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
| ;
| ; jcsample.asm - downsampling (64-bit AVX2)
| ;
| ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
| ; Copyright (C) 2009, 2016, D. R. Commander.
| ; Copyright (C) 2015, Intel Corporation.
| ;
| ; 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
| ;
| ; [TAB8]
|
| %include "jsimdext.inc"
|
| ; --------------------------------------------------------------------------
| SECTION SEG_TEXT
| BITS 64
| ;
| ; Downsample pixel values of a single component.
| ; This version handles the common case of 2:1 horizontal and 1:1 vertical,
| ; without smoothing.
| ;
| ; GLOBAL(void)
| ; jsimd_h2v1_downsample_avx2(JDIMENSION image_width, int max_v_samp_factor,
| ; JDIMENSION v_samp_factor,
| ; JDIMENSION width_in_blocks, JSAMPARRAY input_data,
| ; JSAMPARRAY output_data);
| ;
|
| ; r10d = JDIMENSION image_width
| ; r11 = int max_v_samp_factor
| ; r12d = JDIMENSION v_samp_factor
| ; r13d = JDIMENSION width_in_blocks
| ; r14 = JSAMPARRAY input_data
| ; r15 = JSAMPARRAY output_data
|
| align 32
| GLOBAL_FUNCTION(jsimd_h2v1_downsample_avx2)
|
| EXTN(jsimd_h2v1_downsample_avx2):
| push rbp
| mov rax, rsp
| mov rbp, rsp
| collect_args 6
|
| mov ecx, r13d
| shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols)
| jz near .return
|
| mov edx, r10d
|
| ; -- expand_right_edge
|
| push rcx
| shl rcx, 1 ; output_cols * 2
| sub rcx, rdx
| jle short .expand_end
|
| mov rax, r11
| test rax, rax
| jle short .expand_end
|
| cld
| mov rsi, r14 ; input_data
| .expandloop:
| push rax
| push rcx
|
| mov rdi, JSAMPROW [rsi]
| add rdi, rdx
| mov al, JSAMPLE [rdi-1]
|
| rep stosb
|
| pop rcx
| pop rax
|
| add rsi, byte SIZEOF_JSAMPROW
| dec rax
| jg short .expandloop
|
| .expand_end:
| pop rcx ; output_cols
|
| ; -- h2v1_downsample
|
| mov eax, r12d ; rowctr
| test eax, eax
| jle near .return
|
| mov rdx, 0x00010000 ; bias pattern
| vmovd xmm7, edx
| vpshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
| vperm2i128 ymm7, ymm7, ymm7, 0 ; ymm7={xmm7, xmm7}
| vpcmpeqw ymm6, ymm6, ymm6
| vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
|
| mov rsi, r14 ; input_data
| mov rdi, r15 ; output_data
| .rowloop:
| push rcx
| push rdi
| push rsi
|
| mov rsi, JSAMPROW [rsi] ; inptr
| mov rdi, JSAMPROW [rdi] ; outptr
|
| cmp rcx, byte SIZEOF_YMMWORD
| jae short .columnloop
|
| .columnloop_r24:
| ; rcx can possibly be 8, 16, 24
| cmp rcx, 24
| jne .columnloop_r16
| vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
| vmovdqu xmm1, XMMWORD [rsi+1*SIZEOF_YMMWORD]
| mov rcx, SIZEOF_YMMWORD
| jmp short .downsample
|
| .columnloop_r16:
| cmp rcx, 16
| jne .columnloop_r8
| vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
| vpxor ymm1, ymm1, ymm1
| mov rcx, SIZEOF_YMMWORD
| jmp short .downsample
|
| .columnloop_r8:
| vmovdqu xmm0, XMMWORD[rsi+0*SIZEOF_YMMWORD]
| vpxor ymm1, ymm1, ymm1
| mov rcx, SIZEOF_YMMWORD
| jmp short .downsample
|
| .columnloop:
| vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD]
| vmovdqu ymm1, YMMWORD [rsi+1*SIZEOF_YMMWORD]
|
| .downsample:
| vpsrlw ymm2, ymm0, BYTE_BIT
| vpand ymm0, ymm0, ymm6
| vpsrlw ymm3, ymm1, BYTE_BIT
| vpand ymm1, ymm1, ymm6
|
| vpaddw ymm0, ymm0, ymm2
| vpaddw ymm1, ymm1, ymm3
| vpaddw ymm0, ymm0, ymm7
| vpaddw ymm1, ymm1, ymm7
| vpsrlw ymm0, ymm0, 1
| vpsrlw ymm1, ymm1, 1
|
| vpackuswb ymm0, ymm0, ymm1
| vpermq ymm0, ymm0, 0xd8
|
| vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
|
| sub rcx, byte SIZEOF_YMMWORD ; outcol
| add rsi, byte 2*SIZEOF_YMMWORD ; inptr
| add rdi, byte 1*SIZEOF_YMMWORD ; outptr
| cmp rcx, byte SIZEOF_YMMWORD
| jae short .columnloop
| test rcx, rcx
| jnz near .columnloop_r24
|
| pop rsi
| pop rdi
| pop rcx
|
| add rsi, byte SIZEOF_JSAMPROW ; input_data
| add rdi, byte SIZEOF_JSAMPROW ; output_data
| dec rax ; rowctr
| jg near .rowloop
|
| .return:
| vzeroupper
| uncollect_args 6
| pop rbp
| ret
|
| ; --------------------------------------------------------------------------
| ;
| ; Downsample pixel values of a single component.
| ; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
| ; without smoothing.
| ;
| ; GLOBAL(void)
| ; jsimd_h2v2_downsample_avx2(JDIMENSION image_width, int max_v_samp_factor,
| ; JDIMENSION v_samp_factor,
| ; JDIMENSION width_in_blocks, JSAMPARRAY input_data,
| ; JSAMPARRAY output_data);
| ;
|
| ; r10d = JDIMENSION image_width
| ; r11 = int max_v_samp_factor
| ; r12d = JDIMENSION v_samp_factor
| ; r13d = JDIMENSION width_in_blocks
| ; r14 = JSAMPARRAY input_data
| ; r15 = JSAMPARRAY output_data
|
| align 32
| GLOBAL_FUNCTION(jsimd_h2v2_downsample_avx2)
|
| EXTN(jsimd_h2v2_downsample_avx2):
| push rbp
| mov rax, rsp
| mov rbp, rsp
| collect_args 6
|
| mov ecx, r13d
| shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols)
| jz near .return
|
| mov edx, r10d
|
| ; -- expand_right_edge
|
| push rcx
| shl rcx, 1 ; output_cols * 2
| sub rcx, rdx
| jle short .expand_end
|
| mov rax, r11
| test rax, rax
| jle short .expand_end
|
| cld
| mov rsi, r14 ; input_data
| .expandloop:
| push rax
| push rcx
|
| mov rdi, JSAMPROW [rsi]
| add rdi, rdx
| mov al, JSAMPLE [rdi-1]
|
| rep stosb
|
| pop rcx
| pop rax
|
| add rsi, byte SIZEOF_JSAMPROW
| dec rax
| jg short .expandloop
|
| .expand_end:
| pop rcx ; output_cols
|
| ; -- h2v2_downsample
|
| mov eax, r12d ; rowctr
| test rax, rax
| jle near .return
|
| mov rdx, 0x00020001 ; bias pattern
| vmovd xmm7, edx
| vpcmpeqw ymm6, ymm6, ymm6
| vpshufd xmm7, xmm7, 0x00 ; ymm7={1, 2, 1, 2, 1, 2, 1, 2}
| vperm2i128 ymm7, ymm7, ymm7, 0
| vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..}
|
| mov rsi, r14 ; input_data
| mov rdi, r15 ; output_data
| .rowloop:
| push rcx
| push rdi
| push rsi
|
| mov rdx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
| mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1
| mov rdi, JSAMPROW [rdi] ; outptr
|
| cmp rcx, byte SIZEOF_YMMWORD
| jae short .columnloop
|
| .columnloop_r24:
| cmp rcx, 24
| jne .columnloop_r16
| vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD]
| vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD]
| vmovdqu xmm2, XMMWORD [rdx+1*SIZEOF_YMMWORD]
| vmovdqu xmm3, XMMWORD [rsi+1*SIZEOF_YMMWORD]
| mov rcx, SIZEOF_YMMWORD
| jmp short .downsample
|
| .columnloop_r16:
| cmp rcx, 16
| jne .columnloop_r8
| vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD]
| vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD]
| vpxor ymm2, ymm2, ymm2
| vpxor ymm3, ymm3, ymm3
| mov rcx, SIZEOF_YMMWORD
| jmp short .downsample
|
| .columnloop_r8:
| vmovdqu xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
| vmovdqu xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
| vpxor ymm2, ymm2, ymm2
| vpxor ymm3, ymm3, ymm3
| mov rcx, SIZEOF_YMMWORD
| jmp short .downsample
|
| .columnloop:
| vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD]
| vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD]
| vmovdqu ymm2, YMMWORD [rdx+1*SIZEOF_YMMWORD]
| vmovdqu ymm3, YMMWORD [rsi+1*SIZEOF_YMMWORD]
|
| .downsample:
| vpand ymm4, ymm0, ymm6
| vpsrlw ymm0, ymm0, BYTE_BIT
| vpand ymm5, ymm1, ymm6
| vpsrlw ymm1, ymm1, BYTE_BIT
| vpaddw ymm0, ymm0, ymm4
| vpaddw ymm1, ymm1, ymm5
|
| vpand ymm4, ymm2, ymm6
| vpsrlw ymm2, ymm2, BYTE_BIT
| vpand ymm5, ymm3, ymm6
| vpsrlw ymm3, ymm3, BYTE_BIT
| vpaddw ymm2, ymm2, ymm4
| vpaddw ymm3, ymm3, ymm5
|
| vpaddw ymm0, ymm0, ymm1
| vpaddw ymm2, ymm2, ymm3
| vpaddw ymm0, ymm0, ymm7
| vpaddw ymm2, ymm2, ymm7
| vpsrlw ymm0, ymm0, 2
| vpsrlw ymm2, ymm2, 2
|
| vpackuswb ymm0, ymm0, ymm2
| vpermq ymm0, ymm0, 0xd8
|
| vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0
|
| sub rcx, byte SIZEOF_YMMWORD ; outcol
| add rdx, byte 2*SIZEOF_YMMWORD ; inptr0
| add rsi, byte 2*SIZEOF_YMMWORD ; inptr1
| add rdi, byte 1*SIZEOF_YMMWORD ; outptr
| cmp rcx, byte SIZEOF_YMMWORD
| jae near .columnloop
| test rcx, rcx
| jnz near .columnloop_r24
|
| pop rsi
| pop rdi
| pop rcx
|
| add rsi, byte 2*SIZEOF_JSAMPROW ; input_data
| add rdi, byte 1*SIZEOF_JSAMPROW ; output_data
| dec rax ; rowctr
| jg near .rowloop
|
| .return:
| vzeroupper
| uncollect_args 6
| pop rbp
| ret
|
| ; For some reason, the OS X linker does not honor the request to align the
| ; segment unless we do this.
| align 32
|
|
