aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/libjpeg-turbo/simd/x86_64/jquanti-sse2.asm
blob: 3ee442027a5affa31eee15748b51671414d68b64 (plain) (blame)
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
;
; jquanti.asm - sample data conversion and quantization (64-bit SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright (C) 2009, 2016, D. R. Commander.
; Copyright (C) 2018, Matthias Räncker.
;
; 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

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

; --------------------------------------------------------------------------
    SECTION     SEG_TEXT
    BITS        64
;
; Load data into workspace, applying unsigned->signed conversion
;
; GLOBAL(void)
; jsimd_convsamp_sse2(JSAMPARRAY sample_data, JDIMENSION start_col,
;                     DCTELEM *workspace);
;

; r10 = JSAMPARRAY sample_data
; r11d = JDIMENSION start_col
; r12 = DCTELEM *workspace

    align       32
    GLOBAL_FUNCTION(jsimd_convsamp_sse2)

EXTN(jsimd_convsamp_sse2):
    push        rbp
    mov         rax, rsp
    mov         rbp, rsp
    collect_args 3
    push        rbx

    pxor        xmm6, xmm6              ; xmm6=(all 0's)
    pcmpeqw     xmm7, xmm7
    psllw       xmm7, 7                 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..}

    mov         rsi, r10
    mov         eax, r11d
    mov         rdi, r12
    mov         rcx, DCTSIZE/4
.convloop:
    mov         rbxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW]  ; (JSAMPLE *)
    mov         rdxp, JSAMPROW [rsi+1*SIZEOF_JSAMPROW]  ; (JSAMPLE *)

    movq        xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE]  ; xmm0=(01234567)
    movq        xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE]  ; xmm1=(89ABCDEF)

    mov         rbxp, JSAMPROW [rsi+2*SIZEOF_JSAMPROW]  ; (JSAMPLE *)
    mov         rdxp, JSAMPROW [rsi+3*SIZEOF_JSAMPROW]  ; (JSAMPLE *)

    movq        xmm2, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE]  ; xmm2=(GHIJKLMN)
    movq        xmm3, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE]  ; xmm3=(OPQRSTUV)

    punpcklbw   xmm0, xmm6              ; xmm0=(01234567)
    punpcklbw   xmm1, xmm6              ; xmm1=(89ABCDEF)
    paddw       xmm0, xmm7
    paddw       xmm1, xmm7
    punpcklbw   xmm2, xmm6              ; xmm2=(GHIJKLMN)
    punpcklbw   xmm3, xmm6              ; xmm3=(OPQRSTUV)
    paddw       xmm2, xmm7
    paddw       xmm3, xmm7

    movdqa      XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0
    movdqa      XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1
    movdqa      XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2
    movdqa      XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3

    add         rsi, byte 4*SIZEOF_JSAMPROW
    add         rdi, byte 4*DCTSIZE*SIZEOF_DCTELEM
    dec         rcx
    jnz         short .convloop

    pop         rbx
    uncollect_args 3
    pop         rbp
    ret

; --------------------------------------------------------------------------
;
; Quantize/descale the coefficients, and store into coef_block
;
; This implementation is based on an algorithm described in
;   "How to optimize for the Pentium family of microprocessors"
;   (http://www.agner.org/assem/).
;
; GLOBAL(void)
; jsimd_quantize_sse2(JCOEFPTR coef_block, DCTELEM *divisors,
;                     DCTELEM *workspace);
;

%define RECIPROCAL(m, n, b) \
  XMMBLOCK(DCTSIZE * 0 + (m), (n), (b), SIZEOF_DCTELEM)
%define CORRECTION(m, n, b) \
  XMMBLOCK(DCTSIZE * 1 + (m), (n), (b), SIZEOF_DCTELEM)
%define SCALE(m, n, b) \
  XMMBLOCK(DCTSIZE * 2 + (m), (n), (b), SIZEOF_DCTELEM)

; r10 = JCOEFPTR coef_block
; r11 = DCTELEM *divisors
; r12 = DCTELEM *workspace

    align       32
    GLOBAL_FUNCTION(jsimd_quantize_sse2)

EXTN(jsimd_quantize_sse2):
    push        rbp
    mov         rax, rsp
    mov         rbp, rsp
    collect_args 3

    mov         rsi, r12
    mov         rdx, r11
    mov         rdi, r10
    mov         rax, DCTSIZE2/32
.quantloop:
    movdqa      xmm4, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_DCTELEM)]
    movdqa      xmm5, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_DCTELEM)]
    movdqa      xmm6, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_DCTELEM)]
    movdqa      xmm7, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_DCTELEM)]
    movdqa      xmm0, xmm4
    movdqa      xmm1, xmm5
    movdqa      xmm2, xmm6
    movdqa      xmm3, xmm7
    psraw       xmm4, (WORD_BIT-1)
    psraw       xmm5, (WORD_BIT-1)
    psraw       xmm6, (WORD_BIT-1)
    psraw       xmm7, (WORD_BIT-1)
    pxor        xmm0, xmm4
    pxor        xmm1, xmm5
    pxor        xmm2, xmm6
    pxor        xmm3, xmm7
    psubw       xmm0, xmm4              ; if (xmm0 < 0) xmm0 = -xmm0;
    psubw       xmm1, xmm5              ; if (xmm1 < 0) xmm1 = -xmm1;
    psubw       xmm2, xmm6              ; if (xmm2 < 0) xmm2 = -xmm2;
    psubw       xmm3, xmm7              ; if (xmm3 < 0) xmm3 = -xmm3;

    paddw       xmm0, XMMWORD [CORRECTION(0,0,rdx)]  ; correction + roundfactor
    paddw       xmm1, XMMWORD [CORRECTION(1,0,rdx)]
    paddw       xmm2, XMMWORD [CORRECTION(2,0,rdx)]
    paddw       xmm3, XMMWORD [CORRECTION(3,0,rdx)]
    pmulhuw     xmm0, XMMWORD [RECIPROCAL(0,0,rdx)]  ; reciprocal
    pmulhuw     xmm1, XMMWORD [RECIPROCAL(1,0,rdx)]
    pmulhuw     xmm2, XMMWORD [RECIPROCAL(2,0,rdx)]
    pmulhuw     xmm3, XMMWORD [RECIPROCAL(3,0,rdx)]
    pmulhuw     xmm0, XMMWORD [SCALE(0,0,rdx)]       ; scale
    pmulhuw     xmm1, XMMWORD [SCALE(1,0,rdx)]
    pmulhuw     xmm2, XMMWORD [SCALE(2,0,rdx)]
    pmulhuw     xmm3, XMMWORD [SCALE(3,0,rdx)]

    pxor        xmm0, xmm4
    pxor        xmm1, xmm5
    pxor        xmm2, xmm6
    pxor        xmm3, xmm7
    psubw       xmm0, xmm4
    psubw       xmm1, xmm5
    psubw       xmm2, xmm6
    psubw       xmm3, xmm7
    movdqa      XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0
    movdqa      XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1
    movdqa      XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2
    movdqa      XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3

    add         rsi, byte 32*SIZEOF_DCTELEM
    add         rdx, byte 32*SIZEOF_DCTELEM
    add         rdi, byte 32*SIZEOF_JCOEF
    dec         rax
    jnz         near .quantloop

    uncollect_args 3
    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