DOCUMENTATION_GPU.md

FastR-GPU Compiler Framework for OpenCL JIT Compilation

Threads options

Use the marawacc threads when sapply for Java threads

To enable marawacc threads:

-Dastx.marawacc.threads=true 

Disable with:

-Dastx.marawacc.threads=false

OpenCL optimizations

Optimization for avoiding marshal and unmarshal

Enable PArrays in marshal and unmarshal

-Dastx.marawacc.usePArrays=true

Use Sequence

-Dastx.marawacc.optimizeRSequence=true

Use subtypes of new sequences (Flag and Compass)

-Dastx.marawacc.useTypeOfSequences=true

Use primitive arrays for PArrays (save marshal and unmarshall totally)

-Dastx.marawacc.primArrays=true

Debug

Print R AST to be executed on the GPU.

-Dastx.marawacc.printast=true   # False by default

Enable profiling

-Dastx.marawacc.profiler=true   # false by default

Print the autogenerated kernel (False by default)

-Dmarawacc.printOCLKernel=true

Print the GraalIR that will be used for OpenCL kernel generation

-Dmarawacc.printGraalIR=true

Print OpenCL Information about the platform and kernel compilation

-Dmarawacc.printOCLInfo=true

Dump Graph IR with IGV

-Dmarawacc.dumpGraph=true

Combine with Marawacc options: FastR can set marawacc options too

Choose a specific OpenCL vendor for CPU

$ export GRAAL_CPU_PLATFORM="INTEL"

Chosse a specific OpenCL vendor for GPU

$ export GRAAL_GPU_PLATFORM="NVIDIA"