mirror of
https://github.com/PixarAnimationStudios/OpenSubdiv
synced 2025-01-01 12:50:13 +00:00
Tessellation shader performance improvement
Re-organized the single-crease patch code path in the tessellation control/hull shaders to improve performance in the case where no patches have non-zero crease sharpnesses.
This commit is contained in:
David G. Yu
parent
86026a9fe8
commit
71fcb67509
@ -1120,7 +1120,10 @@ OsdComputePerPatchVertexBSpline(ivec3 patchParam, int ID, vec3 cv[16],
|
||||
|
||||
#if defined OSD_PATCH_ENABLE_SINGLE_CREASE
|
||||
|
||||
mat4 Mj, Ms;
|
||||
vec3 P = vec3(0); // 0 to 1-2^(-Sf)
|
||||
vec3 P1 = vec3(0); // 1-2^(-Sf) to 1-2^(-Sc)
|
||||
vec3 P2 = vec3(0); // 1-2^(-Sc) to 1
|
||||
|
||||
float sharpness = OsdGetPatchSharpness(patchParam);
|
||||
if (sharpness > 0) {
|
||||
float Sf = floor(sharpness);
|
||||
@ -1128,60 +1131,86 @@ OsdComputePerPatchVertexBSpline(ivec3 patchParam, int ID, vec3 cv[16],
|
||||
float Sr = fract(sharpness);
|
||||
mat4 Mf = OsdComputeMs(Sf);
|
||||
mat4 Mc = OsdComputeMs(Sc);
|
||||
Mj = (1-Sr) * Mf + Sr * Mi;
|
||||
Ms = (1-Sr) * Mf + Sr * Mc;
|
||||
mat4 Mj = (1-Sr) * Mf + Sr * Mi;
|
||||
mat4 Ms = (1-Sr) * Mf + Sr * Mc;
|
||||
float s0 = 1 - pow(2, -floor(sharpness));
|
||||
float s1 = 1 - pow(2, -ceil(sharpness));
|
||||
result.vSegments = vec2(s0, s1);
|
||||
} else {
|
||||
Mj = Ms = Mi;
|
||||
result.vSegments = vec2(0);
|
||||
}
|
||||
result.P = vec3(0); // 0 to 1-2^(-Sf)
|
||||
result.P1 = vec3(0); // 1-2^(-Sf) to 1-2^(-Sc)
|
||||
result.P2 = vec3(0); // 1-2^(-Sc) to 1
|
||||
|
||||
mat4 MUi, MUj, MUs;
|
||||
mat4 MVi, MVj, MVs;
|
||||
MUi = MUj = MUs = Q;
|
||||
MVi = MVj = MVs = Q;
|
||||
mat4 MUi = Q, MUj = Q, MUs = Q;
|
||||
mat4 MVi = Q, MVj = Q, MVs = Q;
|
||||
|
||||
int boundaryMask = OsdGetPatchBoundaryMask(patchParam);
|
||||
if ((boundaryMask & 1) != 0) {
|
||||
MVi = OsdFlipMatrix(Mi);
|
||||
MVj = OsdFlipMatrix(Mj);
|
||||
MVs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
if ((boundaryMask & 2) != 0) {
|
||||
MUi = Mi;
|
||||
MUj = Mj;
|
||||
MUs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 4) != 0) {
|
||||
MVi = Mi;
|
||||
MVj = Mj;
|
||||
MVs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 8) != 0) {
|
||||
MUi = OsdFlipMatrix(Mi);
|
||||
MUj = OsdFlipMatrix(Mj);
|
||||
MUs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
int boundaryMask = OsdGetPatchBoundaryMask(patchParam);
|
||||
if ((boundaryMask & 1) != 0) {
|
||||
MVi = OsdFlipMatrix(Mi);
|
||||
MVj = OsdFlipMatrix(Mj);
|
||||
MVs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
if ((boundaryMask & 2) != 0) {
|
||||
MUi = Mi;
|
||||
MUj = Mj;
|
||||
MUs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 4) != 0) {
|
||||
MVi = Mi;
|
||||
MVj = Mj;
|
||||
MVs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 8) != 0) {
|
||||
MUi = OsdFlipMatrix(Mi);
|
||||
MUj = OsdFlipMatrix(Mj);
|
||||
MUs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
|
||||
vec3 Hi[4], Hj[4], Hs[4];
|
||||
for (int l=0; l<4; ++l) {
|
||||
Hi[l] = Hj[l] = Hs[l] = vec3(0);
|
||||
vec3 Hi[4], Hj[4], Hs[4];
|
||||
for (int l=0; l<4; ++l) {
|
||||
Hi[l] = Hj[l] = Hs[l] = vec3(0);
|
||||
for (int k=0; k<4; ++k) {
|
||||
Hi[l] += MUi[i][k] * cv[l*4 + k];
|
||||
Hj[l] += MUj[i][k] * cv[l*4 + k];
|
||||
Hs[l] += MUs[i][k] * cv[l*4 + k];
|
||||
}
|
||||
}
|
||||
for (int k=0; k<4; ++k) {
|
||||
Hi[l] += MUi[i][k] * cv[l*4 + k];
|
||||
Hj[l] += MUj[i][k] * cv[l*4 + k];
|
||||
Hs[l] += MUs[i][k] * cv[l*4 + k];
|
||||
P += MVi[j][k]*Hi[k];
|
||||
P1 += MVj[j][k]*Hj[k];
|
||||
P2 += MVs[j][k]*Hs[k];
|
||||
}
|
||||
} else {
|
||||
result.vSegments = vec2(0);
|
||||
|
||||
mat4 MUi = Q;
|
||||
mat4 MVi = Q;
|
||||
|
||||
int boundaryMask = OsdGetPatchBoundaryMask(patchParam);
|
||||
if ((boundaryMask & 1) != 0) {
|
||||
MVi = OsdFlipMatrix(Mi);
|
||||
}
|
||||
if ((boundaryMask & 2) != 0) {
|
||||
MUi = Mi;
|
||||
}
|
||||
if ((boundaryMask & 4) != 0) {
|
||||
MVi = Mi;
|
||||
}
|
||||
if ((boundaryMask & 8) != 0) {
|
||||
MUi = OsdFlipMatrix(Mi);
|
||||
}
|
||||
|
||||
vec3 Hi[4], Hj[4], Hs[4];
|
||||
for (int l=0; l<4; ++l) {
|
||||
Hi[l] = Hj[l] = Hs[l] = vec3(0);
|
||||
for (int k=0; k<4; ++k) {
|
||||
Hi[l] += MUi[i][k] * cv[l*4 + k];
|
||||
}
|
||||
}
|
||||
for (int k=0; k<4; ++k) {
|
||||
P += MVi[j][k]*Hi[k];
|
||||
}
|
||||
}
|
||||
for (int k=0; k<4; ++k) {
|
||||
result.P += MVi[j][k]*Hi[k];
|
||||
result.P1 += MVj[j][k]*Hj[k];
|
||||
result.P2 += MVs[j][k]*Hs[k];
|
||||
}
|
||||
|
||||
result.P = P;
|
||||
result.P1 = P1;
|
||||
result.P2 = P2;
|
||||
#else
|
||||
OsdComputeBSplineBoundaryPoints(cv, patchParam);
|
||||
|
||||
|
||||
@ -994,7 +994,10 @@ OsdComputePerPatchVertexBSpline(int3 patchParam, int ID, float3 cv[16],
|
||||
|
||||
#if defined OSD_PATCH_ENABLE_SINGLE_CREASE
|
||||
|
||||
float4x4 Mj, Ms;
|
||||
float3 P = float3(0,0,0); // 0 to 1-2^(-Sf)
|
||||
float3 P1 = float3(0,0,0); // 1-2^(-Sf) to 1-2^(-Sc)
|
||||
float3 P2 = float3(0,0,0); // 1-2^(-Sc) to 1
|
||||
|
||||
float sharpness = OsdGetPatchSharpness(patchParam);
|
||||
if (sharpness > 0) {
|
||||
float Sf = floor(sharpness);
|
||||
@ -1002,61 +1005,87 @@ OsdComputePerPatchVertexBSpline(int3 patchParam, int ID, float3 cv[16],
|
||||
float Sr = frac(sharpness);
|
||||
float4x4 Mf = OsdComputeMs(Sf);
|
||||
float4x4 Mc = OsdComputeMs(Sc);
|
||||
Mj = (1-Sr) * Mf + Sr * Mi;
|
||||
Ms = (1-Sr) * Mf + Sr * Mc;
|
||||
float4x4 Mj = (1-Sr) * Mf + Sr * Mi;
|
||||
float4x4 Ms = (1-Sr) * Mf + Sr * Mc;
|
||||
float s0 = 1 - pow(2, -floor(sharpness));
|
||||
float s1 = 1 - pow(2, -ceil(sharpness));
|
||||
result.vSegments = float2(s0, s1);
|
||||
|
||||
} else {
|
||||
Mj = Ms = Mi;
|
||||
result.vSegments = float2(0, 0);
|
||||
}
|
||||
result.P = float3(0,0,0); // 0 to 1-2^(-Sf)
|
||||
result.P1 = float3(0,0,0); // 1-2^(-Sf) to 1-2^(-Sc)
|
||||
result.P2 = float3(0,0,0); // 1-2^(-Sc) to 1
|
||||
float4x4 MUi = Q, MUj = Q, MUs = Q;
|
||||
float4x4 MVi = Q, MVj = Q, MVs = Q;
|
||||
|
||||
float4x4 MUi, MUj, MUs;
|
||||
float4x4 MVi, MVj, MVs;
|
||||
MUi = MUj = MUs = Q;
|
||||
MVi = MVj = MVs = Q;
|
||||
int boundaryMask = OsdGetPatchBoundaryMask(patchParam);
|
||||
if ((boundaryMask & 1) != 0) {
|
||||
MVi = OsdFlipMatrix(Mi);
|
||||
MVj = OsdFlipMatrix(Mj);
|
||||
MVs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
if ((boundaryMask & 2) != 0) {
|
||||
MUi = Mi;
|
||||
MUj = Mj;
|
||||
MUs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 4) != 0) {
|
||||
MVi = Mi;
|
||||
MVj = Mj;
|
||||
MVs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 8) != 0) {
|
||||
MUi = OsdFlipMatrix(Mi);
|
||||
MUj = OsdFlipMatrix(Mj);
|
||||
MUs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
|
||||
int boundaryMask = OsdGetPatchBoundaryMask(patchParam);
|
||||
if ((boundaryMask & 1) != 0) {
|
||||
MVi = OsdFlipMatrix(Mi);
|
||||
MVj = OsdFlipMatrix(Mj);
|
||||
MVs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
if ((boundaryMask & 2) != 0) {
|
||||
MUi = Mi;
|
||||
MUj = Mj;
|
||||
MUs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 4) != 0) {
|
||||
MVi = Mi;
|
||||
MVj = Mj;
|
||||
MVs = Ms;
|
||||
}
|
||||
if ((boundaryMask & 8) != 0) {
|
||||
MUi = OsdFlipMatrix(Mi);
|
||||
MUj = OsdFlipMatrix(Mj);
|
||||
MUs = OsdFlipMatrix(Ms);
|
||||
}
|
||||
|
||||
float3 Hi[4], Hj[4], Hs[4];
|
||||
for (int l=0; l<4; ++l) {
|
||||
Hi[l] = Hj[l] = Hs[l] = float3(0,0,0);
|
||||
float3 Hi[4], Hj[4], Hs[4];
|
||||
for (int l=0; l<4; ++l) {
|
||||
Hi[l] = Hj[l] = Hs[l] = float3(0,0,0);
|
||||
for (int k=0; k<4; ++k) {
|
||||
Hi[l] += MUi[i][k] * cv[l*4 + k];
|
||||
Hj[l] += MUj[i][k] * cv[l*4 + k];
|
||||
Hs[l] += MUs[i][k] * cv[l*4 + k];
|
||||
}
|
||||
}
|
||||
for (int k=0; k<4; ++k) {
|
||||
Hi[l] += MUi[i][k] * cv[l*4 + k];
|
||||
Hj[l] += MUj[i][k] * cv[l*4 + k];
|
||||
Hs[l] += MUs[i][k] * cv[l*4 + k];
|
||||
P += MVi[j][k]*Hi[k];
|
||||
P1 += MVj[j][k]*Hj[k];
|
||||
P2 += MVs[j][k]*Hs[k];
|
||||
}
|
||||
|
||||
} else {
|
||||
result.vSegments = float2(0, 0);
|
||||
|
||||
float4x4 MUi = Q;
|
||||
float4x4 MVi = Q;
|
||||
|
||||
int boundaryMask = OsdGetPatchBoundaryMask(patchParam);
|
||||
if ((boundaryMask & 1) != 0) {
|
||||
MVi = OsdFlipMatrix(Mi);
|
||||
}
|
||||
if ((boundaryMask & 2) != 0) {
|
||||
MUi = Mi;
|
||||
}
|
||||
if ((boundaryMask & 4) != 0) {
|
||||
MVi = Mi;
|
||||
}
|
||||
if ((boundaryMask & 8) != 0) {
|
||||
MUi = OsdFlipMatrix(Mi);
|
||||
}
|
||||
|
||||
float3 Hi[4], Hj[4], Hs[4];
|
||||
for (int l=0; l<4; ++l) {
|
||||
Hi[l] = Hj[l] = Hs[l] = float3(0,0,0);
|
||||
for (int k=0; k<4; ++k) {
|
||||
Hi[l] += MUi[i][k] * cv[l*4 + k];
|
||||
}
|
||||
}
|
||||
for (int k=0; k<4; ++k) {
|
||||
P += MVi[j][k]*Hi[k];
|
||||
}
|
||||
}
|
||||
for (int k=0; k<4; ++k) {
|
||||
result.P += MVi[j][k]*Hi[k];
|
||||
result.P1 += MVj[j][k]*Hj[k];
|
||||
result.P2 += MVs[j][k]*Hs[k];
|
||||
}
|
||||
|
||||
result.P = P;
|
||||
result.P1 = P1;
|
||||
result.P2 = P2;
|
||||
#else
|
||||
OsdComputeBSplineBoundaryPoints(cv, patchParam);
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user