void MarkLoops(Var * proc)
{
InstrBlock * blk;
NumberBlocks(proc->instr);
for(blk = proc->instr; blk != NULL; blk = blk->next) {
blk->jump_type = JUMP_IF;
}
// Test for each block, if it is end of an loop
for(blk = proc->instr; blk != NULL; blk = blk->next) {
if (blk->cond_to != NULL) {
// We jump to some previous spot in the sequence
if (blk->cond_to->seq_no <= blk->seq_no) {
blk->jump_type = JUMP_LOOP; //->cond_to->loop_end = blk;
}
}
if (blk->to != NULL) {
if (blk->to->seq_no <= blk->seq_no) {
blk->jump_type = JUMP_LOOP; //to->loop_end = blk;
}
}
}
}
void Reduction<VoxelType>::Max(VoxelType * inputs, VoxelType * output, int numInputs, cudaStream_t stream) {
int blockSize = NumberThreadsPerBlockThatBestFit(numInputs, OPTIMAL_BLOCK_SIZE_REDUCTION);
if (numInputs > SIZE_SMALL_CUDA_VECTOR) {
int blocks = NumberBlocks(numInputs, blockSize);
if (temporaryBuffer->Length() < blocks) temporaryBuffer->ResizeWithoutPreservingData(blocks);
KernelMax<VoxelType>(stream, blocks, blockSize, inputs, temporaryBuffer->Pointer(), numInputs);
inputs = temporaryBuffer->Pointer();
numInputs = blocks;
blockSize = NumberThreadsPerBlockThatBestFit(numInputs);
}
KernelMax<VoxelType>(stream, 1, blockSize, inputs, output, numInputs);
}
void Reduction<VoxelType>::MinIndex(VoxelType * inputs, VoxelType * output, int * minIndex, int numInputs, cudaStream_t stream) {
int blockSize = NumberThreadsPerBlockThatBestFit(numInputs, OPTIMAL_BLOCK_SIZE_REDUCTION);
int * indexes = nullptr;
if (numInputs > SIZE_SMALL_CUDA_VECTOR) {
int blocks = NumberBlocks(numInputs, blockSize);
int minSizeBuffer = blocks + (int) ceil(blocks * (sizeof(int) / (float) sizeof(float)));
if (temporaryBuffer->Length() < minSizeBuffer) temporaryBuffer->ResizeWithoutPreservingData(minSizeBuffer);
indexes = (int *)(temporaryBuffer->Pointer() + blocks);
KernelMinIndexes<VoxelType>(stream, blocks, blockSize, inputs, temporaryBuffer->Pointer(), indexes, numInputs, nullptr);
inputs = temporaryBuffer->Pointer();
numInputs = blocks;
blockSize = NumberThreadsPerBlockThatBestFit(numInputs);
}
KernelMinIndexes<VoxelType>(stream, 1, blockSize, inputs, output, minIndex, numInputs, indexes);
}
