diff options
Diffstat (limited to 'intern/cycles/kernel/geom/geom_triangle_intersect.h')
-rw-r--r-- | intern/cycles/kernel/geom/geom_triangle_intersect.h | 478 |
1 files changed, 475 insertions, 3 deletions
diff --git a/intern/cycles/kernel/geom/geom_triangle_intersect.h b/intern/cycles/kernel/geom/geom_triangle_intersect.h index a2b1e050e58..f8c671fed14 100644 --- a/intern/cycles/kernel/geom/geom_triangle_intersect.h +++ b/intern/cycles/kernel/geom/geom_triangle_intersect.h @@ -1,4 +1,4 @@ -/* + /* * Copyright 2014, Blender Foundation. * * Licensed under the Apache License, Version 2.0 (the "License"); @@ -70,7 +70,479 @@ ccl_device_inline bool triangle_intersect(KernelGlobals *kg, return false; } -/* Special ray intersection routines for local intersection. In that case we +#define cross256(A,B, C,D) _mm256_fmsub_ps(A,B, _mm256_mul_ps(C,D)) +#if defined(__KERNEL_CUDA__) && __CUDA_ARCH__ < 300 +ccl_device_inline +#else +ccl_device_forceinline +#endif +int ray_triangle_intersect8(KernelGlobals *kg, + float3 ray_P, + float3 ray_dir, + Intersection **isect, + uint visibility, + int object, + __m256 *triA, + __m256 *triB, + __m256 *triC, + int prim_addr, + int prim_num, + uint *num_hits, + uint max_hits, + int *num_hits_in_instance, + float isec_t) +{ + + const unsigned char prim_num_mask = (1 << prim_num) - 1; + + const __m256i zero256 = _mm256_setzero_si256(); + + const __m256 Px256 = _mm256_set1_ps(ray_P.x); + const __m256 Py256 = _mm256_set1_ps(ray_P.y); + const __m256 Pz256 = _mm256_set1_ps(ray_P.z); + + const __m256 dirx256 = _mm256_set1_ps(ray_dir.x); + const __m256 diry256 = _mm256_set1_ps(ray_dir.y); + const __m256 dirz256 = _mm256_set1_ps(ray_dir.z); + + /* Calculate vertices relative to ray origin. */ + /* const float3 v0 = tri_c - P; + const float3 v1 = tri_a - P; + const float3 v2 = tri_b - P; */ + + __m256 v0_x_256 = _mm256_sub_ps(triC[0], Px256); + __m256 v0_y_256 = _mm256_sub_ps(triC[1], Py256); + __m256 v0_z_256 = _mm256_sub_ps(triC[2], Pz256); + + __m256 v1_x_256 = _mm256_sub_ps(triA[0], Px256); + __m256 v1_y_256 = _mm256_sub_ps(triA[1], Py256); + __m256 v1_z_256 = _mm256_sub_ps(triA[2], Pz256); + + __m256 v2_x_256 = _mm256_sub_ps(triB[0], Px256); + __m256 v2_y_256 = _mm256_sub_ps(triB[1], Py256); + __m256 v2_z_256 = _mm256_sub_ps(triB[2], Pz256); + + __m256 v0_v1_x_256 = _mm256_add_ps(v0_x_256, v1_x_256); + __m256 v0_v1_y_256 = _mm256_add_ps(v0_y_256, v1_y_256); + __m256 v0_v1_z_256 = _mm256_add_ps(v0_z_256, v1_z_256); + + __m256 v0_v2_x_256 = _mm256_add_ps(v0_x_256, v2_x_256); + __m256 v0_v2_y_256 = _mm256_add_ps(v0_y_256, v2_y_256); + __m256 v0_v2_z_256 = _mm256_add_ps(v0_z_256, v2_z_256); + + __m256 v1_v2_x_256 = _mm256_add_ps(v1_x_256, v2_x_256); + __m256 v1_v2_y_256 = _mm256_add_ps(v1_y_256, v2_y_256); + __m256 v1_v2_z_256 = _mm256_add_ps(v1_z_256, v2_z_256); + + /* Calculate triangle edges. + const float3 e0 = v2 - v0; + const float3 e1 = v0 - v1; + const float3 e2 = v1 - v2;*/ + + __m256 e0_x_256 = _mm256_sub_ps(v2_x_256, v0_x_256); + __m256 e0_y_256 = _mm256_sub_ps(v2_y_256, v0_y_256); + __m256 e0_z_256 = _mm256_sub_ps(v2_z_256, v0_z_256); + + __m256 e1_x_256 = _mm256_sub_ps(v0_x_256, v1_x_256); + __m256 e1_y_256 = _mm256_sub_ps(v0_y_256, v1_y_256); + __m256 e1_z_256 = _mm256_sub_ps(v0_z_256, v1_z_256); + + __m256 e2_x_256 = _mm256_sub_ps(v1_x_256, v2_x_256); + __m256 e2_y_256 = _mm256_sub_ps(v1_y_256, v2_y_256); + __m256 e2_z_256 = _mm256_sub_ps(v1_z_256, v2_z_256); + + /* Perform edge tests. + const float U = dot(cross(v2 + v0, e0), ray_dir); + const float V = dot(cross(v0 + v1, e1), ray_dir); + const float W = dot(cross(v1 + v2, e2), ray_dir);*/ + + //cross (AyBz - AzBy, AzBx -AxBz, AxBy - AyBx) + __m256 U_x_256 = cross256(v0_v2_y_256, e0_z_256, v0_v2_z_256, e0_y_256); + __m256 U_y_256 = cross256(v0_v2_z_256, e0_x_256, v0_v2_x_256, e0_z_256); + __m256 U_z_256 = cross256(v0_v2_x_256, e0_y_256, v0_v2_y_256, e0_x_256); + //vertical dot + __m256 U_256 = _mm256_mul_ps(U_x_256, dirx256); + U_256 = _mm256_fmadd_ps(U_y_256, diry256, U_256); //_mm256_add_ps(U_256, _mm256_mul_ps(U_y_256, diry256)); + U_256 = _mm256_fmadd_ps(U_z_256, dirz256, U_256); //_mm256_add_ps(U_256, _mm256_mul_ps(U_z_256, dirz256)); + + __m256 V_x_256 = cross256(v0_v1_y_256, e1_z_256, v0_v1_z_256, e1_y_256); + __m256 V_y_256 = cross256(v0_v1_z_256, e1_x_256, v0_v1_x_256, e1_z_256); + __m256 V_z_256 = cross256(v0_v1_x_256, e1_y_256, v0_v1_y_256, e1_x_256); + //vertical dot + __m256 V_256 = _mm256_mul_ps(V_x_256, dirx256); + V_256 = _mm256_fmadd_ps(V_y_256, diry256, V_256);// _mm256_add_ps(V_256, _mm256_mul_ps(V_y_256, diry256)); + V_256 = _mm256_fmadd_ps(V_z_256, dirz256, V_256);// _mm256_add_ps(V_256, _mm256_mul_ps(V_z_256, dirz256)); + + __m256 W_x_256 = cross256(v1_v2_y_256, e2_z_256, v1_v2_z_256, e2_y_256); + __m256 W_y_256 = cross256(v1_v2_z_256, e2_x_256, v1_v2_x_256, e2_z_256); + __m256 W_z_256 = cross256(v1_v2_x_256, e2_y_256, v1_v2_y_256, e2_x_256); + //vertical dot + __m256 W_256 = _mm256_mul_ps(W_x_256, dirx256); + W_256 = _mm256_fmadd_ps(W_y_256, diry256,W_256);//_mm256_add_ps(W_256, _mm256_mul_ps(W_y_256, diry256)); + W_256 = _mm256_fmadd_ps(W_z_256, dirz256,W_256);//_mm256_add_ps(W_256, _mm256_mul_ps(W_z_256, dirz256)); + + //const float minUVW = min(U, min(V, W)); + //const float maxUVW = max(U, max(V, W)); +#if 0 + __m256 minUVW_256 = _mm256_min_ps(U_256, _mm256_min_ps(V_256, W_256)); + __m256 maxUVW_256 = _mm256_max_ps(U_256, _mm256_max_ps(V_256, W_256)); + + //if(minUVW < 0.0f && maxUVW > 0.0f) + __m256i mask_minmaxUVW_256 = _mm256_and_si256( + _mm256_cmpgt_epi32(zero256, _mm256_castps_si256(minUVW_256)), + //_mm256_castps_si256(minUVW_256), + _mm256_cmpgt_epi32(_mm256_castps_si256(maxUVW_256), zero256)); +#else + __m256i U_256_1 = _mm256_srli_epi32(_mm256_castps_si256(U_256), 31); + __m256i V_256_1 = _mm256_srli_epi32(_mm256_castps_si256(V_256), 31); + __m256i W_256_1 = _mm256_srli_epi32(_mm256_castps_si256(W_256), 31); + __m256i UVW_256_1 = _mm256_add_epi32(_mm256_add_epi32(U_256_1, V_256_1), W_256_1); + + const __m256i one256 = _mm256_set1_epi32(1); + const __m256i two256 = _mm256_set1_epi32(2); + + __m256i mask_minmaxUVW_256 = _mm256_or_si256( + _mm256_cmpeq_epi32(one256, UVW_256_1), + _mm256_cmpeq_epi32(two256, UVW_256_1) ); +#endif + + unsigned char mask_minmaxUVW_pos = _mm256_movemask_ps(_mm256_castsi256_ps(mask_minmaxUVW_256)); + if((mask_minmaxUVW_pos & prim_num_mask) == prim_num_mask) { //all bits set + return false; + } + + /* Calculate geometry normal and denominator. */ + // const float3 Ng1 = cross(e1, e0); + //const Vec3vfM Ng1 = stable_triangle_normal(e2,e1,e0); + + __m256 Ng1_x_256 = cross256(e1_y_256, e0_z_256, e1_z_256, e0_y_256); + __m256 Ng1_y_256 = cross256(e1_z_256, e0_x_256, e1_x_256, e0_z_256); + __m256 Ng1_z_256 = cross256(e1_x_256, e0_y_256, e1_y_256, e0_x_256); + + //const float3 Ng = Ng1 + Ng1; + Ng1_x_256 = _mm256_add_ps(Ng1_x_256, Ng1_x_256); + Ng1_y_256 = _mm256_add_ps(Ng1_y_256, Ng1_y_256); + Ng1_z_256 = _mm256_add_ps(Ng1_z_256, Ng1_z_256); + + //const float den = dot3(Ng, dir); + //vertical dot + __m256 den_256 = _mm256_mul_ps(Ng1_x_256, dirx256); + den_256 = _mm256_fmadd_ps(Ng1_y_256, diry256,den_256);//_mm256_add_ps(den_256, _mm256_mul_ps(Ng1_y_256, diry256)); + den_256 = _mm256_fmadd_ps(Ng1_z_256, dirz256,den_256);//_mm256_add_ps(den_256, _mm256_mul_ps(Ng1_z_256, dirz256)); + + // __m256i maskden256 = _mm256_cmpeq_epi32(_mm256_castps_si256(den_256), zero256); + + /* Perform depth test. */ + //const float T = dot3(v0, Ng); + __m256 T_256 = _mm256_mul_ps(Ng1_x_256, v0_x_256); + T_256 = _mm256_fmadd_ps(Ng1_y_256, v0_y_256,T_256);//_mm256_add_ps(T_256, _mm256_mul_ps(Ng1_y_256, v0_y_256)); + T_256 = _mm256_fmadd_ps(Ng1_z_256, v0_z_256,T_256);//_mm256_add_ps(T_256, _mm256_mul_ps(Ng1_z_256, v0_z_256)); + + //const int sign_den = (__float_as_int(den) & 0x80000000); + const __m256i c0x80000000 = _mm256_set1_epi32(0x80000000); + __m256i sign_den_256 = _mm256_and_si256(_mm256_castps_si256(den_256), c0x80000000); + + //const float sign_T = xor_signmask(T, sign_den); + __m256 sign_T_256 = _mm256_castsi256_ps(_mm256_xor_si256(_mm256_castps_si256(T_256), sign_den_256)); + + /*if((sign_T < 0.0f) || mask_minmaxUVW_pos { return false;} */ + unsigned char mask_sign_T = _mm256_movemask_ps(sign_T_256); + if(((mask_minmaxUVW_pos | mask_sign_T) & prim_num_mask) == prim_num_mask) { + return false; + } /**/ + + __m256 xor_signmask_256 = _mm256_castsi256_ps(_mm256_xor_si256(_mm256_castps_si256(den_256), sign_den_256)); + + + ccl_align(32) float den8[8], U8[8], V8[8], T8[8], sign_T8[8], xor_signmask8[8]; + ccl_align(32) unsigned int mask_minmaxUVW8[8]; + + if(visibility == PATH_RAY_SHADOW_OPAQUE){ + __m256i mask_final_256 = _mm256_cmpeq_epi32(mask_minmaxUVW_256, zero256);//~mask_minmaxUVW_256 + + __m256i maskden256 = _mm256_cmpeq_epi32(_mm256_castps_si256(den_256), zero256); + + __m256i mask0 = _mm256_cmpgt_epi32(zero256, _mm256_castps_si256(sign_T_256)); + __m256 rayt_256 = _mm256_set1_ps((*isect)->t); + + __m256i mask1 = _mm256_cmpgt_epi32(_mm256_castps_si256(sign_T_256), + _mm256_castps_si256( + _mm256_mul_ps(_mm256_castsi256_ps(_mm256_xor_si256(_mm256_castps_si256(den_256), sign_den_256)), rayt_256) + ) + ); + /* __m256i mask1 = _mm256_castps_si256(_mm256_cmp_ps(sign_T_256, + _mm256_mul_ps(_mm256_castsi256_ps(_mm256_xor_si256(_mm256_castps_si256(den_256), sign_den_256)), rayt_256), + _CMP_GT_OS + ) );*/ + + mask0 = _mm256_or_si256(mask1, mask0); + //unsigned char mask = _mm256_movemask_ps(_mm256_castsi256_ps(mask0)); + //unsigned char maskden = _mm256_movemask_ps(_mm256_castsi256_ps(maskden256)); + //unsigned char mask_final = ((~mask) & (~maskden) & (~mask_minmaxUVW_pos)); + mask_final_256 = _mm256_andnot_si256(mask0, mask_final_256); //(~mask_minmaxUVW_pos) &(~mask) + mask_final_256 = _mm256_andnot_si256(maskden256, mask_final_256); //(~mask_minmaxUVW_pos) &(~mask) & (~maskden) + + unsigned char mask_final = _mm256_movemask_ps(_mm256_castsi256_ps(mask_final_256)); + if((mask_final & prim_num_mask) == 0) { //all bits NOT set + return false; + } /**/ + + unsigned long i = 0; +#if defined(_MSC_VER) + unsigned char res = _BitScanForward(&i, (unsigned long)mask_final); +#else + i = __builtin_ffs(mask_final)-1; +#endif + + den_256 = _mm256_rcp_ps(den_256); //inv_den + U_256 = _mm256_mul_ps(U_256, den_256); //*inv_den + V_256 = _mm256_mul_ps(V_256, den_256); //*inv_den + T_256 = _mm256_mul_ps(T_256, den_256); //*inv_den + + _mm256_store_ps(U8, U_256); + _mm256_store_ps(V8, V_256); + _mm256_store_ps(T8, T_256); + + + //here we assume (kernel_tex_fetch(__prim_visibility, (prim_addr +i)) & visibility) is always true + + (*isect)->u = U8[i]; + (*isect)->v = V8[i]; + (*isect)->t = T8[i]; + + (*isect)->prim = (prim_addr + i); + (*isect)->object = object; + (*isect)->type = PRIMITIVE_TRIANGLE; + + return true; + } + else { + _mm256_store_ps(den8, den_256); + _mm256_store_ps(U8, U_256); + _mm256_store_ps(V8, V_256); + _mm256_store_ps(T8, T_256); + + _mm256_store_ps(sign_T8, sign_T_256); + _mm256_store_ps(xor_signmask8, xor_signmask_256); + _mm256_store_si256((__m256i*)mask_minmaxUVW8, mask_minmaxUVW_256); + + int ret = false; + + if(visibility == PATH_RAY_SHADOW) { + for(int i = 0; i < prim_num; i++) { + if(!mask_minmaxUVW8[i]) { +#ifdef __VISIBILITY_FLAG__ + if(kernel_tex_fetch(__prim_visibility, (prim_addr + i)) & visibility) +#endif + { + if((sign_T8[i] >= 0.0f) && + (sign_T8[i] <= (*isect)->t * xor_signmask8[i])) + { + if(den8[i]) { + const float inv_den = 1.0f / den8[i]; + + (*isect)->u = U8[i] * inv_den; + (*isect)->v = V8[i] * inv_den; + (*isect)->t = T8[i] * inv_den; + + (*isect)->prim = (prim_addr + i); + (*isect)->object = object; + (*isect)->type = PRIMITIVE_TRIANGLE; + + int prim = kernel_tex_fetch(__prim_index, (*isect)->prim); + int shader = 0; + +#ifdef __HAIR__ + if(kernel_tex_fetch(__prim_type, (*isect)->prim) & PRIMITIVE_ALL_TRIANGLE) +#endif + { + shader = kernel_tex_fetch(__tri_shader, prim); + } +#ifdef __HAIR__ + else { + float4 str = kernel_tex_fetch(__curves, prim); + shader = __float_as_int(str.z); + } +#endif + int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags; + + /* if no transparent shadows, all light is blocked */ + if(!(flag & SD_HAS_TRANSPARENT_SHADOW)) { + return 2; + } + /* if maximum number of hits reached, block all light */ + else if(*num_hits == max_hits) { + return 2; + } + /* move on to next entry in intersections array */ + ret = true; + + (*isect)++; + (*num_hits)++; + + (*num_hits_in_instance)++; + + (*isect)->t = isec_t; + + } //den + } //if sign + } //vis + }//if mask + } //for + } + else { //default case + for(int i = 0; i < prim_num; i++) { + if(!mask_minmaxUVW8[i]) { +#ifdef __VISIBILITY_FLAG__ + if(kernel_tex_fetch(__prim_visibility, (prim_addr + i)) & visibility) +#endif + { + if((sign_T8[i] >= 0.0f) && + (sign_T8[i] <= (*isect)->t * xor_signmask8[i])) + { + if(den8[i]) { + const float inv_den = 1.0f / den8[i]; + + (*isect)->u = U8[i] * inv_den; + (*isect)->v = V8[i] * inv_den; + (*isect)->t = T8[i] * inv_den; + + (*isect)->prim = (prim_addr + i); + (*isect)->object = object; + (*isect)->type = PRIMITIVE_TRIANGLE; + + ret = true; + } //den + } //if sign + } //vis + }//if mask + } //for + } //default + return ret; +}// else PATH_RAY_SHADOW_OPAQUE + +} + +//vz static +ccl_device_inline +int triangle_intersect8(KernelGlobals *kg, + Intersection **isect, + float3 P, + float3 dir, + uint visibility, + int object, + int prim_addr, + int prim_num, + uint *num_hits, + uint max_hits, + int *num_hits_in_instance, + float isec_t) + { + __m128 tri_a[8], tri_b[8], tri_c[8]; + __m256 tritmp[12], tri[12]; + __m256 triA[3], triB[3], triC[3]; + + int i, r; + + uint tri_vindex = kernel_tex_fetch(__prim_tri_index, prim_addr); + for(i = 0; i < prim_num; i++) { + tri_a[i] = *(__m128*)&kg->__prim_tri_verts.data[tri_vindex++]; + tri_b[i] = *(__m128*)&kg->__prim_tri_verts.data[tri_vindex++]; + tri_c[i] = *(__m128*)&kg->__prim_tri_verts.data[tri_vindex++]; + } + //create 9 or 12 placeholders + tri[0] = _mm256_castps128_ps256(tri_a[0]); //_mm256_zextps128_ps256 + tri[1] = _mm256_castps128_ps256(tri_b[0]);//_mm256_zextps128_ps256 + tri[2] = _mm256_castps128_ps256(tri_c[0]);//_mm256_zextps128_ps256 + + tri[3] = _mm256_castps128_ps256(tri_a[1]); //_mm256_zextps128_ps256 + tri[4] = _mm256_castps128_ps256(tri_b[1]);//_mm256_zextps128_ps256 + tri[5] = _mm256_castps128_ps256(tri_c[1]);//_mm256_zextps128_ps256 + + tri[6] = _mm256_castps128_ps256(tri_a[2]); //_mm256_zextps128_ps256 + tri[7] = _mm256_castps128_ps256(tri_b[2]);//_mm256_zextps128_ps256 + tri[8] = _mm256_castps128_ps256(tri_c[2]);//_mm256_zextps128_ps256 + + if(prim_num > 3) { + tri[9] = _mm256_castps128_ps256(tri_a[3]); //_mm256_zextps128_ps256 + tri[10] = _mm256_castps128_ps256(tri_b[3]);//_mm256_zextps128_ps256 + tri[11] = _mm256_castps128_ps256(tri_c[3]);//_mm256_zextps128_ps256 + } + + for(i = 4, r = 0; i < prim_num; i ++, r += 3) { + tri[r] = _mm256_insertf128_ps(tri[r] , tri_a[i], 1); + tri[r + 1] = _mm256_insertf128_ps(tri[r + 1], tri_b[i], 1); + tri[r + 2] = _mm256_insertf128_ps(tri[r + 2], tri_c[i], 1); + } + + //------------------------------------------------ + //0! Xa0 Ya0 Za0 1 Xa4 Ya4 Za4 1 + //1! Xb0 Yb0 Zb0 1 Xb4 Yb4 Zb4 1 + //2! Xc0 Yc0 Zc0 1 Xc4 Yc4 Zc4 1 + + //3! Xa1 Ya1 Za1 1 Xa5 Ya5 Za5 1 + //4! Xb1 Yb1 Zb1 1 Xb5 Yb5 Zb5 1 + //5! Xc1 Yc1 Zc1 1 Xc5 Yc5 Zc5 1 + + //6! Xa2 Ya2 Za2 1 Xa6 Ya6 Za6 1 + //7! Xb2 Yb2 Zb2 1 Xb6 Yb6 Zb6 1 + //8! Xc2 Yc2 Zc2 1 Xc6 Yc6 Zc6 1 + + //9! Xa3 Ya3 Za3 1 Xa7 Ya7 Za7 1 + //10! Xb3 Yb3 Zb3 1 Xb7 Yb7 Zb7 1 + //11! Xc3 Yc3 Zc3 1 Xc7 Yc7 Zc7 1 + + //"transpose" + tritmp[0] = _mm256_unpacklo_ps(tri[0], tri[3]); //0! Xa0 Xa1 Ya0 Ya1 Xa4 Xa5 Ya4 Ya5 + tritmp[1] = _mm256_unpackhi_ps(tri[0], tri[3]); //1! Za0 Za1 1 1 Za4 Za5 1 1 + + tritmp[2] = _mm256_unpacklo_ps(tri[6], tri[9]); //2! Xa2 Xa3 Ya2 Ya3 Xa6 Xa7 Ya6 Ya7 + tritmp[3] = _mm256_unpackhi_ps(tri[6], tri[9]); //3! Za2 Za3 1 1 Za6 Za7 1 1 + + tritmp[4] = _mm256_unpacklo_ps(tri[1], tri[4]); //4! Xb0 Xb1 Yb0 Yb1 Xb4 Xb5 Yb4 Yb5 + tritmp[5] = _mm256_unpackhi_ps(tri[1], tri[4]); //5! Zb0 Zb1 1 1 Zb4 Zb5 1 1 + + tritmp[6] = _mm256_unpacklo_ps(tri[7], tri[10]); //6! Xb2 Xb3 Yb2 Yb3 Xb6 Xb7 Yb6 Yb7 + tritmp[7] = _mm256_unpackhi_ps(tri[7], tri[10]); //7! Zb2 Zb3 1 1 Zb6 Zb7 1 1 + + tritmp[8] = _mm256_unpacklo_ps(tri[2], tri[5]); //8! Xc0 Xc1 Yc0 Yc1 Xc4 Xc5 Yc4 Yc5 + tritmp[9] = _mm256_unpackhi_ps(tri[2], tri[5]); //9! Zc0 Zc1 1 1 Zc4 Zc5 1 1 + + tritmp[10] = _mm256_unpacklo_ps(tri[8], tri[11]); //10! Xc2 Xc3 Yc2 Yc3 Xc6 Xc7 Yc6 Yc7 + tritmp[11] = _mm256_unpackhi_ps(tri[8], tri[11]); //11! Zc2 Zc3 1 1 Zc6 Zc7 1 1 + + /*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/ + triA[0] = _mm256_castpd_ps(_mm256_unpacklo_pd(_mm256_castps_pd(tritmp[0]), _mm256_castps_pd(tritmp[2]))); // Xa0 Xa1 Xa2 Xa3 Xa4 Xa5 Xa6 Xa7 + triA[1] = _mm256_castpd_ps(_mm256_unpackhi_pd(_mm256_castps_pd(tritmp[0]), _mm256_castps_pd(tritmp[2]))); // Ya0 Ya1 Ya2 Ya3 Ya4 Ya5 Ya6 Ya7 + triA[2] = _mm256_castpd_ps(_mm256_unpacklo_pd(_mm256_castps_pd(tritmp[1]), _mm256_castps_pd(tritmp[3]))); // Za0 Za1 Za2 Za3 Za4 Za5 Za6 Za7 + + triB[0] = _mm256_castpd_ps(_mm256_unpacklo_pd(_mm256_castps_pd(tritmp[4]), _mm256_castps_pd(tritmp[6]))); // Xb0 Xb1 Xb2 Xb3 Xb4 Xb5 Xb5 Xb7 + triB[1] = _mm256_castpd_ps(_mm256_unpackhi_pd(_mm256_castps_pd(tritmp[4]), _mm256_castps_pd(tritmp[6]))); // Yb0 Yb1 Yb2 Yb3 Yb4 Yb5 Yb5 Yb7 + triB[2] = _mm256_castpd_ps(_mm256_unpacklo_pd(_mm256_castps_pd(tritmp[5]), _mm256_castps_pd(tritmp[7]))); // Zb0 Zb1 Zb2 Zb3 Zb4 Zb5 Zb5 Zb7 + + triC[0] = _mm256_castpd_ps(_mm256_unpacklo_pd(_mm256_castps_pd(tritmp[8]), _mm256_castps_pd(tritmp[10]))); //Xc0 Xc1 Xc2 Xc3 Xc4 Xc5 Xc6 Xc7 + triC[1] = _mm256_castpd_ps(_mm256_unpackhi_pd(_mm256_castps_pd(tritmp[8]), _mm256_castps_pd(tritmp[10]))); //Yc0 Yc1 Yc2 Yc3 Yc4 Yc5 Yc6 Yc7 + triC[2] = _mm256_castpd_ps(_mm256_unpacklo_pd(_mm256_castps_pd(tritmp[9]), _mm256_castps_pd(tritmp[11]))); //Zc0 Zc1 Zc2 Zc3 Zc4 Zc5 Zc6 Zc7 + + /*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/ + + int result = ray_triangle_intersect8(kg, P, + dir, + isect, + visibility, object, + triA, + triB, + triC, + prim_addr, + prim_num, + num_hits, + max_hits, + num_hits_in_instance, + isec_t); + return result; +} + +/* Special ray intersection routines for subsurface scattering. In that case we * only want to intersect with primitives in the same object, and if case of * multiple hits we pick a single random primitive as the intersection point. * Returns whether traversal should be stopped. @@ -83,7 +555,7 @@ ccl_device_inline bool triangle_intersect_local( float3 P, float3 dir, int object, - int local_object, + int local_object, int prim_addr, float tmax, uint *lcg_state, |