2 ; jidctint.asm - accurate integer IDCT (MMX)
4 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
5 ; Copyright (C) 2016, D. R. Commander.
7 ; Based on the x86 SIMD extension for IJG JPEG library
8 ; Copyright (C) 1999-2006, MIYASAKA Masaru.
9 ; For conditions of distribution and use, see copyright notice in jsimdext.inc
11 ; This file should be assembled with NASM (Netwide Assembler),
12 ; can *not* be assembled with Microsoft's MASM or any compatible
13 ; assembler (including Borland's Turbo Assembler).
14 ; NASM is available from http://nasm.sourceforge.net/ or
15 ; http://sourceforge.net/project/showfiles.php?group_id=6208
17 ; This file contains a slow-but-accurate integer implementation of the
18 ; inverse DCT (Discrete Cosine Transform). The following code is based
19 ; directly on the IJG's original jidctint.c; see the jidctint.c for
24 %include "jsimdext.inc"
27 ; --------------------------------------------------------------------------
32 %define DESCALE_P1 (CONST_BITS - PASS1_BITS)
33 %define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3)
36 F_0_298 equ 2446 ; FIX(0.298631336)
37 F_0_390 equ 3196 ; FIX(0.390180644)
38 F_0_541 equ 4433 ; FIX(0.541196100)
39 F_0_765 equ 6270 ; FIX(0.765366865)
40 F_0_899 equ 7373 ; FIX(0.899976223)
41 F_1_175 equ 9633 ; FIX(1.175875602)
42 F_1_501 equ 12299 ; FIX(1.501321110)
43 F_1_847 equ 15137 ; FIX(1.847759065)
44 F_1_961 equ 16069 ; FIX(1.961570560)
45 F_2_053 equ 16819 ; FIX(2.053119869)
46 F_2_562 equ 20995 ; FIX(2.562915447)
47 F_3_072 equ 25172 ; FIX(3.072711026)
49 ; NASM cannot do compile-time arithmetic on floating-point constants.
50 %define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n))
51 F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336)
52 F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644)
53 F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100)
54 F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865)
55 F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223)
56 F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602)
57 F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110)
58 F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065)
59 F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560)
60 F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869)
61 F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447)
62 F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026)
65 ; --------------------------------------------------------------------------
69 GLOBAL_DATA(jconst_idct_islow_mmx)
71 EXTN(jconst_idct_islow_mmx):
73 PW_F130_F054 times 2 dw (F_0_541 + F_0_765), F_0_541
74 PW_F054_MF130 times 2 dw F_0_541, (F_0_541 - F_1_847)
75 PW_MF078_F117 times 2 dw (F_1_175 - F_1_961), F_1_175
76 PW_F117_F078 times 2 dw F_1_175, (F_1_175 - F_0_390)
77 PW_MF060_MF089 times 2 dw (F_0_298 - F_0_899), -F_0_899
78 PW_MF089_F060 times 2 dw -F_0_899, (F_1_501 - F_0_899)
79 PW_MF050_MF256 times 2 dw (F_2_053 - F_2_562), -F_2_562
80 PW_MF256_F050 times 2 dw -F_2_562, (F_3_072 - F_2_562)
81 PD_DESCALE_P1 times 2 dd 1 << (DESCALE_P1 - 1)
82 PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2 - 1)
83 PB_CENTERJSAMP times 8 db CENTERJSAMPLE
87 ; --------------------------------------------------------------------------
91 ; Perform dequantization and inverse DCT on one block of coefficients.
94 ; jsimd_idct_islow_mmx(void *dct_table, JCOEFPTR coef_block,
95 ; JSAMPARRAY output_buf, JDIMENSION output_col)
98 %define dct_table(b) (b) + 8 ; jpeg_component_info *compptr
99 %define coef_block(b) (b) + 12 ; JCOEFPTR coef_block
100 %define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf
101 %define output_col(b) (b) + 20 ; JDIMENSION output_col
103 %define original_ebp ebp + 0
104 %define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD
107 %define workspace wk(0) - DCTSIZE2 * SIZEOF_JCOEF
108 ; JCOEF workspace[DCTSIZE2]
111 GLOBAL_FUNCTION(jsimd_idct_islow_mmx)
113 EXTN(jsimd_idct_islow_mmx):
115 mov eax, esp ; eax = original ebp
117 and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits
119 mov ebp, esp ; ebp = aligned ebp
122 ; push ecx ; need not be preserved
123 ; push edx ; need not be preserved
127 get_GOT ebx ; get GOT address
129 ; ---- Pass 1: process columns from input, store into work array.
131 ; mov eax, [original_ebp]
132 mov edx, POINTER [dct_table(eax)] ; quantptr
133 mov esi, JCOEFPTR [coef_block(eax)] ; inptr
134 lea edi, [workspace] ; JCOEF *wsptr
135 mov ecx, DCTSIZE/4 ; ctr
138 %ifndef NO_ZERO_COLUMN_TEST_ISLOW_MMX
139 mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
140 or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
143 movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
144 movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
145 por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
146 por mm1, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
147 por mm0, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
148 por mm1, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
149 por mm0, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
156 ; -- AC terms all zero
158 movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
159 pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
161 psllw mm0, PASS1_BITS
163 movq mm2, mm0 ; mm0=in0=(00 01 02 03)
164 punpcklwd mm0, mm0 ; mm0=(00 00 01 01)
165 punpckhwd mm2, mm2 ; mm2=(02 02 03 03)
168 punpckldq mm0, mm0 ; mm0=(00 00 00 00)
169 punpckhdq mm1, mm1 ; mm1=(01 01 01 01)
171 punpckldq mm2, mm2 ; mm2=(02 02 02 02)
172 punpckhdq mm3, mm3 ; mm3=(03 03 03 03)
174 movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0
175 movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0
176 movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1
177 movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm1
178 movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2
179 movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm2
180 movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm3
181 movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm3
189 movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
190 movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
191 pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
192 pmullw mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
193 movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
194 movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
195 pmullw mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
196 pmullw mm3, MMWORD [MMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
199 ; z1 = (z2 + z3) * 0.541196100;
200 ; tmp2 = z1 + z3 * -1.847759065;
201 ; tmp3 = z1 + z2 * 0.765366865;
203 ; (This implementation)
204 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
205 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
207 movq mm4, mm1 ; mm1=in2=z2
209 punpcklwd mm4, mm3 ; mm3=in6=z3
213 pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L
214 pmaddwd mm5, [GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H
215 pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L
216 pmaddwd mm3, [GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H
219 paddw mm0, mm2 ; mm0=in0+in4
220 psubw mm6, mm2 ; mm6=in0-in4
224 punpcklwd mm7, mm0 ; mm7=tmp0L
225 punpckhwd mm2, mm0 ; mm2=tmp0H
226 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
227 psrad mm2, (16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS
230 paddd mm7, mm4 ; mm7=tmp10L
231 psubd mm0, mm4 ; mm0=tmp13L
233 paddd mm2, mm5 ; mm2=tmp10H
234 psubd mm4, mm5 ; mm4=tmp13H
236 movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L
237 movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H
238 movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L
239 movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H
243 punpcklwd mm5, mm6 ; mm5=tmp1L
244 punpckhwd mm7, mm6 ; mm7=tmp1H
245 psrad mm5, (16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS
246 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
249 paddd mm5, mm1 ; mm5=tmp11L
250 psubd mm2, mm1 ; mm2=tmp12L
252 paddd mm7, mm3 ; mm7=tmp11H
253 psubd mm0, mm3 ; mm0=tmp12H
255 movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L
256 movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H
257 movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L
258 movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H
262 movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
263 movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
264 pmullw mm4, MMWORD [MMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
265 pmullw mm6, MMWORD [MMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
266 movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
267 movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
268 pmullw mm1, MMWORD [MMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
269 pmullw mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
273 paddw mm5, mm3 ; mm5=z3
274 paddw mm7, mm1 ; mm7=z4
277 ; z5 = (z3 + z4) * 1.175875602;
278 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
279 ; z3 += z5; z4 += z5;
281 ; (This implementation)
282 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
283 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
291 pmaddwd mm2, [GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L
292 pmaddwd mm0, [GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H
293 pmaddwd mm5, [GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L
294 pmaddwd mm7, [GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H
296 movq MMWORD [wk(10)], mm2 ; wk(10)=z3L
297 movq MMWORD [wk(11)], mm0 ; wk(11)=z3H
300 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
301 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
302 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
303 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
304 ; tmp0 += z1 + z3; tmp1 += z2 + z4;
305 ; tmp2 += z2 + z3; tmp3 += z1 + z4;
307 ; (This implementation)
308 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
309 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
310 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
311 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
312 ; tmp0 += z3; tmp1 += z4;
313 ; tmp2 += z3; tmp3 += z4;
321 pmaddwd mm2, [GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L
322 pmaddwd mm0, [GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H
323 pmaddwd mm3, [GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L
324 pmaddwd mm4, [GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H
326 paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L
327 paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H
328 paddd mm3, mm5 ; mm3=tmp3L
329 paddd mm4, mm7 ; mm4=tmp3H
331 movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L
332 movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H
340 pmaddwd mm2, [GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L
341 pmaddwd mm0, [GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H
342 pmaddwd mm1, [GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L
343 pmaddwd mm6, [GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H
345 paddd mm2, mm5 ; mm2=tmp1L
346 paddd mm0, mm7 ; mm0=tmp1H
347 paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L
348 paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H
350 movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L
351 movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H
353 ; -- Final output stage
355 movq mm5, MMWORD [wk(0)] ; mm5=tmp10L
356 movq mm7, MMWORD [wk(1)] ; mm7=tmp10H
360 paddd mm5, mm3 ; mm5=data0L
361 paddd mm7, mm4 ; mm7=data0H
362 psubd mm2, mm3 ; mm2=data7L
363 psubd mm0, mm4 ; mm0=data7H
365 movq mm3, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm3=[PD_DESCALE_P1]
369 psrad mm5, DESCALE_P1
370 psrad mm7, DESCALE_P1
373 psrad mm2, DESCALE_P1
374 psrad mm0, DESCALE_P1
376 packssdw mm5, mm7 ; mm5=data0=(00 01 02 03)
377 packssdw mm2, mm0 ; mm2=data7=(70 71 72 73)
379 movq mm4, MMWORD [wk(4)] ; mm4=tmp11L
380 movq mm3, MMWORD [wk(5)] ; mm3=tmp11H
384 paddd mm4, mm1 ; mm4=data1L
385 paddd mm3, mm6 ; mm3=data1H
386 psubd mm7, mm1 ; mm7=data6L
387 psubd mm0, mm6 ; mm0=data6H
389 movq mm1, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm1=[PD_DESCALE_P1]
393 psrad mm4, DESCALE_P1
394 psrad mm3, DESCALE_P1
397 psrad mm7, DESCALE_P1
398 psrad mm0, DESCALE_P1
400 packssdw mm4, mm3 ; mm4=data1=(10 11 12 13)
401 packssdw mm7, mm0 ; mm7=data6=(60 61 62 63)
403 movq mm6, mm5 ; transpose coefficients(phase 1)
404 punpcklwd mm5, mm4 ; mm5=(00 10 01 11)
405 punpckhwd mm6, mm4 ; mm6=(02 12 03 13)
406 movq mm1, mm7 ; transpose coefficients(phase 1)
407 punpcklwd mm7, mm2 ; mm7=(60 70 61 71)
408 punpckhwd mm1, mm2 ; mm1=(62 72 63 73)
410 movq mm3, MMWORD [wk(6)] ; mm3=tmp12L
411 movq mm0, MMWORD [wk(7)] ; mm0=tmp12H
412 movq mm4, MMWORD [wk(10)] ; mm4=tmp1L
413 movq mm2, MMWORD [wk(11)] ; mm2=tmp1H
415 movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 01 11)
416 movq MMWORD [wk(1)], mm6 ; wk(1)=(02 12 03 13)
417 movq MMWORD [wk(4)], mm7 ; wk(4)=(60 70 61 71)
418 movq MMWORD [wk(5)], mm1 ; wk(5)=(62 72 63 73)
422 paddd mm3, mm4 ; mm3=data2L
423 paddd mm0, mm2 ; mm0=data2H
424 psubd mm5, mm4 ; mm5=data5L
425 psubd mm6, mm2 ; mm6=data5H
427 movq mm7, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm7=[PD_DESCALE_P1]
431 psrad mm3, DESCALE_P1
432 psrad mm0, DESCALE_P1
435 psrad mm5, DESCALE_P1
436 psrad mm6, DESCALE_P1
438 packssdw mm3, mm0 ; mm3=data2=(20 21 22 23)
439 packssdw mm5, mm6 ; mm5=data5=(50 51 52 53)
441 movq mm1, MMWORD [wk(2)] ; mm1=tmp13L
442 movq mm4, MMWORD [wk(3)] ; mm4=tmp13H
443 movq mm2, MMWORD [wk(8)] ; mm2=tmp0L
444 movq mm7, MMWORD [wk(9)] ; mm7=tmp0H
448 paddd mm1, mm2 ; mm1=data3L
449 paddd mm4, mm7 ; mm4=data3H
450 psubd mm0, mm2 ; mm0=data4L
451 psubd mm6, mm7 ; mm6=data4H
453 movq mm2, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm2=[PD_DESCALE_P1]
457 psrad mm1, DESCALE_P1
458 psrad mm4, DESCALE_P1
461 psrad mm0, DESCALE_P1
462 psrad mm6, DESCALE_P1
464 packssdw mm1, mm4 ; mm1=data3=(30 31 32 33)
465 packssdw mm0, mm6 ; mm0=data4=(40 41 42 43)
467 movq mm7, MMWORD [wk(0)] ; mm7=(00 10 01 11)
468 movq mm2, MMWORD [wk(1)] ; mm2=(02 12 03 13)
470 movq mm4, mm3 ; transpose coefficients(phase 1)
471 punpcklwd mm3, mm1 ; mm3=(20 30 21 31)
472 punpckhwd mm4, mm1 ; mm4=(22 32 23 33)
473 movq mm6, mm0 ; transpose coefficients(phase 1)
474 punpcklwd mm0, mm5 ; mm0=(40 50 41 51)
475 punpckhwd mm6, mm5 ; mm6=(42 52 43 53)
477 movq mm1, mm7 ; transpose coefficients(phase 2)
478 punpckldq mm7, mm3 ; mm7=(00 10 20 30)
479 punpckhdq mm1, mm3 ; mm1=(01 11 21 31)
480 movq mm5, mm2 ; transpose coefficients(phase 2)
481 punpckldq mm2, mm4 ; mm2=(02 12 22 32)
482 punpckhdq mm5, mm4 ; mm5=(03 13 23 33)
484 movq mm3, MMWORD [wk(4)] ; mm3=(60 70 61 71)
485 movq mm4, MMWORD [wk(5)] ; mm4=(62 72 63 73)
487 movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm7
488 movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1
489 movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2
490 movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm5
492 movq mm7, mm0 ; transpose coefficients(phase 2)
493 punpckldq mm0, mm3 ; mm0=(40 50 60 70)
494 punpckhdq mm7, mm3 ; mm7=(41 51 61 71)
495 movq mm1, mm6 ; transpose coefficients(phase 2)
496 punpckldq mm6, mm4 ; mm6=(42 52 62 72)
497 punpckhdq mm1, mm4 ; mm1=(43 53 63 73)
499 movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0
500 movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm7
501 movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm6
502 movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm1
505 add esi, byte 4*SIZEOF_JCOEF ; coef_block
506 add edx, byte 4*SIZEOF_ISLOW_MULT_TYPE ; quantptr
507 add edi, byte 4*DCTSIZE*SIZEOF_JCOEF ; wsptr
511 ; ---- Pass 2: process rows from work array, store into output array.
513 mov eax, [original_ebp]
514 lea esi, [workspace] ; JCOEF *wsptr
515 mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
516 mov eax, JDIMENSION [output_col(eax)]
517 mov ecx, DCTSIZE/4 ; ctr
523 movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
524 movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
525 movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
526 movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
529 ; z1 = (z2 + z3) * 0.541196100;
530 ; tmp2 = z1 + z3 * -1.847759065;
531 ; tmp3 = z1 + z2 * 0.765366865;
533 ; (This implementation)
534 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
535 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
537 movq mm4, mm1 ; mm1=in2=z2
539 punpcklwd mm4, mm3 ; mm3=in6=z3
543 pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L
544 pmaddwd mm5, [GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H
545 pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L
546 pmaddwd mm3, [GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H
549 paddw mm0, mm2 ; mm0=in0+in4
550 psubw mm6, mm2 ; mm6=in0-in4
554 punpcklwd mm7, mm0 ; mm7=tmp0L
555 punpckhwd mm2, mm0 ; mm2=tmp0H
556 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
557 psrad mm2, (16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS
560 paddd mm7, mm4 ; mm7=tmp10L
561 psubd mm0, mm4 ; mm0=tmp13L
563 paddd mm2, mm5 ; mm2=tmp10H
564 psubd mm4, mm5 ; mm4=tmp13H
566 movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L
567 movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H
568 movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L
569 movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H
573 punpcklwd mm5, mm6 ; mm5=tmp1L
574 punpckhwd mm7, mm6 ; mm7=tmp1H
575 psrad mm5, (16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS
576 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
579 paddd mm5, mm1 ; mm5=tmp11L
580 psubd mm2, mm1 ; mm2=tmp12L
582 paddd mm7, mm3 ; mm7=tmp11H
583 psubd mm0, mm3 ; mm0=tmp12H
585 movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L
586 movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H
587 movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L
588 movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H
592 movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
593 movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
594 movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
595 movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
599 paddw mm5, mm3 ; mm5=z3
600 paddw mm7, mm1 ; mm7=z4
603 ; z5 = (z3 + z4) * 1.175875602;
604 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
605 ; z3 += z5; z4 += z5;
607 ; (This implementation)
608 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
609 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
617 pmaddwd mm2, [GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L
618 pmaddwd mm0, [GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H
619 pmaddwd mm5, [GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L
620 pmaddwd mm7, [GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H
622 movq MMWORD [wk(10)], mm2 ; wk(10)=z3L
623 movq MMWORD [wk(11)], mm0 ; wk(11)=z3H
626 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
627 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
628 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
629 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
630 ; tmp0 += z1 + z3; tmp1 += z2 + z4;
631 ; tmp2 += z2 + z3; tmp3 += z1 + z4;
633 ; (This implementation)
634 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
635 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
636 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
637 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
638 ; tmp0 += z3; tmp1 += z4;
639 ; tmp2 += z3; tmp3 += z4;
647 pmaddwd mm2, [GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L
648 pmaddwd mm0, [GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H
649 pmaddwd mm3, [GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L
650 pmaddwd mm4, [GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H
652 paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L
653 paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H
654 paddd mm3, mm5 ; mm3=tmp3L
655 paddd mm4, mm7 ; mm4=tmp3H
657 movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L
658 movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H
666 pmaddwd mm2, [GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L
667 pmaddwd mm0, [GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H
668 pmaddwd mm1, [GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L
669 pmaddwd mm6, [GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H
671 paddd mm2, mm5 ; mm2=tmp1L
672 paddd mm0, mm7 ; mm0=tmp1H
673 paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L
674 paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H
676 movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L
677 movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H
679 ; -- Final output stage
681 movq mm5, MMWORD [wk(0)] ; mm5=tmp10L
682 movq mm7, MMWORD [wk(1)] ; mm7=tmp10H
686 paddd mm5, mm3 ; mm5=data0L
687 paddd mm7, mm4 ; mm7=data0H
688 psubd mm2, mm3 ; mm2=data7L
689 psubd mm0, mm4 ; mm0=data7H
691 movq mm3, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm3=[PD_DESCALE_P2]
695 psrad mm5, DESCALE_P2
696 psrad mm7, DESCALE_P2
699 psrad mm2, DESCALE_P2
700 psrad mm0, DESCALE_P2
702 packssdw mm5, mm7 ; mm5=data0=(00 10 20 30)
703 packssdw mm2, mm0 ; mm2=data7=(07 17 27 37)
705 movq mm4, MMWORD [wk(4)] ; mm4=tmp11L
706 movq mm3, MMWORD [wk(5)] ; mm3=tmp11H
710 paddd mm4, mm1 ; mm4=data1L
711 paddd mm3, mm6 ; mm3=data1H
712 psubd mm7, mm1 ; mm7=data6L
713 psubd mm0, mm6 ; mm0=data6H
715 movq mm1, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm1=[PD_DESCALE_P2]
719 psrad mm4, DESCALE_P2
720 psrad mm3, DESCALE_P2
723 psrad mm7, DESCALE_P2
724 psrad mm0, DESCALE_P2
726 packssdw mm4, mm3 ; mm4=data1=(01 11 21 31)
727 packssdw mm7, mm0 ; mm7=data6=(06 16 26 36)
729 packsswb mm5, mm7 ; mm5=(00 10 20 30 06 16 26 36)
730 packsswb mm4, mm2 ; mm4=(01 11 21 31 07 17 27 37)
732 movq mm6, MMWORD [wk(6)] ; mm6=tmp12L
733 movq mm1, MMWORD [wk(7)] ; mm1=tmp12H
734 movq mm3, MMWORD [wk(10)] ; mm3=tmp1L
735 movq mm0, MMWORD [wk(11)] ; mm0=tmp1H
737 movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 20 30 06 16 26 36)
738 movq MMWORD [wk(1)], mm4 ; wk(1)=(01 11 21 31 07 17 27 37)
742 paddd mm6, mm3 ; mm6=data2L
743 paddd mm1, mm0 ; mm1=data2H
744 psubd mm7, mm3 ; mm7=data5L
745 psubd mm2, mm0 ; mm2=data5H
747 movq mm5, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm5=[PD_DESCALE_P2]
751 psrad mm6, DESCALE_P2
752 psrad mm1, DESCALE_P2
755 psrad mm7, DESCALE_P2
756 psrad mm2, DESCALE_P2
758 packssdw mm6, mm1 ; mm6=data2=(02 12 22 32)
759 packssdw mm7, mm2 ; mm7=data5=(05 15 25 35)
761 movq mm4, MMWORD [wk(2)] ; mm4=tmp13L
762 movq mm3, MMWORD [wk(3)] ; mm3=tmp13H
763 movq mm0, MMWORD [wk(8)] ; mm0=tmp0L
764 movq mm5, MMWORD [wk(9)] ; mm5=tmp0H
768 paddd mm4, mm0 ; mm4=data3L
769 paddd mm3, mm5 ; mm3=data3H
770 psubd mm1, mm0 ; mm1=data4L
771 psubd mm2, mm5 ; mm2=data4H
773 movq mm0, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm0=[PD_DESCALE_P2]
777 psrad mm4, DESCALE_P2
778 psrad mm3, DESCALE_P2
781 psrad mm1, DESCALE_P2
782 psrad mm2, DESCALE_P2
784 movq mm5, [GOTOFF(ebx,PB_CENTERJSAMP)] ; mm5=[PB_CENTERJSAMP]
786 packssdw mm4, mm3 ; mm4=data3=(03 13 23 33)
787 packssdw mm1, mm2 ; mm1=data4=(04 14 24 34)
789 movq mm0, MMWORD [wk(0)] ; mm0=(00 10 20 30 06 16 26 36)
790 movq mm3, MMWORD [wk(1)] ; mm3=(01 11 21 31 07 17 27 37)
792 packsswb mm6, mm1 ; mm6=(02 12 22 32 04 14 24 34)
793 packsswb mm4, mm7 ; mm4=(03 13 23 33 05 15 25 35)
800 movq mm2, mm0 ; transpose coefficients(phase 1)
801 punpcklbw mm0, mm3 ; mm0=(00 01 10 11 20 21 30 31)
802 punpckhbw mm2, mm3 ; mm2=(06 07 16 17 26 27 36 37)
803 movq mm1, mm6 ; transpose coefficients(phase 1)
804 punpcklbw mm6, mm4 ; mm6=(02 03 12 13 22 23 32 33)
805 punpckhbw mm1, mm4 ; mm1=(04 05 14 15 24 25 34 35)
807 movq mm7, mm0 ; transpose coefficients(phase 2)
808 punpcklwd mm0, mm6 ; mm0=(00 01 02 03 10 11 12 13)
809 punpckhwd mm7, mm6 ; mm7=(20 21 22 23 30 31 32 33)
810 movq mm5, mm1 ; transpose coefficients(phase 2)
811 punpcklwd mm1, mm2 ; mm1=(04 05 06 07 14 15 16 17)
812 punpckhwd mm5, mm2 ; mm5=(24 25 26 27 34 35 36 37)
814 movq mm3, mm0 ; transpose coefficients(phase 3)
815 punpckldq mm0, mm1 ; mm0=(00 01 02 03 04 05 06 07)
816 punpckhdq mm3, mm1 ; mm3=(10 11 12 13 14 15 16 17)
817 movq mm4, mm7 ; transpose coefficients(phase 3)
818 punpckldq mm7, mm5 ; mm7=(20 21 22 23 24 25 26 27)
819 punpckhdq mm4, mm5 ; mm4=(30 31 32 33 34 35 36 37)
821 pushpic ebx ; save GOT address
823 mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
824 mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
825 movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm0
826 movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm3
827 mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
828 mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
829 movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm7
830 movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm4
832 poppic ebx ; restore GOT address
834 add esi, byte 4*SIZEOF_JCOEF ; wsptr
835 add edi, byte 4*SIZEOF_JSAMPROW
839 emms ; empty MMX state
843 ; pop edx ; need not be preserved
844 ; pop ecx ; need not be preserved
846 mov esp, ebp ; esp <- aligned ebp
847 pop esp ; esp <- original ebp
851 ; For some reason, the OS X linker does not honor the request to align the
852 ; segment unless we do this.