubd xmm2,xmm7 ; xmm2=data4L
767 psubd xmm0,xmm5 ; xmm0=data4H
769 movdqa xmm7,[rel PD_DESCALE_P2] ; xmm7=[rel PD_DESCALE_P2]
773 psrad xmm3,DESCALE_P2
774 psrad xmm1,DESCALE_P2
777 psrad xmm2,DESCALE_P2
778 psrad xmm0,DESCALE_P2
780 movdqa xmm5,[rel PB_CENTERJSAMP] ; xmm5=[rel PB_CENTERJSAMP]
782 packssdw xmm3,xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73)
783 packssdw xmm2,xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74)
785 movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
786 movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
788 packsswb xmm6,xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
789 packsswb xmm3,xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
796 movdqa xmm0,xmm7 ; transpose coefficients(phase 1)
797 punpcklbw xmm7,xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
798 punpckhbw xmm0,xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
799 movdqa xmm2,xmm6 ; transpose coefficients(phase 1)
800 punpcklbw xmm6,xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
801 punpckhbw xmm2,xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
803 movdqa xmm4,xmm7 ; transpose coefficients(phase 2)
804 punpcklwd xmm7,xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
805 punpckhwd xmm4,xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
806 movdqa xmm5,xmm2 ; transpose coefficients(phase 2)
807 punpcklwd xmm2,xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
808 punpckhwd xmm5,xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
810 movdqa xmm1,xmm7 ; transpose coefficients(phase 3)
811 punpckldq xmm7,xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
812 punpckhdq xmm1,xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
813 movdqa xmm3,xmm4 ; transpose coefficients(phase 3)
814 punpckldq xmm4,xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
815 punpckhdq xmm3,xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
817 pshufd xmm6,xmm7,0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
818 pshufd xmm0,xmm1,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
819 pshufd xmm2,xmm4,0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
820 pshufd xmm5,xmm3,0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
822 mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
823 mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
824 movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm7
825 movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm1
826 mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW]
827 mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW]
828 movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
829 movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
831 mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
832 mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
833 movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
834 movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0
835 mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW]
836 mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW]
837 movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm2
838 movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm5
841 mov rsp,rbp ; rsp <- aligned rbp
842 pop rsp ; rsp <- original rbp
846 ; For some reason, the OS X linker does not honor the request to align the
847 ; segment unless we do this.