Cycles: Use utility function for SSS triangle intersection

This effectively de-duplicates triangle intersection logic implemented
for both regular triangle and SSS triangle.

1 files changed, 20 insertions, 116 deletions

diff --git a/intern/cycles/kernel/geom/geom_triangle_intersect.h b/intern/cycles/kernel/geom/geom_triangle_intersect.h
index ea2ee5c3536..973b3566378 100644
--- a/intern/cycles/kernel/geom/geom_triangle_intersect.h
+++ b/intern/cycles/kernel/geom/geom_triangle_intersect.h
@@ -102,129 +102,27 @@ ccl_device_inline void triangle_intersect_subsurface(
         uint *lcg_state,
         int max_hits)
 {
-	const int kx = isect_precalc->kx;
-	const int ky = isect_precalc->ky;
-	const int kz = isect_precalc->kz;
-	const float Sx = isect_precalc->Sx;
-	const float Sy = isect_precalc->Sy;
-	const float Sz = isect_precalc->Sz;
-
-	/* Calculate vertices relative to ray origin. */
 	const uint tri_vindex = kernel_tex_fetch(__prim_tri_index, prim_addr);
-	const float4 tri_a = kernel_tex_fetch(__prim_tri_verts, tri_vindex+0),
-	             tri_b = kernel_tex_fetch(__prim_tri_verts, tri_vindex+1),
-	             tri_c = kernel_tex_fetch(__prim_tri_verts, tri_vindex+2);
 
 #if defined(__KERNEL_AVX2__) && defined(__KERNEL_SSE__)
-	const avxf avxf_P(P.m128, P.m128);
-
-	const avxf tri_ab = kernel_tex_fetch_avxf(__prim_tri_verts, tri_vindex + 0);
-	const avxf tri_bc = kernel_tex_fetch_avxf(__prim_tri_verts, tri_vindex + 1);
-
-	const avxf AB = tri_ab - avxf_P;
-	const avxf BC = tri_bc - avxf_P;
-
-	const __m256i permuteMask = _mm256_set_epi32(0x3, kz, ky, kx, 0x3, kz, ky, kx);
-
-	const avxf AB_k = shuffle(AB, permuteMask);
-	const avxf BC_k = shuffle(BC, permuteMask);
-
-	/* Akz, Akz, Bkz, Bkz, Bkz, Bkz, Ckz, Ckz */
-	const avxf ABBC_kz = shuffle<2>(AB_k, BC_k);
-
-	/* Akx, Aky, Bkx, Bky, Bkx,Bky, Ckx, Cky */
-	const avxf ABBC_kxy = shuffle<0,1,0,1>(AB_k, BC_k);
-
-	const avxf Sxy(Sy, Sx, Sy, Sx);
-
-	/* Ax, Ay, Bx, By, Bx, By, Cx, Cy */
-	const avxf ABBC_xy = nmadd(ABBC_kz, Sxy, ABBC_kxy);
-
-	float ABBC_kz_array[8];
-	_mm256_storeu_ps((float*)&ABBC_kz_array, ABBC_kz);
-
-	const float A_kz = ABBC_kz_array[0];
-	const float B_kz = ABBC_kz_array[2];
-	const float C_kz = ABBC_kz_array[6];
-
-	/* By, Bx, Cy, Cx, By, Bx, Ay, Ax */
-	const avxf BCBA_yx = permute<3,2,7,6,3,2,1,0>(ABBC_xy);
-
-	const avxf negMask(0,0,0,0,0x80000000, 0x80000000, 0x80000000, 0x80000000);
-
-	/* W           U                             V
-	 * (AxBy-AyBx) (BxCy-ByCx) XX XX (BxBy-ByBx) (CxAy-CyAx) XX XX
-	 */
-	const avxf WUxxxxVxx_neg = _mm256_hsub_ps(ABBC_xy * BCBA_yx, negMask /* Dont care */);
-
-	const avxf WUVWnegWUVW = permute<0,1,5,0,0,1,5,0>(WUxxxxVxx_neg) ^ negMask;
-
-	/* Calculate scaled barycentric coordinates. */
-	float WUVW_array[4];
-	_mm_storeu_ps((float*)&WUVW_array, _mm256_castps256_ps128 (WUVWnegWUVW));
-
-	const float W = WUVW_array[0];
-	const float U = WUVW_array[1];
-	const float V = WUVW_array[2];
-
-	const int WUVW_mask = 0x7 & _mm256_movemask_ps(WUVWnegWUVW);
-	const int WUVW_zero = 0x7 & _mm256_movemask_ps(_mm256_cmp_ps(WUVWnegWUVW,
-	                                               _mm256_setzero_ps(), 0));
-
-	if(!((WUVW_mask == 7) || (WUVW_mask == 0)) && ((WUVW_mask | WUVW_zero) != 7)) {
-		return;
-	}
+	const ssef *verts = (ssef*)&kg->__prim_tri_verts.data[tri_vindex];
 #else
-	const float3 A = make_float3(tri_a.x - P.x, tri_a.y - P.y, tri_a.z - P.z);
-	const float3 B = make_float3(tri_b.x - P.x, tri_b.y - P.y, tri_b.z - P.z);
-	const float3 C = make_float3(tri_c.x - P.x, tri_c.y - P.y, tri_c.z - P.z);
-
-	const float A_kx = IDX(A, kx), A_ky = IDX(A, ky), A_kz = IDX(A, kz);
-	const float B_kx = IDX(B, kx), B_ky = IDX(B, ky), B_kz = IDX(B, kz);
-	const float C_kx = IDX(C, kx), C_ky = IDX(C, ky), C_kz = IDX(C, kz);
-
-	/* Perform shear and scale of vertices. */
-	const float Ax = A_kx - Sx * A_kz;
-	const float Ay = A_ky - Sy * A_kz;
-	const float Bx = B_kx - Sx * B_kz;
-	const float By = B_ky - Sy * B_kz;
-	const float Cx = C_kx - Sx * C_kz;
-	const float Cy = C_ky - Sy * C_kz;
-
-	/* Calculate scaled barycentric coordinates. */
-	float U = Cx * By - Cy * Bx;
-	float V = Ax * Cy - Ay * Cx;
-	float W = Bx * Ay - By * Ax;
-
-	if((U < 0.0f || V < 0.0f || W < 0.0f) &&
-	   (U > 0.0f || V > 0.0f || W > 0.0f))
-	{
-		return;
-	}
+	const float4 tri_a = kernel_tex_fetch(__prim_tri_verts, tri_vindex+0),
+	             tri_b = kernel_tex_fetch(__prim_tri_verts, tri_vindex+1),
+	             tri_c = kernel_tex_fetch(__prim_tri_verts, tri_vindex+2);
+	const float3 verts[3] = {float4_to_float3(tri_a),
+	                         float4_to_float3(tri_b),
+	                         float4_to_float3(tri_c)};
 #endif
-
-	/* Calculate determinant. */
-	float det = U + V + W;
-	if(UNLIKELY(det == 0.0f)) {
-		return;
-	}
-
-	/* Calculate scaled z−coordinates of vertices and use them to calculate
-	 * the hit distance.
-	 */
-	const int sign_det = (__float_as_int(det) & 0x80000000);
-	const float T = (U * A_kz + V * B_kz + W * C_kz) * Sz;
-	const float sign_T = xor_signmask(T, sign_det);
-	if((sign_T < 0.0f) ||
-	   (sign_T > tmax * xor_signmask(det, sign_det)))
+	float t, u, v;
+	if(!ray_triangle_intersect(isect_precalc,
+	                           P, tmax,
+	                           verts,
+	                           &u, &v, &t))
 	{
 		return;
 	}
 
-	/* Normalize U, V, W, and T. */
-	const float inv_det = 1.0f / det;
-
-	const float t = T * inv_det;
 	for(int i = min(max_hits, ss_isect->num_hits) - 1; i >= 0; --i) {
 		if(ss_isect->hits[i].t == t) {
 			return;
@@ -251,11 +149,17 @@ ccl_device_inline void triangle_intersect_subsurface(
 	isect->prim = prim_addr;
 	isect->object = object;
 	isect->type = PRIMITIVE_TRIANGLE;
-	isect->u = U * inv_det;
-	isect->v = V * inv_det;
+	isect->u = u;
+	isect->v = v;
 	isect->t = t;
 
 	/* Record geometric normal. */
+	/* TODO(sergey): Check whether it's faster to re-use ssef verts. */
+#if defined(__KERNEL_AVX2__) && defined(__KERNEL_SSE__)
+	const float4 tri_a = kernel_tex_fetch(__prim_tri_verts, tri_vindex+0),
+	             tri_b = kernel_tex_fetch(__prim_tri_verts, tri_vindex+1),
+	             tri_c = kernel_tex_fetch(__prim_tri_verts, tri_vindex+2);
+#endif
 	/* TODO(sergey): Use float4_to_float3() on just an edges. */
 	const float3 v0 = float4_to_float3(tri_a);
 	const float3 v1 = float4_to_float3(tri_b);