Newer
Older
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
/*
* Fadecandy Firmware: Low-level per-pixel LUT code
* (Included into fadecandy.cpp)
*
* Copyright (c) 2013 Micah Elizabeth Scott
*
* Permission is hereby granted, free of charge, to any person obtaining a copy of
* this software and associated documentation files (the "Software"), to deal in
* the Software without restriction, including without limitation the rights to
* use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
* the Software, and to permit persons to whom the Software is furnished to do so,
* subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
* FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
* COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
* IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
ALWAYS_INLINE static inline
uint32_t FCP_FN(lutInterpolate)(const uint16_t *lut, uint32_t arg)
{
/*
* Using our color LUT for the indicated channel, convert the
* 16-bit intensity "arg" in our input colorspace to a corresponding
* 16-bit intensity in the device colorspace.
*
* Remember that our LUT is 257 entries long. The final entry corresponds to an
* input of 0x10000, which can't quite be reached.
*
* 'arg' is in the range [0, 0xFFFF]
*
* This operation is equivalent to the following:
*
* unsigned index = arg >> 8; // Range [0, 0xFF]
* unsigned alpha = arg & 0xFF; // Range [0, 0xFF]
* unsigned invAlpha = 0x100 - alpha; // Range [1, 0x100]
*
* // Result in range [0, 0xFFFF]
* return (lut[index] * invAlpha + lut[index + 1] * alpha) >> 8;
*
* This is easy to understand, but it turns out to be a serious bottleneck
* in terms of speed and memory bandwidth, as well as register pressure that
* affects the compilation of updatePixel().
*
* To speed this up, we try and do the lut[index] and lut[index+1] portions
* in parallel using the SMUAD instruction. This is a pair of 16x16 multiplies,
* and the results are added together. We can combine this with an unaligned load
* to grab two adjacent entries from the LUT. The remaining complications are:
*
* 1. We wanted unsigned, not signed
* 2. We still need to generate the input values efficiently.
*
* (1) is easy to solve if we're okay with 15-bit precision for the LUT instead
* of 16-bit, which is fine. During LUT preparation, we right-shift each entry
* by 1, keeping them within the positive range of a signed 16-bit int.
*
* For (2), we need to quickly put 'alpha' in the high halfword and invAlpha in
* the low halfword, or vice versa. One fast way to do this is (0x01000000 + x - (x << 16).
*/
#if FCP_INTERPOLATION
uint32_t index = arg >> 8; // Range [0, 0xFF]
// Load lut[index] into low halfword, lut[index+1] into high halfword.
uint32_t pair = *(const uint32_t*)(lut + index);
unsigned alpha = arg & 0xFF; // Range [0, 0xFF]
// Reversed halfword order
uint32_t pairAlpha = (0x01000000 + alpha - (alpha << 16));
return __SMUADX(pairAlpha, pair) >> 7;
#else
// Simpler non-interpolated version
return lut[arg >> 8] << 1;
#endif
}