diff options
author | Brecht Van Lommel <brecht@blender.org> | 2022-08-11 17:53:11 +0300 |
---|---|---|
committer | Brecht Van Lommel <brecht@blender.org> | 2022-08-15 14:48:02 +0300 |
commit | e949d6da5bd347663addd583ff3f0211c96b81c8 (patch) | |
tree | 4a7478de4eb8fbd4c67989d94bc8e7c0353bca94 | |
parent | d8841d0aa341e05c8cb9559b116b7e2a9ec11882 (diff) |
Cycles: simplify handling of ray differentials
* Store compact ray differentials in ShaderData and compute full differentials
on demand. This reduces register pressure on the GPU.
* Remove BSDF differential code that was effectively doing nothing as the
differential orientation was discarded when making it compact.
This gives a 1-5% speedup with RTX A6000 + OptiX in our benchmarks, with the
bigger speedups in simpler scenes.
Renders appear to be identical except for the Both displacement option that
does both displacement and bump.
Differential Revision: https://developer.blender.org/D15677
35 files changed, 163 insertions, 707 deletions
diff --git a/intern/cycles/kernel/closure/bsdf.h b/intern/cycles/kernel/closure/bsdf.h index 7b28018afad..d6b7e7bfa88 100644 --- a/intern/cycles/kernel/closure/bsdf.h +++ b/intern/cycles/kernel/closure/bsdf.h @@ -105,7 +105,6 @@ ccl_device_inline int bsdf_sample(KernelGlobals kg, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private differential3 *domega_in, ccl_private float *pdf) { /* For curves use the smooth normal, particularly for ribbons the geometric @@ -115,304 +114,80 @@ ccl_device_inline int bsdf_sample(KernelGlobals kg, switch (sc->type) { case CLOSURE_BSDF_DIFFUSE_ID: - label = bsdf_diffuse_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_diffuse_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; #ifdef __SVM__ case CLOSURE_BSDF_OREN_NAYAR_ID: - label = bsdf_oren_nayar_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_oren_nayar_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; # ifdef __OSL__ case CLOSURE_BSDF_PHONG_RAMP_ID: - label = bsdf_phong_ramp_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_phong_ramp_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_DIFFUSE_RAMP_ID: - label = bsdf_diffuse_ramp_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_diffuse_ramp_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; # endif case CLOSURE_BSDF_TRANSLUCENT_ID: - label = bsdf_translucent_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_translucent_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_REFLECTION_ID: - label = bsdf_reflection_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_reflection_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_REFRACTION_ID: - label = bsdf_refraction_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_refraction_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_TRANSPARENT_ID: - label = bsdf_transparent_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_transparent_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_MICROFACET_GGX_ID: case CLOSURE_BSDF_MICROFACET_GGX_FRESNEL_ID: case CLOSURE_BSDF_MICROFACET_GGX_CLEARCOAT_ID: case CLOSURE_BSDF_MICROFACET_GGX_REFRACTION_ID: - label = bsdf_microfacet_ggx_sample(kg, - sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_microfacet_ggx_sample(kg, sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_MICROFACET_MULTI_GGX_ID: case CLOSURE_BSDF_MICROFACET_MULTI_GGX_FRESNEL_ID: - label = bsdf_microfacet_multi_ggx_sample(kg, - sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf, - &sd->lcg_state); + label = bsdf_microfacet_multi_ggx_sample( + kg, sc, Ng, sd->I, randu, randv, eval, omega_in, pdf, &sd->lcg_state); break; case CLOSURE_BSDF_MICROFACET_MULTI_GGX_GLASS_ID: case CLOSURE_BSDF_MICROFACET_MULTI_GGX_GLASS_FRESNEL_ID: - label = bsdf_microfacet_multi_ggx_glass_sample(kg, - sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf, - &sd->lcg_state); + label = bsdf_microfacet_multi_ggx_glass_sample( + kg, sc, Ng, sd->I, randu, randv, eval, omega_in, pdf, &sd->lcg_state); break; case CLOSURE_BSDF_MICROFACET_BECKMANN_ID: case CLOSURE_BSDF_MICROFACET_BECKMANN_REFRACTION_ID: - label = bsdf_microfacet_beckmann_sample(kg, - sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_microfacet_beckmann_sample( + kg, sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_ASHIKHMIN_SHIRLEY_ID: - label = bsdf_ashikhmin_shirley_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_ashikhmin_shirley_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_ASHIKHMIN_VELVET_ID: - label = bsdf_ashikhmin_velvet_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_ashikhmin_velvet_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_DIFFUSE_TOON_ID: - label = bsdf_diffuse_toon_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_diffuse_toon_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_GLOSSY_TOON_ID: - label = bsdf_glossy_toon_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_glossy_toon_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_HAIR_REFLECTION_ID: - label = bsdf_hair_reflection_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_hair_reflection_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_HAIR_TRANSMISSION_ID: - label = bsdf_hair_transmission_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_hair_transmission_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_HAIR_PRINCIPLED_ID: - label = bsdf_principled_hair_sample( - kg, sc, sd, randu, randv, eval, omega_in, &domega_in->dx, &domega_in->dy, pdf); + label = bsdf_principled_hair_sample(kg, sc, sd, randu, randv, eval, omega_in, pdf); break; # ifdef __PRINCIPLED__ case CLOSURE_BSDF_PRINCIPLED_DIFFUSE_ID: - label = bsdf_principled_diffuse_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_principled_diffuse_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; case CLOSURE_BSDF_PRINCIPLED_SHEEN_ID: - label = bsdf_principled_sheen_sample(sc, - Ng, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + label = bsdf_principled_sheen_sample(sc, Ng, sd->I, randu, randv, eval, omega_in, pdf); break; # endif /* __PRINCIPLED__ */ #endif diff --git a/intern/cycles/kernel/closure/bsdf_ashikhmin_shirley.h b/intern/cycles/kernel/closure/bsdf_ashikhmin_shirley.h index 2f6ac2dceb0..75995262030 100644 --- a/intern/cycles/kernel/closure/bsdf_ashikhmin_shirley.h +++ b/intern/cycles/kernel/closure/bsdf_ashikhmin_shirley.h @@ -133,14 +133,10 @@ ccl_device_inline void bsdf_ashikhmin_shirley_sample_first_quadrant(float n_x, ccl_device int bsdf_ashikhmin_shirley_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const MicrofacetBsdf *bsdf = (ccl_private const MicrofacetBsdf *)sc; @@ -221,12 +217,6 @@ ccl_device int bsdf_ashikhmin_shirley_sample(ccl_private const ShaderClosure *sc /* leave the rest to eval_reflect */ *eval = bsdf_ashikhmin_shirley_eval_reflect(sc, I, *omega_in, pdf); } - -#ifdef __RAY_DIFFERENTIALS__ - /* just do the reflection thing for now */ - *domega_in_dx = (2.0f * dot(N, dIdx)) * N - dIdx; - *domega_in_dy = (2.0f * dot(N, dIdy)) * N - dIdy; -#endif } return label; diff --git a/intern/cycles/kernel/closure/bsdf_ashikhmin_velvet.h b/intern/cycles/kernel/closure/bsdf_ashikhmin_velvet.h index ee58bd50aa1..9e68ea5d5e5 100644 --- a/intern/cycles/kernel/closure/bsdf_ashikhmin_velvet.h +++ b/intern/cycles/kernel/closure/bsdf_ashikhmin_velvet.h @@ -87,14 +87,10 @@ ccl_device Spectrum bsdf_ashikhmin_velvet_eval_transmit(ccl_private const Shader ccl_device int bsdf_ashikhmin_velvet_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const VelvetBsdf *bsdf = (ccl_private const VelvetBsdf *)sc; @@ -130,12 +126,6 @@ ccl_device int bsdf_ashikhmin_velvet_sample(ccl_private const ShaderClosure *sc, float power = 0.25f * (D * G) / cosNO; *eval = make_spectrum(power); - -#ifdef __RAY_DIFFERENTIALS__ - // TODO: find a better approximation for the retroreflective bounce - *domega_in_dx = (2 * dot(N, dIdx)) * N - dIdx; - *domega_in_dy = (2 * dot(N, dIdy)) * N - dIdy; -#endif } else { *pdf = 0.0f; diff --git a/intern/cycles/kernel/closure/bsdf_diffuse.h b/intern/cycles/kernel/closure/bsdf_diffuse.h index 2a082796043..ec64c375666 100644 --- a/intern/cycles/kernel/closure/bsdf_diffuse.h +++ b/intern/cycles/kernel/closure/bsdf_diffuse.h @@ -51,14 +51,10 @@ ccl_device Spectrum bsdf_diffuse_eval_transmit(ccl_private const ShaderClosure * ccl_device int bsdf_diffuse_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const DiffuseBsdf *bsdf = (ccl_private const DiffuseBsdf *)sc; @@ -69,11 +65,6 @@ ccl_device int bsdf_diffuse_sample(ccl_private const ShaderClosure *sc, if (dot(Ng, *omega_in) > 0.0f) { *eval = make_spectrum(*pdf); -#ifdef __RAY_DIFFERENTIALS__ - // TODO: find a better approximation for the diffuse bounce - *domega_in_dx = (2 * dot(N, dIdx)) * N - dIdx; - *domega_in_dy = (2 * dot(N, dIdy)) * N - dIdy; -#endif } else { *pdf = 0.0f; @@ -115,14 +106,10 @@ ccl_device Spectrum bsdf_translucent_eval_transmit(ccl_private const ShaderClosu ccl_device int bsdf_translucent_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const DiffuseBsdf *bsdf = (ccl_private const DiffuseBsdf *)sc; @@ -133,11 +120,6 @@ ccl_device int bsdf_translucent_sample(ccl_private const ShaderClosure *sc, sample_cos_hemisphere(-N, randu, randv, omega_in, pdf); if (dot(Ng, *omega_in) < 0) { *eval = make_spectrum(*pdf); -#ifdef __RAY_DIFFERENTIALS__ - // TODO: find a better approximation for the diffuse bounce - *domega_in_dx = -((2 * dot(N, dIdx)) * N - dIdx); - *domega_in_dy = -((2 * dot(N, dIdy)) * N - dIdy); -#endif } else { *pdf = 0; diff --git a/intern/cycles/kernel/closure/bsdf_diffuse_ramp.h b/intern/cycles/kernel/closure/bsdf_diffuse_ramp.h index 8b88edb37f7..d7faf5c9e9a 100644 --- a/intern/cycles/kernel/closure/bsdf_diffuse_ramp.h +++ b/intern/cycles/kernel/closure/bsdf_diffuse_ramp.h @@ -71,14 +71,10 @@ ccl_device Spectrum bsdf_diffuse_ramp_eval_transmit(ccl_private const ShaderClos ccl_device int bsdf_diffuse_ramp_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { const DiffuseRampBsdf *bsdf = (const DiffuseRampBsdf *)sc; @@ -89,10 +85,6 @@ ccl_device int bsdf_diffuse_ramp_sample(ccl_private const ShaderClosure *sc, if (dot(Ng, *omega_in) > 0.0f) { *eval = rgb_to_spectrum(bsdf_diffuse_ramp_get_color(bsdf->colors, *pdf * M_PI_F) * M_1_PI_F); -# ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(N, dIdx)) * N - dIdx; - *domega_in_dy = (2 * dot(N, dIdy)) * N - dIdy; -# endif } else { *pdf = 0.0f; diff --git a/intern/cycles/kernel/closure/bsdf_hair.h b/intern/cycles/kernel/closure/bsdf_hair.h index 4179f73e22c..a29f7c444ae 100644 --- a/intern/cycles/kernel/closure/bsdf_hair.h +++ b/intern/cycles/kernel/closure/bsdf_hair.h @@ -151,14 +151,10 @@ ccl_device Spectrum bsdf_hair_transmission_eval_transmit(ccl_private const Shade ccl_device int bsdf_hair_reflection_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const HairBsdf *bsdf = (ccl_private const HairBsdf *)sc; @@ -194,12 +190,6 @@ ccl_device int bsdf_hair_reflection_sample(ccl_private const ShaderClosure *sc, fast_sincosf(phi, &sinphi, &cosphi); *omega_in = (cosphi * costheta_i) * locy - (sinphi * costheta_i) * locx + (sintheta_i)*Tg; - // differentials - TODO: find a better approximation for the reflective bounce -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = 2 * dot(locy, dIdx) * locy - dIdx; - *domega_in_dy = 2 * dot(locy, dIdy) * locy - dIdy; -#endif - *pdf = fabsf(phi_pdf * theta_pdf); if (M_PI_2_F - fabsf(theta_i) < 0.001f) *pdf = 0.0f; @@ -212,14 +202,10 @@ ccl_device int bsdf_hair_reflection_sample(ccl_private const ShaderClosure *sc, ccl_device int bsdf_hair_transmission_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const HairBsdf *bsdf = (ccl_private const HairBsdf *)sc; @@ -255,12 +241,6 @@ ccl_device int bsdf_hair_transmission_sample(ccl_private const ShaderClosure *sc fast_sincosf(phi, &sinphi, &cosphi); *omega_in = (cosphi * costheta_i) * locy - (sinphi * costheta_i) * locx + (sintheta_i)*Tg; - // differentials - TODO: find a better approximation for the transmission bounce -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = 2 * dot(locy, dIdx) * locy - dIdx; - *domega_in_dy = 2 * dot(locy, dIdy) * locy - dIdy; -#endif - *pdf = fabsf(phi_pdf * theta_pdf); if (M_PI_2_F - fabsf(theta_i) < 0.001f) { *pdf = 0.0f; diff --git a/intern/cycles/kernel/closure/bsdf_hair_principled.h b/intern/cycles/kernel/closure/bsdf_hair_principled.h index f78a05ea212..2236bc62050 100644 --- a/intern/cycles/kernel/closure/bsdf_hair_principled.h +++ b/intern/cycles/kernel/closure/bsdf_hair_principled.h @@ -354,8 +354,6 @@ ccl_device int bsdf_principled_hair_sample(KernelGlobals kg, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private PrincipledHairBSDF *bsdf = (ccl_private PrincipledHairBSDF *)sc; @@ -471,12 +469,6 @@ ccl_device int bsdf_principled_hair_sample(KernelGlobals kg, *omega_in = X * sin_theta_i + Y * cos_theta_i * cosf(phi_i) + Z * cos_theta_i * sinf(phi_i); -#ifdef __RAY_DIFFERENTIALS__ - float3 N = safe_normalize(sd->I + *omega_in); - *domega_in_dx = (2 * dot(N, sd->dI.dx)) * N - sd->dI.dx; - *domega_in_dy = (2 * dot(N, sd->dI.dy)) * N - sd->dI.dy; -#endif - return LABEL_GLOSSY | ((p == 0) ? LABEL_REFLECT : LABEL_TRANSMIT); } diff --git a/intern/cycles/kernel/closure/bsdf_microfacet.h b/intern/cycles/kernel/closure/bsdf_microfacet.h index 091fbde5585..04d5ca90bfd 100644 --- a/intern/cycles/kernel/closure/bsdf_microfacet.h +++ b/intern/cycles/kernel/closure/bsdf_microfacet.h @@ -537,14 +537,10 @@ ccl_device int bsdf_microfacet_ggx_sample(KernelGlobals kg, ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const MicrofacetBsdf *bsdf = (ccl_private const MicrofacetBsdf *)sc; @@ -672,11 +668,6 @@ ccl_device int bsdf_microfacet_ggx_sample(KernelGlobals kg, if (bsdf->type == CLOSURE_BSDF_MICROFACET_GGX_CLEARCOAT_ID) { *eval *= 0.25f * bsdf->extra->clearcoat; } - -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(m, dIdx)) * m - dIdx; - *domega_in_dy = (2 * dot(m, dIdy)) * m - dIdy; -#endif } else { *eval = zero_spectrum(); @@ -690,34 +681,13 @@ ccl_device int bsdf_microfacet_ggx_sample(KernelGlobals kg, /* CAUTION: the i and o variables are inverted relative to the paper * eq. 39 - compute actual refractive direction */ float3 R, T; -#ifdef __RAY_DIFFERENTIALS__ - float3 dRdx, dRdy, dTdx, dTdy; -#endif float m_eta = bsdf->ior, fresnel; bool inside; - fresnel = fresnel_dielectric(m_eta, - m, - I, - &R, - &T, -#ifdef __RAY_DIFFERENTIALS__ - dIdx, - dIdy, - &dRdx, - &dRdy, - &dTdx, - &dTdy, -#endif - &inside); + fresnel = fresnel_dielectric(m_eta, m, I, &R, &T, &inside); if (!inside && fresnel != 1.0f) { - *omega_in = T; -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = dTdx; - *domega_in_dy = dTdy; -#endif if (alpha_x * alpha_y <= 1e-7f || fabsf(m_eta - 1.0f) < 1e-4f) { /* some high number for MIS */ @@ -978,14 +948,10 @@ ccl_device int bsdf_microfacet_beckmann_sample(KernelGlobals kg, ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const MicrofacetBsdf *bsdf = (ccl_private const MicrofacetBsdf *)sc; @@ -1076,11 +1042,6 @@ ccl_device int bsdf_microfacet_beckmann_sample(KernelGlobals kg, *eval = make_spectrum(out); } - -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(m, dIdx)) * m - dIdx; - *domega_in_dy = (2 * dot(m, dIdy)) * m - dIdy; -#endif } else { *eval = zero_spectrum(); @@ -1094,35 +1055,14 @@ ccl_device int bsdf_microfacet_beckmann_sample(KernelGlobals kg, /* CAUTION: the i and o variables are inverted relative to the paper * eq. 39 - compute actual refractive direction */ float3 R, T; -#ifdef __RAY_DIFFERENTIALS__ - float3 dRdx, dRdy, dTdx, dTdy; -#endif float m_eta = bsdf->ior, fresnel; bool inside; - fresnel = fresnel_dielectric(m_eta, - m, - I, - &R, - &T, -#ifdef __RAY_DIFFERENTIALS__ - dIdx, - dIdy, - &dRdx, - &dRdy, - &dTdx, - &dTdy, -#endif - &inside); + fresnel = fresnel_dielectric(m_eta, m, I, &R, &T, &inside); if (!inside && fresnel != 1.0f) { *omega_in = T; -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = dTdx; - *domega_in_dy = dTdy; -#endif - if (alpha_x * alpha_y <= 1e-7f || fabsf(m_eta - 1.0f) < 1e-4f) { /* some high number for MIS */ *pdf = 1e6f; diff --git a/intern/cycles/kernel/closure/bsdf_microfacet_multi.h b/intern/cycles/kernel/closure/bsdf_microfacet_multi.h index e6ab116519b..ac37a648a2c 100644 --- a/intern/cycles/kernel/closure/bsdf_microfacet_multi.h +++ b/intern/cycles/kernel/closure/bsdf_microfacet_multi.h @@ -478,14 +478,10 @@ ccl_device int bsdf_microfacet_multi_ggx_sample(KernelGlobals kg, ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf, ccl_private uint *lcg_state) { @@ -510,10 +506,6 @@ ccl_device int bsdf_microfacet_multi_ggx_sample(KernelGlobals kg, } *pdf = 1e6f; *eval = make_spectrum(1e6f); -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(Z, dIdx)) * Z - dIdx; - *domega_in_dy = (2 * dot(Z, dIdy)) * Z - dIdy; -#endif return LABEL_REFLECT | LABEL_SINGULAR; } @@ -551,10 +543,6 @@ ccl_device int bsdf_microfacet_multi_ggx_sample(KernelGlobals kg, *pdf = mf_ggx_pdf(localI, localO, bsdf->alpha_x); *eval *= *pdf; -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(Z, dIdx)) * Z - dIdx; - *domega_in_dy = (2 * dot(Z, dIdy)) * Z - dIdy; -#endif return LABEL_REFLECT | LABEL_GLOSSY; } @@ -662,14 +650,10 @@ ccl_device int bsdf_microfacet_multi_ggx_glass_sample(KernelGlobals kg, ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf, ccl_private uint *lcg_state) { @@ -680,41 +664,17 @@ ccl_device int bsdf_microfacet_multi_ggx_glass_sample(KernelGlobals kg, if (bsdf->alpha_x * bsdf->alpha_y < 1e-7f) { float3 R, T; -#ifdef __RAY_DIFFERENTIALS__ - float3 dRdx, dRdy, dTdx, dTdy; -#endif bool inside; - float fresnel = fresnel_dielectric(bsdf->ior, - Z, - I, - &R, - &T, -#ifdef __RAY_DIFFERENTIALS__ - dIdx, - dIdy, - &dRdx, - &dRdy, - &dTdx, - &dTdy, -#endif - &inside); + float fresnel = fresnel_dielectric(bsdf->ior, Z, I, &R, &T, &inside); *pdf = 1e6f; *eval = make_spectrum(1e6f); if (randu < fresnel) { *omega_in = R; -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = dRdx; - *domega_in_dy = dRdy; -#endif return LABEL_REFLECT | LABEL_SINGULAR; } else { *omega_in = T; -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = dTdx; - *domega_in_dy = dTdy; -#endif return LABEL_TRANSMIT | LABEL_SINGULAR; } } @@ -740,22 +700,9 @@ ccl_device int bsdf_microfacet_multi_ggx_glass_sample(KernelGlobals kg, *omega_in = X * localO.x + Y * localO.y + Z * localO.z; if (localO.z * localI.z > 0.0f) { -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(Z, dIdx)) * Z - dIdx; - *domega_in_dy = (2 * dot(Z, dIdy)) * Z - dIdy; -#endif return LABEL_REFLECT | LABEL_GLOSSY; } else { -#ifdef __RAY_DIFFERENTIALS__ - float cosI = dot(Z, I); - float dnp = max(sqrtf(1.0f - (bsdf->ior * bsdf->ior * (1.0f - cosI * cosI))), 1e-7f); - *domega_in_dx = -(bsdf->ior * dIdx) + - ((bsdf->ior - bsdf->ior * bsdf->ior * cosI / dnp) * dot(dIdx, Z)) * Z; - *domega_in_dy = -(bsdf->ior * dIdy) + - ((bsdf->ior - bsdf->ior * bsdf->ior * cosI / dnp) * dot(dIdy, Z)) * Z; -#endif - return LABEL_TRANSMIT | LABEL_GLOSSY; } } diff --git a/intern/cycles/kernel/closure/bsdf_oren_nayar.h b/intern/cycles/kernel/closure/bsdf_oren_nayar.h index fcfeb9257f1..b85390f0676 100644 --- a/intern/cycles/kernel/closure/bsdf_oren_nayar.h +++ b/intern/cycles/kernel/closure/bsdf_oren_nayar.h @@ -75,14 +75,10 @@ ccl_device Spectrum bsdf_oren_nayar_eval_transmit(ccl_private const ShaderClosur ccl_device int bsdf_oren_nayar_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const OrenNayarBsdf *bsdf = (ccl_private const OrenNayarBsdf *)sc; @@ -90,12 +86,6 @@ ccl_device int bsdf_oren_nayar_sample(ccl_private const ShaderClosure *sc, if (dot(Ng, *omega_in) > 0.0f) { *eval = bsdf_oren_nayar_get_intensity(sc, bsdf->N, I, *omega_in); - -#ifdef __RAY_DIFFERENTIALS__ - // TODO: find a better approximation for the bounce - *domega_in_dx = (2.0f * dot(bsdf->N, dIdx)) * bsdf->N - dIdx; - *domega_in_dy = (2.0f * dot(bsdf->N, dIdy)) * bsdf->N - dIdy; -#endif } else { *pdf = 0.0f; diff --git a/intern/cycles/kernel/closure/bsdf_phong_ramp.h b/intern/cycles/kernel/closure/bsdf_phong_ramp.h index d010e74cb65..4236e77ae6c 100644 --- a/intern/cycles/kernel/closure/bsdf_phong_ramp.h +++ b/intern/cycles/kernel/closure/bsdf_phong_ramp.h @@ -82,14 +82,10 @@ ccl_device float3 bsdf_phong_ramp_eval_transmit(ccl_private const ShaderClosure ccl_device int bsdf_phong_ramp_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const PhongRampBsdf *bsdf = (ccl_private const PhongRampBsdf *)sc; @@ -99,12 +95,6 @@ ccl_device int bsdf_phong_ramp_sample(ccl_private const ShaderClosure *sc, if (cosNO > 0) { // reflect the view vector float3 R = (2 * cosNO) * bsdf->N - I; - -# ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(bsdf->N, dIdx)) * bsdf->N - dIdx; - *domega_in_dy = (2 * dot(bsdf->N, dIdy)) * bsdf->N - dIdy; -# endif - float3 T, B; make_orthonormals(R, &T, &B); float phi = M_2PI_F * randu; diff --git a/intern/cycles/kernel/closure/bsdf_principled_diffuse.h b/intern/cycles/kernel/closure/bsdf_principled_diffuse.h index 90ef252b3b9..39cca1bd970 100644 --- a/intern/cycles/kernel/closure/bsdf_principled_diffuse.h +++ b/intern/cycles/kernel/closure/bsdf_principled_diffuse.h @@ -142,14 +142,10 @@ ccl_device Spectrum bsdf_principled_diffuse_eval_transmit(ccl_private const Shad ccl_device int bsdf_principled_diffuse_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const PrincipledDiffuseBsdf *bsdf = (ccl_private const PrincipledDiffuseBsdf *)sc; @@ -160,12 +156,6 @@ ccl_device int bsdf_principled_diffuse_sample(ccl_private const ShaderClosure *s if (dot(Ng, *omega_in) > 0) { *eval = bsdf_principled_diffuse_compute_brdf(bsdf, N, I, *omega_in, pdf); - -#ifdef __RAY_DIFFERENTIALS__ - // TODO: find a better approximation for the diffuse bounce - *domega_in_dx = -((2 * dot(N, dIdx)) * N - dIdx); - *domega_in_dy = -((2 * dot(N, dIdy)) * N - dIdy); -#endif } else { *pdf = 0.0f; diff --git a/intern/cycles/kernel/closure/bsdf_principled_sheen.h b/intern/cycles/kernel/closure/bsdf_principled_sheen.h index 42a776299eb..fa46f47eb21 100644 --- a/intern/cycles/kernel/closure/bsdf_principled_sheen.h +++ b/intern/cycles/kernel/closure/bsdf_principled_sheen.h @@ -93,14 +93,10 @@ ccl_device Spectrum bsdf_principled_sheen_eval_transmit(ccl_private const Shader ccl_device int bsdf_principled_sheen_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const PrincipledSheenBsdf *bsdf = (ccl_private const PrincipledSheenBsdf *)sc; @@ -113,12 +109,6 @@ ccl_device int bsdf_principled_sheen_sample(ccl_private const ShaderClosure *sc, float3 H = normalize(I + *omega_in); *eval = calculate_principled_sheen_brdf(N, I, *omega_in, H, pdf); - -#ifdef __RAY_DIFFERENTIALS__ - // TODO: find a better approximation for the diffuse bounce - *domega_in_dx = -((2 * dot(N, dIdx)) * N - dIdx); - *domega_in_dy = -((2 * dot(N, dIdy)) * N - dIdy); -#endif } else { *eval = zero_spectrum(); diff --git a/intern/cycles/kernel/closure/bsdf_reflection.h b/intern/cycles/kernel/closure/bsdf_reflection.h index 40d02b13b46..5e6c6cdcde6 100644 --- a/intern/cycles/kernel/closure/bsdf_reflection.h +++ b/intern/cycles/kernel/closure/bsdf_reflection.h @@ -39,14 +39,10 @@ ccl_device Spectrum bsdf_reflection_eval_transmit(ccl_private const ShaderClosur ccl_device int bsdf_reflection_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const MicrofacetBsdf *bsdf = (ccl_private const MicrofacetBsdf *)sc; @@ -57,10 +53,6 @@ ccl_device int bsdf_reflection_sample(ccl_private const ShaderClosure *sc, if (cosNO > 0) { *omega_in = (2 * cosNO) * N - I; if (dot(Ng, *omega_in) > 0) { -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = 2 * dot(N, dIdx) * N - dIdx; - *domega_in_dy = 2 * dot(N, dIdy) * N - dIdy; -#endif /* Some high number for MIS. */ *pdf = 1e6f; *eval = make_spectrum(1e6f); diff --git a/intern/cycles/kernel/closure/bsdf_refraction.h b/intern/cycles/kernel/closure/bsdf_refraction.h index 3faa51025d6..e680a9617db 100644 --- a/intern/cycles/kernel/closure/bsdf_refraction.h +++ b/intern/cycles/kernel/closure/bsdf_refraction.h @@ -39,14 +39,10 @@ ccl_device Spectrum bsdf_refraction_eval_transmit(ccl_private const ShaderClosur ccl_device int bsdf_refraction_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const MicrofacetBsdf *bsdf = (ccl_private const MicrofacetBsdf *)sc; @@ -54,35 +50,15 @@ ccl_device int bsdf_refraction_sample(ccl_private const ShaderClosure *sc, float3 N = bsdf->N; float3 R, T; -#ifdef __RAY_DIFFERENTIALS__ - float3 dRdx, dRdy, dTdx, dTdy; -#endif bool inside; float fresnel; - fresnel = fresnel_dielectric(m_eta, - N, - I, - &R, - &T, -#ifdef __RAY_DIFFERENTIALS__ - dIdx, - dIdy, - &dRdx, - &dRdy, - &dTdx, - &dTdy, -#endif - &inside); + fresnel = fresnel_dielectric(m_eta, N, I, &R, &T, &inside); if (!inside && fresnel != 1.0f) { /* Some high number for MIS. */ *pdf = 1e6f; *eval = make_spectrum(1e6f); *omega_in = T; -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = dTdx; - *domega_in_dy = dTdy; -#endif } else { *pdf = 0.0f; diff --git a/intern/cycles/kernel/closure/bsdf_toon.h b/intern/cycles/kernel/closure/bsdf_toon.h index f2f48417319..c9086823de9 100644 --- a/intern/cycles/kernel/closure/bsdf_toon.h +++ b/intern/cycles/kernel/closure/bsdf_toon.h @@ -83,14 +83,10 @@ ccl_device Spectrum bsdf_diffuse_toon_eval_transmit(ccl_private const ShaderClos ccl_device int bsdf_diffuse_toon_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const ToonBsdf *bsdf = (ccl_private const ToonBsdf *)sc; @@ -104,12 +100,6 @@ ccl_device int bsdf_diffuse_toon_sample(ccl_private const ShaderClosure *sc, if (dot(Ng, *omega_in) > 0.0f) { *eval = make_spectrum(*pdf * bsdf_toon_get_intensity(max_angle, smooth, angle)); - -#ifdef __RAY_DIFFERENTIALS__ - // TODO: find a better approximation for the bounce - *domega_in_dx = (2.0f * dot(bsdf->N, dIdx)) * bsdf->N - dIdx; - *domega_in_dy = (2.0f * dot(bsdf->N, dIdy)) * bsdf->N - dIdy; -#endif } else { *eval = zero_spectrum(); @@ -175,14 +165,10 @@ ccl_device Spectrum bsdf_glossy_toon_eval_transmit(ccl_private const ShaderClosu ccl_device int bsdf_glossy_toon_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { ccl_private const ToonBsdf *bsdf = (ccl_private const ToonBsdf *)sc; @@ -205,11 +191,6 @@ ccl_device int bsdf_glossy_toon_sample(ccl_private const ShaderClosure *sc, /* make sure the direction we chose is still in the right hemisphere */ if (cosNI > 0) { *eval = make_spectrum(*pdf * bsdf_toon_get_intensity(max_angle, smooth, angle)); - -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = (2 * dot(bsdf->N, dIdx)) * bsdf->N - dIdx; - *domega_in_dy = (2 * dot(bsdf->N, dIdy)) * bsdf->N - dIdy; -#endif } else { *pdf = 0.0f; diff --git a/intern/cycles/kernel/closure/bsdf_transparent.h b/intern/cycles/kernel/closure/bsdf_transparent.h index 89b36c709e4..c2aee1e1633 100644 --- a/intern/cycles/kernel/closure/bsdf_transparent.h +++ b/intern/cycles/kernel/closure/bsdf_transparent.h @@ -80,22 +80,14 @@ ccl_device Spectrum bsdf_transparent_eval_transmit(ccl_private const ShaderClosu ccl_device int bsdf_transparent_sample(ccl_private const ShaderClosure *sc, float3 Ng, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { // only one direction is possible *omega_in = -I; -#ifdef __RAY_DIFFERENTIALS__ - *domega_in_dx = -dIdx; - *domega_in_dy = -dIdy; -#endif *pdf = 1; *eval = one_spectrum(); return LABEL_TRANSMIT | LABEL_TRANSPARENT; diff --git a/intern/cycles/kernel/closure/bsdf_util.h b/intern/cycles/kernel/closure/bsdf_util.h index 10f2643721e..3c48b98fed9 100644 --- a/intern/cycles/kernel/closure/bsdf_util.h +++ b/intern/cycles/kernel/closure/bsdf_util.h @@ -15,14 +15,6 @@ ccl_device float fresnel_dielectric(float eta, const float3 I, ccl_private float3 *R, ccl_private float3 *T, -#ifdef __RAY_DIFFERENTIALS__ - const float3 dIdx, - const float3 dIdy, - ccl_private float3 *dRdx, - ccl_private float3 *dRdy, - ccl_private float3 *dTdx, - ccl_private float3 *dTdy, -#endif ccl_private bool *is_inside) { float cos = dot(N, I), neta; @@ -45,28 +37,16 @@ ccl_device float fresnel_dielectric(float eta, // compute reflection *R = (2 * cos) * Nn - I; -#ifdef __RAY_DIFFERENTIALS__ - *dRdx = (2 * dot(Nn, dIdx)) * Nn - dIdx; - *dRdy = (2 * dot(Nn, dIdy)) * Nn - dIdy; -#endif float arg = 1 - (neta * neta * (1 - (cos * cos))); if (arg < 0) { *T = make_float3(0.0f, 0.0f, 0.0f); -#ifdef __RAY_DIFFERENTIALS__ - *dTdx = make_float3(0.0f, 0.0f, 0.0f); - *dTdy = make_float3(0.0f, 0.0f, 0.0f); -#endif return 1; // total internal reflection } else { float dnp = max(sqrtf(arg), 1e-7f); float nK = (neta * cos) - dnp; *T = -(neta * I) + (nK * Nn); -#ifdef __RAY_DIFFERENTIALS__ - *dTdx = -(neta * dIdx) + ((neta - neta * neta * cos / dnp) * dot(dIdx, Nn)) * Nn; - *dTdy = -(neta * dIdy) + ((neta - neta * neta * cos / dnp) * dot(dIdy, Nn)) * Nn; -#endif // compute Fresnel terms float cosTheta1 = cos; // N.R float cosTheta2 = -dot(Nn, *T); diff --git a/intern/cycles/kernel/closure/volume.h b/intern/cycles/kernel/closure/volume.h index 10494be87cb..9dbb5154457 100644 --- a/intern/cycles/kernel/closure/volume.h +++ b/intern/cycles/kernel/closure/volume.h @@ -101,14 +101,10 @@ henyey_greenstrein_sample(float3 D, float g, float randu, float randv, ccl_priva ccl_device int volume_henyey_greenstein_sample(ccl_private const ShaderVolumeClosure *svc, float3 I, - float3 dIdx, - float3 dIdy, float randu, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private float3 *domega_in_dx, - ccl_private float3 *domega_in_dy, ccl_private float *pdf) { float g = svc->g; @@ -117,12 +113,6 @@ ccl_device int volume_henyey_greenstein_sample(ccl_private const ShaderVolumeClo *omega_in = henyey_greenstrein_sample(-I, g, randu, randv, pdf); *eval = make_spectrum(*pdf); /* perfect importance sampling */ -#ifdef __RAY_DIFFERENTIALS__ - /* todo: implement ray differential estimation */ - *domega_in_dx = make_float3(0.0f, 0.0f, 0.0f); - *domega_in_dy = make_float3(0.0f, 0.0f, 0.0f); -#endif - return LABEL_VOLUME_SCATTER; } @@ -142,20 +132,9 @@ ccl_device int volume_phase_sample(ccl_private const ShaderData *sd, float randv, ccl_private Spectrum *eval, ccl_private float3 *omega_in, - ccl_private differential3 *domega_in, ccl_private float *pdf) { - return volume_henyey_greenstein_sample(svc, - sd->I, - sd->dI.dx, - sd->dI.dy, - randu, - randv, - eval, - omega_in, - &domega_in->dx, - &domega_in->dy, - pdf); + return volume_henyey_greenstein_sample(svc, sd->I, randu, randv, eval, omega_in, pdf); } /* Volume sampling utilities. */ diff --git a/intern/cycles/kernel/geom/shader_data.h b/intern/cycles/kernel/geom/shader_data.h index 5af89b45f20..028c03ace1d 100644 --- a/intern/cycles/kernel/geom/shader_data.h +++ b/intern/cycles/kernel/geom/shader_data.h @@ -123,9 +123,9 @@ ccl_device_inline void shader_setup_from_ray(KernelGlobals kg, #ifdef __RAY_DIFFERENTIALS__ /* differentials */ - differential_transfer_compact(&sd->dP, ray->dP, ray->D, ray->dD, sd->Ng, sd->ray_length); - differential_incoming_compact(&sd->dI, ray->D, ray->dD); - differential_dudv(&sd->du, &sd->dv, sd->dPdu, sd->dPdv, sd->dP, sd->Ng); + sd->dP = differential_transfer_compact(ray->dP, ray->D, ray->dD, sd->ray_length); + sd->dI = differential_incoming_compact(ray->dD); + differential_dudv_compact(&sd->du, &sd->dv, sd->dPdu, sd->dPdv, sd->dP, sd->Ng); #endif } @@ -240,8 +240,8 @@ ccl_device_inline void shader_setup_from_sample(KernelGlobals kg, #ifdef __RAY_DIFFERENTIALS__ /* no ray differentials here yet */ - sd->dP = differential3_zero(); - sd->dI = differential3_zero(); + sd->dP = differential_zero_compact(); + sd->dI = differential_zero_compact(); sd->du = differential_zero(); sd->dv = differential_zero(); #endif @@ -348,8 +348,8 @@ ccl_device void shader_setup_from_curve(KernelGlobals kg, /* No ray differentials currently. */ #ifdef __RAY_DIFFERENTIALS__ - sd->dP = differential3_zero(); - sd->dI = differential3_zero(); + sd->dP = differential_zero_compact(); + sd->dI = differential_zero_compact(); sd->du = differential_zero(); sd->dv = differential_zero(); #endif @@ -391,8 +391,8 @@ ccl_device_inline void shader_setup_from_background(KernelGlobals kg, #ifdef __RAY_DIFFERENTIALS__ /* differentials */ - sd->dP = differential3_zero(); /* TODO: ray->dP */ - differential_incoming(&sd->dI, sd->dP); + sd->dP = differential_zero_compact(); /* TODO: ray->dP */ + sd->dI = differential_zero_compact(); sd->du = differential_zero(); sd->dv = differential_zero(); #endif @@ -433,8 +433,8 @@ ccl_device_inline void shader_setup_from_volume(KernelGlobals kg, # ifdef __RAY_DIFFERENTIALS__ /* differentials */ - sd->dP = differential3_zero(); /* TODO ray->dD */ - differential_incoming(&sd->dI, sd->dP); + sd->dP = differential_zero_compact(); /* TODO ray->dD */ + sd->dI = differential_zero_compact(); sd->du = differential_zero(); sd->dv = differential_zero(); # endif diff --git a/intern/cycles/kernel/integrator/shade_surface.h b/intern/cycles/kernel/integrator/shade_surface.h index 4eb50171532..f42e2979b3b 100644 --- a/intern/cycles/kernel/integrator/shade_surface.h +++ b/intern/cycles/kernel/integrator/shade_surface.h @@ -362,11 +362,10 @@ ccl_device_forceinline int integrate_surface_bsdf_bssrdf_bounce( float bsdf_pdf; BsdfEval bsdf_eval ccl_optional_struct_init; float3 bsdf_omega_in ccl_optional_struct_init; - differential3 bsdf_domega_in ccl_optional_struct_init; int label; label = shader_bsdf_sample_closure( - kg, sd, sc, bsdf_u, bsdf_v, &bsdf_eval, &bsdf_omega_in, &bsdf_domega_in, &bsdf_pdf); + kg, sd, sc, bsdf_u, bsdf_v, &bsdf_eval, &bsdf_omega_in, &bsdf_pdf); if (bsdf_pdf == 0.0f || bsdf_eval_is_zero(&bsdf_eval)) { return LABEL_NONE; @@ -385,7 +384,6 @@ ccl_device_forceinline int integrate_surface_bsdf_bssrdf_bounce( 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(bsdf_domega_in); #endif } diff --git a/intern/cycles/kernel/integrator/shade_volume.h b/intern/cycles/kernel/integrator/shade_volume.h index 0d35011c359..599454c5cb2 100644 --- a/intern/cycles/kernel/integrator/shade_volume.h +++ b/intern/cycles/kernel/integrator/shade_volume.h @@ -871,17 +871,9 @@ ccl_device_forceinline bool integrate_volume_phase_scatter( 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 = shader_volume_phase_sample( + kg, sd, phases, phase_u, phase_v, &phase_eval, &phase_omega_in, &phase_pdf); if (phase_pdf == 0.0f || bsdf_eval_is_zero(&phase_eval)) { return false; @@ -894,7 +886,6 @@ ccl_device_forceinline bool integrate_volume_phase_scatter( 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; diff --git a/intern/cycles/kernel/integrator/shader_eval.h b/intern/cycles/kernel/integrator/shader_eval.h index d9c14b25a4b..e6b0d0a6466 100644 --- a/intern/cycles/kernel/integrator/shader_eval.h +++ b/intern/cycles/kernel/integrator/shader_eval.h @@ -339,7 +339,6 @@ ccl_device int shader_bsdf_sample_closure(KernelGlobals kg, float randv, ccl_private BsdfEval *bsdf_eval, ccl_private float3 *omega_in, - ccl_private differential3 *domega_in, ccl_private float *pdf) { /* BSSRDF should already have been handled elsewhere. */ @@ -349,7 +348,7 @@ ccl_device int shader_bsdf_sample_closure(KernelGlobals kg, Spectrum eval = zero_spectrum(); *pdf = 0.0f; - label = bsdf_sample(kg, sd, sc, randu, randv, &eval, omega_in, domega_in, pdf); + label = bsdf_sample(kg, sd, sc, randu, randv, &eval, omega_in, pdf); if (*pdf != 0.0f) { bsdf_eval_init(bsdf_eval, sc->type, eval * sc->weight); @@ -708,7 +707,6 @@ ccl_device int shader_volume_phase_sample(KernelGlobals kg, float randv, ccl_private BsdfEval *phase_eval, ccl_private float3 *omega_in, - ccl_private differential3 *domega_in, ccl_private float *pdf) { int sampled = 0; @@ -751,7 +749,7 @@ ccl_device int shader_volume_phase_sample(KernelGlobals kg, Spectrum eval = zero_spectrum(); *pdf = 0.0f; - label = volume_phase_sample(sd, svc, randu, randv, &eval, omega_in, domega_in, pdf); + label = volume_phase_sample(sd, svc, randu, randv, &eval, omega_in, pdf); if (*pdf != 0.0f) { bsdf_eval_init(phase_eval, CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID, eval); @@ -767,14 +765,13 @@ ccl_device int shader_phase_sample_closure(KernelGlobals kg, float randv, ccl_private BsdfEval *phase_eval, ccl_private float3 *omega_in, - ccl_private differential3 *domega_in, ccl_private float *pdf) { int label; Spectrum eval = zero_spectrum(); *pdf = 0.0f; - label = volume_phase_sample(sd, sc, randu, randv, &eval, omega_in, domega_in, pdf); + label = volume_phase_sample(sd, sc, randu, randv, &eval, omega_in, pdf); if (*pdf != 0.0f) bsdf_eval_init(phase_eval, CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID, eval); diff --git a/intern/cycles/kernel/osl/services.cpp b/intern/cycles/kernel/osl/services.cpp index 6b7981b7f3a..6766fe2ce89 100644 --- a/intern/cycles/kernel/osl/services.cpp +++ b/intern/cycles/kernel/osl/services.cpp @@ -1102,8 +1102,9 @@ bool OSLRenderServices::get_background_attribute(const KernelGlobalsCPU *kg, ndc[0] = camera_world_to_ndc(kg, sd, sd->P); if (derivatives) { - ndc[1] = camera_world_to_ndc(kg, sd, sd->P + sd->dP.dx) - ndc[0]; - ndc[2] = camera_world_to_ndc(kg, sd, sd->P + sd->dP.dy) - ndc[0]; + const differential3 dP = differential_from_compact(sd->Ng, sd->dP); + ndc[1] = camera_world_to_ndc(kg, sd, sd->P + dP.dx) - ndc[0]; + ndc[2] = camera_world_to_ndc(kg, sd, sd->P + dP.dy) - ndc[0]; } } @@ -1755,11 +1756,13 @@ bool OSLRenderServices::getmessage(OSL::ShaderGlobals *sg, return set_attribute_float3(sd->Ng, type, derivatives, val); } else if (name == u_P) { - float3 f[3] = {sd->P, sd->dP.dx, sd->dP.dy}; + const differential3 dP = differential_from_compact(sd->Ng, sd->dP); + float3 f[3] = {sd->P, dP.dx, dP.dy}; return set_attribute_float3(f, type, derivatives, val); } else if (name == u_I) { - float3 f[3] = {sd->I, sd->dI.dx, sd->dI.dy}; + const differential3 dI = differential_from_compact(sd->I, sd->dI); + float3 f[3] = {sd->I, dI.dx, dI.dy}; return set_attribute_float3(f, type, derivatives, val); } else if (name == u_u) { diff --git a/intern/cycles/kernel/osl/shader.cpp b/intern/cycles/kernel/osl/shader.cpp index af96c0070e3..5862b6a8a2b 100644 --- a/intern/cycles/kernel/osl/shader.cpp +++ b/intern/cycles/kernel/osl/shader.cpp @@ -17,6 +17,8 @@ #include "kernel/osl/globals.h" #include "kernel/osl/services.h" #include "kernel/osl/shader.h" + +#include "kernel/util/differential.h" // clang-format on #include "scene/attribute.h" @@ -79,13 +81,16 @@ static void shaderdata_to_shaderglobals(const KernelGlobalsCPU *kg, { OSL::ShaderGlobals *globals = &tdata->globals; + const differential3 dP = differential_from_compact(sd->Ng, sd->dP); + const differential3 dI = differential_from_compact(sd->I, sd->dI); + /* copy from shader data to shader globals */ globals->P = TO_VEC3(sd->P); - globals->dPdx = TO_VEC3(sd->dP.dx); - globals->dPdy = TO_VEC3(sd->dP.dy); + globals->dPdx = TO_VEC3(dP.dx); + globals->dPdy = TO_VEC3(dP.dy); globals->I = TO_VEC3(sd->I); - globals->dIdx = TO_VEC3(sd->dI.dx); - globals->dIdy = TO_VEC3(sd->dI.dy); + globals->dIdx = TO_VEC3(dI.dx); + globals->dIdy = TO_VEC3(dI.dy); globals->N = TO_VEC3(sd->N); globals->Ng = TO_VEC3(sd->Ng); globals->u = sd->u; @@ -183,9 +188,10 @@ void OSLShader::eval_surface(const KernelGlobalsCPU *kg, /* automatic bump shader */ if (kg->osl->bump_state[shader]) { /* save state */ - float3 P = sd->P; - float3 dPdx = sd->dP.dx; - float3 dPdy = sd->dP.dy; + const float3 P = sd->P; + const float dP = sd->dP; + const OSL::Vec3 dPdx = globals->dPdx; + const OSL::Vec3 dPdy = globals->dPdy; /* set state as if undisplaced */ if (sd->flag & SD_HAS_DISPLACEMENT) { @@ -199,17 +205,20 @@ void OSLShader::eval_surface(const KernelGlobalsCPU *kg, (void)found; assert(found); + differential3 tmp_dP; memcpy(&sd->P, data, sizeof(float) * 3); - memcpy(&sd->dP.dx, data + 3, sizeof(float) * 3); - memcpy(&sd->dP.dy, data + 6, sizeof(float) * 3); + memcpy(&tmp_dP.dx, data + 3, sizeof(float) * 3); + memcpy(&tmp_dP.dy, data + 6, sizeof(float) * 3); object_position_transform(kg, sd, &sd->P); - object_dir_transform(kg, sd, &sd->dP.dx); - object_dir_transform(kg, sd, &sd->dP.dy); + object_dir_transform(kg, sd, &tmp_dP.dx); + object_dir_transform(kg, sd, &tmp_dP.dy); + + sd->dP = differential_make_compact(tmp_dP); globals->P = TO_VEC3(sd->P); - globals->dPdx = TO_VEC3(sd->dP.dx); - globals->dPdy = TO_VEC3(sd->dP.dy); + globals->dPdx = TO_VEC3(tmp_dP.dx); + globals->dPdy = TO_VEC3(tmp_dP.dy); } /* execute bump shader */ @@ -217,8 +226,7 @@ void OSLShader::eval_surface(const KernelGlobalsCPU *kg, /* reset state */ sd->P = P; - sd->dP.dx = dPdx; - sd->dP.dy = dPdy; + sd->dP = dP; globals->P = TO_VEC3(P); globals->dPdx = TO_VEC3(dPdx); diff --git a/intern/cycles/kernel/svm/attribute.h b/intern/cycles/kernel/svm/attribute.h index a3609d8b4b0..5f0d1609f08 100644 --- a/intern/cycles/kernel/svm/attribute.h +++ b/intern/cycles/kernel/svm/attribute.h @@ -140,6 +140,16 @@ ccl_device_noinline void svm_node_attr(KernelGlobals kg, } } +ccl_device_forceinline float3 svm_node_bump_P_dx(const ccl_private ShaderData *sd) +{ + return sd->P + differential_from_compact(sd->Ng, sd->dP).dx; +} + +ccl_device_forceinline float3 svm_node_bump_P_dy(const ccl_private ShaderData *sd) +{ + return sd->P + differential_from_compact(sd->Ng, sd->dP).dy; +} + ccl_device_noinline void svm_node_attr_bump_dx(KernelGlobals kg, ccl_private ShaderData *sd, ccl_private float *stack, @@ -167,7 +177,7 @@ ccl_device_noinline void svm_node_attr_bump_dx(KernelGlobals kg, if (node.y == ATTR_STD_GENERATED && desc.element == ATTR_ELEMENT_NONE) { /* No generated attribute, fall back to object coordinates. */ - float3 f = sd->P + sd->dP.dx; + float3 f = svm_node_bump_P_dx(sd); if (sd->object != OBJECT_NONE) { object_inverse_position_transform(kg, sd, &f); } @@ -265,7 +275,7 @@ ccl_device_noinline void svm_node_attr_bump_dy(KernelGlobals kg, if (node.y == ATTR_STD_GENERATED && desc.element == ATTR_ELEMENT_NONE) { /* No generated attribute, fall back to object coordinates. */ - float3 f = sd->P + sd->dP.dy; + float3 f = svm_node_bump_P_dy(sd); if (sd->object != OBJECT_NONE) { object_inverse_position_transform(kg, sd, &f); } diff --git a/intern/cycles/kernel/svm/bump.h b/intern/cycles/kernel/svm/bump.h index 566c45f5f25..1009a6a4241 100644 --- a/intern/cycles/kernel/svm/bump.h +++ b/intern/cycles/kernel/svm/bump.h @@ -14,23 +14,21 @@ ccl_device_noinline void svm_node_enter_bump_eval(KernelGlobals kg, { /* save state */ stack_store_float3(stack, offset + 0, sd->P); - stack_store_float3(stack, offset + 3, sd->dP.dx); - stack_store_float3(stack, offset + 6, sd->dP.dy); + stack_store_float(stack, offset + 3, sd->dP); /* set state as if undisplaced */ const AttributeDescriptor desc = find_attribute(kg, sd, ATTR_STD_POSITION_UNDISPLACED); if (desc.offset != ATTR_STD_NOT_FOUND) { - float3 P, dPdx, dPdy; - P = primitive_surface_attribute_float3(kg, sd, desc, &dPdx, &dPdy); + differential3 dP; + float3 P = primitive_surface_attribute_float3(kg, sd, desc, &dP.dx, &dP.dy); object_position_transform(kg, sd, &P); - object_dir_transform(kg, sd, &dPdx); - object_dir_transform(kg, sd, &dPdy); + object_dir_transform(kg, sd, &dP.dx); + object_dir_transform(kg, sd, &dP.dy); sd->P = P; - sd->dP.dx = dPdx; - sd->dP.dy = dPdy; + sd->dP = differential_make_compact(dP); } } @@ -41,8 +39,7 @@ ccl_device_noinline void svm_node_leave_bump_eval(KernelGlobals kg, { /* restore state */ sd->P = stack_load_float3(stack, offset + 0); - sd->dP.dx = stack_load_float3(stack, offset + 3); - sd->dP.dy = stack_load_float3(stack, offset + 6); + sd->dP = stack_load_float(stack, offset + 3); } CCL_NAMESPACE_END diff --git a/intern/cycles/kernel/svm/displace.h b/intern/cycles/kernel/svm/displace.h index 128023263fd..230f8c73820 100644 --- a/intern/cycles/kernel/svm/displace.h +++ b/intern/cycles/kernel/svm/displace.h @@ -24,18 +24,17 @@ ccl_device_noinline void svm_node_set_bump(KernelGlobals kg, float3 normal_in = stack_valid(normal_offset) ? stack_load_float3(stack, normal_offset) : sd->N; - float3 dPdx = sd->dP.dx; - float3 dPdy = sd->dP.dy; + differential3 dP = differential_from_compact(sd->Ng, sd->dP); if (use_object_space) { object_inverse_normal_transform(kg, sd, &normal_in); - object_inverse_dir_transform(kg, sd, &dPdx); - object_inverse_dir_transform(kg, sd, &dPdy); + object_inverse_dir_transform(kg, sd, &dP.dx); + object_inverse_dir_transform(kg, sd, &dP.dy); } /* get surface tangents from normal */ - float3 Rx = cross(dPdy, normal_in); - float3 Ry = cross(normal_in, dPdx); + float3 Rx = cross(dP.dy, normal_in); + float3 Ry = cross(normal_in, dP.dx); /* get bump values */ uint c_offset, x_offset, y_offset, strength_offset; @@ -46,7 +45,7 @@ ccl_device_noinline void svm_node_set_bump(KernelGlobals kg, float h_y = stack_load_float(stack, y_offset); /* compute surface gradient and determinant */ - float det = dot(dPdx, Rx); + float det = dot(dP.dx, Rx); float3 surfgrad = (h_x - h_c) * Rx + (h_y - h_c) * Ry; float absdet = fabsf(det); diff --git a/intern/cycles/kernel/svm/geometry.h b/intern/cycles/kernel/svm/geometry.h index bbefdcfa755..cbd87d84409 100644 --- a/intern/cycles/kernel/svm/geometry.h +++ b/intern/cycles/kernel/svm/geometry.h @@ -54,7 +54,7 @@ ccl_device_noinline void svm_node_geometry_bump_dx(KernelGlobals kg, switch (type) { case NODE_GEOM_P: - data = sd->P + sd->dP.dx; + data = svm_node_bump_P_dx(sd); break; case NODE_GEOM_uv: data = make_float3(1.0f - sd->u - sd->du.dx - sd->v - sd->dv.dx, sd->u + sd->du.dx, 0.0f); @@ -81,7 +81,7 @@ ccl_device_noinline void svm_node_geometry_bump_dy(KernelGlobals kg, switch (type) { case NODE_GEOM_P: - data = sd->P + sd->dP.dy; + data = svm_node_bump_P_dy(sd); break; case NODE_GEOM_uv: data = make_float3(1.0f - sd->u - sd->du.dy - sd->v - sd->dv.dy, sd->u + sd->du.dy, 0.0f); diff --git a/intern/cycles/kernel/svm/tex_coord.h b/intern/cycles/kernel/svm/tex_coord.h index 2a0130e11d4..8154c542e6f 100644 --- a/intern/cycles/kernel/svm/tex_coord.h +++ b/intern/cycles/kernel/svm/tex_coord.h @@ -106,7 +106,7 @@ ccl_device_noinline int svm_node_tex_coord_bump_dx(KernelGlobals kg, switch (type) { case NODE_TEXCO_OBJECT: { - data = sd->P + sd->dP.dx; + data = svm_node_bump_P_dx(sd); if (node.w == 0) { if (sd->object != OBJECT_NONE) { object_inverse_position_transform(kg, sd, &data); @@ -130,9 +130,9 @@ ccl_device_noinline int svm_node_tex_coord_bump_dx(KernelGlobals kg, Transform tfm = kernel_data.cam.worldtocamera; if (sd->object != OBJECT_NONE) - data = transform_point(&tfm, sd->P + sd->dP.dx); + data = transform_point(&tfm, svm_node_bump_P_dx(sd)); else - data = transform_point(&tfm, sd->P + sd->dP.dx + camera_position(kg)); + data = transform_point(&tfm, svm_node_bump_P_dx(sd) + camera_position(kg)); break; } case NODE_TEXCO_WINDOW: { @@ -140,7 +140,7 @@ ccl_device_noinline int svm_node_tex_coord_bump_dx(KernelGlobals kg, kernel_data.cam.type == CAMERA_ORTHOGRAPHIC) data = camera_world_to_ndc(kg, sd, sd->ray_P); else - data = camera_world_to_ndc(kg, sd, sd->P + sd->dP.dx); + data = camera_world_to_ndc(kg, sd, svm_node_bump_P_dx(sd)); data.z = 0.0f; break; } @@ -160,7 +160,7 @@ ccl_device_noinline int svm_node_tex_coord_bump_dx(KernelGlobals kg, break; } case NODE_TEXCO_VOLUME_GENERATED: { - data = sd->P + sd->dP.dx; + data = svm_node_bump_P_dx(sd); # ifdef __VOLUME__ if (sd->object != OBJECT_NONE) @@ -191,7 +191,7 @@ ccl_device_noinline int svm_node_tex_coord_bump_dy(KernelGlobals kg, switch (type) { case NODE_TEXCO_OBJECT: { - data = sd->P + sd->dP.dy; + data = svm_node_bump_P_dy(sd); if (node.w == 0) { if (sd->object != OBJECT_NONE) { object_inverse_position_transform(kg, sd, &data); @@ -215,9 +215,9 @@ ccl_device_noinline int svm_node_tex_coord_bump_dy(KernelGlobals kg, Transform tfm = kernel_data.cam.worldtocamera; if (sd->object != OBJECT_NONE) - data = transform_point(&tfm, sd->P + sd->dP.dy); + data = transform_point(&tfm, svm_node_bump_P_dy(sd)); else - data = transform_point(&tfm, sd->P + sd->dP.dy + camera_position(kg)); + data = transform_point(&tfm, svm_node_bump_P_dy(sd) + camera_position(kg)); break; } case NODE_TEXCO_WINDOW: { @@ -225,7 +225,7 @@ ccl_device_noinline int svm_node_tex_coord_bump_dy(KernelGlobals kg, kernel_data.cam.type == CAMERA_ORTHOGRAPHIC) data = camera_world_to_ndc(kg, sd, sd->ray_P); else - data = camera_world_to_ndc(kg, sd, sd->P + sd->dP.dy); + data = camera_world_to_ndc(kg, sd, svm_node_bump_P_dy(sd)); data.z = 0.0f; break; } @@ -245,7 +245,7 @@ ccl_device_noinline int svm_node_tex_coord_bump_dy(KernelGlobals kg, break; } case NODE_TEXCO_VOLUME_GENERATED: { - data = sd->P + sd->dP.dy; + data = svm_node_bump_P_dy(sd); # ifdef __VOLUME__ if (sd->object != OBJECT_NONE) diff --git a/intern/cycles/kernel/svm/types.h b/intern/cycles/kernel/svm/types.h index 12d0ec141e6..98dfe6a4375 100644 --- a/intern/cycles/kernel/svm/types.h +++ b/intern/cycles/kernel/svm/types.h @@ -12,7 +12,7 @@ CCL_NAMESPACE_BEGIN /* SVM stack offsets with this value indicate that it's not on the stack */ #define SVM_STACK_INVALID 255 -#define SVM_BUMP_EVAL_STATE_SIZE 9 +#define SVM_BUMP_EVAL_STATE_SIZE 4 /* Nodes */ diff --git a/intern/cycles/kernel/svm/wireframe.h b/intern/cycles/kernel/svm/wireframe.h index e5fe08e5d04..91fadf4cfc4 100644 --- a/intern/cycles/kernel/svm/wireframe.h +++ b/intern/cycles/kernel/svm/wireframe.h @@ -14,6 +14,7 @@ CCL_NAMESPACE_BEGIN ccl_device_inline float wireframe(KernelGlobals kg, ccl_private ShaderData *sd, + const differential3 dP, float size, int pixel_size, ccl_private float3 *P) @@ -46,8 +47,8 @@ ccl_device_inline float wireframe(KernelGlobals kg, if (pixel_size) { // Project the derivatives of P to the viewing plane defined // by I so we have a measure of how big is a pixel at this point - float pixelwidth_x = len(sd->dP.dx - dot(sd->dP.dx, sd->I) * sd->I); - float pixelwidth_y = len(sd->dP.dy - dot(sd->dP.dy, sd->I) * sd->I); + float pixelwidth_x = len(dP.dx - dot(dP.dx, sd->I) * sd->I); + float pixelwidth_y = len(dP.dy - dot(dP.dy, sd->I) * sd->I); // Take the average of both axis' length pixelwidth = (pixelwidth_x + pixelwidth_y) * 0.5f; } @@ -86,16 +87,17 @@ ccl_device_noinline void svm_node_wireframe(KernelGlobals kg, int pixel_size = (int)use_pixel_size; /* Calculate wireframe */ - float f = wireframe(kg, sd, size, pixel_size, &sd->P); + const differential3 dP = differential_from_compact(sd->Ng, sd->dP); + float f = wireframe(kg, sd, dP, size, pixel_size, &sd->P); /* TODO(sergey): Think of faster way to calculate derivatives. */ if (bump_offset == NODE_BUMP_OFFSET_DX) { - float3 Px = sd->P - sd->dP.dx; - f += (f - wireframe(kg, sd, size, pixel_size, &Px)) / len(sd->dP.dx); + float3 Px = sd->P - dP.dx; + f += (f - wireframe(kg, sd, dP, size, pixel_size, &Px)) / len(dP.dx); } else if (bump_offset == NODE_BUMP_OFFSET_DY) { - float3 Py = sd->P - sd->dP.dy; - f += (f - wireframe(kg, sd, size, pixel_size, &Py)) / len(sd->dP.dy); + float3 Py = sd->P - dP.dy; + f += (f - wireframe(kg, sd, dP, size, pixel_size, &Py)) / len(dP.dy); } if (stack_valid(out_fac)) diff --git a/intern/cycles/kernel/types.h b/intern/cycles/kernel/types.h index 252c27a40f0..59ea6c64be7 100644 --- a/intern/cycles/kernel/types.h +++ b/intern/cycles/kernel/types.h @@ -873,10 +873,10 @@ typedef struct ccl_align(16) ShaderData float ray_length; #ifdef __RAY_DIFFERENTIALS__ - /* differential of P. these are orthogonal to Ng, not N */ - differential3 dP; - /* differential of I */ - differential3 dI; + /* Radius of differential of P. */ + float dP; + /* Radius of differential of I. */ + float dI; /* differential of u, v */ differential du; differential dv; diff --git a/intern/cycles/kernel/util/differential.h b/intern/cycles/kernel/util/differential.h index 3682e91ea66..aad9bb6bb22 100644 --- a/intern/cycles/kernel/util/differential.h +++ b/intern/cycles/kernel/util/differential.h @@ -101,53 +101,59 @@ ccl_device differential3 differential3_zero() return d; } -/* Compact ray differentials that are just a scale to reduce memory usage and - * access cost in GPU. +/* Compact ray differentials that are just a radius to reduce memory usage and access cost + * on GPUs, basically cone tracing. * - * See above for more accurate reference implementations. - * - * TODO: also store the more compact version in ShaderData and recompute where - * needed? */ + * See above for more accurate reference implementations of ray differentials. */ ccl_device_forceinline float differential_zero_compact() { return 0.0f; } -ccl_device_forceinline float differential_make_compact(const differential3 D) +ccl_device_forceinline float differential_make_compact(const float dD) { - return 0.5f * (len(D.dx) + len(D.dy)); + return dD; } -ccl_device_forceinline void differential_transfer_compact(ccl_private differential3 *surface_dP, - const float ray_dP, - const float3 /* ray_D */, - const float ray_dD, - const float3 surface_Ng, - const float ray_t) +ccl_device_forceinline float differential_make_compact(const differential3 dD) { - /* ray differential transfer through homogeneous medium, to - * compute dPdx/dy at a shading point from the incoming ray */ - float scale = ray_dP + ray_t * ray_dD; + return 0.5f * (len(dD.dx) + len(dD.dy)); +} - float3 dx, dy; - make_orthonormals(surface_Ng, &dx, &dy); - surface_dP->dx = dx * scale; - surface_dP->dy = dy * scale; +ccl_device_forceinline float differential_incoming_compact(const float dD) +{ + return dD; } -ccl_device_forceinline void differential_incoming_compact(ccl_private differential3 *dI, - const float3 D, - const float dD) +ccl_device_forceinline float differential_transfer_compact(const float ray_dP, + const float3 /* ray_D */, + const float ray_dD, + const float ray_t) { - /* compute dIdx/dy at a shading point, we just need to negate the - * differential of the ray direction */ + return ray_dP + ray_t * ray_dD; +} +ccl_device_forceinline differential3 differential_from_compact(const float3 D, const float dD) +{ float3 dx, dy; make_orthonormals(D, &dx, &dy); - dI->dx = dD * dx; - dI->dy = dD * dy; + differential3 d; + d.dx = dD * dx; + d.dy = dD * dy; + return d; +} + +ccl_device void differential_dudv_compact(ccl_private differential *du, + ccl_private differential *dv, + float3 dPdu, + float3 dPdv, + float dP, + float3 Ng) +{ + /* TODO: can we speed this up? */ + differential_dudv(du, dv, dPdu, dPdv, differential_from_compact(Ng, dP), Ng); } CCL_NAMESPACE_END diff --git a/intern/cycles/scene/camera.cpp b/intern/cycles/scene/camera.cpp index eec269ab542..d9e574873bd 100644 --- a/intern/cycles/scene/camera.cpp +++ b/intern/cycles/scene/camera.cpp @@ -772,10 +772,7 @@ float Camera::world_to_raster_size(float3 P) #endif /* TODO: would it help to use more accurate differentials here? */ - differential3 dP; - differential_transfer_compact(&dP, ray.dP, ray.D, ray.dD, ray.D, dist); - - return max(len(dP.dx), len(dP.dy)); + return differential_transfer_compact(ray.dP, ray.D, ray.dD, dist); } return res; |