diff options
Diffstat (limited to 'intern/cycles/kernel/integrator/shade_volume.h')
| -rw-r--r-- | intern/cycles/kernel/integrator/shade_volume.h | 279 |
1 files changed, 139 insertions, 140 deletions
diff --git a/intern/cycles/kernel/integrator/shade_volume.h b/intern/cycles/kernel/integrator/shade_volume.h index 4a5015946aa..aaef92729d6 100644 --- a/intern/cycles/kernel/integrator/shade_volume.h +++ b/intern/cycles/kernel/integrator/shade_volume.h @@ -3,12 +3,13 @@ #pragma once -#include "kernel/film/accumulate.h" -#include "kernel/film/passes.h" +#include "kernel/film/data_passes.h" +#include "kernel/film/denoising_passes.h" +#include "kernel/film/light_passes.h" #include "kernel/integrator/intersect_closest.h" #include "kernel/integrator/path_state.h" -#include "kernel/integrator/shader_eval.h" +#include "kernel/integrator/volume_shader.h" #include "kernel/integrator/volume_stack.h" #include "kernel/light/light.h" @@ -29,13 +30,13 @@ typedef enum VolumeIntegrateEvent { typedef struct VolumeIntegrateResult { /* Throughput and offset for direct light scattering. */ bool direct_scatter; - float3 direct_throughput; + Spectrum direct_throughput; float direct_t; ShaderVolumePhases direct_phases; /* Throughput and offset for indirect light scattering. */ bool indirect_scatter; - float3 indirect_throughput; + Spectrum indirect_throughput; float indirect_t; ShaderVolumePhases indirect_phases; } VolumeIntegrateResult; @@ -52,19 +53,19 @@ typedef struct VolumeIntegrateResult { * sigma_t = sigma_a + sigma_s */ typedef struct VolumeShaderCoefficients { - float3 sigma_t; - float3 sigma_s; - float3 emission; + Spectrum sigma_t; + Spectrum sigma_s; + Spectrum emission; } VolumeShaderCoefficients; /* Evaluate shader to get extinction coefficient at P. */ ccl_device_inline bool shadow_volume_shader_sample(KernelGlobals kg, IntegratorShadowState state, ccl_private ShaderData *ccl_restrict sd, - ccl_private float3 *ccl_restrict extinction) + ccl_private Spectrum *ccl_restrict extinction) { VOLUME_READ_LAMBDA(integrator_state_read_shadow_volume_stack(state, i)) - shader_eval_volume<true>(kg, state, sd, PATH_RAY_SHADOW, volume_read_lambda_pass); + volume_shader_eval<true>(kg, state, sd, PATH_RAY_SHADOW, volume_read_lambda_pass); if (!(sd->flag & SD_EXTINCTION)) { return false; @@ -83,15 +84,16 @@ ccl_device_inline bool volume_shader_sample(KernelGlobals kg, { const uint32_t path_flag = INTEGRATOR_STATE(state, path, flag); VOLUME_READ_LAMBDA(integrator_state_read_volume_stack(state, i)) - shader_eval_volume<false>(kg, state, sd, path_flag, volume_read_lambda_pass); + volume_shader_eval<false>(kg, state, sd, path_flag, volume_read_lambda_pass); if (!(sd->flag & (SD_EXTINCTION | SD_SCATTER | SD_EMISSION))) { return false; } - coeff->sigma_s = zero_float3(); - coeff->sigma_t = (sd->flag & SD_EXTINCTION) ? sd->closure_transparent_extinction : zero_float3(); - coeff->emission = (sd->flag & SD_EMISSION) ? sd->closure_emission_background : zero_float3(); + coeff->sigma_s = zero_spectrum(); + coeff->sigma_t = (sd->flag & SD_EXTINCTION) ? sd->closure_transparent_extinction : + zero_spectrum(); + coeff->emission = (sd->flag & SD_EMISSION) ? sd->closure_emission_background : zero_spectrum(); if (sd->flag & SD_SCATTER) { for (int i = 0; i < sd->num_closure; i++) { @@ -114,7 +116,8 @@ ccl_device_inline bool volume_shader_sample(KernelGlobals kg, ccl_device_forceinline void volume_step_init(KernelGlobals kg, ccl_private const RNGState *rng_state, const float object_step_size, - float t, + const float tmin, + const float tmax, ccl_private float *step_size, ccl_private float *step_shade_offset, ccl_private float *steps_offset, @@ -122,7 +125,7 @@ ccl_device_forceinline void volume_step_init(KernelGlobals kg, { if (object_step_size == FLT_MAX) { /* Homogeneous volume. */ - *step_size = t; + *step_size = tmax - tmin; *step_shade_offset = 0.0f; *steps_offset = 1.0f; *max_steps = 1; @@ -130,6 +133,7 @@ ccl_device_forceinline void volume_step_init(KernelGlobals kg, else { /* Heterogeneous volume. */ *max_steps = kernel_data.integrator.volume_max_steps; + const float t = tmax - tmin; float step = min(object_step_size, t); /* compute exact steps in advance for malloc */ @@ -141,11 +145,11 @@ ccl_device_forceinline void volume_step_init(KernelGlobals kg, /* Perform shading at this offset within a step, to integrate over * over the entire step segment. */ - *step_shade_offset = path_state_rng_1D_hash(kg, rng_state, 0x1e31d8a4); + *step_shade_offset = path_state_rng_1D(kg, rng_state, PRNG_VOLUME_SHADE_OFFSET); /* Shift starting point of all segment by this random amount to avoid * banding artifacts from the volume bounding shape. */ - *steps_offset = path_state_rng_1D_hash(kg, rng_state, 0x3d22c7b3); + *steps_offset = path_state_rng_1D(kg, rng_state, PRNG_VOLUME_OFFSET); } } @@ -160,12 +164,12 @@ ccl_device_forceinline void volume_step_init(KernelGlobals kg, ccl_device void volume_shadow_homogeneous(KernelGlobals kg, IntegratorState state, ccl_private Ray *ccl_restrict ray, ccl_private ShaderData *ccl_restrict sd, - ccl_global float3 *ccl_restrict throughput) + ccl_global Spectrum *ccl_restrict throughput) { - float3 sigma_t = zero_float3(); + Spectrum sigma_t = zero_spectrum(); if (shadow_volume_shader_sample(kg, state, sd, &sigma_t)) { - *throughput *= volume_color_transmittance(sigma_t, ray->t); + *throughput *= volume_color_transmittance(sigma_t, ray->tmax - ray->tmin); } } # endif @@ -176,14 +180,14 @@ ccl_device void volume_shadow_heterogeneous(KernelGlobals kg, IntegratorShadowState state, ccl_private Ray *ccl_restrict ray, ccl_private ShaderData *ccl_restrict sd, - ccl_private float3 *ccl_restrict throughput, + ccl_private Spectrum *ccl_restrict throughput, const float object_step_size) { /* Load random number state. */ RNGState rng_state; shadow_path_state_rng_load(state, &rng_state); - float3 tp = *throughput; + Spectrum tp = *throughput; /* Prepare for stepping. * For shadows we do not offset all segments, since the starting point is @@ -194,7 +198,8 @@ ccl_device void volume_shadow_heterogeneous(KernelGlobals kg, volume_step_init(kg, &rng_state, object_step_size, - ray->t, + ray->tmin, + ray->tmax, &step_size, &step_shade_offset, &unused, @@ -202,17 +207,17 @@ ccl_device void volume_shadow_heterogeneous(KernelGlobals kg, const float steps_offset = 1.0f; /* compute extinction at the start */ - float t = 0.0f; + float t = ray->tmin; - float3 sum = zero_float3(); + Spectrum sum = zero_spectrum(); for (int i = 0; i < max_steps; i++) { /* advance to new position */ - float new_t = min(ray->t, (i + steps_offset) * step_size); + float new_t = min(ray->tmax, ray->tmin + (i + steps_offset) * step_size); float dt = new_t - t; float3 new_P = ray->P + ray->D * (t + dt * step_shade_offset); - float3 sigma_t = zero_float3(); + Spectrum sigma_t = zero_spectrum(); /* compute attenuation over segment */ sd->P = new_P; @@ -222,20 +227,19 @@ ccl_device void volume_shadow_heterogeneous(KernelGlobals kg, * check then. */ sum += (-sigma_t * dt); if ((i & 0x07) == 0) { /* TODO: Other interval? */ - tp = *throughput * exp3(sum); + tp = *throughput * exp(sum); /* stop if nearly all light is blocked */ - if (tp.x < VOLUME_THROUGHPUT_EPSILON && tp.y < VOLUME_THROUGHPUT_EPSILON && - tp.z < VOLUME_THROUGHPUT_EPSILON) + if (reduce_max(tp) < VOLUME_THROUGHPUT_EPSILON) break; } } /* stop if at the end of the volume */ t = new_t; - if (t == ray->t) { + if (t == ray->tmax) { /* Update throughput in case we haven't done it above */ - tp = *throughput * exp3(sum); + tp = *throughput * exp(sum); break; } } @@ -257,15 +261,16 @@ ccl_device float volume_equiangular_sample(ccl_private const Ray *ccl_restrict r const float xi, ccl_private float *pdf) { - const float t = ray->t; + const float tmin = ray->tmin; + const float tmax = ray->tmax; const float delta = dot((light_P - ray->P), ray->D); const float D = safe_sqrtf(len_squared(light_P - ray->P) - delta * delta); if (UNLIKELY(D == 0.0f)) { *pdf = 0.0f; return 0.0f; } - const float theta_a = -atan2f(delta, D); - const float theta_b = atan2f(t - delta, D); + const float theta_a = atan2f(tmin - delta, D); + const float theta_b = atan2f(tmax - delta, D); const float t_ = D * tanf((xi * theta_b) + (1 - xi) * theta_a); if (UNLIKELY(theta_b == theta_a)) { *pdf = 0.0f; @@ -273,7 +278,7 @@ ccl_device float volume_equiangular_sample(ccl_private const Ray *ccl_restrict r } *pdf = D / ((theta_b - theta_a) * (D * D + t_ * t_)); - return min(t, delta + t_); /* min is only for float precision errors */ + return clamp(delta + t_, tmin, tmax); /* clamp is only for float precision errors */ } ccl_device float volume_equiangular_pdf(ccl_private const Ray *ccl_restrict ray, @@ -286,11 +291,12 @@ ccl_device float volume_equiangular_pdf(ccl_private const Ray *ccl_restrict ray, return 0.0f; } - const float t = ray->t; + const float tmin = ray->tmin; + const float tmax = ray->tmax; const float t_ = sample_t - delta; - const float theta_a = -atan2f(delta, D); - const float theta_b = atan2f(t - delta, D); + const float theta_a = atan2f(tmin - delta, D); + const float theta_b = atan2f(tmax - delta, D); if (UNLIKELY(theta_b == theta_a)) { return 0.0f; } @@ -310,11 +316,12 @@ ccl_device float volume_equiangular_cdf(ccl_private const Ray *ccl_restrict ray, return 0.0f; } - const float t = ray->t; + const float tmin = ray->tmin; + const float tmax = ray->tmax; const float t_ = sample_t - delta; - const float theta_a = -atan2f(delta, D); - const float theta_b = atan2f(t - delta, D); + const float theta_a = atan2f(tmin - delta, D); + const float theta_b = atan2f(tmax - delta, D); if (UNLIKELY(theta_b == theta_a)) { return 0.0f; } @@ -328,22 +335,22 @@ ccl_device float volume_equiangular_cdf(ccl_private const Ray *ccl_restrict ray, /* Distance sampling */ ccl_device float volume_distance_sample(float max_t, - float3 sigma_t, + Spectrum sigma_t, int channel, float xi, - ccl_private float3 *transmittance, - ccl_private float3 *pdf) + ccl_private Spectrum *transmittance, + ccl_private Spectrum *pdf) { /* xi is [0, 1[ so log(0) should never happen, division by zero is * avoided because sample_sigma_t > 0 when SD_SCATTER is set */ float sample_sigma_t = volume_channel_get(sigma_t, channel); - float3 full_transmittance = volume_color_transmittance(sigma_t, max_t); + Spectrum full_transmittance = volume_color_transmittance(sigma_t, max_t); float sample_transmittance = volume_channel_get(full_transmittance, channel); float sample_t = min(max_t, -logf(1.0f - xi * (1.0f - sample_transmittance)) / sample_sigma_t); *transmittance = volume_color_transmittance(sigma_t, sample_t); - *pdf = safe_divide_color(sigma_t * *transmittance, one_float3() - full_transmittance); + *pdf = safe_divide_color(sigma_t * *transmittance, one_spectrum() - full_transmittance); /* todo: optimization: when taken together with hit/miss decision, * the full_transmittance cancels out drops out and xi does not @@ -352,33 +359,36 @@ ccl_device float volume_distance_sample(float max_t, return sample_t; } -ccl_device float3 volume_distance_pdf(float max_t, float3 sigma_t, float sample_t) +ccl_device Spectrum volume_distance_pdf(float max_t, Spectrum sigma_t, float sample_t) { - float3 full_transmittance = volume_color_transmittance(sigma_t, max_t); - float3 transmittance = volume_color_transmittance(sigma_t, sample_t); + Spectrum full_transmittance = volume_color_transmittance(sigma_t, max_t); + Spectrum transmittance = volume_color_transmittance(sigma_t, sample_t); - return safe_divide_color(sigma_t * transmittance, one_float3() - full_transmittance); + return safe_divide_color(sigma_t * transmittance, one_spectrum() - full_transmittance); } /* Emission */ -ccl_device float3 volume_emission_integrate(ccl_private VolumeShaderCoefficients *coeff, - int closure_flag, - float3 transmittance, - float t) +ccl_device Spectrum volume_emission_integrate(ccl_private VolumeShaderCoefficients *coeff, + int closure_flag, + Spectrum transmittance, + float t) { /* integral E * exp(-sigma_t * t) from 0 to t = E * (1 - exp(-sigma_t * t))/sigma_t * this goes to E * t as sigma_t goes to zero * * todo: we should use an epsilon to avoid precision issues near zero sigma_t */ - float3 emission = coeff->emission; + Spectrum emission = coeff->emission; if (closure_flag & SD_EXTINCTION) { - float3 sigma_t = coeff->sigma_t; + Spectrum sigma_t = coeff->sigma_t; - emission.x *= (sigma_t.x > 0.0f) ? (1.0f - transmittance.x) / sigma_t.x : t; - emission.y *= (sigma_t.y > 0.0f) ? (1.0f - transmittance.y) / sigma_t.y : t; - emission.z *= (sigma_t.z > 0.0f) ? (1.0f - transmittance.z) / sigma_t.z : t; + FOREACH_SPECTRUM_CHANNEL (i) { + GET_SPECTRUM_CHANNEL(emission, i) *= (GET_SPECTRUM_CHANNEL(sigma_t, i) > 0.0f) ? + (1.0f - GET_SPECTRUM_CHANNEL(transmittance, i)) / + GET_SPECTRUM_CHANNEL(sigma_t, i) : + t; + } } else emission *= t; @@ -390,8 +400,8 @@ ccl_device float3 volume_emission_integrate(ccl_private VolumeShaderCoefficients typedef struct VolumeIntegrateState { /* Volume segment extents. */ - float start_t; - float end_t; + float tmin; + float tmax; /* If volume is absorption-only up to this point, and no probabilistic * scattering or termination has been used yet. */ @@ -413,27 +423,27 @@ ccl_device_forceinline void volume_integrate_step_scattering( ccl_private const Ray *ray, const float3 equiangular_light_P, ccl_private const VolumeShaderCoefficients &ccl_restrict coeff, - const float3 transmittance, + const Spectrum transmittance, ccl_private VolumeIntegrateState &ccl_restrict vstate, ccl_private VolumeIntegrateResult &ccl_restrict result) { /* Pick random color channel, we use the Veach one-sample * model with balance heuristic for the channels. */ - const float3 albedo = safe_divide_color(coeff.sigma_s, coeff.sigma_t); - float3 channel_pdf; + const Spectrum albedo = safe_divide_color(coeff.sigma_s, coeff.sigma_t); + Spectrum channel_pdf; const int channel = volume_sample_channel( albedo, result.indirect_throughput, vstate.rphase, &channel_pdf); /* Equiangular sampling for direct lighting. */ if (vstate.direct_sample_method == VOLUME_SAMPLE_EQUIANGULAR && !result.direct_scatter) { - if (result.direct_t >= vstate.start_t && result.direct_t <= vstate.end_t && + if (result.direct_t >= vstate.tmin && result.direct_t <= vstate.tmax && vstate.equiangular_pdf > VOLUME_SAMPLE_PDF_CUTOFF) { - const float new_dt = result.direct_t - vstate.start_t; - const float3 new_transmittance = volume_color_transmittance(coeff.sigma_t, new_dt); + const float new_dt = result.direct_t - vstate.tmin; + const Spectrum new_transmittance = volume_color_transmittance(coeff.sigma_t, new_dt); result.direct_scatter = true; result.direct_throughput *= coeff.sigma_s * new_transmittance / vstate.equiangular_pdf; - shader_copy_volume_phases(&result.direct_phases, sd); + volume_shader_copy_phases(&result.direct_phases, sd); /* Multiple importance sampling. */ if (vstate.use_mis) { @@ -458,10 +468,10 @@ ccl_device_forceinline void volume_integrate_step_scattering( /* compute sampling distance */ const float sample_sigma_t = volume_channel_get(coeff.sigma_t, channel); const float new_dt = -logf(1.0f - vstate.rscatter) / sample_sigma_t; - const float new_t = vstate.start_t + new_dt; + const float new_t = vstate.tmin + new_dt; /* transmittance and pdf */ - const float3 new_transmittance = volume_color_transmittance(coeff.sigma_t, new_dt); + const Spectrum new_transmittance = volume_color_transmittance(coeff.sigma_t, new_dt); const float distance_pdf = dot(channel_pdf, coeff.sigma_t * new_transmittance); if (vstate.distance_pdf * distance_pdf > VOLUME_SAMPLE_PDF_CUTOFF) { @@ -469,7 +479,7 @@ ccl_device_forceinline void volume_integrate_step_scattering( result.indirect_scatter = true; result.indirect_t = new_t; result.indirect_throughput *= coeff.sigma_s * new_transmittance / distance_pdf; - shader_copy_volume_phases(&result.indirect_phases, sd); + volume_shader_copy_phases(&result.indirect_phases, sd); if (vstate.direct_sample_method != VOLUME_SAMPLE_EQUIANGULAR) { /* If using distance sampling for direct light, just copy parameters @@ -477,7 +487,7 @@ ccl_device_forceinline void volume_integrate_step_scattering( result.direct_scatter = true; result.direct_t = result.indirect_t; result.direct_throughput = result.indirect_throughput; - shader_copy_volume_phases(&result.direct_phases, sd); + volume_shader_copy_phases(&result.direct_phases, sd); /* Multiple importance sampling. */ if (vstate.use_mis) { @@ -528,7 +538,8 @@ ccl_device_forceinline void volume_integrate_heterogeneous( volume_step_init(kg, rng_state, object_step_size, - ray->t, + ray->tmin, + ray->tmax, &step_size, &step_shade_offset, &steps_offset, @@ -536,11 +547,11 @@ ccl_device_forceinline void volume_integrate_heterogeneous( /* Initialize volume integration state. */ VolumeIntegrateState vstate ccl_optional_struct_init; - vstate.start_t = 0.0f; - vstate.end_t = 0.0f; + vstate.tmin = ray->tmin; + vstate.tmax = ray->tmin; vstate.absorption_only = true; - vstate.rscatter = path_state_rng_1D(kg, rng_state, PRNG_SCATTER_DISTANCE); - vstate.rphase = path_state_rng_1D(kg, rng_state, PRNG_PHASE_CHANNEL); + vstate.rscatter = path_state_rng_1D(kg, rng_state, PRNG_VOLUME_SCATTER_DISTANCE); + vstate.rphase = path_state_rng_1D(kg, rng_state, PRNG_VOLUME_PHASE_CHANNEL); /* Multiple importance sampling: pick between equiangular and distance sampling strategy. */ vstate.direct_sample_method = direct_sample_method; @@ -559,7 +570,7 @@ ccl_device_forceinline void volume_integrate_heterogeneous( vstate.distance_pdf = 1.0f; /* Initialize volume integration result. */ - const float3 throughput = INTEGRATOR_STATE(state, path, throughput); + const Spectrum throughput = INTEGRATOR_STATE(state, path, throughput); result.direct_throughput = throughput; result.indirect_throughput = throughput; @@ -572,14 +583,14 @@ ccl_device_forceinline void volume_integrate_heterogeneous( # ifdef __DENOISING_FEATURES__ const bool write_denoising_features = (INTEGRATOR_STATE(state, path, flag) & PATH_RAY_DENOISING_FEATURES); - float3 accum_albedo = zero_float3(); + Spectrum accum_albedo = zero_spectrum(); # endif - float3 accum_emission = zero_float3(); + Spectrum accum_emission = zero_spectrum(); for (int i = 0; i < max_steps; i++) { /* Advance to new position */ - vstate.end_t = min(ray->t, (i + steps_offset) * step_size); - const float shade_t = vstate.start_t + (vstate.end_t - vstate.start_t) * step_shade_offset; + vstate.tmax = min(ray->tmax, ray->tmin + (i + steps_offset) * step_size); + const float shade_t = vstate.tmin + (vstate.tmax - vstate.tmin) * step_shade_offset; sd->P = ray->P + ray->D * shade_t; /* compute segment */ @@ -588,17 +599,17 @@ ccl_device_forceinline void volume_integrate_heterogeneous( const int closure_flag = sd->flag; /* Evaluate transmittance over segment. */ - const float dt = (vstate.end_t - vstate.start_t); - const float3 transmittance = (closure_flag & SD_EXTINCTION) ? - volume_color_transmittance(coeff.sigma_t, dt) : - one_float3(); + const float dt = (vstate.tmax - vstate.tmin); + const Spectrum transmittance = (closure_flag & SD_EXTINCTION) ? + volume_color_transmittance(coeff.sigma_t, dt) : + one_spectrum(); /* Emission. */ if (closure_flag & SD_EMISSION) { /* Only write emission before indirect light scatter position, since we terminate * stepping at that point if we have already found a direct light scatter position. */ if (!result.indirect_scatter) { - const float3 emission = volume_emission_integrate( + const Spectrum emission = volume_emission_integrate( &coeff, closure_flag, transmittance, dt); accum_emission += result.indirect_throughput * emission; } @@ -609,8 +620,8 @@ ccl_device_forceinline void volume_integrate_heterogeneous( # ifdef __DENOISING_FEATURES__ /* Accumulate albedo for denoising features. */ if (write_denoising_features && (closure_flag & SD_SCATTER)) { - const float3 albedo = safe_divide_color(coeff.sigma_s, coeff.sigma_t); - accum_albedo += result.indirect_throughput * albedo * (one_float3() - transmittance); + const Spectrum albedo = safe_divide_color(coeff.sigma_s, coeff.sigma_t); + accum_albedo += result.indirect_throughput * albedo * (one_spectrum() - transmittance); } # endif @@ -626,13 +637,13 @@ ccl_device_forceinline void volume_integrate_heterogeneous( /* Stop if nearly all light blocked. */ if (!result.indirect_scatter) { - if (max3(result.indirect_throughput) < VOLUME_THROUGHPUT_EPSILON) { - result.indirect_throughput = zero_float3(); + if (reduce_max(result.indirect_throughput) < VOLUME_THROUGHPUT_EPSILON) { + result.indirect_throughput = zero_spectrum(); break; } } else if (!result.direct_scatter) { - if (max3(result.direct_throughput) < VOLUME_THROUGHPUT_EPSILON) { + if (reduce_max(result.direct_throughput) < VOLUME_THROUGHPUT_EPSILON) { break; } } @@ -645,28 +656,27 @@ ccl_device_forceinline void volume_integrate_heterogeneous( } /* Stop if at the end of the volume. */ - vstate.start_t = vstate.end_t; - if (vstate.start_t == ray->t) { + vstate.tmin = vstate.tmax; + if (vstate.tmin == ray->tmax) { break; } } /* Write accumulated emission. */ if (!is_zero(accum_emission)) { - kernel_accum_emission( + film_write_volume_emission( kg, state, accum_emission, render_buffer, object_lightgroup(kg, sd->object)); } # ifdef __DENOISING_FEATURES__ /* Write denoising features. */ if (write_denoising_features) { - kernel_write_denoising_features_volume( + film_write_denoising_features_volume( kg, state, accum_albedo, result.indirect_scatter, render_buffer); } # endif /* __DENOISING_FEATURES__ */ } -# ifdef __EMISSION__ /* Path tracing: sample point on light and evaluate light shader, then * queue shadow ray to be traced. */ ccl_device_forceinline bool integrate_volume_sample_light( @@ -684,11 +694,10 @@ ccl_device_forceinline bool integrate_volume_sample_light( /* Sample position on a light. */ const uint32_t path_flag = INTEGRATOR_STATE(state, path, flag); const uint bounce = INTEGRATOR_STATE(state, path, bounce); - float light_u, light_v; - path_state_rng_2D(kg, rng_state, PRNG_LIGHT_U, &light_u, &light_v); + const float2 rand_light = path_state_rng_2D(kg, rng_state, PRNG_LIGHT); if (!light_distribution_sample_from_volume_segment( - kg, light_u, light_v, sd->time, sd->P, bounce, path_flag, ls)) { + kg, rand_light.x, rand_light.y, sd->time, sd->P, bounce, path_flag, ls)) { return false; } @@ -708,7 +717,7 @@ ccl_device_forceinline void integrate_volume_direct_light( ccl_private const RNGState *ccl_restrict rng_state, const float3 P, ccl_private const ShaderVolumePhases *ccl_restrict phases, - ccl_private const float3 throughput, + ccl_private const Spectrum throughput, ccl_private LightSample *ccl_restrict ls) { PROFILING_INIT(kg, PROFILING_SHADE_VOLUME_DIRECT_LIGHT); @@ -725,11 +734,10 @@ ccl_device_forceinline void integrate_volume_direct_light( { const uint32_t path_flag = INTEGRATOR_STATE(state, path, flag); const uint bounce = INTEGRATOR_STATE(state, path, bounce); - float light_u, light_v; - path_state_rng_2D(kg, rng_state, PRNG_LIGHT_U, &light_u, &light_v); + const float2 rand_light = path_state_rng_2D(kg, rng_state, PRNG_LIGHT); if (!light_distribution_sample_from_position( - kg, light_u, light_v, sd->time, P, bounce, path_flag, ls)) { + kg, rand_light.x, rand_light.y, sd->time, P, bounce, path_flag, ls)) { return; } } @@ -746,21 +754,21 @@ ccl_device_forceinline void integrate_volume_direct_light( * non-constant light sources. */ ShaderDataTinyStorage emission_sd_storage; ccl_private ShaderData *emission_sd = AS_SHADER_DATA(&emission_sd_storage); - const float3 light_eval = light_sample_shader_eval(kg, state, emission_sd, ls, sd->time); + const Spectrum light_eval = light_sample_shader_eval(kg, state, emission_sd, ls, sd->time); if (is_zero(light_eval)) { return; } /* Evaluate BSDF. */ BsdfEval phase_eval ccl_optional_struct_init; - const float phase_pdf = shader_volume_phase_eval(kg, sd, phases, ls->D, &phase_eval); + const float phase_pdf = volume_shader_phase_eval(kg, sd, phases, ls->D, &phase_eval); if (ls->shader & SHADER_USE_MIS) { float mis_weight = light_sample_mis_weight_nee(kg, ls->pdf, phase_pdf); bsdf_eval_mul(&phase_eval, mis_weight); } - bsdf_eval_mul3(&phase_eval, light_eval / ls->pdf); + bsdf_eval_mul(&phase_eval, light_eval / ls->pdf); /* Path termination. */ const float terminate = path_state_rng_light_termination(kg, rng_state); @@ -774,8 +782,8 @@ ccl_device_forceinline void integrate_volume_direct_light( const bool is_light = light_sample_is_light(ls); /* Branch off shadow kernel. */ - INTEGRATOR_SHADOW_PATH_INIT( - shadow_state, state, DEVICE_KERNEL_INTEGRATOR_INTERSECT_SHADOW, shadow); + IntegratorShadowState shadow_state = integrator_shadow_path_init( + kg, state, DEVICE_KERNEL_INTEGRATOR_INTERSECT_SHADOW, false); /* Write shadow ray and associated state to global memory. */ integrator_state_write_shadow_ray(kg, shadow_state, &ray); @@ -789,11 +797,11 @@ ccl_device_forceinline void integrate_volume_direct_light( const uint16_t transparent_bounce = INTEGRATOR_STATE(state, path, transparent_bounce); uint32_t shadow_flag = INTEGRATOR_STATE(state, path, flag); shadow_flag |= (is_light) ? PATH_RAY_SHADOW_FOR_LIGHT : 0; - const float3 throughput_phase = throughput * bsdf_eval_sum(&phase_eval); + const Spectrum throughput_phase = throughput * bsdf_eval_sum(&phase_eval); if (kernel_data.kernel_features & KERNEL_FEATURE_LIGHT_PASSES) { - packed_float3 pass_diffuse_weight; - packed_float3 pass_glossy_weight; + PackedSpectrum pass_diffuse_weight; + PackedSpectrum pass_glossy_weight; if (shadow_flag & PATH_RAY_ANY_PASS) { /* Indirect bounce, use weights from earlier surface or volume bounce. */ @@ -803,8 +811,8 @@ ccl_device_forceinline void integrate_volume_direct_light( else { /* Direct light, no diffuse/glossy distinction needed for volumes. */ shadow_flag |= PATH_RAY_VOLUME_PASS; - pass_diffuse_weight = packed_float3(one_float3()); - pass_glossy_weight = packed_float3(zero_float3()); + pass_diffuse_weight = one_spectrum(); + pass_glossy_weight = zero_spectrum(); } INTEGRATOR_STATE_WRITE(shadow_state, shadow_path, pass_diffuse_weight) = pass_diffuse_weight; @@ -842,7 +850,6 @@ ccl_device_forceinline void integrate_volume_direct_light( integrator_state_copy_volume_stack_to_shadow(kg, shadow_state, state); } -# endif /* Path tracing: scatter in new direction using phase function */ ccl_device_forceinline bool integrate_volume_phase_scatter( @@ -854,24 +861,15 @@ ccl_device_forceinline bool integrate_volume_phase_scatter( { PROFILING_INIT(kg, PROFILING_SHADE_VOLUME_INDIRECT_LIGHT); - float phase_u, phase_v; - path_state_rng_2D(kg, rng_state, PRNG_BSDF_U, &phase_u, &phase_v); + const float2 rand_phase = path_state_rng_2D(kg, rng_state, PRNG_VOLUME_PHASE); /* Phase closure, sample direction. */ float phase_pdf; BsdfEval phase_eval ccl_optional_struct_init; float3 phase_omega_in ccl_optional_struct_init; - differential3 phase_domega_in ccl_optional_struct_init; - - const int label = shader_volume_phase_sample(kg, - sd, - phases, - phase_u, - phase_v, - &phase_eval, - &phase_omega_in, - &phase_domega_in, - &phase_pdf); + + const int label = volume_shader_phase_sample( + kg, sd, phases, rand_phase, &phase_eval, &phase_omega_in, &phase_pdf); if (phase_pdf == 0.0f || bsdf_eval_is_zero(&phase_eval)) { return false; @@ -880,28 +878,27 @@ ccl_device_forceinline bool integrate_volume_phase_scatter( /* Setup ray. */ INTEGRATOR_STATE_WRITE(state, ray, P) = sd->P; INTEGRATOR_STATE_WRITE(state, ray, D) = normalize(phase_omega_in); - INTEGRATOR_STATE_WRITE(state, ray, t) = FLT_MAX; + INTEGRATOR_STATE_WRITE(state, ray, tmin) = 0.0f; + INTEGRATOR_STATE_WRITE(state, ray, tmax) = FLT_MAX; # ifdef __RAY_DIFFERENTIALS__ INTEGRATOR_STATE_WRITE(state, ray, dP) = differential_make_compact(sd->dP); - INTEGRATOR_STATE_WRITE(state, ray, dD) = differential_make_compact(phase_domega_in); # endif // Save memory by storing last hit prim and object in isect INTEGRATOR_STATE_WRITE(state, isect, prim) = sd->prim; INTEGRATOR_STATE_WRITE(state, isect, object) = sd->object; /* Update throughput. */ - const float3 throughput = INTEGRATOR_STATE(state, path, throughput); - const float3 throughput_phase = throughput * bsdf_eval_sum(&phase_eval) / phase_pdf; + const Spectrum throughput = INTEGRATOR_STATE(state, path, throughput); + const Spectrum throughput_phase = throughput * bsdf_eval_sum(&phase_eval) / phase_pdf; INTEGRATOR_STATE_WRITE(state, path, throughput) = throughput_phase; if (kernel_data.kernel_features & KERNEL_FEATURE_LIGHT_PASSES) { - INTEGRATOR_STATE_WRITE(state, path, pass_diffuse_weight) = one_float3(); - INTEGRATOR_STATE_WRITE(state, path, pass_glossy_weight) = zero_float3(); + INTEGRATOR_STATE_WRITE(state, path, pass_diffuse_weight) = one_spectrum(); + INTEGRATOR_STATE_WRITE(state, path, pass_glossy_weight) = zero_spectrum(); } /* Update path state */ INTEGRATOR_STATE_WRITE(state, path, mis_ray_pdf) = phase_pdf; - INTEGRATOR_STATE_WRITE(state, path, mis_ray_t) = 0.0f; INTEGRATOR_STATE_WRITE(state, path, min_ray_pdf) = fminf( phase_pdf, INTEGRATOR_STATE(state, path, min_ray_pdf)); @@ -1021,7 +1018,7 @@ ccl_device void integrator_shade_volume(KernelGlobals kg, integrator_state_read_isect(kg, state, &isect); /* Set ray length to current segment. */ - ray.t = (isect.prim != PRIM_NONE) ? isect.t : FLT_MAX; + ray.tmax = (isect.prim != PRIM_NONE) ? isect.t : FLT_MAX; /* Clean volume stack for background rays. */ if (isect.prim == PRIM_NONE) { @@ -1032,13 +1029,15 @@ ccl_device void integrator_shade_volume(KernelGlobals kg, if (event == VOLUME_PATH_SCATTERED) { /* Queue intersect_closest kernel. */ - INTEGRATOR_PATH_NEXT(DEVICE_KERNEL_INTEGRATOR_SHADE_VOLUME, + integrator_path_next(kg, + state, + DEVICE_KERNEL_INTEGRATOR_SHADE_VOLUME, DEVICE_KERNEL_INTEGRATOR_INTERSECT_CLOSEST); return; } else if (event == VOLUME_PATH_MISSED) { /* End path. */ - INTEGRATOR_PATH_TERMINATE(DEVICE_KERNEL_INTEGRATOR_SHADE_VOLUME); + integrator_path_terminate(kg, state, DEVICE_KERNEL_INTEGRATOR_SHADE_VOLUME); return; } else { |