blender/intern/cycles/kernel/kernel_light.h
Sergey Sharybin ed5df50192 Cycles: Fix/workaround for toggling world MIS causing CUDA to fail
Seems it's just another issue with the compiler, worked around by explicitly
telling not to inline some function.

In theory we can unify this with CPU, but we're quite close to the release
so better be safe than sorry.
2015-03-03 18:48:37 +05:00

668 lines
18 KiB
C++

/*
* 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.
*/
CCL_NAMESPACE_BEGIN
/* Light Sample result */
typedef struct LightSample {
float3 P; /* position on light, or direction for distant light */
float3 Ng; /* normal on light */
float3 D; /* direction from shading point to light */
float t; /* distance to light (FLT_MAX for distant light) */
float u, v; /* parametric coordinate on primitive */
float pdf; /* light sampling probability density function */
float eval_fac; /* intensity multiplier */
int object; /* object id for triangle/curve lights */
int prim; /* primitive id for triangle/curve lights */
int shader; /* shader id */
int lamp; /* lamp id */
LightType type; /* type of light */
} LightSample;
/* Background Light */
#ifdef __BACKGROUND_MIS__
/* TODO(sergey): In theory it should be all fine to use noinline for all
* devices, but we're so close to the release so better not screw things
* up for CPU at least.
*/
#ifdef __KERNEL_GPU__
ccl_device_noinline
#else
ccl_device
#endif
float3 background_light_sample(KernelGlobals *kg, float randu, float randv, float *pdf)
{
/* for the following, the CDF values are actually a pair of floats, with the
* function value as X and the actual CDF as Y. The last entry's function
* value is the CDF total. */
int res = kernel_data.integrator.pdf_background_res;
int cdf_count = res + 1;
/* this is basically std::lower_bound as used by pbrt */
int first = 0;
int count = res;
while(count > 0) {
int step = count >> 1;
int middle = first + step;
if(kernel_tex_fetch(__light_background_marginal_cdf, middle).y < randv) {
first = middle + 1;
count -= step + 1;
}
else
count = step;
}
int index_v = max(0, first - 1);
kernel_assert(index_v >= 0 && index_v < res);
float2 cdf_v = kernel_tex_fetch(__light_background_marginal_cdf, index_v);
float2 cdf_next_v = kernel_tex_fetch(__light_background_marginal_cdf, index_v + 1);
float2 cdf_last_v = kernel_tex_fetch(__light_background_marginal_cdf, res);
/* importance-sampled V direction */
float dv = (randv - cdf_v.y) / (cdf_next_v.y - cdf_v.y);
float v = (index_v + dv) / res;
/* this is basically std::lower_bound as used by pbrt */
first = 0;
count = res;
while(count > 0) {
int step = count >> 1;
int middle = first + step;
if(kernel_tex_fetch(__light_background_conditional_cdf, index_v * cdf_count + middle).y < randu) {
first = middle + 1;
count -= step + 1;
}
else
count = step;
}
int index_u = max(0, first - 1);
kernel_assert(index_u >= 0 && index_u < res);
float2 cdf_u = kernel_tex_fetch(__light_background_conditional_cdf, index_v * cdf_count + index_u);
float2 cdf_next_u = kernel_tex_fetch(__light_background_conditional_cdf, index_v * cdf_count + index_u + 1);
float2 cdf_last_u = kernel_tex_fetch(__light_background_conditional_cdf, index_v * cdf_count + res);
/* importance-sampled U direction */
float du = (randu - cdf_u.y) / (cdf_next_u.y - cdf_u.y);
float u = (index_u + du) / res;
/* compute pdf */
float denom = cdf_last_u.x * cdf_last_v.x;
float sin_theta = sinf(M_PI_F * v);
if(sin_theta == 0.0f || denom == 0.0f)
*pdf = 0.0f;
else
*pdf = (cdf_u.x * cdf_v.x)/(M_2PI_F * M_PI_F * sin_theta * denom);
*pdf *= kernel_data.integrator.pdf_lights;
/* compute direction */
return -equirectangular_to_direction(u, v);
}
/* TODO(sergey): Same as above, after the release we should consider using
* 'noinline' for all devices.
*/
#ifdef __KERNEL_GPU__
ccl_device_noinline
#else
ccl_device
#endif
float background_light_pdf(KernelGlobals *kg, float3 direction)
{
float2 uv = direction_to_equirectangular(direction);
int res = kernel_data.integrator.pdf_background_res;
float sin_theta = sinf(uv.y * M_PI_F);
if(sin_theta == 0.0f)
return 0.0f;
int index_u = clamp(float_to_int(uv.x * res), 0, res - 1);
int index_v = clamp(float_to_int(uv.y * res), 0, res - 1);
/* pdfs in V direction */
float2 cdf_last_u = kernel_tex_fetch(__light_background_conditional_cdf, index_v * (res + 1) + res);
float2 cdf_last_v = kernel_tex_fetch(__light_background_marginal_cdf, res);
float denom = cdf_last_u.x * cdf_last_v.x;
if(denom == 0.0f)
return 0.0f;
/* pdfs in U direction */
float2 cdf_u = kernel_tex_fetch(__light_background_conditional_cdf, index_v * (res + 1) + index_u);
float2 cdf_v = kernel_tex_fetch(__light_background_marginal_cdf, index_v);
float pdf = (cdf_u.x * cdf_v.x)/(M_2PI_F * M_PI_F * sin_theta * denom);
return pdf * kernel_data.integrator.pdf_lights;
}
#endif
/* Regular Light */
ccl_device float3 disk_light_sample(float3 v, float randu, float randv)
{
float3 ru, rv;
make_orthonormals(v, &ru, &rv);
to_unit_disk(&randu, &randv);
return ru*randu + rv*randv;
}
ccl_device float3 distant_light_sample(float3 D, float radius, float randu, float randv)
{
return normalize(D + disk_light_sample(D, randu, randv)*radius);
}
ccl_device float3 sphere_light_sample(float3 P, float3 center, float radius, float randu, float randv)
{
return disk_light_sample(normalize(P - center), randu, randv)*radius;
}
/* Uses the following paper:
*
* Carlos Urena et al.
* An Area-Preserving Parametrization for Spherical Rectangles.
*
* https://www.solidangle.com/research/egsr2013_spherical_rectangle.pdf
*
* Note: light_p is modified when sample_coord is true.
*/
ccl_device float area_light_sample(float3 P,
float3 *light_p,
float3 axisu, float3 axisv,
float randu, float randv,
bool sample_coord)
{
/* In our name system we're using P for the center,
* which is o in the paper.
*/
float3 corner = *light_p - axisu * 0.5f - axisv * 0.5f;
float axisu_len, axisv_len;
/* Compute local reference system R. */
float3 x = normalize_len(axisu, &axisu_len);
float3 y = normalize_len(axisv, &axisv_len);
float3 z = cross(x, y);
/* Compute rectangle coords in local reference system. */
float3 dir = corner - P;
float z0 = dot(dir, z);
/* Flip 'z' to make it point against Q. */
if(z0 > 0.0f) {
z *= -1.0f;
z0 *= -1.0f;
}
float x0 = dot(dir, x);
float y0 = dot(dir, y);
float x1 = x0 + axisu_len;
float y1 = y0 + axisv_len;
/* Create vectors to four vertices. */
float3 v00 = make_float3(x0, y0, z0);
float3 v01 = make_float3(x0, y1, z0);
float3 v10 = make_float3(x1, y0, z0);
float3 v11 = make_float3(x1, y1, z0);
/* Compute normals to edges. */
float3 n0 = normalize(cross(v00, v10));
float3 n1 = normalize(cross(v10, v11));
float3 n2 = normalize(cross(v11, v01));
float3 n3 = normalize(cross(v01, v00));
/* Compute internal angles (gamma_i). */
float g0 = safe_acosf(-dot(n0, n1));
float g1 = safe_acosf(-dot(n1, n2));
float g2 = safe_acosf(-dot(n2, n3));
float g3 = safe_acosf(-dot(n3, n0));
/* Compute predefined constants. */
float b0 = n0.z;
float b1 = n2.z;
float b0sq = b0 * b0;
float k = M_2PI_F - g2 - g3;
/* Compute solid angle from internal angles. */
float S = g0 + g1 - k;
if(sample_coord) {
/* Compute cu. */
float au = randu * S + k;
float fu = (cosf(au) * b0 - b1) / sinf(au);
float cu = 1.0f / sqrtf(fu * fu + b0sq) * (fu > 0.0f ? 1.0f : -1.0f);
cu = clamp(cu, -1.0f, 1.0f);
/* Compute xu. */
float xu = -(cu * z0) / sqrtf(1.0f - cu * cu);
xu = clamp(xu, x0, x1);
/* Compute yv. */
float z0sq = z0 * z0;
float y0sq = y0 * y0;
float y1sq = y1 * y1;
float d = sqrtf(xu * xu + z0sq);
float h0 = y0 / sqrtf(d * d + y0sq);
float h1 = y1 / sqrtf(d * d + y1sq);
float hv = h0 + randv * (h1 - h0), hv2 = hv * hv;
float yv = (hv2 < 1.0f - 1e-6f) ? (hv * d) / sqrtf(1.0f - hv2) : y1;
/* Transform (xu, yv, z0) to world coords. */
*light_p = P + xu * x + yv * y + z0 * z;
}
/* return pdf */
if(S != 0.0f)
return 1.0f / S;
else
return 0.0f;
}
ccl_device float spot_light_attenuation(float4 data1, float4 data2, LightSample *ls)
{
float3 dir = make_float3(data2.y, data2.z, data2.w);
float3 I = ls->Ng;
float spot_angle = data1.w;
float spot_smooth = data2.x;
float attenuation = dot(dir, I);
if(attenuation <= spot_angle) {
attenuation = 0.0f;
}
else {
float t = attenuation - spot_angle;
if(t < spot_smooth && spot_smooth != 0.0f)
attenuation *= smoothstepf(t/spot_smooth);
}
return attenuation;
}
ccl_device float lamp_light_pdf(KernelGlobals *kg, const float3 Ng, const float3 I, float t)
{
float cos_pi = dot(Ng, I);
if(cos_pi <= 0.0f)
return 0.0f;
return t*t/cos_pi;
}
ccl_device void lamp_light_sample(KernelGlobals *kg, int lamp,
float randu, float randv, float3 P, LightSample *ls)
{
float4 data0 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 0);
float4 data1 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 1);
LightType type = (LightType)__float_as_int(data0.x);
ls->type = type;
ls->shader = __float_as_int(data1.x);
ls->object = PRIM_NONE;
ls->prim = PRIM_NONE;
ls->lamp = lamp;
ls->u = randu;
ls->v = randv;
if(type == LIGHT_DISTANT) {
/* distant light */
float3 lightD = make_float3(data0.y, data0.z, data0.w);
float3 D = lightD;
float radius = data1.y;
float invarea = data1.w;
if(radius > 0.0f)
D = distant_light_sample(D, radius, randu, randv);
ls->P = D;
ls->Ng = D;
ls->D = -D;
ls->t = FLT_MAX;
float costheta = dot(lightD, D);
ls->pdf = invarea/(costheta*costheta*costheta);
ls->eval_fac = ls->pdf*kernel_data.integrator.inv_pdf_lights;
}
#ifdef __BACKGROUND_MIS__
else if(type == LIGHT_BACKGROUND) {
/* infinite area light (e.g. light dome or env light) */
float3 D = background_light_sample(kg, randu, randv, &ls->pdf);
ls->P = D;
ls->Ng = D;
ls->D = -D;
ls->t = FLT_MAX;
ls->eval_fac = 1.0f;
}
#endif
else {
ls->P = make_float3(data0.y, data0.z, data0.w);
if(type == LIGHT_POINT || type == LIGHT_SPOT) {
float radius = data1.y;
if(radius > 0.0f)
/* sphere light */
ls->P += sphere_light_sample(P, ls->P, radius, randu, randv);
ls->D = normalize_len(ls->P - P, &ls->t);
ls->Ng = -ls->D;
float invarea = data1.z;
ls->eval_fac = (0.25f*M_1_PI_F)*invarea;
ls->pdf = invarea;
if(type == LIGHT_SPOT) {
/* spot light attenuation */
float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
ls->eval_fac *= spot_light_attenuation(data1, data2, ls);
}
ls->pdf *= lamp_light_pdf(kg, ls->Ng, -ls->D, ls->t);
}
else {
/* area light */
float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
float4 data3 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 3);
float3 axisu = make_float3(data1.y, data1.z, data1.w);
float3 axisv = make_float3(data2.y, data2.z, data2.w);
float3 D = make_float3(data3.y, data3.z, data3.w);
ls->pdf = area_light_sample(P, &ls->P,
axisu, axisv,
randu, randv,
true);
ls->Ng = D;
ls->D = normalize_len(ls->P - P, &ls->t);
float invarea = data2.x;
ls->eval_fac = 0.25f*invarea;
if(dot(ls->D, D) > 0.0f)
ls->pdf = 0.0f;
}
ls->eval_fac *= kernel_data.integrator.inv_pdf_lights;
}
}
ccl_device bool lamp_light_eval(KernelGlobals *kg, int lamp, float3 P, float3 D, float t, LightSample *ls)
{
float4 data0 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 0);
float4 data1 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 1);
LightType type = (LightType)__float_as_int(data0.x);
ls->type = type;
ls->shader = __float_as_int(data1.x);
ls->object = PRIM_NONE;
ls->prim = PRIM_NONE;
ls->lamp = lamp;
/* todo: missing texture coordinates */
ls->u = 0.0f;
ls->v = 0.0f;
if(!(ls->shader & SHADER_USE_MIS))
return false;
if(type == LIGHT_DISTANT) {
/* distant light */
float radius = data1.y;
if(radius == 0.0f)
return false;
if(t != FLT_MAX)
return false;
/* a distant light is infinitely far away, but equivalent to a disk
* shaped light exactly 1 unit away from the current shading point.
*
* radius t^2/cos(theta)
* <----------> t = sqrt(1^2 + tan(theta)^2)
* tan(th) area = radius*radius*pi
* <----->
* \ | (1 + tan(theta)^2)/cos(theta)
* \ | (1 + tan(acos(cos(theta)))^2)/cos(theta)
* t \th| 1 simplifies to
* \-| 1/(cos(theta)^3)
* \| magic!
* P
*/
float3 lightD = make_float3(data0.y, data0.z, data0.w);
float costheta = dot(-lightD, D);
float cosangle = data1.z;
if(costheta < cosangle)
return false;
ls->P = -D;
ls->Ng = -D;
ls->D = D;
ls->t = FLT_MAX;
/* compute pdf */
float invarea = data1.w;
ls->pdf = invarea/(costheta*costheta*costheta);
ls->eval_fac = ls->pdf;
}
else if(type == LIGHT_POINT || type == LIGHT_SPOT) {
float3 lightP = make_float3(data0.y, data0.z, data0.w);
float radius = data1.y;
/* sphere light */
if(radius == 0.0f)
return false;
if(!ray_aligned_disk_intersect(P, D, t,
lightP, radius, &ls->P, &ls->t))
return false;
ls->Ng = -D;
ls->D = D;
float invarea = data1.z;
ls->eval_fac = (0.25f*M_1_PI_F)*invarea;
ls->pdf = invarea;
if(type == LIGHT_SPOT) {
/* spot light attenuation */
float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
ls->eval_fac *= spot_light_attenuation(data1, data2, ls);
if(ls->eval_fac == 0.0f)
return false;
}
/* compute pdf */
if(ls->t != FLT_MAX)
ls->pdf *= lamp_light_pdf(kg, ls->Ng, -ls->D, ls->t);
}
else if(type == LIGHT_AREA) {
/* area light */
float4 data2 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 2);
float4 data3 = kernel_tex_fetch(__light_data, lamp*LIGHT_SIZE + 3);
float invarea = data2.x;
if(invarea == 0.0f)
return false;
float3 axisu = make_float3(data1.y, data1.z, data1.w);
float3 axisv = make_float3(data2.y, data2.z, data2.w);
float3 Ng = make_float3(data3.y, data3.z, data3.w);
/* one sided */
if(dot(D, Ng) >= 0.0f)
return false;
ls->P = make_float3(data0.y, data0.z, data0.w);
if(!ray_quad_intersect(P, D, t,
ls->P, axisu, axisv, &ls->P, &ls->t))
return false;
ls->D = D;
ls->Ng = Ng;
ls->pdf = area_light_sample(P, &ls->P, axisu, axisv, 0, 0, false);
ls->eval_fac = 0.25f*invarea;
}
else
return false;
return true;
}
/* Triangle Light */
ccl_device void object_transform_light_sample(KernelGlobals *kg, LightSample *ls, int object, float time)
{
#ifdef __INSTANCING__
/* instance transform */
if(object >= 0) {
#ifdef __OBJECT_MOTION__
Transform itfm;
Transform tfm = object_fetch_transform_motion_test(kg, object, time, &itfm);
#else
Transform tfm = object_fetch_transform(kg, object, OBJECT_TRANSFORM);
#endif
ls->P = transform_point(&tfm, ls->P);
ls->Ng = normalize(transform_direction(&tfm, ls->Ng));
}
#endif
}
ccl_device void triangle_light_sample(KernelGlobals *kg, int prim, int object,
float randu, float randv, float time, LightSample *ls)
{
float u, v;
/* compute random point in triangle */
randu = sqrtf(randu);
u = 1.0f - randu;
v = randv*randu;
/* triangle, so get position, normal, shader */
triangle_point_normal(kg, object, prim, u, v, &ls->P, &ls->Ng, &ls->shader);
ls->object = object;
ls->prim = prim;
ls->lamp = LAMP_NONE;
ls->shader |= SHADER_USE_MIS;
ls->t = 0.0f;
ls->u = u;
ls->v = v;
ls->type = LIGHT_TRIANGLE;
ls->eval_fac = 1.0f;
object_transform_light_sample(kg, ls, object, time);
}
ccl_device float triangle_light_pdf(KernelGlobals *kg,
const float3 Ng, const float3 I, float t)
{
float pdf = kernel_data.integrator.pdf_triangles;
float cos_pi = fabsf(dot(Ng, I));
if(cos_pi == 0.0f)
return 0.0f;
return t*t*pdf/cos_pi;
}
/* Light Distribution */
ccl_device int light_distribution_sample(KernelGlobals *kg, float randt)
{
/* this is basically std::upper_bound as used by pbrt, to find a point light or
* triangle to emit from, proportional to area. a good improvement would be to
* also sample proportional to power, though it's not so well defined with
* OSL shaders. */
int first = 0;
int len = kernel_data.integrator.num_distribution + 1;
while(len > 0) {
int half_len = len >> 1;
int middle = first + half_len;
if(randt < kernel_tex_fetch(__light_distribution, middle).x) {
len = half_len;
}
else {
first = middle + 1;
len = len - half_len - 1;
}
}
/* clamping should not be needed but float rounding errors seem to
* make this fail on rare occasions */
return clamp(first-1, 0, kernel_data.integrator.num_distribution-1);
}
/* Generic Light */
ccl_device bool light_select_reached_max_bounces(KernelGlobals *kg, int index, int bounce)
{
float4 data4 = kernel_tex_fetch(__light_data, index*LIGHT_SIZE + 4);
return (bounce > __float_as_int(data4.x));
}
ccl_device void light_sample(KernelGlobals *kg, float randt, float randu, float randv, float time, float3 P, int bounce, LightSample *ls)
{
/* sample index */
int index = light_distribution_sample(kg, randt);
/* fetch light data */
float4 l = kernel_tex_fetch(__light_distribution, index);
int prim = __float_as_int(l.y);
if(prim >= 0) {
int object = __float_as_int(l.w);
int shader_flag = __float_as_int(l.z);
triangle_light_sample(kg, prim, object, randu, randv, time, ls);
/* compute incoming direction, distance and pdf */
ls->D = normalize_len(ls->P - P, &ls->t);
ls->pdf = triangle_light_pdf(kg, ls->Ng, -ls->D, ls->t);
ls->shader |= shader_flag;
}
else {
int lamp = -prim-1;
if(UNLIKELY(light_select_reached_max_bounces(kg, lamp, bounce))) {
ls->pdf = 0.0f;
return;
}
lamp_light_sample(kg, lamp, randu, randv, P, ls);
}
}
ccl_device int light_select_num_samples(KernelGlobals *kg, int index)
{
float4 data3 = kernel_tex_fetch(__light_data, index*LIGHT_SIZE + 3);
return __float_as_int(data3.x);
}
CCL_NAMESPACE_END