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
|
// -*- c++ -*-
unsigned char* pDest;
const unsigned char* pSrcP;
const unsigned char* pSrc;
const unsigned char* pBob;
const unsigned char* pBobP;
int64_t Max_Mov = 0x0404040404040404ull;
int64_t DiffThres = 0x0f0f0f0f0f0f0f0full;
int64_t YMask = 0x00ff00ff00ff00ffull; // keeps only luma
int64_t UVMask = 0xff00ff00ff00ff00ull; // keeps only chroma
int64_t TENS = 0x0a0a0a0a0a0a0a0aull;
int64_t FOURS = 0x0404040404040404ull;
int64_t ONES = 0x0101010101010101ull;
int64_t Min_Vals = 0x0000000000000000ull;
int64_t Max_Vals = 0x0000000000000000ull;
int64_t ShiftMask = 0xfefffefffefffeffull;
// long is int32 on ARCH_368, int64 on ARCH_AMD64. Declaring it this way
// saves a lot of xor's to delete 64bit garbage.
#if defined(DBL_RESIZE) || defined(USE_FOR_DSCALER)
long src_pitch2 = src_pitch; // even & odd lines are not longerleaved in DScaler
#else
long src_pitch2 = 2 * src_pitch; // even & odd lines are longerleaved in Avisynth
#endif
long dst_pitch2 = 2 * dst_pitch;
long y;
#ifdef IS_SSE2
long Last8 = (rowsize-16); // ofs to last 16 bytes in row for SSE2
#else
long Last8 = (rowsize-8); // ofs to last 8 bytes in row
#endif
long dst_pitchw = dst_pitch; // local stor so asm can ref
pSrc = pWeaveSrc; // polongs 1 weave line above
pSrcP = pWeaveSrcP; // "
#ifdef DBL_RESIZE
#ifdef USE_VERTICAL_FILTER
pDest = pWeaveDest + dst_pitch2;
#else
pDest = pWeaveDest + 3*dst_pitch;
#endif
#else
#ifdef USE_VERTICAL_FILTER
pDest = pWeaveDest + dst_pitch;
#else
pDest = pWeaveDest + dst_pitch2;
#endif
#endif
if (TopFirst)
{
pBob = pCopySrc + src_pitch2; // remember one weave line just copied previously
pBobP = pCopySrcP + src_pitch2;
}
else
{
pBob = pCopySrc;
pBobP = pCopySrcP;
}
#ifndef _pBob
#define _pBob "%0"
#define _src_pitch2 "%1"
#define _ShiftMask "%2"
#define _pDest "%3"
#define _dst_pitchw "%4"
#define _Last8 "%5"
#define _pSrc "%6"
#define _pSrcP "%7"
#define _pBobP "%8"
#define _DiffThres "%9"
#define _Min_Vals "%10"
#define _Max_Vals "%11"
#define _FOURS "%12"
#define _TENS "%13"
#define _ONES "%14"
#define _UVMask "%15"
#define _Max_Mov "%16"
#define _YMask "%17"
#define _oldbx "%18"
#endif
long oldbx;
for (y=1; y < FldHeight-1; y++)
{
// pretend it's indented -->>
__asm__ __volatile__
(
// Loop general reg usage
//
// XAX - pBobP, then pDest
// XBX - pBob
// XCX - src_pitch2
// XDX - current offset
// XDI - prev weave pixels, 1 line up
// XSI - next weave pixels, 1 line up
// Save "XBX" (-fPIC)
MOVX " %%" XBX ", " _oldbx "\n\t"
#ifdef IS_SSE2
// sse2 code deleted for now
#else
// simple bob first 8 bytes
MOVX " " _pBob ", %%" XBX "\n\t"
MOVX " " _src_pitch2 ", %%" XCX "\n\t"
#ifdef USE_VERTICAL_FILTER
"movq (%%" XBX "), %%mm0\n\t"
"movq (%%" XBX ", %%" XCX "), %%mm1\n\t" //, qword ptr["XBX"+"XCX"]
"movq %%mm0, %%mm2\n\t"
V_PAVGB ("%%mm2", "%%mm1", "%%mm3", _ShiftMask) // halfway between
V_PAVGB ("%%mm0", "%%mm2", "%%mm3", _ShiftMask) // 1/4 way
V_PAVGB ("%%mm1", "%%mm2", "%%mm3", _ShiftMask) // 3/4 way
MOVX " " _pDest ", %%" XDI "\n\t"
MOVX " " _dst_pitchw ", %%" XAX "\n\t"
V_MOVNTQ ("(%%" XDI ")", "%%mm0")
V_MOVNTQ ("(%%" XDI ", %%" XAX ")", "%%mm1") // qword ptr["XDI"+"XAX"], mm1
// simple bob last 8 bytes
MOVX " " _Last8 ", %%" XDX "\n\t"
LEAX " (%%" XBX ", %%" XDX "), %%" XSI "\n\t" // ["XBX"+"XDX"]
"movq (%%" XSI "), %%mm0\n\t"
"movq (%%" XSI ", %%" XCX "), %%mm1\n\t" // qword ptr["XSI"+"XCX"]
"movq %%mm0, %%mm2\n\t"
V_PAVGB ("%%mm2", "%%mm1", "%%mm3", _ShiftMask) // halfway between
V_PAVGB ("%%mm0", "%%mm2", "%%mm3", _ShiftMask) // 1/4 way
V_PAVGB ("%%mm1", "%%mm2", "%%mm3", _ShiftMask) // 3/4 way
ADDX " %%" XDX ", %%" XDI "\n\t" // last 8 bytes of dest
V_MOVNTQ ("%%" XDI "", "%%mm0")
V_MOVNTQ ("(%%" XDI ", %%" XAX ")", "%%mm1") // qword ptr["XDI"+"XAX"], mm1)
#else
"movq (%%" XBX "), %%mm0\n\t"
// pavgb mm0, qword ptr["XBX"+"XCX"]
V_PAVGB ("%%mm0", "(%%" XBX ", %%" XCX ")", "%%mm2", _ShiftMask) // qword ptr["XBX"+"XCX"], mm2, ShiftMask)
MOVX " " _pDest ", %%" XDI "\n\t"
V_MOVNTQ ("(%%" XDI ")", "%%mm0")
// simple bob last 8 bytes
MOVX " " _Last8 ", %%" XDX "\n\t"
LEAX " (%%" XBX ", %%" XDX "), %%" XSI "\n\t" //"XSI", ["XBX"+"XDX"]
"movq (%%" XSI "), %%mm0\n\t"
// pavgb mm0, qword ptr["XSI"+"XCX"]
V_PAVGB ("%%mm0", "(%%" XSI ", %%" XCX ")", "%%mm2", _ShiftMask) // qword ptr["XSI"+"XCX"], mm2, ShiftMask)
V_MOVNTQ ("(%%" XDI ", %%" XDX ")", "%%mm0") // qword ptr["XDI"+"XDX"], mm0)
#endif
// now loop and get the middle qwords
MOVX " " _pSrc ", %%" XSI "\n\t"
MOVX " " _pSrcP ", %%" XDI "\n\t"
MOVX " $8, %%" XDX "\n\t" // curr offset longo all lines
"1:\n\t"
MOVX " " _pBobP ", %%" XAX "\n\t"
ADDX " $8, %%" XDI "\n\t"
ADDX " $8, %%" XSI "\n\t"
ADDX " $8, %%" XBX "\n\t"
ADDX " %%" XDX ", %%" XAX "\n\t"
#ifdef USE_STRANGE_BOB
#include "StrangeBob.inc"
#else
#include "WierdBob.inc"
#endif
// For non-SSE2:
// through out most of the rest of this loop we will malongain
// mm4 our min bob value
// mm5 best weave pixels so far
// mm6 our max Bob value
// mm7 best weighted pixel ratings so far
// We will keep a slight bias to using the weave pixels
// from the current location, by rating them by the min distance
// from the Bob value instead of the avg distance from that value.
// our best and only rating so far
"pcmpeqb %%mm7, %%mm7\n\t" // ffff, say we didn't find anything good yet
#endif
|
