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
|
/*
* Copyright 2011-2013 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef __UTIL_HALF_H__
#define __UTIL_HALF_H__
#include "util_types.h"
#ifdef __KERNEL_SSE2__
#include "util_simd.h"
#endif
CCL_NAMESPACE_BEGIN
/* Half Floats */
#ifdef __KERNEL_OPENCL__
#define float4_store_half(h, f, scale) vstore_half4(f * (scale), 0, h);
#else
/* CUDA has its own half data type, no need to define then */
#ifndef __KERNEL_CUDA__
typedef unsigned short half;
#endif
struct half4 { half x, y, z, w; };
#ifdef __KERNEL_CUDA__
ccl_device_inline void float4_store_half(half *h, float4 f, float scale)
{
h[0] = __float2half(f.x * scale);
h[1] = __float2half(f.y * scale);
h[2] = __float2half(f.z * scale);
h[3] = __float2half(f.w * scale);
}
#else
ccl_device_inline void float4_store_half(half *h, float4 f, float scale)
{
#ifndef __KERNEL_SSE2__
for(int i = 0; i < 4; i++) {
/* optimized float to half for pixels:
* assumes no negative, no nan, no inf, and sets denormal to 0 */
union { uint i; float f; } in;
float fscale = f[i] * scale;
in.f = (fscale > 0.0f)? ((fscale < 65504.0f)? fscale: 65504.0f): 0.0f;
int x = in.i;
int absolute = x & 0x7FFFFFFF;
int Z = absolute + 0xC8000000;
int result = (absolute < 0x38800000)? 0: Z;
int rshift = (result >> 13);
h[i] = (rshift & 0x7FFF);
}
#else
/* same as above with SSE */
ssef fscale = load4f(f) * scale;
ssef x = min(max(fscale, 0.0f), 65504.0f);
#ifdef __KERNEL_AVX2__
ssei rpack = _mm_cvtps_ph(x, 0);
#else
ssei absolute = cast(x) & 0x7FFFFFFF;
ssei Z = absolute + 0xC8000000;
ssei result = andnot(absolute < 0x38800000, Z);
ssei rshift = (result >> 13) & 0x7FFF;
ssei rpack = _mm_packs_epi32(rshift, rshift);
#endif
_mm_storel_pi((__m64*)h, _mm_castsi128_ps(rpack));
#endif
}
ccl_device_inline float half_to_float(half h)
{
float f;
*((int*) &f) = ((h & 0x8000) << 16) | (((h & 0x7c00) + 0x1C000) << 13) | ((h & 0x03FF) << 13);
return f;
}
ccl_device_inline float4 half4_to_float4(half4 h)
{
float4 f;
f.x = half_to_float(h.x);
f.y = half_to_float(h.y);
f.z = half_to_float(h.z);
f.w = half_to_float(h.w);
return f;
}
#endif
#endif
CCL_NAMESPACE_END
#endif /* __UTIL_HALF_H__ */
|
