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
|
;
; Copyright (c) 2016, Alliance for Open Media. All rights reserved
;
; This source code is subject to the terms of the BSD 2 Clause License and
; the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
; was not distributed with this source code in the LICENSE file, you can
; obtain it at www.aomedia.org/license/software. If the Alliance for Open
; Media Patent License 1.0 was not distributed with this source code in the
; PATENTS file, you can obtain it at www.aomedia.org/license/patent.
;
;
%define private_prefix av1
%include "third_party/x86inc/x86inc.asm"
SECTION .text
; int64_t av1_block_error(int16_t *coeff, int16_t *dqcoeff, intptr_t block_size,
; int64_t *ssz)
INIT_XMM sse2
cglobal block_error, 3, 3, 8, uqc, dqc, size, ssz
pxor m4, m4 ; sse accumulator
pxor m6, m6 ; ssz accumulator
pxor m5, m5 ; dedicated zero register
lea uqcq, [uqcq+sizeq*2]
lea dqcq, [dqcq+sizeq*2]
neg sizeq
.loop:
mova m2, [uqcq+sizeq*2]
mova m0, [dqcq+sizeq*2]
mova m3, [uqcq+sizeq*2+mmsize]
mova m1, [dqcq+sizeq*2+mmsize]
psubw m0, m2
psubw m1, m3
; individual errors are max. 15bit+sign, so squares are 30bit, and
; thus the sum of 2 should fit in a 31bit integer (+ unused sign bit)
pmaddwd m0, m0
pmaddwd m1, m1
pmaddwd m2, m2
pmaddwd m3, m3
; accumulate in 64bit
punpckldq m7, m0, m5
punpckhdq m0, m5
paddq m4, m7
punpckldq m7, m1, m5
paddq m4, m0
punpckhdq m1, m5
paddq m4, m7
punpckldq m7, m2, m5
paddq m4, m1
punpckhdq m2, m5
paddq m6, m7
punpckldq m7, m3, m5
paddq m6, m2
punpckhdq m3, m5
paddq m6, m7
paddq m6, m3
add sizeq, mmsize
jl .loop
; accumulate horizontally and store in return value
movhlps m5, m4
movhlps m7, m6
paddq m4, m5
paddq m6, m7
%if ARCH_X86_64
movq rax, m4
movq [sszq], m6
%else
mov eax, sszm
pshufd m5, m4, 0x1
movq [eax], m6
movd eax, m4
movd edx, m5
%endif
RET
; Compute the sum of squared difference between two int16_t vectors.
; int64_t av1_block_error_fp(int16_t *coeff, int16_t *dqcoeff,
; intptr_t block_size)
INIT_XMM sse2
cglobal block_error_fp, 3, 3, 6, uqc, dqc, size
pxor m4, m4 ; sse accumulator
pxor m5, m5 ; dedicated zero register
lea uqcq, [uqcq+sizeq*2]
lea dqcq, [dqcq+sizeq*2]
neg sizeq
.loop:
mova m2, [uqcq+sizeq*2]
mova m0, [dqcq+sizeq*2]
mova m3, [uqcq+sizeq*2+mmsize]
mova m1, [dqcq+sizeq*2+mmsize]
psubw m0, m2
psubw m1, m3
; individual errors are max. 15bit+sign, so squares are 30bit, and
; thus the sum of 2 should fit in a 31bit integer (+ unused sign bit)
pmaddwd m0, m0
pmaddwd m1, m1
; accumulate in 64bit
punpckldq m3, m0, m5
punpckhdq m0, m5
paddq m4, m3
punpckldq m3, m1, m5
paddq m4, m0
punpckhdq m1, m5
paddq m4, m3
paddq m4, m1
add sizeq, mmsize
jl .loop
; accumulate horizontally and store in return value
movhlps m5, m4
paddq m4, m5
%if ARCH_X86_64
movq rax, m4
%else
pshufd m5, m4, 0x1
movd eax, m4
movd edx, m5
%endif
RET
|
