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
|
;
; jcqnts2f-64.asm - sample data conversion and quantization (64-bit SSE & SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright 2009 D. R. Commander
;
; Based on
; 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"
%include "jdct.inc"
; --------------------------------------------------------------------------
SECTION SEG_TEXT
BITS 64
;
; Load data into workspace, applying unsigned->signed conversion
;
; GLOBAL(void)
; jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col,
; FAST_FLOAT * workspace);
;
; r10 = JSAMPARRAY sample_data
; r11 = JDIMENSION start_col
; r12 = FAST_FLOAT * workspace
align 16
global EXTN(jsimd_convsamp_float_sse2)
EXTN(jsimd_convsamp_float_sse2):
push rbp
mov rax,rsp
mov rbp,rsp
collect_args
push rbx
pcmpeqw xmm7,xmm7
psllw xmm7,7
packsswb xmm7,xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..)
mov rsi, r10
mov rax, r11
mov rdi, r12
mov rcx, DCTSIZE/2
.convloop:
mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *)
mov rdx, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *)
movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE]
movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE]
psubb xmm0,xmm7 ; xmm0=(01234567)
psubb xmm1,xmm7 ; xmm1=(89ABCDEF)
punpcklbw xmm0,xmm0 ; xmm0=(*0*1*2*3*4*5*6*7)
punpcklbw xmm1,xmm1 ; xmm1=(*8*9*A*B*C*D*E*F)
punpcklwd xmm2,xmm0 ; xmm2=(***0***1***2***3)
punpckhwd xmm0,xmm0 ; xmm0=(***4***5***6***7)
punpcklwd xmm3,xmm1 ; xmm3=(***8***9***A***B)
punpckhwd xmm1,xmm1 ; xmm1=(***C***D***E***F)
psrad xmm2,(DWORD_BIT-BYTE_BIT) ; xmm2=(0123)
psrad xmm0,(DWORD_BIT-BYTE_BIT) ; xmm0=(4567)
cvtdq2ps xmm2,xmm2 ; xmm2=(0123)
cvtdq2ps xmm0,xmm0 ; xmm0=(4567)
psrad xmm3,(DWORD_BIT-BYTE_BIT) ; xmm3=(89AB)
psrad xmm1,(DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF)
cvtdq2ps xmm3,xmm3 ; xmm3=(89AB)
cvtdq2ps xmm1,xmm1 ; xmm1=(CDEF)
movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm2
movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0
movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3
movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1
add rsi, byte 2*SIZEOF_JSAMPROW
add rdi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT
dec rcx
jnz short .convloop
pop rbx
uncollect_args
pop rbp
ret
; --------------------------------------------------------------------------
;
; Quantize/descale the coefficients, and store into coef_block
;
; GLOBAL(void)
; jsimd_quantize_float_sse2 (JCOEFPTR coef_block, FAST_FLOAT * divisors,
; FAST_FLOAT * workspace);
;
; r10 = JCOEFPTR coef_block
; r11 = FAST_FLOAT * divisors
; r12 = FAST_FLOAT * workspace
align 16
global EXTN(jsimd_quantize_float_sse2)
EXTN(jsimd_quantize_float_sse2):
push rbp
mov rax,rsp
mov rbp,rsp
collect_args
mov rsi, r12
mov rdx, r11
mov rdi, r10
mov rax, DCTSIZE2/16
.quantloop:
movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)]
movaps xmm1, XMMWORD [XMMBLOCK(0,1,rsi,SIZEOF_FAST_FLOAT)]
mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)]
mulps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)]
movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)]
movaps xmm3, XMMWORD [XMMBLOCK(1,1,rsi,SIZEOF_FAST_FLOAT)]
mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)]
mulps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)]
cvtps2dq xmm0,xmm0
cvtps2dq xmm1,xmm1
cvtps2dq xmm2,xmm2
cvtps2dq xmm3,xmm3
packssdw xmm0,xmm1
packssdw xmm2,xmm3
movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_JCOEF)], xmm0
movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_JCOEF)], xmm2
add rsi, byte 16*SIZEOF_FAST_FLOAT
add rdx, byte 16*SIZEOF_FAST_FLOAT
add rdi, byte 16*SIZEOF_JCOEF
dec rax
jnz short .quantloop
uncollect_args
pop rbp
ret
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
align 16
|
