void GatherCollisionObjectAndShapeData (RaycastGatheredObjectData* gatheredObjectData, RaycastTask_LocalStoreMemory* lsMemPtr, ppu_address_t objectWrapper)
{
register int dmaSize;
register ppu_address_t dmaPpuAddress2;
/* DMA Collision object wrapper into local store */
dmaSize = sizeof(SpuCollisionObjectWrapper);
dmaPpuAddress2 = objectWrapper;
cellDmaGet(&lsMemPtr->gCollisionObjectWrapper, dmaPpuAddress2, dmaSize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
/* DMA Collision object into local store */
dmaSize = sizeof(btCollisionObject);
dmaPpuAddress2 = lsMemPtr->getCollisionObjectWrapper()->getCollisionObjectPtr();
cellDmaGet(&lsMemPtr->gColObj, dmaPpuAddress2 , dmaSize, DMA_TAG(2), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(2));
/* Gather information about collision object and shape */
gatheredObjectData->m_worldTransform = lsMemPtr->getColObj()->getWorldTransform();
gatheredObjectData->m_collisionMargin = lsMemPtr->getCollisionObjectWrapper()->getCollisionMargin ();
gatheredObjectData->m_shapeType = lsMemPtr->getCollisionObjectWrapper()->getShapeType ();
gatheredObjectData->m_collisionShape = (ppu_address_t)lsMemPtr->getColObj()->getCollisionShape();
gatheredObjectData->m_spuCollisionShape = (void*)&lsMemPtr->gCollisionShape.collisionShape;
/* DMA shape data */
dmaCollisionShape (gatheredObjectData->m_spuCollisionShape, gatheredObjectData->m_collisionShape, 1, gatheredObjectData->m_shapeType);
cellDmaWaitTagStatusAll(DMA_MASK(1));
if (btBroadphaseProxy::isConvex (gatheredObjectData->m_shapeType))
{
btConvexInternalShape* spuConvexShape = (btConvexInternalShape*)gatheredObjectData->m_spuCollisionShape;
gatheredObjectData->m_primitiveDimensions = spuConvexShape->getImplicitShapeDimensions ();
} else {
gatheredObjectData->m_primitiveDimensions = btVector3(1.0, 1.0, 1.0);
}
}
void processDecodeSet(unsigned int uiPtr)
{
SpursSpeexTaskOutput spuOutput;
cellDmaGet(&gviSpursSpeexTaskDesc, uiPtr, sizeof(SpursSpeexTaskDesc), DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] CMD_SAMPLE_TASK_DECODESET_COMMAND\n");
if (gviSpursSpeexTaskDesc.mDebugPause)
{
snPause();
}
cellDmaLargeGet(gviSpursSpeexStateBuffer, (uint64_t)gviSpursSpeexTaskDesc.mSpeexStateBuffer, SPEEX_DECODER_STATE_BUFFER_SIZE, DMA_TAG(1), 0,0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
gviSpursSpeexDecodeSet(&spuOutput);
if (spuOutput.mSpeexReturnCode < 0)
{
spuDebugPrintf("SPU: failed to encode, ret = %d\n", spuOutput.mSpeexReturnCode);
}
cellDmaPut(&spuOutput, (uint64_t)gviSpursSpeexTaskDesc.mSpeexTaskOutput, sizeof(SpursSpeexTaskOutput), DMA_TAG(1),
0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
cellDmaLargePut(gviSpursSpeexStateBuffer, (uint64_t)gviSpursSpeexTaskDesc.mSpeexStateBuffer, SPEEX_DECODER_STATE_BUFFER_SIZE, DMA_TAG(1), 0,0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] buffer dma done\n");
}
void processDecodeInit(unsigned int uiPtr)
{
SpursSpeexTaskOutput spuOutput;
cellDmaGet(&gviSpursSpeexTaskDesc, uiPtr, sizeof(SpursSpeexTaskDesc), DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] CMD_SAMPLE_TASK_DECODE_INIT_COMMAND\n");
if (gviSpursSpeexTaskDesc.mDebugPause)
{
snPause();
}
gviSpursSpeexDecoderInitialize(&spuOutput);
if (spuOutput.mSpeexReturnCode < 0)
{
spuDebugPrintf("[Speex][SPU] failed to initialize decoder, ret = %d\n", spuOutput.mSpeexReturnCode);
}
cellDmaPut(&spuOutput, (uint64_t)gviSpursSpeexTaskDesc.mSpeexTaskOutput, sizeof(SpursSpeexTaskOutput), DMA_TAG(1),
0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
cellDmaLargePut(gviSpursSpeexStateBuffer, (uint64_t)gviSpursSpeexTaskDesc.mSpeexStateBuffer,
gviSpursSpeexTaskDesc.mSpeexStateBufferSize, DMA_TAG(1), 0,0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] buffer dma done\n");
}
void procesEncodeInit(unsigned int uiPtr)
{
SpursSpeexTaskOutput spuOutput;
//spuDebugPrintf("[Speex][SPU] CMD_SAMPLE_TASK_ENCODE_INIT_COMMAND\n");
cellDmaGet(&gviSpursSpeexTaskDesc, uiPtr, sizeof(SpursSpeexTaskDesc), DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
if (gviSpursSpeexTaskDesc.mDebugPause)
{
snPause();
}
gviSpursSpeexEncoderInitialize(&spuOutput);
if (spuOutput.mSpeexReturnCode < 0)
{
spuDebugPrintf("[Speex][SPU] failed to initialize encoder, ret = %d\n", spuOutput.mSpeexReturnCode);
}
//spuDebugPrintf("[Speex][SPU] done with initializing things for speex, now returning data via DMA put\n");
//printGlobalTaskDescData();
cellDmaPut(&spuOutput, (uint64_t)gviSpursSpeexTaskDesc.mSpeexTaskOutput, sizeof(SpursSpeexTaskOutput), DMA_TAG(1),
0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] task dma done\n");
cellDmaLargePut(gviSpursSpeexStateBuffer, (uint64_t)gviSpursSpeexTaskDesc.mSpeexStateBuffer, SPEEX_ENCODER_STATE_BUFFER_SIZE, DMA_TAG(1), 0,0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] buffer dma done\n");
}
void gviSpursSpeexEncode(SpursSpeexTaskOutput *spuTaskOut)
{
short *inBuffer;
float *speexBuffer;
char *outBuffer;
unsigned int i;
spuTaskOut->mSpeexEncodedFrameSize = 0;
spuTaskOut->mSpeexInitialized = 1;
spuTaskOut->mSpeexSamplesPerFrame = 0;
spuTaskOut->mSpeexReturnCode = 0;
spuTaskOut->mSpeexOutBufferSize = 0;
speexBuffer = (float *)memalign(16, gviSpursSpeexTaskDesc.mInputBufferSize * sizeof(float));
inBuffer = (short *)memalign(16, gviSpursSpeexTaskDesc.mInputBufferSize * sizeof(short));
outBuffer = (char *)memalign(16, gviSpursSpeexTaskDesc.mOutputBufferSize);
memset(speexBuffer, 0, gviSpursSpeexTaskDesc.mInputBufferSize * sizeof(float));
memset(inBuffer, 0, gviSpursSpeexTaskDesc.mInputBufferSize * sizeof(short));
memset(outBuffer, 0, gviSpursSpeexTaskDesc.mOutputBufferSize);
cellDmaGet(inBuffer, (uint64_t)gviSpursSpeexTaskDesc.mInputBuffer, gviSpursSpeexTaskDesc.mInputBufferSize * sizeof(short), DMA_TAG(1), 0,0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
// convert the input to floats for encoding
for(i = 0 ; i < gviSpursSpeexTaskDesc.mInputBufferSize ; i++)
speexBuffer[i] = inBuffer[i];
// (re)initialize the bits struct
speex_bits_init_buffer(&gviSpursSpeexBits,gviSpursSpeexBitsBuffer,sizeof(gviSpursSpeexBitsBuffer));
// flush the bits
speex_bits_reset(&gviSpursSpeexBits);
// encode the frame
speex_encode(gviSpursSpeexStateBuffer, speexBuffer, &gviSpursSpeexBits);
// write the bits to the output
spuTaskOut->mSpeexOutBufferSize = speex_bits_write(&gviSpursSpeexBits, (char *)outBuffer, gviSpursSpeexTaskDesc.mEncodedFrameSize);
//spuDebugPrintf("[Speex][SPU] transferring data back, output size should be: %d\n", gviSpursSpeexTaskDesc.mOutputBufferSize>16?gviSpursSpeexTaskDesc.mOutputBufferSize:16);
cellDmaPut(outBuffer, (uint64_t)gviSpursSpeexTaskDesc.mOutputBuffer, gviSpursSpeexTaskDesc.mOutputBufferSize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] done transferring data back\n");
free(speexBuffer);
free(inBuffer);
free(outBuffer);
spuTaskOut->mSpeexReturnCode = 0;
}
void gviSpursSpeexDecodeAdd(SpursSpeexTaskOutput *spuTaskOut)
{
char *inBuffer;
float *speexBuffer;
short *outBuffer;
int rcode;
unsigned int i;
//spuDebugPrintf("[Speex][SPU] allocating buffers for decoding\n");
speexBuffer = (float *)memalign(16, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(float));
outBuffer = (short *)memalign(16, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(short));
inBuffer = (char *)memalign(16, gviSpursSpeexTaskDesc.mInputBufferSize);
memset(speexBuffer, 0, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(float));
memset(outBuffer, 0, gviSpursSpeexTaskDesc.mOutputBufferSize);
memset(inBuffer, 0, gviSpursSpeexTaskDesc.mInputBufferSize * sizeof(short));
//spuDebugPrintf("[Speex][SPU] done allocating, getting input data, inbuffer size: %d\n", gSpuSampleTaskDesc.mInputBufferSize);
cellDmaGet(inBuffer, (uint64_t)gviSpursSpeexTaskDesc.mInputBuffer, gviSpursSpeexTaskDesc.mInputBufferSize, DMA_TAG(1), 0,0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
// spuDebugPrintf("[Speex][SPU] done getting input data, preparing for speex to decode\n");
// read the data into the bits
// (re)initialize the bits struct
speex_bits_init_buffer(&gviSpursSpeexBits,gviSpursSpeexBitsBuffer,sizeof(gviSpursSpeexBitsBuffer));
speex_bits_read_from(&gviSpursSpeexBits, (char *)inBuffer, gviSpursSpeexTaskDesc.mEncodedFrameSize);
// decode it
rcode = speex_decode((void *)gviSpursSpeexStateBuffer, &gviSpursSpeexBits, speexBuffer);
assert(rcode == 0);
//spuDebugPrintf("[Speex][SPU] done with speex decode\n");
// convert the output from floats
for(i = 0 ; i < gviSpursSpeexTaskDesc.mOutputBufferSize ; i++)
outBuffer[i] = (short)speexBuffer[i];
//spuDebugPrintf("[Speex][SPU] transferring data back\n");
cellDmaPut(outBuffer, (uint64_t)gviSpursSpeexTaskDesc.mOutputBuffer, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(short), DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//spuDebugPrintf("[Speex][SPU] done transferring data back\n");
free(speexBuffer);
free(inBuffer);
free(outBuffer);
spuTaskOut->mSpeexReturnCode = 0;
}
SIMD_FORCE_INLINE void small_cache_read_triple( void* ls0, ppu_address_t ea0,
void* ls1, ppu_address_t ea1,
void* ls2, ppu_address_t ea2,
size_t size)
{
btAssert(size<16);
ATTRIBUTE_ALIGNED16(char tmpBuffer0[32]);
ATTRIBUTE_ALIGNED16(char tmpBuffer1[32]);
ATTRIBUTE_ALIGNED16(char tmpBuffer2[32]);
uint32_t i;
///make sure last 4 bits are the same, for cellDmaSmallGet
char* localStore0 = (char*)ls0;
uint32_t last4BitsOffset = ea0 & 0x0f;
char* tmpTarget0 = tmpBuffer0 + last4BitsOffset;
#ifdef __SPU__
cellDmaSmallGet(tmpTarget0,ea0,size,DMA_TAG(1),0,0);
#else
tmpTarget0 = (char*)cellDmaSmallGetReadOnly(tmpTarget0,ea0,size,DMA_TAG(1),0,0);
#endif
char* localStore1 = (char*)ls1;
last4BitsOffset = ea1 & 0x0f;
char* tmpTarget1 = tmpBuffer1 + last4BitsOffset;
#ifdef __SPU__
cellDmaSmallGet(tmpTarget1,ea1,size,DMA_TAG(1),0,0);
#else
tmpTarget1 = (char*)cellDmaSmallGetReadOnly(tmpTarget1,ea1,size,DMA_TAG(1),0,0);
#endif
char* localStore2 = (char*)ls2;
last4BitsOffset = ea2 & 0x0f;
char* tmpTarget2 = tmpBuffer2 + last4BitsOffset;
#ifdef __SPU__
cellDmaSmallGet(tmpTarget2,ea2,size,DMA_TAG(1),0,0);
#else
tmpTarget2 = (char*)cellDmaSmallGetReadOnly(tmpTarget2,ea2,size,DMA_TAG(1),0,0);
#endif
cellDmaWaitTagStatusAll( DMA_MASK(1) );
//this is slowish, perhaps memcpy on SPU is smarter?
for (i=0; btLikely( i<size );i++)
{
localStore0[i] = tmpTarget0[i];
localStore1[i] = tmpTarget1[i];
localStore2[i] = tmpTarget2[i];
}
}
void
performRaycastAgainstConvex (RaycastGatheredObjectData* gatheredObjectData, const SpuRaycastTaskWorkUnit& workUnit, SpuRaycastTaskWorkUnitOut* workUnitOut, RaycastTask_LocalStoreMemory* lsMemPtr)
{
SpuVoronoiSimplexSolver simplexSolver;
btTransform rayFromTrans, rayToTrans;
rayFromTrans.setIdentity ();
rayFromTrans.setOrigin (workUnit.rayFrom);
rayToTrans.setIdentity ();
rayToTrans.setOrigin (workUnit.rayTo);
SpuCastResult result;
/* Load the vertex data if the shape is a convex hull */
/* XXX: We might be loading the shape twice */
ATTRIBUTE_ALIGNED16(char convexHullShape[sizeof(btConvexHullShape)]);
if (gatheredObjectData->m_shapeType == CONVEX_HULL_SHAPE_PROXYTYPE)
{
register int dmaSize;
register ppu_address_t dmaPpuAddress2;
dmaSize = sizeof(btConvexHullShape);
dmaPpuAddress2 = gatheredObjectData->m_collisionShape;
cellDmaGet(&convexHullShape, dmaPpuAddress2, dmaSize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
dmaConvexVertexData (&lsMemPtr->convexVertexData, (btConvexHullShape*)&convexHullShape);
cellDmaWaitTagStatusAll(DMA_MASK(2)); // dmaConvexVertexData uses dma channel 2!
lsMemPtr->convexVertexData.gSpuConvexShapePtr = gatheredObjectData->m_spuCollisionShape;
lsMemPtr->convexVertexData.gConvexPoints = &lsMemPtr->convexVertexData.g_convexPointBuffer[0];
}
/* performRaycast */
SpuSubsimplexRayCast caster (gatheredObjectData->m_spuCollisionShape, &lsMemPtr->convexVertexData, gatheredObjectData->m_shapeType, gatheredObjectData->m_collisionMargin, &simplexSolver);
bool r = caster.calcTimeOfImpact (rayFromTrans, rayToTrans, gatheredObjectData->m_worldTransform, gatheredObjectData->m_worldTransform,result);
if (r)
{
workUnitOut->hitFraction = result.m_fraction;
workUnitOut->hitNormal = result.m_normal;
}
}
void gviSpursSpeexDecodeSet(SpursSpeexTaskOutput *spuTaskOut)
{
char *inBuffer;
float *speexBuffer;
short *outBuffer;
int rcode;
unsigned int i;
speexBuffer = (float *)memalign(16, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(float));
outBuffer = (short *)memalign(16, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(short));
inBuffer = (char *)memalign(16, gviSpursSpeexTaskDesc.mInputBufferSize);
memset(speexBuffer, 0, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(float));
memset(inBuffer, 0, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(short));
memset(outBuffer, 0, gviSpursSpeexTaskDesc.mInputBufferSize);
cellDmaGet(inBuffer, (uint64_t)gviSpursSpeexTaskDesc.mInputBuffer, gviSpursSpeexTaskDesc.mInputBufferSize, DMA_TAG(1), 0,0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
// read the data into the bits
speex_bits_read_from(&gviSpursSpeexBits, (char *)inBuffer, gviSpursSpeexTaskDesc.mEncodedFrameSize);
// decode it
rcode = speex_decode((void *)gviSpursSpeexStateBuffer, &gviSpursSpeexBits, speexBuffer);
assert(rcode == 0);
// convert the output from floats
for(i = 0 ; i < gviSpursSpeexTaskDesc.mOutputBufferSize ; i++)
// Expanded to remove warnings in VS2K5
outBuffer[i] = (short)speexBuffer[i];
cellDmaPut(outBuffer, (uint64_t)gviSpursSpeexTaskDesc.mOutputBuffer, gviSpursSpeexTaskDesc.mOutputBufferSize * sizeof(short), DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
free(speexBuffer);
free(inBuffer);
free(outBuffer);
spuTaskOut->mSpeexReturnCode = 0;
}
///this unalignedDma should not be frequently used, only for small data. It handles alignment and performs check on size (<16 bytes)
int stallingUnalignedDmaSmallGet(void *ls, uint64_t ea, uint32_t size)
{
btAssert(size<32);
ATTRIBUTE_ALIGNED16(char tmpBuffer[32]);
char* mainMem = (char*)ea;
char* localStore = (char*)ls;
uint32_t i;
///make sure last 4 bits are the same, for cellDmaSmallGet
uint32_t last4BitsOffset = ea & 0x0f;
char* tmpTarget = tmpBuffer + last4BitsOffset;
#if defined (__SPU__) || defined (USE_LIBSPE2)
int remainingSize = size;
//#define FORCE_cellDmaUnalignedGet 1
#ifdef FORCE_cellDmaUnalignedGet
cellDmaUnalignedGet(tmpTarget,ea,size,DMA_TAG(1),0,0);
#else
char* remainingTmpTarget = tmpTarget;
uint64_t remainingEa = ea;
while (remainingSize)
{
switch (remainingSize)
{
case 1:
case 2:
case 4:
case 8:
case 16:
{
mfc_get(remainingTmpTarget,remainingEa,remainingSize,DMA_TAG(1),0,0);
remainingSize=0;
break;
}
default:
{
//spu_printf("unaligned DMA with non-natural size:%d\n",remainingSize);
int actualSize = 0;
if (remainingSize > 16)
actualSize = 16;
else
if (remainingSize >8)
actualSize=8;
else
if (remainingSize >4)
actualSize=4;
else
if (remainingSize >2)
actualSize=2;
mfc_get(remainingTmpTarget,remainingEa,actualSize,DMA_TAG(1),0,0);
remainingSize-=actualSize;
remainingTmpTarget+=actualSize;
remainingEa += actualSize;
}
}
}
#endif//FORCE_cellDmaUnalignedGet
#else
//copy into final destination
#ifdef USE_MEMCPY
memcpy(tmpTarget,mainMem,size);
#else
for ( i=0;i<size;i++)
{
tmpTarget[i] = mainMem[i];
}
#endif //USE_MEMCPY
#endif
cellDmaWaitTagStatusAll(DMA_MASK(1));
//this is slowish, perhaps memcpy on SPU is smarter?
for (i=0; btLikely( i<size );i++)
{
localStore[i] = tmpTarget[i];
}
return 0;
}
void processRaycastTask(void* userPtr, void* lsMemory)
{
RaycastTask_LocalStoreMemory* localMemory = (RaycastTask_LocalStoreMemory*)lsMemory;
SpuRaycastTaskDesc* taskDescPtr = (SpuRaycastTaskDesc*)userPtr;
SpuRaycastTaskDesc& taskDesc = *taskDescPtr;
SpuCollisionObjectWrapper* cows = (SpuCollisionObjectWrapper*)taskDesc.spuCollisionObjectsWrappers;
//spu_printf("in processRaycastTask %d\n", taskDesc.numSpuCollisionObjectWrappers);
/* for each object */
RaycastGatheredObjectData gatheredObjectData;
for (int objectId = 0; objectId < taskDesc.numSpuCollisionObjectWrappers; objectId++)
{
//spu_printf("%d / %d\n", objectId, taskDesc.numSpuCollisionObjectWrappers);
/* load initial collision shape */
GatherCollisionObjectAndShapeData (&gatheredObjectData, localMemory, (ppu_address_t)&cows[objectId]);
if (btBroadphaseProxy::isConcave (gatheredObjectData.m_shapeType))
{
SpuRaycastTaskWorkUnitOut tWorkUnitsOut[SPU_RAYCAST_WORK_UNITS_PER_TASK];
for (int rayId = 0; rayId < taskDesc.numWorkUnits; rayId++)
{
tWorkUnitsOut[rayId].hitFraction = 1.0;
}
performRaycastAgainstConcave (&gatheredObjectData, &taskDesc.workUnits[0], &tWorkUnitsOut[0], taskDesc.numWorkUnits, localMemory);
for (int rayId = 0; rayId < taskDesc.numWorkUnits; rayId++)
{
const SpuRaycastTaskWorkUnit& workUnit = taskDesc.workUnits[rayId];
if (tWorkUnitsOut[rayId].hitFraction == 1.0)
continue;
ATTRIBUTE_ALIGNED16(SpuRaycastTaskWorkUnitOut workUnitOut);
dmaLoadRayOutput ((ppu_address_t)workUnit.output, &workUnitOut, 1);
cellDmaWaitTagStatusAll(DMA_MASK(1));
/* XXX Only support taking the closest hit for now */
if (tWorkUnitsOut[rayId].hitFraction < workUnitOut.hitFraction)
{
workUnitOut.hitFraction = tWorkUnitsOut[rayId].hitFraction;
workUnitOut.hitNormal = tWorkUnitsOut[rayId].hitNormal;
}
/* write ray cast data back */
dmaStoreRayOutput ((ppu_address_t)workUnit.output, &workUnitOut, 1);
cellDmaWaitTagStatusAll(DMA_MASK(1));
}
} else if (btBroadphaseProxy::isConvex (gatheredObjectData.m_shapeType)) {
btVector3 objectBoxMin, objectBoxMax;
computeAabb (objectBoxMin, objectBoxMax, (btConvexInternalShape*)gatheredObjectData.m_spuCollisionShape, gatheredObjectData.m_collisionShape, gatheredObjectData.m_shapeType, gatheredObjectData.m_worldTransform);
for (unsigned int rayId = 0; rayId < taskDesc.numWorkUnits; rayId++)
{
const SpuRaycastTaskWorkUnit& workUnit = taskDesc.workUnits[rayId];
btScalar ignored_param = 1.0;
btVector3 ignored_normal;
if (btRayAabb(workUnit.rayFrom, workUnit.rayTo, objectBoxMin, objectBoxMax, ignored_param, ignored_normal))
{
ATTRIBUTE_ALIGNED16(SpuRaycastTaskWorkUnitOut workUnitOut);
SpuRaycastTaskWorkUnitOut tWorkUnitOut;
tWorkUnitOut.hitFraction = 1.0;
performRaycastAgainstConvex (&gatheredObjectData, workUnit, &tWorkUnitOut, localMemory);
if (tWorkUnitOut.hitFraction == 1.0)
continue;
dmaLoadRayOutput ((ppu_address_t)workUnit.output, &workUnitOut, 1);
cellDmaWaitTagStatusAll(DMA_MASK(1));
/* XXX Only support taking the closest hit for now */
if (tWorkUnitOut.hitFraction < workUnitOut.hitFraction)
{
workUnitOut.hitFraction = tWorkUnitOut.hitFraction;
workUnitOut.hitNormal = tWorkUnitOut.hitNormal;
/* write ray cast data back */
dmaStoreRayOutput ((ppu_address_t)workUnit.output, &workUnitOut, 1);
cellDmaWaitTagStatusAll(DMA_MASK(1));
}
}
}
} else if (btBroadphaseProxy::isCompound (gatheredObjectData.m_shapeType)) {
for (unsigned int rayId = 0; rayId < taskDesc.numWorkUnits; rayId++)
{
const SpuRaycastTaskWorkUnit& workUnit = taskDesc.workUnits[rayId];
ATTRIBUTE_ALIGNED16(SpuRaycastTaskWorkUnitOut workUnitOut);
SpuRaycastTaskWorkUnitOut tWorkUnitOut;
tWorkUnitOut.hitFraction = 1.0;
performRaycastAgainstCompound (&gatheredObjectData, workUnit, &tWorkUnitOut, localMemory);
if (tWorkUnitOut.hitFraction == 1.0)
continue;
dmaLoadRayOutput ((ppu_address_t)workUnit.output, &workUnitOut, 1);
cellDmaWaitTagStatusAll(DMA_MASK(1));
/* XXX Only support taking the closest hit for now */
if (tWorkUnitOut.hitFraction < workUnitOut.hitFraction)
{
workUnitOut.hitFraction = tWorkUnitOut.hitFraction;
workUnitOut.hitNormal = tWorkUnitOut.hitNormal;
}
/* write ray cast data back */
dmaStoreRayOutput ((ppu_address_t)workUnit.output, &workUnitOut, 1);
cellDmaWaitTagStatusAll(DMA_MASK(1));
}
}
}
}
void performRaycastAgainstConcave (RaycastGatheredObjectData* gatheredObjectData, const SpuRaycastTaskWorkUnit* workUnits, SpuRaycastTaskWorkUnitOut* workUnitsOut, int numWorkUnits, RaycastTask_LocalStoreMemory* lsMemPtr)
{
//order: first collision shape is convex, second concave. m_isSwapped is true, if the original order was opposite
register int dmaSize;
register ppu_address_t dmaPpuAddress2;
btBvhTriangleMeshShape* trimeshShape = (btBvhTriangleMeshShape*)gatheredObjectData->m_spuCollisionShape;
//need the mesh interface, for access to triangle vertices
dmaBvhShapeData (&(lsMemPtr->bvhShapeData), trimeshShape);
unsigned short int quantizedQueryAabbMin[SPU_RAYCAST_WORK_UNITS_PER_TASK][3];
unsigned short int quantizedQueryAabbMax[SPU_RAYCAST_WORK_UNITS_PER_TASK][3];
btVector3 rayFromInTriangleSpace[SPU_RAYCAST_WORK_UNITS_PER_TASK];
btVector3 rayToInTriangleSpace[SPU_RAYCAST_WORK_UNITS_PER_TASK];
/* Calculate the AABB for the ray in the triangle mesh shape */
btTransform rayInTriangleSpace;
rayInTriangleSpace = gatheredObjectData->m_worldTransform.inverse();
for (int i = 0; i < numWorkUnits; i++)
{
btVector3 aabbMin;
btVector3 aabbMax;
rayFromInTriangleSpace[i] = rayInTriangleSpace(workUnits[i].rayFrom);
rayToInTriangleSpace[i] = rayInTriangleSpace(workUnits[i].rayTo);
aabbMin = rayFromInTriangleSpace[i];
aabbMin.setMin (rayToInTriangleSpace[i]);
aabbMax = rayFromInTriangleSpace[i];
aabbMax.setMax (rayToInTriangleSpace[i]);
lsMemPtr->bvhShapeData.getOptimizedBvh()->quantizeWithClamp(quantizedQueryAabbMin[i],aabbMin,0);
lsMemPtr->bvhShapeData.getOptimizedBvh()->quantizeWithClamp(quantizedQueryAabbMax[i],aabbMax,1);
}
QuantizedNodeArray& nodeArray = lsMemPtr->bvhShapeData.getOptimizedBvh()->getQuantizedNodeArray();
//spu_printf("SPU: numNodes = %d\n",nodeArray.size());
BvhSubtreeInfoArray& subTrees = lsMemPtr->bvhShapeData.getOptimizedBvh()->getSubtreeInfoArray();
#ifdef CALLBACK_ALL
spuRaycastNodeCallback nodeCallback (gatheredObjectData, workUnits, workUnitsOut, numWorkUnits, lsMemPtr);
#else
spuRaycastNodeCallback1 nodeCallback (gatheredObjectData, workUnits, workUnitsOut, lsMemPtr);
#endif
IndexedMeshArray& indexArray = lsMemPtr->bvhShapeData.gTriangleMeshInterfacePtr->getIndexedMeshArray();
//spu_printf("SPU:indexArray.size() = %d\n",indexArray.size());
// spu_printf("SPU: numSubTrees = %d\n",subTrees.size());
//not likely to happen
if (subTrees.size() && indexArray.size() == 1)
{
///DMA in the index info
dmaBvhIndexedMesh (&lsMemPtr->bvhShapeData.gIndexMesh, indexArray, 0 /* index into indexArray */, 1 /* dmaTag */);
cellDmaWaitTagStatusAll(DMA_MASK(1));
//display the headers
int numBatch = subTrees.size();
for (int i=0;i<numBatch;)
{
// BEN: TODO - can reorder DMA transfers for less stall
int remaining = subTrees.size() - i;
int nextBatch = remaining < MAX_SPU_SUBTREE_HEADERS ? remaining : MAX_SPU_SUBTREE_HEADERS;
dmaBvhSubTreeHeaders (&lsMemPtr->bvhShapeData.gSubtreeHeaders[0], (ppu_address_t)(&subTrees[i]), nextBatch, 1);
cellDmaWaitTagStatusAll(DMA_MASK(1));
// spu_printf("nextBatch = %d\n",nextBatch);
for (int j=0;j<nextBatch;j++)
{
const btBvhSubtreeInfo& subtree = lsMemPtr->bvhShapeData.gSubtreeHeaders[j];
unsigned int overlap = 1;
for (int boxId = 0; boxId < numWorkUnits; boxId++)
{
overlap = spuTestQuantizedAabbAgainstQuantizedAabb(quantizedQueryAabbMin[boxId],quantizedQueryAabbMax[boxId],subtree.m_quantizedAabbMin,subtree.m_quantizedAabbMax);
if (overlap)
break;
}
if (overlap)
{
btAssert(subtree.m_subtreeSize);
//dma the actual nodes of this subtree
dmaBvhSubTreeNodes (&lsMemPtr->bvhShapeData.gSubtreeNodes[0], subtree, nodeArray, 2);
cellDmaWaitTagStatusAll(DMA_MASK(2));
/* Walk this subtree */
{
spuWalkStacklessQuantizedTreeAgainstRays(lsMemPtr,
&nodeCallback,
&rayFromInTriangleSpace[0],
&rayToInTriangleSpace[0],
numWorkUnits,
&quantizedQueryAabbMin[0][0],&quantizedQueryAabbMax[0][0],
&lsMemPtr->bvhShapeData.gSubtreeNodes[0], 0, subtree.m_subtreeSize);
}
}
// spu_printf("subtreeSize = %d\n",gSubtreeHeaders[j].m_subtreeSize);
}
// unsigned short int m_quantizedAabbMin[3];
// unsigned short int m_quantizedAabbMax[3];
// int m_rootNodeIndex;
// int m_subtreeSize;
i+=nextBatch;
}
//pre-fetch first tree, then loop and double buffer
}
}
//-- MAIN METHOD
void processSampleTask(void* userPtr, void* lsMemory)
{
// BT_PROFILE("processSampleTask");
SampleTask_LocalStoreMemory* localMemory = (SampleTask_LocalStoreMemory*)lsMemory;
SpuSampleTaskDesc* taskDescPtr = (SpuSampleTaskDesc*)userPtr;
SpuSampleTaskDesc& taskDesc = *taskDescPtr;
switch (taskDesc.m_sampleCommand)
{
case CMD_SAMPLE_INTEGRATE_BODIES:
{
btTransform predictedTrans;
btCollisionObject** eaPtr = (btCollisionObject**)taskDesc.m_mainMemoryPtr;
int batchSize = taskDesc.m_sampleValue;
if (batchSize>MAX_NUM_BODIES)
{
spu_printf("SPU Error: exceed number of bodies, see MAX_NUM_BODIES in SpuSampleTask.cpp\n");
break;
}
int dmaArraySize = batchSize*sizeof(void*);
uint64_t ppuArrayAddress = reinterpret_cast<uint64_t>(eaPtr);
// spu_printf("array location is at %llx, batchSize = %d, DMA size = %d\n",ppuArrayAddress,batchSize,dmaArraySize);
if (dmaArraySize>=16)
{
cellDmaLargeGet((void*)&localMemory->gPointerArray[0], ppuArrayAddress , dmaArraySize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
} else
{
stallingUnalignedDmaSmallGet((void*)&localMemory->gPointerArray[0], ppuArrayAddress , dmaArraySize);
}
for ( int i=0;i<batchSize;i++)
{
///DMA rigid body
void* localPtr = &localMemory->gLocalRigidBody[0];
void* shortAdd = localMemory->gPointerArray[i];
uint64_t ppuRigidBodyAddress = reinterpret_cast<uint64_t>(shortAdd);
// spu_printf("cellDmaGet at CMD_SAMPLE_INTEGRATE_BODIES from %llx to %llx\n",ppuRigidBodyAddress,localPtr);
int dmaBodySize = sizeof(btRigidBody);
cellDmaGet((void*)localPtr, ppuRigidBodyAddress , dmaBodySize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
float timeStep = 1.f/60.f;
btRigidBody* body = (btRigidBody*) localPtr;//btRigidBody::upcast(colObj);
if (body)
{
if (body->isActive() && (!body->isStaticOrKinematicObject()))
{
body->predictIntegratedTransform(timeStep, predictedTrans);
body->proceedToTransform( predictedTrans);
void* ptr = (void*)localPtr;
// spu_printf("cellDmaLargePut from %llx to LS %llx\n",ptr,ppuRigidBodyAddress);
cellDmaLargePut(ptr, ppuRigidBodyAddress , dmaBodySize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
}
}
}
break;
}
case CMD_SAMPLE_PREDICT_MOTION_BODIES:
{
btTransform predictedTrans;
btCollisionObject** eaPtr = (btCollisionObject**)taskDesc.m_mainMemoryPtr;
int batchSize = taskDesc.m_sampleValue;
int dmaArraySize = batchSize*sizeof(void*);
if (batchSize>MAX_NUM_BODIES)
{
spu_printf("SPU Error: exceed number of bodies, see MAX_NUM_BODIES in SpuSampleTask.cpp\n");
break;
}
uint64_t ppuArrayAddress = reinterpret_cast<uint64_t>(eaPtr);
// spu_printf("array location is at %llx, batchSize = %d, DMA size = %d\n",ppuArrayAddress,batchSize,dmaArraySize);
if (dmaArraySize>=16)
{
cellDmaLargeGet((void*)&localMemory->gPointerArray[0], ppuArrayAddress , dmaArraySize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
} else
{
stallingUnalignedDmaSmallGet((void*)&localMemory->gPointerArray[0], ppuArrayAddress , dmaArraySize);
}
for ( int i=0;i<batchSize;i++)
{
///DMA rigid body
void* localPtr = &localMemory->gLocalRigidBody[0];
void* shortAdd = localMemory->gPointerArray[i];
uint64_t ppuRigidBodyAddress = reinterpret_cast<uint64_t>(shortAdd);
// spu_printf("cellDmaGet at CMD_SAMPLE_INTEGRATE_BODIES from %llx to %llx\n",ppuRigidBodyAddress,localPtr);
int dmaBodySize = sizeof(btRigidBody);
cellDmaGet((void*)localPtr, ppuRigidBodyAddress , dmaBodySize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
float timeStep = 1.f/60.f;
btRigidBody* body = (btRigidBody*) localPtr;//btRigidBody::upcast(colObj);
if (body)
{
if (!body->isStaticOrKinematicObject())
{
if (body->isActive())
{
body->integrateVelocities( timeStep);
//damping
body->applyDamping(timeStep);
body->predictIntegratedTransform(timeStep,body->getInterpolationWorldTransform());
void* ptr = (void*)localPtr;
cellDmaLargePut(ptr, ppuRigidBodyAddress , dmaBodySize, DMA_TAG(1), 0, 0);
cellDmaWaitTagStatusAll(DMA_MASK(1));
}
}
}
}
break;
}
default:
{
}
};
}