Commit 0232a4f9 authored by Stelle, George Widgery's avatar Stelle, George Widgery Committed by George Stelle
Browse files

Fixed run size

parent 175e7002

clang/lib/Headers/kitsune.h

+1 −1
Original line number Diff line number Diff line
@@ -3,7 +3,7 @@ 
#include <stdint.h>



#define ocl_mmap(a, n) __kitsune_opencl_mmap_marker((void*)a, n)

void __kitsune_opencl_mmap_marker(void* ptr, uint64_t n);

extern "C" void __kitsune_opencl_mmap_marker(void* ptr, uint64_t n);

#define spawn _kitsune_spawn

#define sync _kitsune_sync

#define forall _kitsune_forall

llvm/lib/Transforms/Tapir/OpenCLABI.cpp

+11 −2
Original line number Diff line number Diff line
@@ -367,6 +367,7 @@ void SPIRVLoop::processOutlinedLoopCall(TapirLoopInfo &TL, TaskOutlineInfo &TOI, 
  PassManager->add(createReassociatePass());

  PassManager->add(createGVNPass());

  PassManager->add(createCFGSimplificationPass());

  PassManager->add(createLoopVectorizePass());

  PassManager->add(createSLPVectorizerPass());

  //PassManager->add(createBreakCriticalEdgesPass());

  PassManager->add(createConstantPropagationPass());
@@ -477,7 +478,15 @@ void SPIRVLoop::processOutlinedLoopCall(TapirLoopInfo &TL, TaskOutlineInfo &TOI, 
  }





  Value *RunSize = B.CreateSub(TripCount, ConstantInt::get(TripCount->getType(), 1));

  Value *Grainsize = TL.getGrainsize() ?  

    ConstantInt::get(TripCount->getType(), TL.getGrainsize()) :

    OrderedInputs[2]; 



  Value *RunSizeQ = B.CreateUDiv(TripCount, Grainsize);

  Value *RunRem = B.CreateURem(TripCount, Grainsize);

  Value *IsRem = B.CreateICmp(ICmpInst::ICMP_UGT, RunRem, ConstantInt::get(RunRem->getType(), 0)); 

  Value *IsRemAdd = B.CreateZExt(IsRem, RunSizeQ->getType()); 

  Value *RunSize = B.CreateAdd(RunSizeQ, IsRemAdd);  

  

  B.CreateCall(KitsuneGPUSetRunSize, { KernelID, RunSize });