170 lines
5.4 KiB
HLSL
170 lines
5.4 KiB
HLSL
#ifndef _SAMPLING_SOBOLSAMPLING_HLSL_
|
|
#define _SAMPLING_SOBOLSAMPLING_HLSL_
|
|
|
|
#define SOBOL_MATRIX_SIZE 52
|
|
#define SOBOL_MATRICES_COUNT 1024
|
|
|
|
#include "SamplingResources.hlsl"
|
|
#include "Hashes.hlsl"
|
|
#include "Common.hlsl"
|
|
|
|
// HLSLcc cannot correctly translate `reversebits(x)` for large unsigned integers.
|
|
// Therefore, when using HLSLcc, we use our own implementation. https://jira.unity3d.com/browse/GFXFEAT-629
|
|
#ifdef UNITY_COMPILER_HLSLCC
|
|
uint ReverseBitsSafe(uint x)
|
|
{
|
|
x = ((x >> 1) & 0x55555555u) | ((x & 0x55555555u) << 1);
|
|
x = ((x >> 2) & 0x33333333u) | ((x & 0x33333333u) << 2);
|
|
x = ((x >> 4) & 0x0f0f0f0fu) | ((x & 0x0f0f0f0fu) << 4);
|
|
x = ((x >> 8) & 0x00ff00ffu) | ((x & 0x00ff00ffu) << 8);
|
|
x = ((x >> 16) & 0xffffu) | ((x & 0xffffu) << 16);
|
|
return x;
|
|
}
|
|
#else
|
|
#define ReverseBitsSafe reversebits
|
|
#endif
|
|
|
|
// See https://psychopath.io/post/2021_01_30_building_a_better_lk_hash
|
|
uint LaineKarrasPermutation(uint x, uint seed)
|
|
{
|
|
x ^= x * 0x3d20adea;
|
|
x += seed;
|
|
x *= (seed >> 16) | 1;
|
|
x ^= x * 0x05526c56;
|
|
x ^= x * 0x53a22864;
|
|
return x;
|
|
}
|
|
|
|
uint NestedUniformOwenScramble(uint x, uint seed)
|
|
{
|
|
x = ReverseBitsSafe(x);
|
|
x = LaineKarrasPermutation(x, seed);
|
|
x = ReverseBitsSafe(x);
|
|
return x;
|
|
}
|
|
|
|
//See https://psychopath.io/post/2022_08_14_a_fast_hash_for_base_4_owen_scrambling
|
|
uint LaineKarrasStylePermutationBase4(uint x, uint seed)
|
|
{
|
|
x ^= x * 0x3d20adeau;
|
|
x ^= (x >> 1) & (x << 1) & 0x55555555u;
|
|
x += seed;
|
|
x *= (seed >> 16) | 1;
|
|
x ^= (x >> 1) & (x << 1) & 0x55555555u;
|
|
x ^= x * 0x05526c56u;
|
|
x ^= x * 0x53a22864u;
|
|
return x;
|
|
}
|
|
|
|
uint NestedUniformScrambleBase4(uint x, uint seed)
|
|
{
|
|
x = ReverseBitsSafe(x);
|
|
x = LaineKarrasStylePermutationBase4(x, seed);
|
|
x = ReverseBitsSafe(x);
|
|
return x;
|
|
}
|
|
|
|
// "Insert" a 0 bit after each of the 16 low bits of x.
|
|
// Ref: https://fgiesen.wordpress.com/2009/12/13/decoding-morton-codes/
|
|
uint Part1By1(uint x)
|
|
{
|
|
x &= 0x0000ffff; // x = ---- ---- ---- ---- fedc ba98 7654 3210
|
|
x = (x ^ (x << 8)) & 0x00ff00ff; // x = ---- ---- fedc ba98 ---- ---- 7654 3210
|
|
x = (x ^ (x << 4)) & 0x0f0f0f0f; // x = ---- fedc ---- ba98 ---- 7654 ---- 3210
|
|
x = (x ^ (x << 2)) & 0x33333333; // x = --fe --dc --ba --98 --76 --54 --32 --10
|
|
x = (x ^ (x << 1)) & 0x55555555; // x = -f-e -d-c -b-a -9-8 -7-6 -5-4 -3-2 -1-0
|
|
return x;
|
|
}
|
|
|
|
uint EncodeMorton2D(uint2 coord)
|
|
{
|
|
return (Part1By1(coord.y) << 1) + Part1By1(coord.x);
|
|
}
|
|
|
|
uint SobolSampleUint(uint index, int dimension, uint indexMaxBitCount)
|
|
{
|
|
uint result = 0;
|
|
|
|
#ifdef UNIFIED_RT_BACKEND_HARDWARE
|
|
// Raytracing tracing shaders compile only with unrolled loop, but unrolling fails in the compute path.
|
|
// TODO: test again when Unity updates the shader compiler
|
|
[unroll]
|
|
#endif
|
|
for (uint i = dimension * SOBOL_MATRIX_SIZE; i < dimension * SOBOL_MATRIX_SIZE + indexMaxBitCount; index >>= 1, ++i)
|
|
{
|
|
result ^= _SobolMatricesBuffer[i] * (index & 1);
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
uint SobolSampleUint(uint2 index64, uint dimension)
|
|
{
|
|
uint result = 0;
|
|
uint i = dimension * SOBOL_MATRIX_SIZE;
|
|
|
|
for (; i < dimension * SOBOL_MATRIX_SIZE + 32; index64.x >>= 1, ++i)
|
|
{
|
|
result ^= _SobolMatricesBuffer[i] * (index64.x & 1);
|
|
}
|
|
|
|
for (; i < dimension * SOBOL_MATRIX_SIZE + 32+20; index64.y >>= 1, ++i)
|
|
{
|
|
result ^= _SobolMatricesBuffer[i] * (index64.y & 1);
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
uint GetMortonShuffledSampleIndex(uint index, uint pixelMortonIndex, uint mortonShuffleSeed, uint log2SamplesPerPixel)
|
|
{
|
|
uint mortonIndexShuffled = NestedUniformScrambleBase4(pixelMortonIndex, mortonShuffleSeed);
|
|
|
|
uint sampleIndexForPixel = (mortonIndexShuffled << log2SamplesPerPixel) | index;
|
|
return sampleIndexForPixel;
|
|
}
|
|
|
|
uint2 GetMortonShuffledSampleIndex64(uint index, uint pixelMortonIndex, uint mortonShuffleSeed, uint log2SamplesPerPixel)
|
|
{
|
|
uint mortonIndexShuffled = NestedUniformScrambleBase4(pixelMortonIndex, mortonShuffleSeed);
|
|
|
|
uint2 sampleIndex64;
|
|
sampleIndex64.x = (mortonIndexShuffled << log2SamplesPerPixel) | index;
|
|
sampleIndex64.y = mortonIndexShuffled >> (32 - log2SamplesPerPixel);
|
|
return sampleIndex64;
|
|
}
|
|
|
|
float GetSobolSample(uint index, int dimension)
|
|
{
|
|
uint result = SobolSampleUint(index, dimension, 20);
|
|
return UintToFloat01(result);
|
|
}
|
|
|
|
float GetOwenScrambledSobolSample(uint sampleIndex, uint dim, uint valueScrambleSeed)
|
|
{
|
|
uint sobolUInt = SobolSampleUint(sampleIndex, dim, 20);
|
|
uint result = NestedUniformOwenScramble(sobolUInt, valueScrambleSeed);
|
|
return UintToFloat01(result);
|
|
}
|
|
|
|
float GetOwenScrambledZShuffledSobolSample(uint sampleIndex, uint dim, uint maxDim, uint pixelMortonCode, uint log2SamplesPerPixel)
|
|
{
|
|
const uint kSeedValueScramble = 0xab773au;
|
|
uint mortonShuffleSeed = LowBiasHash32(dim, 0);
|
|
uint valueScrambleSeed = LowBiasHash32(dim, kSeedValueScramble);
|
|
|
|
#ifdef QRNG_GLOBAL_SOBOL_ENHANCED_TILING
|
|
uint2 shuffledSampleIndex = GetMortonShuffledSampleIndex64(sampleIndex, pixelMortonCode, mortonShuffleSeed, log2SamplesPerPixel);
|
|
uint sobolUInt = SobolSampleUint(shuffledSampleIndex, dim % maxDim);
|
|
#else
|
|
uint shuffledSampleIndex = GetMortonShuffledSampleIndex(sampleIndex, pixelMortonCode, mortonShuffleSeed, log2SamplesPerPixel);
|
|
uint sobolUInt = SobolSampleUint(shuffledSampleIndex, dim % maxDim, 32);
|
|
#endif
|
|
|
|
uint result = NestedUniformOwenScramble(sobolUInt, valueScrambleSeed);
|
|
return UintToFloat01(result);
|
|
}
|
|
|
|
|
|
#endif // UNITY_SOBOL_SAMPLING_INCLUDED
|