summaryrefslogtreecommitdiff
path: root/media/libjpeg/simd/jfdctflt-sse.asm
blob: e7ede26c0c12a0fa48600ca9a4922a635b8e567b (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
;
; jfdctflt.asm - floating-point FDCT (SSE)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
;
; Based on the x86 SIMD extension for IJG JPEG library
; Copyright (C) 1999-2006, MIYASAKA Masaru.
; For conditions of distribution and use, see copyright notice in jsimdext.inc
;
; This file should be assembled with NASM (Netwide Assembler),
; can *not* be assembled with Microsoft's MASM or any compatible
; assembler (including Borland's Turbo Assembler).
; NASM is available from http://nasm.sourceforge.net/ or
; http://sourceforge.net/project/showfiles.php?group_id=6208
;
; This file contains a floating-point implementation of the forward DCT
; (Discrete Cosine Transform). The following code is based directly on
; the IJG's original jfdctflt.c; see the jfdctflt.c for more details.
;
; [TAB8]

%include "jsimdext.inc"
%include "jdct.inc"

; --------------------------------------------------------------------------

%macro  unpcklps2 2     ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5)
        shufps  %1,%2,0x44
%endmacro

%macro  unpckhps2 2     ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7)
        shufps  %1,%2,0xEE
%endmacro

; --------------------------------------------------------------------------
        SECTION SEG_CONST

        alignz  16
        global  EXTN(jconst_fdct_float_sse)

EXTN(jconst_fdct_float_sse):

PD_0_382        times 4 dd  0.382683432365089771728460
PD_0_707        times 4 dd  0.707106781186547524400844
PD_0_541        times 4 dd  0.541196100146196984399723
PD_1_306        times 4 dd  1.306562964876376527856643

        alignz  16

; --------------------------------------------------------------------------
        SECTION SEG_TEXT
        BITS    32
;
; Perform the forward DCT on one block of samples.
;
; GLOBAL(void)
; jsimd_fdct_float_sse (FAST_FLOAT *data)
;

%define data(b)         (b)+8           ; FAST_FLOAT *data

%define original_ebp    ebp+0
%define wk(i)           ebp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
%define WK_NUM          2

        align   16
        global  EXTN(jsimd_fdct_float_sse)

EXTN(jsimd_fdct_float_sse):
        push    ebp
        mov     eax,esp                         ; eax = original ebp
        sub     esp, byte 4
        and     esp, byte (-SIZEOF_XMMWORD)     ; align to 128 bits
        mov     [esp],eax
        mov     ebp,esp                         ; ebp = aligned ebp
        lea     esp, [wk(0)]
        pushpic ebx
;       push    ecx             ; need not be preserved
;       push    edx             ; need not be preserved
;       push    esi             ; unused
;       push    edi             ; unused

        get_GOT ebx             ; get GOT address

        ; ---- Pass 1: process rows.

        mov     edx, POINTER [data(eax)]        ; (FAST_FLOAT *)
        mov     ecx, DCTSIZE/4
        alignx  16,7
.rowloop:

        movaps  xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm2, XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm3, XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)]

        ; xmm0=(20 21 22 23), xmm2=(24 25 26 27)
        ; xmm1=(30 31 32 33), xmm3=(34 35 36 37)

        movaps   xmm4,xmm0              ; transpose coefficients(phase 1)
        unpcklps xmm0,xmm1              ; xmm0=(20 30 21 31)
        unpckhps xmm4,xmm1              ; xmm4=(22 32 23 33)
        movaps   xmm5,xmm2              ; transpose coefficients(phase 1)
        unpcklps xmm2,xmm3              ; xmm2=(24 34 25 35)
        unpckhps xmm5,xmm3              ; xmm5=(26 36 27 37)

        movaps  xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)]

        ; xmm6=(00 01 02 03), xmm1=(04 05 06 07)
        ; xmm7=(10 11 12 13), xmm3=(14 15 16 17)

        movaps  XMMWORD [wk(0)], xmm4   ; wk(0)=(22 32 23 33)
        movaps  XMMWORD [wk(1)], xmm2   ; wk(1)=(24 34 25 35)

        movaps   xmm4,xmm6              ; transpose coefficients(phase 1)
        unpcklps xmm6,xmm7              ; xmm6=(00 10 01 11)
        unpckhps xmm4,xmm7              ; xmm4=(02 12 03 13)
        movaps   xmm2,xmm1              ; transpose coefficients(phase 1)
        unpcklps xmm1,xmm3              ; xmm1=(04 14 05 15)
        unpckhps xmm2,xmm3              ; xmm2=(06 16 07 17)

        movaps    xmm7,xmm6             ; transpose coefficients(phase 2)
        unpcklps2 xmm6,xmm0             ; xmm6=(00 10 20 30)=data0
        unpckhps2 xmm7,xmm0             ; xmm7=(01 11 21 31)=data1
        movaps    xmm3,xmm2             ; transpose coefficients(phase 2)
        unpcklps2 xmm2,xmm5             ; xmm2=(06 16 26 36)=data6
        unpckhps2 xmm3,xmm5             ; xmm3=(07 17 27 37)=data7

        movaps  xmm0,xmm7
        movaps  xmm5,xmm6
        subps   xmm7,xmm2               ; xmm7=data1-data6=tmp6
        subps   xmm6,xmm3               ; xmm6=data0-data7=tmp7
        addps   xmm0,xmm2               ; xmm0=data1+data6=tmp1
        addps   xmm5,xmm3               ; xmm5=data0+data7=tmp0

        movaps  xmm2, XMMWORD [wk(0)]   ; xmm2=(22 32 23 33)
        movaps  xmm3, XMMWORD [wk(1)]   ; xmm3=(24 34 25 35)
        movaps  XMMWORD [wk(0)], xmm7   ; wk(0)=tmp6
        movaps  XMMWORD [wk(1)], xmm6   ; wk(1)=tmp7

        movaps    xmm7,xmm4             ; transpose coefficients(phase 2)
        unpcklps2 xmm4,xmm2             ; xmm4=(02 12 22 32)=data2
        unpckhps2 xmm7,xmm2             ; xmm7=(03 13 23 33)=data3
        movaps    xmm6,xmm1             ; transpose coefficients(phase 2)
        unpcklps2 xmm1,xmm3             ; xmm1=(04 14 24 34)=data4
        unpckhps2 xmm6,xmm3             ; xmm6=(05 15 25 35)=data5

        movaps  xmm2,xmm7
        movaps  xmm3,xmm4
        addps   xmm7,xmm1               ; xmm7=data3+data4=tmp3
        addps   xmm4,xmm6               ; xmm4=data2+data5=tmp2
        subps   xmm2,xmm1               ; xmm2=data3-data4=tmp4
        subps   xmm3,xmm6               ; xmm3=data2-data5=tmp5

        ; -- Even part

        movaps  xmm1,xmm5
        movaps  xmm6,xmm0
        subps   xmm5,xmm7               ; xmm5=tmp13
        subps   xmm0,xmm4               ; xmm0=tmp12
        addps   xmm1,xmm7               ; xmm1=tmp10
        addps   xmm6,xmm4               ; xmm6=tmp11

        addps   xmm0,xmm5
        mulps   xmm0,[GOTOFF(ebx,PD_0_707)] ; xmm0=z1

        movaps  xmm7,xmm1
        movaps  xmm4,xmm5
        subps   xmm1,xmm6               ; xmm1=data4
        subps   xmm5,xmm0               ; xmm5=data6
        addps   xmm7,xmm6               ; xmm7=data0
        addps   xmm4,xmm0               ; xmm4=data2

        movaps  XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)], xmm1
        movaps  XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)], xmm5
        movaps  XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
        movaps  XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4

        ; -- Odd part

        movaps  xmm6, XMMWORD [wk(0)]   ; xmm6=tmp6
        movaps  xmm0, XMMWORD [wk(1)]   ; xmm0=tmp7

        addps   xmm2,xmm3               ; xmm2=tmp10
        addps   xmm3,xmm6               ; xmm3=tmp11
        addps   xmm6,xmm0               ; xmm6=tmp12, xmm0=tmp7

        mulps   xmm3,[GOTOFF(ebx,PD_0_707)] ; xmm3=z3

        movaps  xmm1,xmm2               ; xmm1=tmp10
        subps   xmm2,xmm6
        mulps   xmm2,[GOTOFF(ebx,PD_0_382)] ; xmm2=z5
        mulps   xmm1,[GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
        mulps   xmm6,[GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
        addps   xmm1,xmm2               ; xmm1=z2
        addps   xmm6,xmm2               ; xmm6=z4

        movaps  xmm5,xmm0
        subps   xmm0,xmm3               ; xmm0=z13
        addps   xmm5,xmm3               ; xmm5=z11

        movaps  xmm7,xmm0
        movaps  xmm4,xmm5
        subps   xmm0,xmm1               ; xmm0=data3
        subps   xmm5,xmm6               ; xmm5=data7
        addps   xmm7,xmm1               ; xmm7=data5
        addps   xmm4,xmm6               ; xmm4=data1

        movaps  XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
        movaps  XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)], xmm5
        movaps  XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)], xmm7
        movaps  XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4

        add     edx, 4*DCTSIZE*SIZEOF_FAST_FLOAT
        dec     ecx
        jnz     near .rowloop

        ; ---- Pass 2: process columns.

        mov     edx, POINTER [data(eax)]        ; (FAST_FLOAT *)
        mov     ecx, DCTSIZE/4
        alignx  16,7
.columnloop:

        movaps  xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm2, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)]

        ; xmm0=(02 12 22 32), xmm2=(42 52 62 72)
        ; xmm1=(03 13 23 33), xmm3=(43 53 63 73)

        movaps   xmm4,xmm0              ; transpose coefficients(phase 1)
        unpcklps xmm0,xmm1              ; xmm0=(02 03 12 13)
        unpckhps xmm4,xmm1              ; xmm4=(22 23 32 33)
        movaps   xmm5,xmm2              ; transpose coefficients(phase 1)
        unpcklps xmm2,xmm3              ; xmm2=(42 43 52 53)
        unpckhps xmm5,xmm3              ; xmm5=(62 63 72 73)

        movaps  xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm1, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)]
        movaps  xmm3, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)]

        ; xmm6=(00 10 20 30), xmm1=(40 50 60 70)
        ; xmm7=(01 11 21 31), xmm3=(41 51 61 71)

        movaps  XMMWORD [wk(0)], xmm4   ; wk(0)=(22 23 32 33)
        movaps  XMMWORD [wk(1)], xmm2   ; wk(1)=(42 43 52 53)

        movaps   xmm4,xmm6              ; transpose coefficients(phase 1)
        unpcklps xmm6,xmm7              ; xmm6=(00 01 10 11)
        unpckhps xmm4,xmm7              ; xmm4=(20 21 30 31)
        movaps   xmm2,xmm1              ; transpose coefficients(phase 1)
        unpcklps xmm1,xmm3              ; xmm1=(40 41 50 51)
        unpckhps xmm2,xmm3              ; xmm2=(60 61 70 71)

        movaps    xmm7,xmm6             ; transpose coefficients(phase 2)
        unpcklps2 xmm6,xmm0             ; xmm6=(00 01 02 03)=data0
        unpckhps2 xmm7,xmm0             ; xmm7=(10 11 12 13)=data1
        movaps    xmm3,xmm2             ; transpose coefficients(phase 2)
        unpcklps2 xmm2,xmm5             ; xmm2=(60 61 62 63)=data6
        unpckhps2 xmm3,xmm5             ; xmm3=(70 71 72 73)=data7

        movaps  xmm0,xmm7
        movaps  xmm5,xmm6
        subps   xmm7,xmm2               ; xmm7=data1-data6=tmp6
        subps   xmm6,xmm3               ; xmm6=data0-data7=tmp7
        addps   xmm0,xmm2               ; xmm0=data1+data6=tmp1
        addps   xmm5,xmm3               ; xmm5=data0+data7=tmp0

        movaps  xmm2, XMMWORD [wk(0)]   ; xmm2=(22 23 32 33)
        movaps  xmm3, XMMWORD [wk(1)]   ; xmm3=(42 43 52 53)
        movaps  XMMWORD [wk(0)], xmm7   ; wk(0)=tmp6
        movaps  XMMWORD [wk(1)], xmm6   ; wk(1)=tmp7

        movaps    xmm7,xmm4             ; transpose coefficients(phase 2)
        unpcklps2 xmm4,xmm2             ; xmm4=(20 21 22 23)=data2
        unpckhps2 xmm7,xmm2             ; xmm7=(30 31 32 33)=data3
        movaps    xmm6,xmm1             ; transpose coefficients(phase 2)
        unpcklps2 xmm1,xmm3             ; xmm1=(40 41 42 43)=data4
        unpckhps2 xmm6,xmm3             ; xmm6=(50 51 52 53)=data5

        movaps  xmm2,xmm7
        movaps  xmm3,xmm4
        addps   xmm7,xmm1               ; xmm7=data3+data4=tmp3
        addps   xmm4,xmm6               ; xmm4=data2+data5=tmp2
        subps   xmm2,xmm1               ; xmm2=data3-data4=tmp4
        subps   xmm3,xmm6               ; xmm3=data2-data5=tmp5

        ; -- Even part

        movaps  xmm1,xmm5
        movaps  xmm6,xmm0
        subps   xmm5,xmm7               ; xmm5=tmp13
        subps   xmm0,xmm4               ; xmm0=tmp12
        addps   xmm1,xmm7               ; xmm1=tmp10
        addps   xmm6,xmm4               ; xmm6=tmp11

        addps   xmm0,xmm5
        mulps   xmm0,[GOTOFF(ebx,PD_0_707)] ; xmm0=z1

        movaps  xmm7,xmm1
        movaps  xmm4,xmm5
        subps   xmm1,xmm6               ; xmm1=data4
        subps   xmm5,xmm0               ; xmm5=data6
        addps   xmm7,xmm6               ; xmm7=data0
        addps   xmm4,xmm0               ; xmm4=data2

        movaps  XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)], xmm1
        movaps  XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)], xmm5
        movaps  XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7
        movaps  XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4

        ; -- Odd part

        movaps  xmm6, XMMWORD [wk(0)]   ; xmm6=tmp6
        movaps  xmm0, XMMWORD [wk(1)]   ; xmm0=tmp7

        addps   xmm2,xmm3               ; xmm2=tmp10
        addps   xmm3,xmm6               ; xmm3=tmp11
        addps   xmm6,xmm0               ; xmm6=tmp12, xmm0=tmp7

        mulps   xmm3,[GOTOFF(ebx,PD_0_707)] ; xmm3=z3

        movaps  xmm1,xmm2               ; xmm1=tmp10
        subps   xmm2,xmm6
        mulps   xmm2,[GOTOFF(ebx,PD_0_382)] ; xmm2=z5
        mulps   xmm1,[GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196)
        mulps   xmm6,[GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562)
        addps   xmm1,xmm2               ; xmm1=z2
        addps   xmm6,xmm2               ; xmm6=z4

        movaps  xmm5,xmm0
        subps   xmm0,xmm3               ; xmm0=z13
        addps   xmm5,xmm3               ; xmm5=z11

        movaps  xmm7,xmm0
        movaps  xmm4,xmm5
        subps   xmm0,xmm1               ; xmm0=data3
        subps   xmm5,xmm6               ; xmm5=data7
        addps   xmm7,xmm1               ; xmm7=data5
        addps   xmm4,xmm6               ; xmm4=data1

        movaps  XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0
        movaps  XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)], xmm5
        movaps  XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)], xmm7
        movaps  XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4

        add     edx, byte 4*SIZEOF_FAST_FLOAT
        dec     ecx
        jnz     near .columnloop

;       pop     edi             ; unused
;       pop     esi             ; unused
;       pop     edx             ; need not be preserved
;       pop     ecx             ; need not be preserved
        poppic  ebx
        mov     esp,ebp         ; esp <- aligned ebp
        pop     esp             ; esp <- original ebp
        pop     ebp
        ret

; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
        align   16