Page MenuHome
Differential D9491 Diff 30842 intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h

Changeset View

Standalone View

View Options

intern/cycles/kernel/kernels/cuda/kernel_cuda_image.h

Show All 18 Lines
# include "nanovdb/util/SampleFromVoxels.h" # include "nanovdb/util/SampleFromVoxels.h"
#endif #endif
/* w0, w1, w2, and w3 are the four cubic B-spline basis functions. */ /* w0, w1, w2, and w3 are the four cubic B-spline basis functions. */
ccl_device float cubic_w0(float a) ccl_device float cubic_w0(float a)
{ {
return (1.0f / 6.0f) * (a * (a * (-a + 3.0f) - 3.0f) + 1.0f); return (1.0f / 6.0f) * (a * (a * (-a + 3.0f) - 3.0f) + 1.0f);
} }
ccl_device float cubic_w1(float a) ccl_device float cubic_w1(float a)
{ {
return (1.0f / 6.0f) * (a * a * (3.0f * a - 6.0f) + 4.0f); return (1.0f / 6.0f) * (a * a * (3.0f * a - 6.0f) + 4.0f);
} }
ccl_device float cubic_w2(float a) ccl_device float cubic_w2(float a)
{ {
return (1.0f / 6.0f) * (a * (a * (-3.0f * a + 3.0f) + 3.0f) + 1.0f); return (1.0f / 6.0f) * (a * (a * (-3.0f * a + 3.0f) + 3.0f) + 1.0f);
} }
ccl_device float cubic_w3(float a) ccl_device float cubic_w3(float a)
{ {
return (1.0f / 6.0f) * (a * a * a); return (1.0f / 6.0f) * (a * a * a);
} }
/* g0 and g1 are the two amplitude functions. */ /* g0 and g1 are the two amplitude functions. */
ccl_device float cubic_g0(float a) ccl_device float cubic_g0(float a)
{ {
return cubic_w0(a) + cubic_w1(a); return cubic_w0(a) + cubic_w1(a);
} }
ccl_device float cubic_g1(float a) ccl_device float cubic_g1(float a)
{ {
return cubic_w2(a) + cubic_w3(a); return cubic_w2(a) + cubic_w3(a);
} }
/* h0 and h1 are the two offset functions */ /* h0 and h1 are the two offset functions */
ccl_device float cubic_h0(float a) ccl_device float cubic_h0(float a)
{ {
/* Note +0.5 offset to compensate for CUDA linear filtering convention. */ return (cubic_w1(a) / cubic_g0(a)) - 1.0f;
return -1.0f + cubic_w1(a) / (cubic_w0(a) + cubic_w1(a)) + 0.5f;
} }
ccl_device float cubic_h1(float a) ccl_device float cubic_h1(float a)
{ {
return 1.0f + cubic_w3(a) / (cubic_w2(a) + cubic_w3(a)) + 0.5f; return (cubic_w3(a) / cubic_g1(a)) + 1.0f;
} }
/* Fast bicubic texture lookup using 4 bilinear lookups, adapted from CUDA samples. */ /* Fast bicubic texture lookup using 4 bilinear lookups, adapted from CUDA samples. */
template<typename T> template<typename T>
ccl_device T kernel_tex_image_interp_bicubic(const TextureInfo &info, float x, float y) ccl_device T kernel_tex_image_interp_bicubic(const TextureInfo &info, float x, float y)
{ {
CUtexObject tex = (CUtexObject)info.data; CUtexObject tex = (CUtexObject)info.data;
x = (x * info.width) - 0.5f; x = (x * info.width) - 0.5f;
y = (y * info.height) - 0.5f; y = (y * info.height) - 0.5f;
float px = floor(x); float px = floor(x);
float py = floor(y); float py = floor(y);
float fx = x - px; float fx = x - px;
float fy = y - py; float fy = y - py;
float g0x = cubic_g0(fx); float g0x = cubic_g0(fx);
float g1x = cubic_g1(fx); float g1x = cubic_g1(fx);
float x0 = (px + cubic_h0(fx)) / info.width; /* Note +0.5 offset to compensate for CUDA linear filtering convention. */
float x1 = (px + cubic_h1(fx)) / info.width; float x0 = (px + cubic_h0(fx) + 0.5f) / info.width;
float y0 = (py + cubic_h0(fy)) / info.height; float x1 = (px + cubic_h1(fx) + 0.5f) / info.width;
float y1 = (py + cubic_h1(fy)) / info.height; float y0 = (py + cubic_h0(fy) + 0.5f) / info.height;
float y1 = (py + cubic_h1(fy) + 0.5f) / info.height;
return cubic_g0(fy) * (g0x * tex2D<T>(tex, x0, y0) + g1x * tex2D<T>(tex, x1, y0)) + return cubic_g0(fy) * (g0x * tex2D<T>(tex, x0, y0) + g1x * tex2D<T>(tex, x1, y0)) +
cubic_g1(fy) * (g0x * tex2D<T>(tex, x0, y1) + g1x * tex2D<T>(tex, x1, y1)); cubic_g1(fy) * (g0x * tex2D<T>(tex, x0, y1) + g1x * tex2D<T>(tex, x1, y1));
} }
/* Fast tricubic texture lookup using 8 trilinear lookups. */ /* Fast tricubic texture lookup using 8 trilinear lookups. */
template<typename T> template<typename T>
ccl_device T kernel_tex_image_interp_bicubic_3d(const TextureInfo &info, float x, float y, float z) ccl_device T kernel_tex_image_interp_tricubic(const TextureInfo &info, float x, float y, float z)
{ {
CUtexObject tex = (CUtexObject)info.data; CUtexObject tex = (CUtexObject)info.data;
x = (x * info.width) - 0.5f; x = (x * info.width) - 0.5f;
y = (y * info.height) - 0.5f; y = (y * info.height) - 0.5f;
z = (z * info.depth) - 0.5f; z = (z * info.depth) - 0.5f;
float px = floor(x); float px = floor(x);
float py = floor(y); float py = floor(y);
float pz = floor(z); float pz = floor(z);
float fx = x - px; float fx = x - px;
float fy = y - py; float fy = y - py;
float fz = z - pz; float fz = z - pz;
float g0x = cubic_g0(fx); float g0x = cubic_g0(fx);
float g1x = cubic_g1(fx); float g1x = cubic_g1(fx);
float g0y = cubic_g0(fy); float g0y = cubic_g0(fy);
float g1y = cubic_g1(fy); float g1y = cubic_g1(fy);
float g0z = cubic_g0(fz); float g0z = cubic_g0(fz);
float g1z = cubic_g1(fz); float g1z = cubic_g1(fz);
float x0 = (px + cubic_h0(fx)) / info.width; /* Note +0.5 offset to compensate for CUDA linear filtering convention. */
float x1 = (px + cubic_h1(fx)) / info.width; float x0 = (px + cubic_h0(fx) + 0.5f) / info.width;
float y0 = (py + cubic_h0(fy)) / info.height; float x1 = (px + cubic_h1(fx) + 0.5f) / info.width;
float y1 = (py + cubic_h1(fy)) / info.height; float y0 = (py + cubic_h0(fy) + 0.5f) / info.height;
float z0 = (pz + cubic_h0(fz)) / info.depth; float y1 = (py + cubic_h1(fy) + 0.5f) / info.height;
float z1 = (pz + cubic_h1(fz)) / info.depth; float z0 = (pz + cubic_h0(fz) + 0.5f) / info.depth;
float z1 = (pz + cubic_h1(fz) + 0.5f) / info.depth;
return g0z * (g0y * (g0x * tex3D<T>(tex, x0, y0, z0) + g1x * tex3D<T>(tex, x1, y0, z0)) + return g0z * (g0y * (g0x * tex3D<T>(tex, x0, y0, z0) + g1x * tex3D<T>(tex, x1, y0, z0)) +
g1y * (g0x * tex3D<T>(tex, x0, y1, z0) + g1x * tex3D<T>(tex, x1, y1, z0))) + g1y * (g0x * tex3D<T>(tex, x0, y1, z0) + g1x * tex3D<T>(tex, x1, y1, z0))) +
g1z * (g0y * (g0x * tex3D<T>(tex, x0, y0, z1) + g1x * tex3D<T>(tex, x1, y0, z1)) + g1z * (g0y * (g0x * tex3D<T>(tex, x0, y0, z1) + g1x * tex3D<T>(tex, x1, y0, z1)) +
g1y * (g0x * tex3D<T>(tex, x0, y1, z1) + g1x * tex3D<T>(tex, x1, y1, z1))); g1y * (g0x * tex3D<T>(tex, x0, y1, z1) + g1x * tex3D<T>(tex, x1, y1, z1)));
} }
#ifdef WITH_NANOVDB #ifdef WITH_NANOVDB
template<typename T, typename S>
ccl_device T kernel_tex_image_interp_tricubic_nanovdb(S &s, float x, float y, float z)
{
float px = floor(x);
float py = floor(y);
float pz = floor(z);
float fx = x - px;
float fy = y - py;
float fz = z - pz;
float g0x = cubic_g0(fx);
float g1x = cubic_g1(fx);
float g0y = cubic_g0(fy);
float g1y = cubic_g1(fy);
float g0z = cubic_g0(fz);
float g1z = cubic_g1(fz);
float x0 = px + cubic_h0(fx);
float x1 = px + cubic_h1(fx);
float y0 = py + cubic_h0(fy);
float y1 = py + cubic_h1(fy);
float z0 = pz + cubic_h0(fz);
float z1 = pz + cubic_h1(fz);
using namespace nanovdb;
return g0z * (g0y * (g0x * s(Vec3f(x0, y0, z0)) + g1x * s(Vec3f(x1, y0, z0))) +
g1y * (g0x * s(Vec3f(x0, y1, z0)) + g1x * s(Vec3f(x1, y1, z0)))) +
g1z * (g0y * (g0x * s(Vec3f(x0, y0, z1)) + g1x * s(Vec3f(x1, y0, z1))) +
g1y * (g0x * s(Vec3f(x0, y1, z1)) + g1x * s(Vec3f(x1, y1, z1))));
}
template<typename T> template<typename T>
ccl_device_inline T kernel_tex_image_interp_nanovdb( ccl_device_inline T kernel_tex_image_interp_nanovdb(
const TextureInfo &info, float x, float y, float z, uint interpolation) const TextureInfo &info, float x, float y, float z, uint interpolation)
{ {
const nanovdb::Vec3f xyz(x, y, z); using namespace nanovdb;
nanovdb::NanoGrid<T> *const grid = (nanovdb::NanoGrid<T> *)info.data; typedef ReadAccessor<NanoRoot<T>> ReadAccessorT;
const nanovdb::NanoRoot<T> &root = grid->tree().root();
NanoGrid<T> *const grid = (NanoGrid<T> *)info.data;
const NanoRoot<T> &root = grid->tree().root();
typedef nanovdb::ReadAccessor<nanovdb::NanoRoot<T>> ReadAccessorT;
switch (interpolation) { switch (interpolation) {
case INTERPOLATION_CLOSEST: case INTERPOLATION_CLOSEST:
return nanovdb::SampleFromVoxels<ReadAccessorT, 0, false>(root)(xyz); return NearestNeighborSampler<ReadAccessorT, false>(root)(Vec3f(x, y, z));
case INTERPOLATION_LINEAR: case INTERPOLATION_LINEAR:
return nanovdb::SampleFromVoxels<ReadAccessorT, 1, false>(root)(xyz); return TrilinearSampler<ReadAccessorT, false>(root)(Vec3f(x - 0.5f, y - 0.5f, z - 0.5f));
default: default:
return nanovdb::SampleFromVoxels<ReadAccessorT, 3, false>(root)(xyz); TrilinearSampler<ReadAccessorT, false> s(root);
return kernel_tex_image_interp_tricubic_nanovdb<T>(s, x - 0.5f, y - 0.5f, z - 0.5f);
} }
} }
#endif #endif
ccl_device float4 kernel_tex_image_interp(KernelGlobals *kg, int id, float x, float y) ccl_device float4 kernel_tex_image_interp(KernelGlobals *kg, int id, float x, float y)
{ {
const TextureInfo &info = kernel_tex_fetch(__texture_info, id); const TextureInfo &info = kernel_tex_fetch(__texture_info, id);
▲ Show 20 Lines • Show All 52 Lines • ▼ Show 20 Lines if (texture_type == IMAGE_DATA_TYPE_NANOVDB_FLOAT3) {
nanovdb::Vec3f f = kernel_tex_image_interp_nanovdb<nanovdb::Vec3f>( nanovdb::Vec3f f = kernel_tex_image_interp_nanovdb<nanovdb::Vec3f>(
info, x, y, z, interpolation); info, x, y, z, interpolation);
return make_float4(f[0], f[1], f[2], 1.0f); return make_float4(f[0], f[1], f[2], 1.0f);
} }
#endif #endif
if (texture_type == IMAGE_DATA_TYPE_FLOAT4 || texture_type == IMAGE_DATA_TYPE_BYTE4 || if (texture_type == IMAGE_DATA_TYPE_FLOAT4 || texture_type == IMAGE_DATA_TYPE_BYTE4 ||
texture_type == IMAGE_DATA_TYPE_HALF4 || texture_type == IMAGE_DATA_TYPE_USHORT4) { texture_type == IMAGE_DATA_TYPE_HALF4 || texture_type == IMAGE_DATA_TYPE_USHORT4) {
if (interpolation == INTERPOLATION_CUBIC) { if (interpolation == INTERPOLATION_CUBIC) {
return kernel_tex_image_interp_bicubic_3d<float4>(info, x, y, z); return kernel_tex_image_interp_tricubic<float4>(info, x, y, z);
} }
else { else {
CUtexObject tex = (CUtexObject)info.data; CUtexObject tex = (CUtexObject)info.data;
return tex3D<float4>(tex, x, y, z); return tex3D<float4>(tex, x, y, z);
} }
} }
else { else {
float f; float f;
if (interpolation == INTERPOLATION_CUBIC) { if (interpolation == INTERPOLATION_CUBIC) {
f = kernel_tex_image_interp_bicubic_3d<float>(info, x, y, z); f = kernel_tex_image_interp_tricubic<float>(info, x, y, z);
} }
else { else {
CUtexObject tex = (CUtexObject)info.data; CUtexObject tex = (CUtexObject)info.data;
f = tex3D<float>(tex, x, y, z); f = tex3D<float>(tex, x, y, z);
} }
return make_float4(f, f, f, 1.0f); return make_float4(f, f, f, 1.0f);
} }
} }