Gitiles
Code Review Sign In
review.blissroms.org / platform_external_libjpeg-turbo / 8683f0eff7e805b2431be011f18535c2ff0f0344 / . / simd / i386 / jidctint-mmx.asm
blob: 4f07f567f55f5427b9e1603b409639fd45843dd1 [file] [log] [blame]
Leon Scroggins III3993b372018-07-16 10:43:45 -0400[diff] [blame]1;
2; jidctint.asm - accurate integer IDCT (MMX)
3;
4; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
5; Copyright (C) 2016, D. R. Commander.
6;
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
10;
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
16;
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
20; more details.
Leon Scroggins III3993b372018-07-16 10:43:45 -0400[diff] [blame]21
22%include "jsimdext.inc"
23%include "jdct.inc"
24
25; --------------------------------------------------------------------------
26
27%define CONST_BITS 13
28%define PASS1_BITS 2
29
30%define DESCALE_P1 (CONST_BITS - PASS1_BITS)
31%define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3)
32
33%if CONST_BITS == 13
34F_0_298 equ 2446 ; FIX(0.298631336)
35F_0_390 equ 3196 ; FIX(0.390180644)
36F_0_541 equ 4433 ; FIX(0.541196100)
37F_0_765 equ 6270 ; FIX(0.765366865)
38F_0_899 equ 7373 ; FIX(0.899976223)
39F_1_175 equ 9633 ; FIX(1.175875602)
40F_1_501 equ 12299 ; FIX(1.501321110)
41F_1_847 equ 15137 ; FIX(1.847759065)
42F_1_961 equ 16069 ; FIX(1.961570560)
43F_2_053 equ 16819 ; FIX(2.053119869)
44F_2_562 equ 20995 ; FIX(2.562915447)
45F_3_072 equ 25172 ; FIX(3.072711026)
46%else
47; NASM cannot do compile-time arithmetic on floating-point constants.
48%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n))
49F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336)
50F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644)
51F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100)
52F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865)
53F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223)
54F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602)
55F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110)
56F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065)
57F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560)
58F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869)
59F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447)
60F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026)
61%endif
62
63; --------------------------------------------------------------------------
64 SECTION SEG_CONST
65
66 alignz 32
67 GLOBAL_DATA(jconst_idct_islow_mmx)
68
69EXTN(jconst_idct_islow_mmx):
70
71PW_F130_F054 times 2 dw (F_0_541 + F_0_765), F_0_541
72PW_F054_MF130 times 2 dw F_0_541, (F_0_541 - F_1_847)
73PW_MF078_F117 times 2 dw (F_1_175 - F_1_961), F_1_175
74PW_F117_F078 times 2 dw F_1_175, (F_1_175 - F_0_390)
75PW_MF060_MF089 times 2 dw (F_0_298 - F_0_899), -F_0_899
76PW_MF089_F060 times 2 dw -F_0_899, (F_1_501 - F_0_899)
77PW_MF050_MF256 times 2 dw (F_2_053 - F_2_562), -F_2_562
78PW_MF256_F050 times 2 dw -F_2_562, (F_3_072 - F_2_562)
79PD_DESCALE_P1 times 2 dd 1 << (DESCALE_P1 - 1)
80PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2 - 1)
81PB_CENTERJSAMP times 8 db CENTERJSAMPLE
82
83 alignz 32
84
85; --------------------------------------------------------------------------
86 SECTION SEG_TEXT
87 BITS 32
88;
89; Perform dequantization and inverse DCT on one block of coefficients.
90;
91; GLOBAL(void)
92; jsimd_idct_islow_mmx(void *dct_table, JCOEFPTR coef_block,
93; JSAMPARRAY output_buf, JDIMENSION output_col)
94;
95
96%define dct_table(b) (b) + 8 ; jpeg_component_info *compptr
97%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block
98%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf
99%define output_col(b) (b) + 20 ; JDIMENSION output_col
100
101%define original_ebp ebp + 0
102%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD
103 ; mmword wk[WK_NUM]
104%define WK_NUM 12
105%define workspace wk(0) - DCTSIZE2 * SIZEOF_JCOEF
106 ; JCOEF workspace[DCTSIZE2]
107
108 align 32
109 GLOBAL_FUNCTION(jsimd_idct_islow_mmx)
110
111EXTN(jsimd_idct_islow_mmx):
112 push ebp
113 mov eax, esp ; eax = original ebp
114 sub esp, byte 4
115 and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits
116 mov [esp], eax
117 mov ebp, esp ; ebp = aligned ebp
118 lea esp, [workspace]
119 push ebx
120; push ecx ; need not be preserved
121; push edx ; need not be preserved
122 push esi
123 push edi
124
125 get_GOT ebx ; get GOT address
126
127 ; ---- Pass 1: process columns from input, store into work array.
128
129; mov eax, [original_ebp]
130 mov edx, POINTER [dct_table(eax)] ; quantptr
131 mov esi, JCOEFPTR [coef_block(eax)] ; inptr
132 lea edi, [workspace] ; JCOEF *wsptr
133 mov ecx, DCTSIZE/4 ; ctr
134 alignx 16, 7
135.columnloop:
136%ifndef NO_ZERO_COLUMN_TEST_ISLOW_MMX
Jake Weinstein8683f0e2020-08-21 18:48:40 +0200[diff] [blame^]137 mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
138 or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
Leon Scroggins III3993b372018-07-16 10:43:45 -0400[diff] [blame]139 jnz short .columnDCT
140
141 movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
142 movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
143 por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
144 por mm1, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
145 por mm0, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
146 por mm1, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
147 por mm0, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
148 por mm1, mm0
149 packsswb mm1, mm1
150 movd eax, mm1
151 test eax, eax
152 jnz short .columnDCT
153
154 ; -- AC terms all zero
155
156 movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
157 pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
158
159 psllw mm0, PASS1_BITS
160
161 movq mm2, mm0 ; mm0=in0=(00 01 02 03)
162 punpcklwd mm0, mm0 ; mm0=(00 00 01 01)
163 punpckhwd mm2, mm2 ; mm2=(02 02 03 03)
164
165 movq mm1, mm0
166 punpckldq mm0, mm0 ; mm0=(00 00 00 00)
167 punpckhdq mm1, mm1 ; mm1=(01 01 01 01)
168 movq mm3, mm2
169 punpckldq mm2, mm2 ; mm2=(02 02 02 02)
170 punpckhdq mm3, mm3 ; mm3=(03 03 03 03)
171
172 movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0
173 movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0
174 movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1
175 movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm1
176 movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2
177 movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm2
178 movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm3
179 movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm3
180 jmp near .nextcolumn
181 alignx 16, 7
182%endif
183.columnDCT:
184
185 ; -- Even part
186
187 movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
188 movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
189 pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
190 pmullw mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
191 movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
192 movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
193 pmullw mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
194 pmullw mm3, MMWORD [MMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
195
196 ; (Original)
197 ; z1 = (z2 + z3) * 0.541196100;
198 ; tmp2 = z1 + z3 * -1.847759065;
199 ; tmp3 = z1 + z2 * 0.765366865;
200 ;
201 ; (This implementation)
202 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
203 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
204
205 movq mm4, mm1 ; mm1=in2=z2
206 movq mm5, mm1
207 punpcklwd mm4, mm3 ; mm3=in6=z3
208 punpckhwd mm5, mm3
209 movq mm1, mm4
210 movq mm3, mm5
211 pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L
212 pmaddwd mm5, [GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H
213 pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L
214 pmaddwd mm3, [GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H
215
216 movq mm6, mm0
217 paddw mm0, mm2 ; mm0=in0+in4
218 psubw mm6, mm2 ; mm6=in0-in4
219
220 pxor mm7, mm7
221 pxor mm2, mm2
222 punpcklwd mm7, mm0 ; mm7=tmp0L
223 punpckhwd mm2, mm0 ; mm2=tmp0H
224 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
225 psrad mm2, (16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS
226
227 movq mm0, mm7
228 paddd mm7, mm4 ; mm7=tmp10L
229 psubd mm0, mm4 ; mm0=tmp13L
230 movq mm4, mm2
231 paddd mm2, mm5 ; mm2=tmp10H
232 psubd mm4, mm5 ; mm4=tmp13H
233
234 movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L
235 movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H
236 movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L
237 movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H
238
239 pxor mm5, mm5
240 pxor mm7, mm7
241 punpcklwd mm5, mm6 ; mm5=tmp1L
242 punpckhwd mm7, mm6 ; mm7=tmp1H
243 psrad mm5, (16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS
244 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
245
246 movq mm2, mm5
247 paddd mm5, mm1 ; mm5=tmp11L
248 psubd mm2, mm1 ; mm2=tmp12L
249 movq mm0, mm7
250 paddd mm7, mm3 ; mm7=tmp11H
251 psubd mm0, mm3 ; mm0=tmp12H
252
253 movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L
254 movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H
255 movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L
256 movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H
257
258 ; -- Odd part
259
260 movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
261 movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
262 pmullw mm4, MMWORD [MMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
263 pmullw mm6, MMWORD [MMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
264 movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
265 movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
266 pmullw mm1, MMWORD [MMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
267 pmullw mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)]
268
269 movq mm5, mm6
270 movq mm7, mm4
271 paddw mm5, mm3 ; mm5=z3
272 paddw mm7, mm1 ; mm7=z4
273
274 ; (Original)
275 ; z5 = (z3 + z4) * 1.175875602;
276 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
277 ; z3 += z5; z4 += z5;
278 ;
279 ; (This implementation)
280 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
281 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
282
283 movq mm2, mm5
284 movq mm0, mm5
285 punpcklwd mm2, mm7
286 punpckhwd mm0, mm7
287 movq mm5, mm2
288 movq mm7, mm0
289 pmaddwd mm2, [GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L
290 pmaddwd mm0, [GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H
291 pmaddwd mm5, [GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L
292 pmaddwd mm7, [GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H
293
294 movq MMWORD [wk(10)], mm2 ; wk(10)=z3L
295 movq MMWORD [wk(11)], mm0 ; wk(11)=z3H
296
297 ; (Original)
298 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
299 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
300 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
301 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
302 ; tmp0 += z1 + z3; tmp1 += z2 + z4;
303 ; tmp2 += z2 + z3; tmp3 += z1 + z4;
304 ;
305 ; (This implementation)
306 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
307 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
308 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
309 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
310 ; tmp0 += z3; tmp1 += z4;
311 ; tmp2 += z3; tmp3 += z4;
312
313 movq mm2, mm3
314 movq mm0, mm3
315 punpcklwd mm2, mm4
316 punpckhwd mm0, mm4
317 movq mm3, mm2
318 movq mm4, mm0
319 pmaddwd mm2, [GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L
320 pmaddwd mm0, [GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H
321 pmaddwd mm3, [GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L
322 pmaddwd mm4, [GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H
323
324 paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L
325 paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H
326 paddd mm3, mm5 ; mm3=tmp3L
327 paddd mm4, mm7 ; mm4=tmp3H
328
329 movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L
330 movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H
331
332 movq mm2, mm1
333 movq mm0, mm1
334 punpcklwd mm2, mm6
335 punpckhwd mm0, mm6
336 movq mm1, mm2
337 movq mm6, mm0
338 pmaddwd mm2, [GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L
339 pmaddwd mm0, [GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H
340 pmaddwd mm1, [GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L
341 pmaddwd mm6, [GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H
342
343 paddd mm2, mm5 ; mm2=tmp1L
344 paddd mm0, mm7 ; mm0=tmp1H
345 paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L
346 paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H
347
348 movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L
349 movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H
350
351 ; -- Final output stage
352
353 movq mm5, MMWORD [wk(0)] ; mm5=tmp10L
354 movq mm7, MMWORD [wk(1)] ; mm7=tmp10H
355
356 movq mm2, mm5
357 movq mm0, mm7
358 paddd mm5, mm3 ; mm5=data0L
359 paddd mm7, mm4 ; mm7=data0H
360 psubd mm2, mm3 ; mm2=data7L
361 psubd mm0, mm4 ; mm0=data7H
362
363 movq mm3, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm3=[PD_DESCALE_P1]
364
365 paddd mm5, mm3
366 paddd mm7, mm3
367 psrad mm5, DESCALE_P1
368 psrad mm7, DESCALE_P1
369 paddd mm2, mm3
370 paddd mm0, mm3
371 psrad mm2, DESCALE_P1
372 psrad mm0, DESCALE_P1
373
374 packssdw mm5, mm7 ; mm5=data0=(00 01 02 03)
375 packssdw mm2, mm0 ; mm2=data7=(70 71 72 73)
376
377 movq mm4, MMWORD [wk(4)] ; mm4=tmp11L
378 movq mm3, MMWORD [wk(5)] ; mm3=tmp11H
379
380 movq mm7, mm4
381 movq mm0, mm3
382 paddd mm4, mm1 ; mm4=data1L
383 paddd mm3, mm6 ; mm3=data1H
384 psubd mm7, mm1 ; mm7=data6L
385 psubd mm0, mm6 ; mm0=data6H
386
387 movq mm1, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm1=[PD_DESCALE_P1]
388
389 paddd mm4, mm1
390 paddd mm3, mm1
391 psrad mm4, DESCALE_P1
392 psrad mm3, DESCALE_P1
393 paddd mm7, mm1
394 paddd mm0, mm1
395 psrad mm7, DESCALE_P1
396 psrad mm0, DESCALE_P1
397
398 packssdw mm4, mm3 ; mm4=data1=(10 11 12 13)
399 packssdw mm7, mm0 ; mm7=data6=(60 61 62 63)
400
401 movq mm6, mm5 ; transpose coefficients(phase 1)
402 punpcklwd mm5, mm4 ; mm5=(00 10 01 11)
403 punpckhwd mm6, mm4 ; mm6=(02 12 03 13)
404 movq mm1, mm7 ; transpose coefficients(phase 1)
405 punpcklwd mm7, mm2 ; mm7=(60 70 61 71)
406 punpckhwd mm1, mm2 ; mm1=(62 72 63 73)
407
408 movq mm3, MMWORD [wk(6)] ; mm3=tmp12L
409 movq mm0, MMWORD [wk(7)] ; mm0=tmp12H
410 movq mm4, MMWORD [wk(10)] ; mm4=tmp1L
411 movq mm2, MMWORD [wk(11)] ; mm2=tmp1H
412
413 movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 01 11)
414 movq MMWORD [wk(1)], mm6 ; wk(1)=(02 12 03 13)
415 movq MMWORD [wk(4)], mm7 ; wk(4)=(60 70 61 71)
416 movq MMWORD [wk(5)], mm1 ; wk(5)=(62 72 63 73)
417
418 movq mm5, mm3
419 movq mm6, mm0
420 paddd mm3, mm4 ; mm3=data2L
421 paddd mm0, mm2 ; mm0=data2H
422 psubd mm5, mm4 ; mm5=data5L
423 psubd mm6, mm2 ; mm6=data5H
424
425 movq mm7, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm7=[PD_DESCALE_P1]
426
427 paddd mm3, mm7
428 paddd mm0, mm7
429 psrad mm3, DESCALE_P1
430 psrad mm0, DESCALE_P1
431 paddd mm5, mm7
432 paddd mm6, mm7
433 psrad mm5, DESCALE_P1
434 psrad mm6, DESCALE_P1
435
436 packssdw mm3, mm0 ; mm3=data2=(20 21 22 23)
437 packssdw mm5, mm6 ; mm5=data5=(50 51 52 53)
438
439 movq mm1, MMWORD [wk(2)] ; mm1=tmp13L
440 movq mm4, MMWORD [wk(3)] ; mm4=tmp13H
441 movq mm2, MMWORD [wk(8)] ; mm2=tmp0L
442 movq mm7, MMWORD [wk(9)] ; mm7=tmp0H
443
444 movq mm0, mm1
445 movq mm6, mm4
446 paddd mm1, mm2 ; mm1=data3L
447 paddd mm4, mm7 ; mm4=data3H
448 psubd mm0, mm2 ; mm0=data4L
449 psubd mm6, mm7 ; mm6=data4H
450
451 movq mm2, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm2=[PD_DESCALE_P1]
452
453 paddd mm1, mm2
454 paddd mm4, mm2
455 psrad mm1, DESCALE_P1
456 psrad mm4, DESCALE_P1
457 paddd mm0, mm2
458 paddd mm6, mm2
459 psrad mm0, DESCALE_P1
460 psrad mm6, DESCALE_P1
461
462 packssdw mm1, mm4 ; mm1=data3=(30 31 32 33)
463 packssdw mm0, mm6 ; mm0=data4=(40 41 42 43)
464
465 movq mm7, MMWORD [wk(0)] ; mm7=(00 10 01 11)
466 movq mm2, MMWORD [wk(1)] ; mm2=(02 12 03 13)
467
468 movq mm4, mm3 ; transpose coefficients(phase 1)
469 punpcklwd mm3, mm1 ; mm3=(20 30 21 31)
470 punpckhwd mm4, mm1 ; mm4=(22 32 23 33)
471 movq mm6, mm0 ; transpose coefficients(phase 1)
472 punpcklwd mm0, mm5 ; mm0=(40 50 41 51)
473 punpckhwd mm6, mm5 ; mm6=(42 52 43 53)
474
475 movq mm1, mm7 ; transpose coefficients(phase 2)
476 punpckldq mm7, mm3 ; mm7=(00 10 20 30)
477 punpckhdq mm1, mm3 ; mm1=(01 11 21 31)
478 movq mm5, mm2 ; transpose coefficients(phase 2)
479 punpckldq mm2, mm4 ; mm2=(02 12 22 32)
480 punpckhdq mm5, mm4 ; mm5=(03 13 23 33)
481
482 movq mm3, MMWORD [wk(4)] ; mm3=(60 70 61 71)
483 movq mm4, MMWORD [wk(5)] ; mm4=(62 72 63 73)
484
485 movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm7
486 movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1
487 movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2
488 movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm5
489
490 movq mm7, mm0 ; transpose coefficients(phase 2)
491 punpckldq mm0, mm3 ; mm0=(40 50 60 70)
492 punpckhdq mm7, mm3 ; mm7=(41 51 61 71)
493 movq mm1, mm6 ; transpose coefficients(phase 2)
494 punpckldq mm6, mm4 ; mm6=(42 52 62 72)
495 punpckhdq mm1, mm4 ; mm1=(43 53 63 73)
496
497 movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0
498 movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm7
499 movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm6
500 movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm1
501
502.nextcolumn:
503 add esi, byte 4*SIZEOF_JCOEF ; coef_block
504 add edx, byte 4*SIZEOF_ISLOW_MULT_TYPE ; quantptr
505 add edi, byte 4*DCTSIZE*SIZEOF_JCOEF ; wsptr
506 dec ecx ; ctr
507 jnz near .columnloop
508
509 ; ---- Pass 2: process rows from work array, store into output array.
510
511 mov eax, [original_ebp]
512 lea esi, [workspace] ; JCOEF *wsptr
513 mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
514 mov eax, JDIMENSION [output_col(eax)]
515 mov ecx, DCTSIZE/4 ; ctr
516 alignx 16, 7
517.rowloop:
518
519 ; -- Even part
520
521 movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)]
522 movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)]
523 movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)]
524 movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)]
525
526 ; (Original)
527 ; z1 = (z2 + z3) * 0.541196100;
528 ; tmp2 = z1 + z3 * -1.847759065;
529 ; tmp3 = z1 + z2 * 0.765366865;
530 ;
531 ; (This implementation)
532 ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065);
533 ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100;
534
535 movq mm4, mm1 ; mm1=in2=z2
536 movq mm5, mm1
537 punpcklwd mm4, mm3 ; mm3=in6=z3
538 punpckhwd mm5, mm3
539 movq mm1, mm4
540 movq mm3, mm5
541 pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L
542 pmaddwd mm5, [GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H
543 pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L
544 pmaddwd mm3, [GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H
545
546 movq mm6, mm0
547 paddw mm0, mm2 ; mm0=in0+in4
548 psubw mm6, mm2 ; mm6=in0-in4
549
550 pxor mm7, mm7
551 pxor mm2, mm2
552 punpcklwd mm7, mm0 ; mm7=tmp0L
553 punpckhwd mm2, mm0 ; mm2=tmp0H
554 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
555 psrad mm2, (16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS
556
557 movq mm0, mm7
558 paddd mm7, mm4 ; mm7=tmp10L
559 psubd mm0, mm4 ; mm0=tmp13L
560 movq mm4, mm2
561 paddd mm2, mm5 ; mm2=tmp10H
562 psubd mm4, mm5 ; mm4=tmp13H
563
564 movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L
565 movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H
566 movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L
567 movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H
568
569 pxor mm5, mm5
570 pxor mm7, mm7
571 punpcklwd mm5, mm6 ; mm5=tmp1L
572 punpckhwd mm7, mm6 ; mm7=tmp1H
573 psrad mm5, (16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS
574 psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS
575
576 movq mm2, mm5
577 paddd mm5, mm1 ; mm5=tmp11L
578 psubd mm2, mm1 ; mm2=tmp12L
579 movq mm0, mm7
580 paddd mm7, mm3 ; mm7=tmp11H
581 psubd mm0, mm3 ; mm0=tmp12H
582
583 movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L
584 movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H
585 movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L
586 movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H
587
588 ; -- Odd part
589
590 movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)]
591 movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)]
592 movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)]
593 movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)]
594
595 movq mm5, mm6
596 movq mm7, mm4
597 paddw mm5, mm3 ; mm5=z3
598 paddw mm7, mm1 ; mm7=z4
599
600 ; (Original)
601 ; z5 = (z3 + z4) * 1.175875602;
602 ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644;
603 ; z3 += z5; z4 += z5;
604 ;
605 ; (This implementation)
606 ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602;
607 ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644);
608
609 movq mm2, mm5
610 movq mm0, mm5
611 punpcklwd mm2, mm7
612 punpckhwd mm0, mm7
613 movq mm5, mm2
614 movq mm7, mm0
615 pmaddwd mm2, [GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L
616 pmaddwd mm0, [GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H
617 pmaddwd mm5, [GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L
618 pmaddwd mm7, [GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H
619
620 movq MMWORD [wk(10)], mm2 ; wk(10)=z3L
621 movq MMWORD [wk(11)], mm0 ; wk(11)=z3H
622
623 ; (Original)
624 ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2;
625 ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869;
626 ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110;
627 ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447;
628 ; tmp0 += z1 + z3; tmp1 += z2 + z4;
629 ; tmp2 += z2 + z3; tmp3 += z1 + z4;
630 ;
631 ; (This implementation)
632 ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223;
633 ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447;
634 ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447);
635 ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223);
636 ; tmp0 += z3; tmp1 += z4;
637 ; tmp2 += z3; tmp3 += z4;
638
639 movq mm2, mm3
640 movq mm0, mm3
641 punpcklwd mm2, mm4
642 punpckhwd mm0, mm4
643 movq mm3, mm2
644 movq mm4, mm0
645 pmaddwd mm2, [GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L
646 pmaddwd mm0, [GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H
647 pmaddwd mm3, [GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L
648 pmaddwd mm4, [GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H
649
650 paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L
651 paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H
652 paddd mm3, mm5 ; mm3=tmp3L
653 paddd mm4, mm7 ; mm4=tmp3H
654
655 movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L
656 movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H
657
658 movq mm2, mm1
659 movq mm0, mm1
660 punpcklwd mm2, mm6
661 punpckhwd mm0, mm6
662 movq mm1, mm2
663 movq mm6, mm0
664 pmaddwd mm2, [GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L
665 pmaddwd mm0, [GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H
666 pmaddwd mm1, [GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L
667 pmaddwd mm6, [GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H
668
669 paddd mm2, mm5 ; mm2=tmp1L
670 paddd mm0, mm7 ; mm0=tmp1H
671 paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L
672 paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H
673
674 movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L
675 movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H
676
677 ; -- Final output stage
678
679 movq mm5, MMWORD [wk(0)] ; mm5=tmp10L
680 movq mm7, MMWORD [wk(1)] ; mm7=tmp10H
681
682 movq mm2, mm5
683 movq mm0, mm7
684 paddd mm5, mm3 ; mm5=data0L
685 paddd mm7, mm4 ; mm7=data0H
686 psubd mm2, mm3 ; mm2=data7L
687 psubd mm0, mm4 ; mm0=data7H
688
689 movq mm3, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm3=[PD_DESCALE_P2]
690
691 paddd mm5, mm3
692 paddd mm7, mm3
693 psrad mm5, DESCALE_P2
694 psrad mm7, DESCALE_P2
695 paddd mm2, mm3
696 paddd mm0, mm3
697 psrad mm2, DESCALE_P2
698 psrad mm0, DESCALE_P2
699
700 packssdw mm5, mm7 ; mm5=data0=(00 10 20 30)
701 packssdw mm2, mm0 ; mm2=data7=(07 17 27 37)
702
703 movq mm4, MMWORD [wk(4)] ; mm4=tmp11L
704 movq mm3, MMWORD [wk(5)] ; mm3=tmp11H
705
706 movq mm7, mm4
707 movq mm0, mm3
708 paddd mm4, mm1 ; mm4=data1L
709 paddd mm3, mm6 ; mm3=data1H
710 psubd mm7, mm1 ; mm7=data6L
711 psubd mm0, mm6 ; mm0=data6H
712
713 movq mm1, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm1=[PD_DESCALE_P2]
714
715 paddd mm4, mm1
716 paddd mm3, mm1
717 psrad mm4, DESCALE_P2
718 psrad mm3, DESCALE_P2
719 paddd mm7, mm1
720 paddd mm0, mm1
721 psrad mm7, DESCALE_P2
722 psrad mm0, DESCALE_P2
723
724 packssdw mm4, mm3 ; mm4=data1=(01 11 21 31)
725 packssdw mm7, mm0 ; mm7=data6=(06 16 26 36)
726
727 packsswb mm5, mm7 ; mm5=(00 10 20 30 06 16 26 36)
728 packsswb mm4, mm2 ; mm4=(01 11 21 31 07 17 27 37)
729
730 movq mm6, MMWORD [wk(6)] ; mm6=tmp12L
731 movq mm1, MMWORD [wk(7)] ; mm1=tmp12H
732 movq mm3, MMWORD [wk(10)] ; mm3=tmp1L
733 movq mm0, MMWORD [wk(11)] ; mm0=tmp1H
734
735 movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 20 30 06 16 26 36)
736 movq MMWORD [wk(1)], mm4 ; wk(1)=(01 11 21 31 07 17 27 37)
737
738 movq mm7, mm6
739 movq mm2, mm1
740 paddd mm6, mm3 ; mm6=data2L
741 paddd mm1, mm0 ; mm1=data2H
742 psubd mm7, mm3 ; mm7=data5L
743 psubd mm2, mm0 ; mm2=data5H
744
745 movq mm5, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm5=[PD_DESCALE_P2]
746
747 paddd mm6, mm5
748 paddd mm1, mm5
749 psrad mm6, DESCALE_P2
750 psrad mm1, DESCALE_P2
751 paddd mm7, mm5
752 paddd mm2, mm5
753 psrad mm7, DESCALE_P2
754 psrad mm2, DESCALE_P2
755
756 packssdw mm6, mm1 ; mm6=data2=(02 12 22 32)
757 packssdw mm7, mm2 ; mm7=data5=(05 15 25 35)
758
759 movq mm4, MMWORD [wk(2)] ; mm4=tmp13L
760 movq mm3, MMWORD [wk(3)] ; mm3=tmp13H
761 movq mm0, MMWORD [wk(8)] ; mm0=tmp0L
762 movq mm5, MMWORD [wk(9)] ; mm5=tmp0H
763
764 movq mm1, mm4
765 movq mm2, mm3
766 paddd mm4, mm0 ; mm4=data3L
767 paddd mm3, mm5 ; mm3=data3H
768 psubd mm1, mm0 ; mm1=data4L
769 psubd mm2, mm5 ; mm2=data4H
770
771 movq mm0, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm0=[PD_DESCALE_P2]
772
773 paddd mm4, mm0
774 paddd mm3, mm0
775 psrad mm4, DESCALE_P2
776 psrad mm3, DESCALE_P2
777 paddd mm1, mm0
778 paddd mm2, mm0
779 psrad mm1, DESCALE_P2
780 psrad mm2, DESCALE_P2
781
782 movq mm5, [GOTOFF(ebx,PB_CENTERJSAMP)] ; mm5=[PB_CENTERJSAMP]
783
784 packssdw mm4, mm3 ; mm4=data3=(03 13 23 33)
785 packssdw mm1, mm2 ; mm1=data4=(04 14 24 34)
786
787 movq mm0, MMWORD [wk(0)] ; mm0=(00 10 20 30 06 16 26 36)
788 movq mm3, MMWORD [wk(1)] ; mm3=(01 11 21 31 07 17 27 37)
789
790 packsswb mm6, mm1 ; mm6=(02 12 22 32 04 14 24 34)
791 packsswb mm4, mm7 ; mm4=(03 13 23 33 05 15 25 35)
792
793 paddb mm0, mm5
794 paddb mm3, mm5
795 paddb mm6, mm5
796 paddb mm4, mm5
797
798 movq mm2, mm0 ; transpose coefficients(phase 1)
799 punpcklbw mm0, mm3 ; mm0=(00 01 10 11 20 21 30 31)
800 punpckhbw mm2, mm3 ; mm2=(06 07 16 17 26 27 36 37)
801 movq mm1, mm6 ; transpose coefficients(phase 1)
802 punpcklbw mm6, mm4 ; mm6=(02 03 12 13 22 23 32 33)
803 punpckhbw mm1, mm4 ; mm1=(04 05 14 15 24 25 34 35)
804
805 movq mm7, mm0 ; transpose coefficients(phase 2)
806 punpcklwd mm0, mm6 ; mm0=(00 01 02 03 10 11 12 13)
807 punpckhwd mm7, mm6 ; mm7=(20 21 22 23 30 31 32 33)
808 movq mm5, mm1 ; transpose coefficients(phase 2)
809 punpcklwd mm1, mm2 ; mm1=(04 05 06 07 14 15 16 17)
810 punpckhwd mm5, mm2 ; mm5=(24 25 26 27 34 35 36 37)
811
812 movq mm3, mm0 ; transpose coefficients(phase 3)
813 punpckldq mm0, mm1 ; mm0=(00 01 02 03 04 05 06 07)
814 punpckhdq mm3, mm1 ; mm3=(10 11 12 13 14 15 16 17)
815 movq mm4, mm7 ; transpose coefficients(phase 3)
816 punpckldq mm7, mm5 ; mm7=(20 21 22 23 24 25 26 27)
817 punpckhdq mm4, mm5 ; mm4=(30 31 32 33 34 35 36 37)
818
819 pushpic ebx ; save GOT address
820
821 mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
822 mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
823 movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm0
824 movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm3
825 mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW]
826 mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW]
827 movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm7
828 movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm4
829
830 poppic ebx ; restore GOT address
831
832 add esi, byte 4*SIZEOF_JCOEF ; wsptr
833 add edi, byte 4*SIZEOF_JSAMPROW
834 dec ecx ; ctr
835 jnz near .rowloop
836
837 emms ; empty MMX state
838
839 pop edi
840 pop esi
841; pop edx ; need not be preserved
842; pop ecx ; need not be preserved
843 pop ebx
844 mov esp, ebp ; esp <- aligned ebp
845 pop esp ; esp <- original ebp
846 pop ebp
847 ret
848
849; For some reason, the OS X linker does not honor the request to align the
850; segment unless we do this.
851 align 32