Ignore:
Timestamp:
Apr 22, 2017, 4:03:25 PM (2 years ago)
Author:
nmedfort
Message:

Removed non-functional CUDA code from icgrep and consolidated grep and multigrep mode into a single function; allowed segment parallel pipeline to utilize process as its initial thread; modified MMapSourceKernel to map and perform mmap directly and advise the OS to drop consumed data streams.

File:
1 edited

Legend:

Unmodified
Added
Removed
  • icGREP/icgrep-devel/icgrep/kernels/radix64.cpp

    r5402 r5418  
    4141// a continous buffer for the full segment (number of blocks).
    4242
    43 void expand3_4Kernel::generateDoSegmentMethod(Value *doFinal, const std::vector<Value *> &) {
     43void expand3_4Kernel::generateDoSegmentMethod() {
    4444
    4545    BasicBlock * expand2_3entry = iBuilder->GetInsertBlock();
     
    9595    // process in multiples of 3 full blocks of data.
    9696    //
    97     Value * loopDivisor = iBuilder->CreateSelect(doFinal, triplePackSize, tripleBlockSize);
     97    Value * loopDivisor = iBuilder->CreateSelect(getIsFinal(), triplePackSize, tripleBlockSize);
    9898    Value * excessItems = iBuilder->CreateURem(itemsAvail, loopDivisor);
    9999    Value * loopItemsToDo = iBuilder->CreateSub(itemsAvail, excessItems);
     
    162162
    163163    // Except for final segment processing, we are done.
    164     iBuilder->CreateCondBr(doFinal, expand3_4_final, expand3_4_exit);
     164    iBuilder->CreateCondBr(getIsFinal(), expand3_4_final, expand3_4_exit);
    165165
    166166    // Final segment processing.   Less than a triplePack remains.
Note: See TracChangeset for help on using the changeset viewer.