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
|
// Copyright 2023 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
#include "common/DeviceRT.ih"
#include "rkcommon/math/vec.ih"
// c++ shared
#include "VarianceShared.h"
OSPRAY_BEGIN_ISPC_NAMESPACE
inline void Variance_kernel(
const vec2ui itemIndex, const FrameBufferView *uniform fbvSelf)
{
uniform const vec2ui &rtDims = fbvSelf->viewDims;
uniform const vec2ui &fbDims = fbvSelf->fbDims;
const vec4f *uniform colorBuffer = fbvSelf->colorBufferInput;
const LiveVariance *uniform self = (const LiveVariance *uniform)fbvSelf;
uniform const vec2ui rtSize = self->rtSize;
const vec4f *uniform varianceBuffer = self->varianceBuffer;
const vec2ui rtLower = itemIndex * rtSize;
// Iterate over all pixels within single render task and calculate maximum
// variance over collected samples
float v = 0.f;
int cnt = 0;
for (uniform uint32 rtY = 0; rtY < rtSize.y; rtY++) {
// Skip if out of range
const uint32 fbY = rtLower.y + rtY;
if (fbY >= fbDims.y)
continue;
for (uniform uint32 rtX = 0; rtX < rtSize.x; rtX++) {
// Skip if out of range
const uint32 fbX = rtLower.x + rtX;
if (fbX >= fbDims.x)
continue;
const uint32 linearId = fbY * fbDims.x + fbX;
const vec4f color = colorBuffer[linearId];
const vec4f variance = varianceBuffer[linearId];
const float den2 = reduce_add(make_vec3f(color))
+ (1.f - color.w); // invert alpha (bright alpha is more important)
if (den2 > 0.f) {
cnt++;
const vec4f diff = absf(color - variance);
v += reduce_add(diff) * rsqrtf(den2);
}
}
}
if (cnt)
v *= rcp((float)cnt); // avg
const uint32 lId = itemIndex.y * rtDims.x + itemIndex.x;
// scale to be backwards compatible with the old default tile size of 64x64
self->taskVarianceBuffer[lId] = v * 64.f;
}
DEFINE_FRAMEOP_KERNEL_LAUNCHER(Variance_kernel);
OSPRAY_END_ISPC_NAMESPACE
|
