archive/blender-archive
Archived
1
1
Fork 0
Code Pull Requests Packages Activity
This repository has been archived on 2023-10-09. You can view files and clone it, but cannot push or open issues or pull requests.
Files
138b0c970efbfb732ddee0530e073537582b6df8
blender-archive/intern/cycles/kernel/kernel_jitter.h

231 lines
5.4 KiB
C++
Raw Normal View History

Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
/*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* Copyright 2011-2013 Blender Foundation
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* 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
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* http://www.apache.org/licenses/LICENSE-2.0
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
*
Cycles: relicense GNU GPL source code to Apache version 2.0. More information in this post: http://code.blender.org/ Thanks to all contributes for giving their permission!
2013-08-18 14:16:15 +00:00
* 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
Cleanup: Fix Cycles Apache header. This was already mixed a bit, but the dot belongs there.
2014-12-25 02:50:24 +01:00
* limitations under the License.
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
*/
Move include outside of the CCL namespace
2014-08-27 00:10:48 +06:00
/* TODO(sergey): Consider moving portable ctz/clz stuff to util. */
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
CCL_NAMESPACE_BEGIN
/* "Correlated Multi-Jittered Sampling"
* Andrew Kensler, Pixar Technical Memo 13-01, 2013 */
/* todo: find good value, suggested 64 gives pattern on cornell box ceiling */
#define CMJ_RANDOM_OFFSET_LIMIT 4096
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device_inline bool cmj_is_pow2(int i)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
return (i > 1) && ((i & (i - 1)) == 0);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device_inline int cmj_fast_mod_pow2(int a, int b)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
return (a & (b - 1));
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
Fix T46671: Cycles assert with CMJ sample function With current formulation of cmj_fast_div_pow2() it should always return 0 in the case of first argument is zero and no assert really needed anymore.
2015-11-03 18:44:33 +05:00
/* b must be > 1 */
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device_inline int cmj_fast_div_pow2(int a, int b)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
kernel_assert(b > 1);
Cleanup: add zero bit counting functions
2019-08-26 15:08:46 +02:00
return a >> count_trailing_zeros(b);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device_inline uint cmj_w_mask(uint w)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
kernel_assert(w > 1);
Cleanup: add zero bit counting functions
2019-08-26 15:08:46 +02:00
return ((1 << (32 - count_leading_zeros(w))) - 1);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device_inline uint cmj_permute(uint i, uint l, uint p)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
uint w = l - 1;
if ((l & w) == 0) {
/* l is a power of two (fast) */
i ^= p;
i *= 0xe170893d;
i ^= p >> 16;
i ^= (i & w) >> 4;
i ^= p >> 8;
i *= 0x0929eb3f;
i ^= p >> 23;
i ^= (i & w) >> 1;
i *= 1 | p >> 27;
i *= 0x6935fa69;
i ^= (i & w) >> 11;
i *= 0x74dcb303;
i ^= (i & w) >> 2;
i *= 0x9e501cc3;
i ^= (i & w) >> 2;
i *= 0xc860a3df;
i &= w;
i ^= i >> 5;
return (i + p) & w;
}
else {
/* l is not a power of two (slow) */
w = cmj_w_mask(w);
do {
i ^= p;
i *= 0xe170893d;
i ^= p >> 16;
i ^= (i & w) >> 4;
i ^= p >> 8;
i *= 0x0929eb3f;
i ^= p >> 23;
i ^= (i & w) >> 1;
i *= 1 | p >> 27;
i *= 0x6935fa69;
i ^= (i & w) >> 11;
i *= 0x74dcb303;
i ^= (i & w) >> 2;
i *= 0x9e501cc3;
i ^= (i & w) >> 2;
i *= 0xc860a3df;
i &= w;
i ^= i >> 5;
} while (i >= l);
return (i + p) % l;
}
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device_inline uint cmj_hash(uint i, uint p)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
i ^= p;
i ^= i >> 17;
i ^= i >> 10;
i *= 0xb36534e5;
i ^= i >> 12;
i ^= i >> 21;
i *= 0x93fc4795;
i ^= 0xdf6e307f;
i ^= i >> 17;
i *= 1 | p >> 18;
return i;
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
Cycles: Fix correlation issues in certain cases There were two cases where correlation issues were obvious: - File from T38710 was giving issues in 2.78a again - File from T50116 was having totally different shadow between sample 1 and sample 32. Use some more simplified version of CMJ hash which seems to give nice randomized value which solves the correlation. This commit will break all unit test files, but it's a bug fix so perhaps OK to commit this. This also fixes T41143: Sobol gives nonuniform noise Proper science paper about hash function is coming. Reviewers: brecht Reviewed By: brecht Subscribers: lukasstockner97 Differential Revision: https://developer.blender.org/D2385
2016-11-26 17:30:02 +01:00
ccl_device_inline uint cmj_hash_simple(uint i, uint p)
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
i = (i ^ 61) ^ p;
i += i << 3;
i ^= i >> 4;
i *= 0x27d4eb2d;
return i;
Cycles: Fix correlation issues in certain cases There were two cases where correlation issues were obvious: - File from T38710 was giving issues in 2.78a again - File from T50116 was having totally different shadow between sample 1 and sample 32. Use some more simplified version of CMJ hash which seems to give nice randomized value which solves the correlation. This commit will break all unit test files, but it's a bug fix so perhaps OK to commit this. This also fixes T41143: Sobol gives nonuniform noise Proper science paper about hash function is coming. Reviewers: brecht Reviewed By: brecht Subscribers: lukasstockner97 Differential Revision: https://developer.blender.org/D2385
2016-11-26 17:30:02 +01:00
}
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device_inline float cmj_randfloat(uint i, uint p)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
return cmj_hash(i, p) * (1.0f / 4294967808.0f);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
#ifdef __CMJ__
Cycles: change __device and similar qualifiers to ccl_device in kernel code. This to avoids build conflicts with libc++ on FreeBSD, these __ prefixed values are reserved for compilers. I apologize to anyone who has patches or branches and has to go through the pain of merging this change, it may be easiest to do these same replacements in your code and then apply/merge the patch. Ref T37477.
2013-11-16 00:17:10 +01:00
ccl_device float cmj_sample_1D(int s, int N, int p)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
kernel_assert(s < N);
Fix #36545: crash with branched path tracing, correlated multi-jittered sampling and subsurface scattering.
2013-08-23 23:04:50 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
uint x = cmj_permute(s, N, p * 0x68bc21eb);
float jx = cmj_randfloat(s, p * 0x967a889b);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
float invN = 1.0f / N;
return (x + jx) * invN;
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
Cycles: Use more stable version of integer square root function Old code was working quite unreliable in combination with fast math flag, especially when compiling with Clang. It seems we were hitting result of the following bug submitted to Clang [1]. Basically, it was happening so that (int)sqrtf(64) was 7 when Cycles is built with Clang but was correct 8 when built with GCC. This commit works this around. Annoying, but don't see other way to keep sampling pattern the same for Clang and GCC. [1] https://bugs.llvm.org//show_bug.cgi?id=24063
2017-05-09 16:12:06 +02:00
/* TODO(sergey): Do some extra tests and consider moving to util_math.h. */
ccl_device_inline int cmj_isqrt(int value)
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
# if defined(__KERNEL_CUDA__)
return float_to_int(__fsqrt_ru(value));
# elif defined(__KERNEL_GPU__)
return float_to_int(sqrtf(value));
# else
/* This is a work around for fast-math on CPU which might replace sqrtf()
* with am approximated version.
*/
return float_to_int(sqrtf(value) + 1e-6f);
# endif
Cycles: Use more stable version of integer square root function Old code was working quite unreliable in combination with fast math flag, especially when compiling with Clang. It seems we were hitting result of the following bug submitted to Clang [1]. Basically, it was happening so that (int)sqrtf(64) was 7 when Cycles is built with Clang but was correct 8 when built with GCC. This commit works this around. Annoying, but don't see other way to keep sampling pattern the same for Clang and GCC. [1] https://bugs.llvm.org//show_bug.cgi?id=24063
2017-05-09 16:12:06 +02:00
}
ccl_device void cmj_sample_2D(int s, int N, int p, float *fx, float *fy)
{
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
kernel_assert(s < N);
Cycles: Use more stable version of integer square root function Old code was working quite unreliable in combination with fast math flag, especially when compiling with Clang. It seems we were hitting result of the following bug submitted to Clang [1]. Basically, it was happening so that (int)sqrtf(64) was 7 when Cycles is built with Clang but was correct 8 when built with GCC. This commit works this around. Annoying, but don't see other way to keep sampling pattern the same for Clang and GCC. [1] https://bugs.llvm.org//show_bug.cgi?id=24063
2017-05-09 16:12:06 +02:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
int m = cmj_isqrt(N);
int n = (N - 1) / m + 1;
float invN = 1.0f / N;
float invm = 1.0f / m;
float invn = 1.0f / n;
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
s = cmj_permute(s, N, p * 0x51633e2d);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
int sdivm, smodm;
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
if (cmj_is_pow2(m)) {
sdivm = cmj_fast_div_pow2(s, m);
smodm = cmj_fast_mod_pow2(s, m);
}
else {
/* Doing s*inmv gives precision issues here. */
sdivm = s / m;
smodm = s - sdivm * m;
}
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
uint sx = cmj_permute(smodm, m, p * 0x68bc21eb);
uint sy = cmj_permute(sdivm, n, p * 0x02e5be93);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
float jx = cmj_randfloat(s, p * 0x967a889b);
float jy = cmj_randfloat(s, p * 0x368cc8b7);
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
ClangFormat: apply to source, most of intern Apply clang format as proposed in T53211. For details on usage and instructions for migrating branches without conflicts, see: https://wiki.blender.org/wiki/Tools/ClangFormat
2019-04-17 06:17:24 +02:00
*fx = (sx + (sy + jx) * invn) * invm;
*fy = (s + jy) * invN;
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
}
#endif
Cycles: add Progressive Multi-Jitter sampling pattern This sampling pattern is particularly suited to adaptive sampling, and will be used for that upcoming feature. Based on "Progressive Multi-Jittered Sample Sequences" by Per Christensen, Andrew Kensler and Charlie Kilpatrick. Ref D4686
2020-03-02 15:12:44 +01:00
ccl_device float pmj_sample_1D(KernelGlobals *kg, int sample, int rng_hash, int dimension)
{
/* Fallback to random */
Fix T74537: Fixed out of bounds memory access in Cycles' PMJ sampler.
2020-03-08 21:31:47 +01:00
if (sample >= NUM_PMJ_SAMPLES) {
Cycles: add Progressive Multi-Jitter sampling pattern This sampling pattern is particularly suited to adaptive sampling, and will be used for that upcoming feature. Based on "Progressive Multi-Jittered Sample Sequences" by Per Christensen, Andrew Kensler and Charlie Kilpatrick. Ref D4686
2020-03-02 15:12:44 +01:00
int p = rng_hash + dimension;
return cmj_randfloat(sample, p);
}
uint tmp_rng = cmj_hash_simple(dimension, rng_hash);
int index = ((dimension % NUM_PMJ_PATTERNS) * NUM_PMJ_SAMPLES + sample) * 2;
return __uint_as_float(kernel_tex_fetch(__sample_pattern_lut, index) ^ (tmp_rng & 0x007fffff)) -
1.0f;
}
ccl_device void pmj_sample_2D(
KernelGlobals *kg, int sample, int rng_hash, int dimension, float *fx, float *fy)
{
Fix T74537: Fixed out of bounds memory access in Cycles' PMJ sampler.
2020-03-08 21:31:47 +01:00
if (sample >= NUM_PMJ_SAMPLES) {
Cycles: add Progressive Multi-Jitter sampling pattern This sampling pattern is particularly suited to adaptive sampling, and will be used for that upcoming feature. Based on "Progressive Multi-Jittered Sample Sequences" by Per Christensen, Andrew Kensler and Charlie Kilpatrick. Ref D4686
2020-03-02 15:12:44 +01:00
int p = rng_hash + dimension;
*fx = cmj_randfloat(sample, p);
*fy = cmj_randfloat(sample, p + 1);
Fix T74537: Fixed out of bounds memory access in Cycles' PMJ sampler.
2020-03-08 21:31:47 +01:00
return;
Cycles: add Progressive Multi-Jitter sampling pattern This sampling pattern is particularly suited to adaptive sampling, and will be used for that upcoming feature. Based on "Progressive Multi-Jittered Sample Sequences" by Per Christensen, Andrew Kensler and Charlie Kilpatrick. Ref D4686
2020-03-02 15:12:44 +01:00
}
uint tmp_rng = cmj_hash_simple(dimension, rng_hash);
int index = ((dimension % NUM_PMJ_PATTERNS) * NUM_PMJ_SAMPLES + sample) * 2;
*fx = __uint_as_float(kernel_tex_fetch(__sample_pattern_lut, index) ^ (tmp_rng & 0x007fffff)) -
1.0f;
tmp_rng = cmj_hash_simple(dimension + 1, rng_hash);
*fy = __uint_as_float(kernel_tex_fetch(__sample_pattern_lut, index + 1) ^
(tmp_rng & 0x007fffff)) -
1.0f;
}
Cycles: experimental correlated multi-jittered sampling pattern that can be used instead of sobol. So far one doesn't seem to be consistently better or worse than the other for the same number of samples but more testing is needed. The random number generator itself is slower than sobol for most number of samples, except 16, 64, 256, .. because they can be computed faster. This can probably be optimized, but we can do that when/if this actually turns out to be useful. Paper this implementation is based on: http://graphics.pixar.com/library/MultiJitteredSampling/ Also includes some refactoring of RNG code, fixing a Sobol correlation issue with the first BSDF and < 16 samples, skipping some unneeded RNG calls and using a simpler unit square to unit disk function.
2013-06-07 16:06:22 +00:00
CCL_NAMESPACE_END
Copy Permalink