Ignore:
Timestamp:
Feb 27, 2017, 2:30:14 PM (2 years ago)
Author:
nmedfort
Message:

Update to BlockOrientedKernel? to move the indirect branch out of the StrideLoopBody? and into StrideLoopDone? to simplify branch prediction.

File:
1 edited

Legend:

Unmodified
Added
Removed
  • icGREP/icgrep-devel/icgrep/toolchain.cpp

    r5350 r5351  
    1313#include <llvm/ADT/SmallString.h>                  // for SmallString
    1414#include <llvm/IR/LegacyPassManager.h>             // for PassManager
     15#include <llvm/IR/Verifier.h>
    1516#include <llvm/InitializePasses.h>                 // for initializeCodeGen
    1617#include <llvm/PassRegistry.h>                     // for PassRegistry
     
    2122#include <llvm/Transforms/Scalar.h>
    2223#include <llvm/Transforms/Utils/Local.h>
    23 #ifndef NDEBUG
    24 #include <llvm/IR/Verifier.h>
    25 #endif
    2624#include <object_cache.h>
    27 namespace llvm { class Module; }
    2825#ifdef CUDA_ENABLED
    2926#include <IR_Gen/llvm2ptx.h>
     
    6764int ThreadNum;
    6865bool EnableAsserts;
     66
     67
    6968
    7069static cl::opt<int, true> BlockSizeOption("BlockSize", cl::location(BlockSize), cl::init(0), cl::desc("specify a block size (defaults to widest SIMD register width in bits)."), cl::cat(CodeGenOptions));
Note: See TracChangeset for help on using the changeset viewer.