fixes in ray-convex for GPU

This commit is contained in:
erwincoumans
2013-06-20 16:27:36 -07:00
parent 47403c7e6f
commit d935eb2520
5 changed files with 133 additions and 139 deletions

View File

@@ -234,8 +234,8 @@ GpuRaytraceScene::GpuRaytraceScene()
m_raytraceData = new GpuRaytraceInternalData; m_raytraceData = new GpuRaytraceInternalData;
m_raytraceData->m_texId = new GLuint; m_raytraceData->m_texId = new GLuint;
m_raytraceData->textureWidth = 1024;//1024; m_raytraceData->textureWidth = 256;//1024;//1024;
m_raytraceData->textureHeight = 1024; m_raytraceData->textureHeight = 256;//1024;
//create new texture //create new texture
glGenTextures(1, m_raytraceData->m_texId); glGenTextures(1, m_raytraceData->m_texId);
@@ -471,9 +471,9 @@ void GpuRaytraceScene::renderScene()
if (hit) if (hit)
{ {
m_raytraceData->m_texels[(i)*3+0] = 255; m_raytraceData->m_texels[(i)*3+0] = 128+128.f*hits[i].m_hitNormal.x;
m_raytraceData->m_texels[(i)*3+1] = 0; m_raytraceData->m_texels[(i)*3+1] = 128+128.f*hits[i].m_hitNormal.y;
m_raytraceData->m_texels[(i)*3+2] = 0; m_raytraceData->m_texels[(i)*3+2] = 128+128.f*hits[i].m_hitNormal.z;
} else } else
{ {
m_raytraceData->m_texels[(i)*3+0] = 0; m_raytraceData->m_texels[(i)*3+0] = 0;

View File

@@ -610,7 +610,7 @@ cl_program b3OpenCLUtils_compileCLProgramFromString(cl_context clContext, cl_dev
int fileUpToDate = 0; int fileUpToDate = 0;
int binaryFileValid=0; int binaryFileValid=0;
if (clFileNameForCaching) if (!disableBinaryCaching && clFileNameForCaching)
{ {
clGetDeviceInfo(device, CL_DEVICE_NAME, 256, &deviceName, NULL); clGetDeviceInfo(device, CL_DEVICE_NAME, 256, &deviceName, NULL);
clGetDeviceInfo(device, CL_DRIVER_VERSION, 256, &driverVersion, NULL); clGetDeviceInfo(device, CL_DRIVER_VERSION, 256, &driverVersion, NULL);
@@ -918,7 +918,7 @@ cl_program b3OpenCLUtils_compileCLProgramFromString(cl_context clContext, cl_dev
} }
if( clFileNameForCaching ) if( !disableBinaryCaching && clFileNameForCaching )
{ // write to binary { // write to binary
cl_uint numAssociatedDevices; cl_uint numAssociatedDevices;

View File

@@ -209,6 +209,8 @@ void b3GpuRaycast::castRays(const b3AlignedObjectArray<b3RayInfo>& rays, b3Align
int numBodies,const struct b3RigidBodyCL* bodies, int numCollidables, const struct b3Collidable* collidables, const struct b3GpuNarrowPhaseInternalData* narrowphaseData) int numBodies,const struct b3RigidBodyCL* bodies, int numCollidables, const struct b3Collidable* collidables, const struct b3GpuNarrowPhaseInternalData* narrowphaseData)
{ {
//castRaysHost(rays,hitResults,numBodies,bodies,numCollidables,collidables,narrowphaseData);
B3_PROFILE("castRaysGPU"); B3_PROFILE("castRaysGPU");
b3OpenCLArray<b3RayInfo> gpuRays(m_data->m_context,m_data->m_q); b3OpenCLArray<b3RayInfo> gpuRays(m_data->m_context,m_data->m_q);

View File

@@ -14,12 +14,12 @@ typedef struct
typedef struct typedef struct
{ {
float m_hitFraction; float m_hitFraction;
int m_hitResult0; int m_hitResult0;
int m_hitResult1; int m_hitResult1;
int m_hitResult2; int m_hitResult2;
float4 m_hitPoint; float4 m_hitPoint;
float4 m_hitNormal; float4 m_hitNormal;
} b3RayHit; } b3RayHit;
typedef struct typedef struct
@@ -82,20 +82,18 @@ typedef struct
typedef float4 Quaternion; typedef float4 Quaternion;
__inline __inline
Quaternion qtMul(Quaternion a, Quaternion b); Quaternion qtMul(Quaternion a, Quaternion b);
__inline __inline
Quaternion qtNormalize(Quaternion in); Quaternion qtNormalize(Quaternion in);
__inline
float4 qtRotate(Quaternion q, float4 vec);
__inline
Quaternion qtInvert(Quaternion q);
__inline __inline
float dot3F4(float4 a, float4 b) Quaternion qtInvert(Quaternion q);
__inline
float dot3F4(float4 a, float4 b)
{ {
float4 a1 = (float4)(a.xyz,0.f); float4 a1 = (float4)(a.xyz,0.f);
float4 b1 = (float4)(b.xyz,0.f); float4 b1 = (float4)(b.xyz,0.f);
@@ -104,65 +102,56 @@ float dot3F4(float4 a, float4 b)
__inline __inline
Quaternion qtMul(Quaternion a, Quaternion b) Quaternion qtMul(Quaternion a, Quaternion b)
{ {
Quaternion ans; Quaternion ans;
ans = cross( a, b ); ans = cross( a, b );
ans += a.w*b+b.w*a; ans += a.w*b+b.w*a;
// ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z); // ans.w = a.w*b.w - (a.x*b.x+a.y*b.y+a.z*b.z);
ans.w = a.w*b.w - dot3F4(a, b); ans.w = a.w*b.w - dot3F4(a, b);
return ans; return ans;
} }
__inline __inline
Quaternion qtNormalize(Quaternion in) Quaternion qtNormalize(Quaternion in)
{ {
return fast_normalize(in); return fast_normalize(in);
// in /= length( in ); // in /= length( in );
// return in; // return in;
} }
__inline __inline
float4 qtRotate(Quaternion q, float4 vec) float4 qtRotate(Quaternion q, float4 vec)
{ {
Quaternion qInv = qtInvert( q ); Quaternion qInv = qtInvert( q );
float4 vcpy = vec; float4 vcpy = vec;
vcpy.w = 0.f; vcpy.w = 0.f;
float4 out = qtMul(qtMul(q,vcpy),qInv); float4 out = qtMul(q,vcpy);
out = qtMul(out,qInv);
return out; return out;
} }
__inline __inline
Quaternion qtInvert(Quaternion q) Quaternion qtInvert(Quaternion q)
{ {
return (Quaternion)(-q.xyz, q.w); return (Quaternion)(-q.xyz, q.w);
} }
__inline __inline
float4 qtInvRotate(const Quaternion q, float4 vec) float4 qtInvRotate(const Quaternion q, float4 vec)
{ {
return qtRotate( qtInvert( q ), vec ); return qtRotate( qtInvert( q ), vec );
} }
__inline
float4 transform(const float4* p, const float4* translation, const Quaternion* orientation)
{
return qtRotate( *orientation, *p ) + (*translation);
}
void trInverse(float4 translationIn, Quaternion orientationIn, void trInverse(float4 translationIn, Quaternion orientationIn,
float4* translationOut, Quaternion* orientationOut) float4* translationOut, Quaternion* orientationOut)
{ {
*orientationOut = qtInvert(orientationIn); *orientationOut = qtInvert(orientationIn);
*translationOut = qtRotate(*orientationOut, -translationIn); *translationOut = qtRotate(*orientationOut, -translationIn);
} }
void trMul(float4 translationA, Quaternion orientationA,
float4 translationB, Quaternion orientationB,
float4* translationOut, Quaternion* orientationOut)
{
*orientationOut = qtMul(orientationA,orientationB);
*translationOut = transform(&translationB,&translationA,&orientationA);
}
@@ -171,10 +160,10 @@ bool rayConvex(float4 rayFromLocal, float4 rayToLocal, int numFaces, int faceOff
{ {
rayFromLocal.w = 0.f; rayFromLocal.w = 0.f;
rayToLocal.w = 0.f; rayToLocal.w = 0.f;
bool result = true; bool result = true;
float exitFraction = *hitFraction; float exitFraction = hitFraction[0];
float enterFraction = -0.1f; float enterFraction = -0.3f;
float4 curHitNormal = (float4)(0,0,0,0); float4 curHitNormal = (float4)(0,0,0,0);
for (int i=0;i<numFaces && result;i++) for (int i=0;i<numFaces && result;i++)
{ {
@@ -211,12 +200,15 @@ bool rayConvex(float4 rayFromLocal, float4 rayToLocal, int numFaces, int faceOff
result = false; result = false;
} }
result = result && (enterFraction < 0.f); if (enterFraction < 0.f)
{
result = false;
}
if (result) if (result)
{ {
*hitFraction = enterFraction; hitFraction[0] = enterFraction;
*hitNormal = curHitNormal; hitNormal[0] = curHitNormal;
} }
return result; return result;
} }
@@ -238,26 +230,26 @@ bool sphere_intersect(float4 spherePos, float radius, float4 rayFrom, float4 ra
float D = B * B - A*C; float D = B * B - A*C;
if (D > 0.0) if (D > 0.0f)
{ {
float t = (-B - sqrt(D))/A; float t = (-B - sqrt(D))/A;
if ( (t >= 0.0f) && (t < (*hitFraction)) ) if ( (t >= 0.0f) && (t < (*hitFraction)) )
{ {
*hitFraction = t; *hitFraction = t;
return true; return true;
} }
} }
return false; return false;
} }
float4 setInterpolate3(float4 from, float4 to, float t) float4 setInterpolate3(float4 from, float4 to, float t)
{ {
float s = 1.0f - t; float s = 1.0f - t;
float4 result; float4 result;
result = s * from + t * to; result = s * from + t * to;
result.w = 0.f; result.w = 0.f;
return result; return result;
} }
__kernel void rayCastKernel( __kernel void rayCastKernel(
@@ -272,72 +264,74 @@ __kernel void rayCastKernel(
{ {
int i = get_global_id(0); int i = get_global_id(0);
if (i<numRays) if (i>=numRays)
return;
hitResults[i].m_hitFraction = 1.f;
float4 rayFrom = rays[i].m_from;
float4 rayTo = rays[i].m_to;
float hitFraction = 1.f;
float4 hitPoint;
float4 hitNormal;
int hitBodyIndex= -1;
int cachedCollidableIndex = -1;
Collidable cachedCollidable;
for (int b=0;b<numBodies;b++)
{ {
hitResults[i].m_hitFraction = 1.f;
float4 rayFrom = rays[i].m_from; Body body = bodies[b];
float4 rayTo = rays[i].m_to; float4 pos = body.m_pos;
float hitFraction = 1.f; float4 orn = body.m_quat;
float4 hitPoint; if (cachedCollidableIndex != body.m_collidableIdx)
float4 hitNormal;
int hitBodyIndex= -1;
int cachedCollidableIndex = -1;
Collidable cachedCollidable;
for (int b=0;b<numBodies;b++)
{ {
cachedCollidableIndex = body.m_collidableIdx;
float4 pos = bodies[b].m_pos; cachedCollidable = collidables[cachedCollidableIndex];
float4 orn = bodies[b].m_quat;
if (cachedCollidableIndex !=bodies[b].m_collidableIdx)
{
cachedCollidableIndex = bodies[b].m_collidableIdx;
cachedCollidable = collidables[cachedCollidableIndex];
}
if (cachedCollidable.m_shapeType == SHAPE_SPHERE)
{
float radius = cachedCollidable.m_radius;
if (sphere_intersect(pos, radius, rayFrom, rayTo, &hitFraction))
{
hitBodyIndex = b;
hitPoint = setInterpolate3(rayFrom, rayTo,hitFraction);
hitNormal = (float4) (hitPoint-bodies[b].m_pos);
}
}
if (cachedCollidable.m_shapeType == SHAPE_CONVEX_HULL)
{
float4 invPos = (float4)(0,0,0,0);
float4 invOrn = (float4)(0,0,0,0);
float4 rayFromLocal = (float4)(0,0,0,0);
float4 rayToLocal = (float4)(0,0,0,0);
trInverse(pos,orn, &invPos, &invOrn);
rayFromLocal = transform(&rayFrom, &invPos, &invOrn);
rayToLocal = transform(&rayTo, &invPos, &invOrn);
int numFaces = convexShapes[cachedCollidable.m_shapeIndex].m_numFaces;
int faceOffset = convexShapes[cachedCollidable.m_shapeIndex].m_faceOffset;
if (rayConvex(rayFromLocal, rayToLocal, numFaces, faceOffset,faces, &hitFraction, &hitNormal))
{
hitBodyIndex = b;
}
}
} }
if (cachedCollidable.m_shapeType == SHAPE_CONVEX_HULL)
if (hitBodyIndex>=0)
{ {
hitResults[i].m_hitFraction = hitFraction;
hitResults[i].m_hitPoint = hitPoint; float4 invPos = (float4)(0,0,0,0);
hitResults[i].m_hitNormal = normalize(hitNormal); float4 invOrn = (float4)(0,0,0,0);
hitResults[i].m_hitResult0 = hitBodyIndex; float4 rayFromLocal = (float4)(0,0,0,0);
float4 rayToLocal = (float4)(0,0,0,0);
invOrn = qtInvert(orn);
invPos = qtRotate(invOrn, -pos);
rayFromLocal = qtRotate( invOrn, rayFrom ) + invPos;
rayToLocal = qtRotate( invOrn, rayTo) + invPos;
rayFromLocal.w = 0.f;
rayToLocal.w = 0.f;
int numFaces = convexShapes[cachedCollidable.m_shapeIndex].m_numFaces;
int faceOffset = convexShapes[cachedCollidable.m_shapeIndex].m_faceOffset;
if (numFaces)
{
if (rayConvex(rayFromLocal, rayToLocal, numFaces, faceOffset,faces, &hitFraction, &hitNormal))
{
hitBodyIndex = b;
}
}
}
if (cachedCollidable.m_shapeType == SHAPE_SPHERE)
{
float radius = cachedCollidable.m_radius;
if (sphere_intersect(pos, radius, rayFrom, rayTo, &hitFraction))
{
hitBodyIndex = b;
hitPoint = setInterpolate3(rayFrom, rayTo,hitFraction);
hitNormal = (float4) (hitPoint-bodies[b].m_pos);
}
} }
} }
if (hitBodyIndex>=0)
{
hitResults[i].m_hitFraction = hitFraction;
hitResults[i].m_hitPoint = hitPoint;
hitResults[i].m_hitNormal = normalize(hitNormal);
hitResults[i].m_hitResult0 = hitBodyIndex;
}
} }

View File

@@ -176,7 +176,7 @@ static const char* rayCastKernelCL= \
" bool result = true;\n" " bool result = true;\n"
" \n" " \n"
" float exitFraction = *hitFraction;\n" " float exitFraction = *hitFraction;\n"
" float enterFraction = -0.1f;\n" " float enterFraction = -0.3f;\n"
" float4 curHitNormal = (float4)(0,0,0,0);\n" " float4 curHitNormal = (float4)(0,0,0,0);\n"
" for (int i=0;i<numFaces && result;i++)\n" " for (int i=0;i<numFaces && result;i++)\n"
" {\n" " {\n"
@@ -285,19 +285,17 @@ static const char* rayCastKernelCL= \
" float4 hitNormal;\n" " float4 hitNormal;\n"
" int hitBodyIndex= -1;\n" " int hitBodyIndex= -1;\n"
" \n" " \n"
" int cachedCollidableIndex = -1; \n" " \n"
" Collidable cachedCollidable;\n" " \n"
" \n" " \n"
" for (int b=0;b<numBodies;b++)\n" " for (int b=0;b<numBodies;b++)\n"
" {\n" " {\n"
" \n" " \n"
" float4 pos = bodies[b].m_pos;\n" " float4 pos = bodies[b].m_pos;\n"
" float4 orn = bodies[b].m_quat;\n" " float4 orn = bodies[b].m_quat;\n"
" if (cachedCollidableIndex !=bodies[b].m_collidableIdx)\n" " \n"
" {\n" " int cachedCollidableIndex = bodies[b].m_collidableIdx; \n"
" cachedCollidableIndex = bodies[b].m_collidableIdx;\n" " Collidable cachedCollidable = collidables[cachedCollidableIndex];\n"
" cachedCollidable = collidables[cachedCollidableIndex];\n"
" }\n"
" \n" " \n"
" if (cachedCollidable.m_shapeType == SHAPE_SPHERE)\n" " if (cachedCollidable.m_shapeType == SHAPE_SPHERE)\n"
" {\n" " {\n"