#line 0 "osd/mtlPatchBoxSplineTriangle.metal" // // Copyright 2015-2019 Pixar // // Licensed under the terms set forth in the LICENSE.txt file available at // https://opensubdiv.org/license. // //---------------------------------------------------------- // Patches.Common //---------------------------------------------------------- #define offsetof_(X, Y) &(((device X*)nullptr)->Y) #define OSD_IS_ADAPTIVE (OSD_PATCH_REGULAR || OSD_PATCH_BOX_SPLINE_TRIANGLE || OSD_PATCH_GREGORY_BASIS || OSD_PATCH_GREGORY_TRIANGLE || OSD_PATCH_GREGORY || OSD_PATCH_GREGORY_BOUNDARY) #ifndef OSD_NUM_ELEMENTS #define OSD_NUM_ELEMENTS 3 #endif struct OsdPerVertexGregory { float3 P; short3 clipFlag; int valence; float3 e0; float3 e1; #if OSD_PATCH_GREGORY_BOUNDARY int zerothNeighbor; float3 org; #endif float3 r[OSD_MAX_VALENCE]; }; struct OsdPerPatchVertexGregory { packed_float3 P; packed_float3 Ep; packed_float3 Em; packed_float3 Fp; packed_float3 Fm; }; //---------------------------------------------------------- // HLSL->Metal Compatibility //---------------------------------------------------------- float4 mul(float4x4 a, float4 b) { return a * b; } float3 mul(float4x4 a, float3 b) { float3x3 m(a[0].xyz, a[1].xyz, a[2].xyz); return m * b; } struct HullVertex { float4 position; #if OSD_ENABLE_PATCH_CULL short3 clipFlag; #endif float3 GetPosition() threadgroup { return position.xyz; } void SetPosition(float3 v) threadgroup { position.xyz = v; } }; // XXXdyu all downstream data can be handled by client code struct OsdPatchVertex { float3 position; float3 normal; float3 tangent; float3 bitangent; float4 patchCoord; //u, v, faceLevel, faceId float2 tessCoord; // tesscoord.st #if OSD_COMPUTE_NORMAL_DERIVATIVES float3 Nu; float3 Nv; #endif #if OSD_PATCH_ENABLE_SINGLE_CREASE float2 vSegments; #endif }; struct OsdPerPatchTessFactors { float4 tessOuterLo; float4 tessOuterHi; }; using OsdPatchParamBufferType = packed_int3; #if OSD_PATCH_REGULAR || OSD_PATCH_BOX_SPLINE_TRIANGLE using PatchVertexType = HullVertex; using PerPatchVertexType = OsdPerPatchVertexBezier; #elif OSD_PATCH_GREGORY || OSD_PATCH_GREGORY_BOUNDARY using PatchVertexType = OsdPerVertexGregory; using PerPatchVertexType = OsdPerPatchVertexGregory; #elif OSD_PATCH_GREGORY_BASIS || OSD_PATCH_GREGORY_TRIANGLE using PatchVertexType = HullVertex; using PerPatchVertexType = OsdPerPatchVertexGregoryBasis; #else using PatchVertexType = OsdInputVertexType; using PerPatchVertexType = OsdInputVertexType; #endif //Shared buffers used by OSD that are common to all kernels struct OsdPatchParamBufferSet { const device OsdInputVertexType* vertexBuffer [[buffer(VERTEX_BUFFER_INDEX)]]; const device unsigned* indexBuffer [[buffer(CONTROL_INDICES_BUFFER_INDEX)]]; const device OsdPatchParamBufferType* patchParamBuffer [[buffer(OSD_PATCHPARAM_BUFFER_INDEX)]]; device PerPatchVertexType* perPatchVertexBuffer [[buffer(OSD_PERPATCHVERTEX_BUFFER_INDEX)]]; #if !USE_PTVS_FACTORS device OsdPerPatchTessFactors* patchTessBuffer [[buffer(OSD_PERPATCHTESSFACTORS_BUFFER_INDEX)]]; #endif #if OSD_PATCH_GREGORY || OSD_PATCH_GREGORY_BOUNDARY const device int* quadOffsetBuffer [[buffer(OSD_QUADOFFSET_BUFFER_INDEX)]]; const device int* valenceBuffer [[buffer(OSD_VALENCE_BUFFER_INDEX)]]; #endif const constant unsigned& kernelExecutionLimit [[buffer(OSD_KERNELLIMIT_BUFFER_INDEX)]]; }; //Shared buffers used by OSD that are common to all PTVS implementations struct OsdVertexBufferSet { const device OsdInputVertexType* vertexBuffer [[buffer(VERTEX_BUFFER_INDEX)]]; const device unsigned* indexBuffer [[buffer(CONTROL_INDICES_BUFFER_INDEX)]]; const device OsdPatchParamBufferType* patchParamBuffer [[buffer(OSD_PATCHPARAM_BUFFER_INDEX)]]; device PerPatchVertexType* perPatchVertexBuffer [[buffer(OSD_PERPATCHVERTEX_BUFFER_INDEX)]]; #if !USE_PTVS_FACTORS device OsdPerPatchTessFactors* patchTessBuffer [[buffer(OSD_PERPATCHTESSFACTORS_BUFFER_INDEX)]]; #endif }; // These are stored in OsdPatchParamBuffer indexed by the value returned // from OsdGetPatchIndex() which is a function of the current PrimitiveID // along with an optional client provided offset. int3 OsdGetPatchParam(int patchIndex, const device OsdPatchParamBufferType* osdPatchParamBuffer) { #if OSD_PATCH_ENABLE_SINGLE_CREASE return int3(osdPatchParamBuffer[patchIndex]); #else auto p = osdPatchParamBuffer[patchIndex]; return int3(p[0], p[1], 0); #endif } int OsdGetPatchIndex(int primitiveId) { return primitiveId; } // ---------------------------------------------------------------------------- // patch culling // ---------------------------------------------------------------------------- bool OsdCullPerPatchVertex( threadgroup PatchVertexType* patch, float4x4 ModelViewMatrix ) { #if OSD_ENABLE_BACKPATCH_CULL && OSD_PATCH_REGULAR auto v0 = float3(ModelViewMatrix * patch[5].position); auto v3 = float3(ModelViewMatrix * patch[6].position); auto v12 = float3(ModelViewMatrix * patch[9].position); auto n = normalize(cross(v3 - v0, v12 - v0)); v0 = normalize(v0 + v3 + v12); if(dot(v0, n) > 0.6f) { return false; } #endif #if OSD_ENABLE_PATCH_CULL short3 clipFlag = short3(0,0,0); for(int i = 0; i < CONTROL_POINTS_PER_PATCH; ++i) { clipFlag |= patch[i].clipFlag; } if (any(clipFlag != short3(3,3,3))) { return false; } #endif return true; } // ---------------------------------------------------------------------------- // Legacy Gregory // ---------------------------------------------------------------------------- #if defined(OSD_PATCH_GREGORY) || defined(OSD_PATCH_GREGORY_BOUNDARY) // precomputed catmark coefficient table up to valence 29 constant float OsdCatmarkCoefficient[30] = { 0, 0, 0, 0.812816, 0.500000, 0.363644, 0.287514, 0.238688, 0.204544, 0.179229, 0.159657, 0.144042, 0.131276, 0.120632, 0.111614, 0.103872, 0.09715, 0.0912559, 0.0860444, 0.0814022, 0.0772401, 0.0734867, 0.0700842, 0.0669851, 0.0641504, 0.0615475, 0.0591488, 0.0569311, 0.0548745, 0.0529621 }; float OsdComputeCatmarkCoefficient(int valence) { #if OSD_MAX_VALENCE < 30 return OsdCatmarkCoefficient[valence]; #else if (valence < 30) { return OsdCatmarkCoefficient[valence]; } else { float t = 2.0f / float(valence); return 1.0f / (valence * (cospi(t) + 5.0f + sqrt((cospi(t) + 9) * (cospi(t) + 1)))/16.0f); } #endif } float cosfn(int n, int j) { return cospi((2.0f * j)/float(n)); } float sinfn(int n, int j) { return sinpi((2.0f * j)/float(n)); } #ifndef OSD_MAX_VALENCE #define OSD_MAX_VALENCE 4 #endif template float3 OsdReadVertex(int vertexIndex, OsdVertexBuffer osdVertexBuffer) { int index = (vertexIndex /*+ OsdBaseVertex()*/); return osdVertexBuffer[index].position; } template int OsdReadVertexValence(int vertexID, OsdValenceBuffer osdValenceBuffer) { int index = int(vertexID * (2 * OSD_MAX_VALENCE + 1)); return osdValenceBuffer[index]; } template int OsdReadVertexIndex(int vertexID, int valenceVertex, OsdValenceBuffer osdValenceBuffer) { int index = int(vertexID * (2 * OSD_MAX_VALENCE + 1) + 1 + valenceVertex); return osdValenceBuffer[index]; } template int OsdReadQuadOffset(int primitiveID, int offsetVertex, OsdQuadOffsetBuffer osdQuadOffsetBuffer) { int index = int(4*primitiveID + offsetVertex); return osdQuadOffsetBuffer[index]; } void OsdComputePerVertexGregory(unsigned vID, float3 P, threadgroup OsdPerVertexGregory& v, OsdPatchParamBufferSet osdBuffers) { v.clipFlag = short3(0,0,0); int ivalence = OsdReadVertexValence(vID, osdBuffers.valenceBuffer); v.valence = ivalence; int valence = abs(ivalence); float3 f[OSD_MAX_VALENCE]; float3 pos = P; float3 opos = float3(0,0,0); #ifdef OSD_PATCH_GREGORY_BOUNDARY v.org = pos; int boundaryEdgeNeighbors[2]; int currNeighbor = 0; int ibefore = 0; int zerothNeighbor = 0; #endif for (int i=0; i 2) { v.P = (OsdReadVertex(boundaryEdgeNeighbors[0], osdBuffers.vertexBuffer) + OsdReadVertex(boundaryEdgeNeighbors[1], osdBuffers.vertexBuffer) + 4.0f * pos)/6.0f; } else { v.P = pos; } v.e0 = (OsdReadVertex(boundaryEdgeNeighbors[0], osdBuffers.vertexBuffer) - OsdReadVertex(boundaryEdgeNeighbors[1], osdBuffers.vertexBuffer))/6.0; float k = float(float(valence) - 1.0f); //k is the number of faces float c = cospi(1.0/k); float s = sinpi(1.0/k); float gamma = -(4.0f*s)/(3.0f*k+c); float alpha_0k = -((1.0f+2.0f*c)*sqrt(1.0f+c))/((3.0f*k+c)*sqrt(1.0f-c)); float beta_0 = s/(3.0f*k + c); int idx_diagonal = OsdReadVertexIndex(vID, 2*zerothNeighbor + 1, osdBuffers.valenceBuffer); float3 diagonal = OsdReadVertex(idx_diagonal, osdBuffers.vertexBuffer); v.e1 = gamma * pos + alpha_0k * OsdReadVertex(boundaryEdgeNeighbors[0], osdBuffers.vertexBuffer) + alpha_0k * OsdReadVertex(boundaryEdgeNeighbors[1], osdBuffers.vertexBuffer) + beta_0 * diagonal; for (int x=1; x> 8) & 0xff; int start_m = OsdReadQuadOffset(primitiveID, im, osdBuffers.quadOffsetBuffer) & 0xff; int prev_p = (OsdReadQuadOffset(primitiveID, ip, osdBuffers.quadOffsetBuffer) >> 8) & 0xff; int np = abs(v[ip].valence); int nm = abs(v[im].valence); // Control Vertices based on : // "Approximating Subdivision Surfaces with Gregory Patches // for Hardware Tessellation" // Loop, Schaefer, Ni, Castano (ACM ToG Siggraph Asia 2009) // // P3 e3- e2+ P2 // O--------O--------O--------O // | | | | // | | | | // | | f3- | f2+ | // | O O | // e3+ O------O O------O e2- // | f3+ f2- | // | | // | | // | f0- f1+ | // e0- O------O O------O e1+ // | O O | // | | f0+ | f1- | // | | | | // | | | | // O--------O--------O--------O // P0 e0+ e1- P1 // #ifdef OSD_PATCH_GREGORY_BOUNDARY float3 Em_ip; if (v[ip].valence < -2) { int j = (np + prev_p - v[ip].zerothNeighbor) % np; Em_ip = v[ip].P + cospi(j/float(np-1))*v[ip].e0 + sinpi(j/float(np-1))*v[ip].e1; } else { Em_ip = v[ip].P + v[ip].e0*cosfn(np, prev_p) + v[ip].e1*sinfn(np, prev_p); } float3 Ep_im; if (v[im].valence < -2) { int j = (nm + start_m - v[im].zerothNeighbor) % nm; Ep_im = v[im].P + cospi(j/float(nm-1))*v[im].e0 + sinpi(j/float(nm-1))*v[im].e1; } else { Ep_im = v[im].P + v[im].e0*cosfn(nm, start_m) + v[im].e1*sinfn(nm, start_m); } if (v[i].valence < 0) { n = (n-1)*2; } if (v[im].valence < 0) { nm = (nm-1)*2; } if (v[ip].valence < 0) { np = (np-1)*2; } if (v[i].valence > 2) { result.Ep = v[i].P + (v[i].e0*cosfn(n, start) + v[i].e1*sinfn(n, start)); result.Em = v[i].P + (v[i].e0*cosfn(n, prev) + v[i].e1*sinfn(n, prev)); float s1=3-2*cosfn(n,1)-cosfn(np,1); float s2=2*cosfn(n,1); result.Fp = (cosfn(np,1)*v[i].P + s1*result.Ep + s2*Em_ip + v[i].r[start])/3.0f; s1 = 3.0f-2.0f*cospi(2.0f/float(n))-cospi(2.0f/float(nm)); result.Fm = (cosfn(nm,1)*v[i].P + s1*result.Em + s2*Ep_im - v[i].r[prev])/3.0f; } else if (v[i].valence < -2) { int j = (valence + start - v[i].zerothNeighbor) % valence; result.Ep = v[i].P + cospi(j/float(valence-1))*v[i].e0 + sinpi(j/float(valence-1))*v[i].e1; j = (valence + prev - v[i].zerothNeighbor) % valence; result.Em = v[i].P + cospi(j/float(valence-1))*v[i].e0 + sinpi(j/float(valence-1))*v[i].e1; float3 Rp = ((-2.0f * v[i].org - 1.0f * v[im].org) + (2.0f * v[ip].org + 1.0f * v[(i+2)%4].org))/3.0f; float3 Rm = ((-2.0f * v[i].org - 1.0f * v[ip].org) + (2.0f * v[im].org + 1.0f * v[(i+2)%4].org))/3.0f; float s1 = 3-2*cosfn(n,1)-cosfn(np,1); float s2 = 2*cosfn(n,1); result.Fp = (cosfn(np,1)*v[i].P + s1*result.Ep + s2*Em_ip + v[i].r[start])/3.0f; s1 = 3.0f-2.0f*cospi(2.0f/float(n))-cospi(2.0f/float(nm)); result.Fm = (cosfn(nm,1)*v[i].P + s1*result.Em + s2*Ep_im - v[i].r[prev])/3.0f; if (v[im].valence < 0) { s1 = 3-2*cosfn(n,1)-cosfn(np,1); result.Fp = result.Fm = (cosfn(np,1)*v[i].P + s1*result.Ep + s2*Em_ip + v[i].r[start])/3.0f; } else if (v[ip].valence < 0) { s1 = 3.0f-2.0f*cospi(2.0f/n)-cospi(2.0f/nm); result.Fm = result.Fp = (cosfn(nm,1)*v[i].P + s1*result.Em + s2*Ep_im - v[i].r[prev])/3.0f; } } else if (v[i].valence == -2) { result.Ep = (2.0f * v[i].org + v[ip].org)/3.0f; result.Em = (2.0f * v[i].org + v[im].org)/3.0f; result.Fp = result.Fm = (4.0f * v[i].org + v[(i+2)%n].org + 2.0f * v[ip].org + 2.0f * v[im].org)/9.0f; } #else // not OSD_PATCH_GREGORY_BOUNDARY result.Ep = v[i].P + v[i].e0 * cosfn(n, start) + v[i].e1*sinfn(n, start); result.Em = v[i].P + v[i].e0 * cosfn(n, prev ) + v[i].e1*sinfn(n, prev ); float3 Em_ip = v[ip].P + v[ip].e0*cosfn(np, prev_p) + v[ip].e1*sinfn(np, prev_p); float3 Ep_im = v[im].P + v[im].e0*cosfn(nm, start_m) + v[im].e1*sinfn(nm, start_m); float s1 = 3-2*cosfn(n,1)-cosfn(np,1); float s2 = 2*cosfn(n,1); result.Fp = (cosfn(np,1)*v[i].P + s1*result.Ep + s2*Em_ip + v[i].r[start])/3.0f; s1 = 3.0f-2.0f*cospi(2.0f/float(n))-cospi(2.0f/float(nm)); result.Fm = (cosfn(nm,1)*v[i].P + s1*result.Em +s2*Ep_im - v[i].r[prev])/3.0f; #endif } #endif // OSD_PATCH_GREGORY || OSD_PATCH_GREGORY_BOUNDARY