cs526-cp2

Environment

Quadro P2200 and Intel(R) Xeon(R) W-2245 CPU with CUDA 11.6 and Ubuntu 21.10

Requirements

LLVM configured as cmake -S llvm -B build -G Ninja -DCMAKE_INSTALL_PREFIX="~/llvm" -DLLVM_ENABLE_PROJECTS="clang"
cmake --build build && cmake --install build
CUDA installed at /usr/local/cuda
sudo apt install mesa-common-dev
llvm installed or linked to ./llvm, or set Makefile's LLVMROOT accordingly
libclc (https://github.com/llvm-mirror/libclc) built (only nvptx) and installed to LLVMROOT
Example configure script (change --prefix and --with-llvm-config accordingly):
./configure.py --prefix=/home/hwnam/llvm --with-llvm-config=/home/hwnam/llvm/bin/llvm-config nvptx--nvidiacl nvptx64--nvidiacl

Step-by-Step guide to setup environment

1. Download LLVM 12.0.0 release (llvm-project-12.0.0.src.tar.xz) from https://github.com/llvm/llvm-project/releases/tag/llvmorg-12.0.0
2. Unzip LLVM folder and let's call the unzipped folder llvm
3. Configure by cmake -S llvm -B build -G Ninja -DCMAKE_INSTALL_PREFIX="~/llvm" -DLLVM_ENABLE_PROJECTS="clang" (can change the prefix to anywhere you want your LLVMROOT in the ./tests/Makefile to be).
4. Inside llvm, build by cmake --build build && cmake --install build
5. Install the CUDA 11.6 Toolkit from https://developer.nvidia.com/cuda-downloads using the corresponding runfile and following the command there.
6. Say CUDA is installed at /usr/local/cuda, then this is your CUDAROOT in ./tests/Makefile
7. sudo apt install mesa-common-dev
8. Git clone our code from https://github.com/hwnam831/cs526-cp2.git
9. Make sure LLVMROOT and CUDAROOT is updated in the cs526-cp2/tests/Makefile.
10. Git clone libclc from https://github.com/llvm-mirror/libclc.git
11. Inside libclc, run ./configure.py --prefix=~/llvm --with-llvm-config=~/llvm/bin/llvm-config nvptx--nvidiacl nvptx64--nvidiacl
12. Then make && make install
13. Inside cs526-cp2, run ./cmake.sh
14. Go ./tests
15. Follow the "To Run Test Guide" below to run test.

• You could have the following directory structure:
  ├── cs526-cp2
  ├── llvm
  ├── libclc
  

Test Suite

The list of algorithms tested in the paper (https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.870.3097&rep=rep1&type=pdf):

transpose matrix vector (tmv)
martrix multiplication (matmul)
matrix-vector multiplication (mv)
vector-vector multiplication (vv)
reduction (reduction)
matrix equation solver (strsm)
convolution (conv)
matrix transpose (transpose)
reconstruct image (demosaic)
find the regional maxima (imregionmax)

To Run Test Guide

cd tests

• Run all tests: make all
• Clean all output files: make cleanAll
• Run individial test (take tmv for example): cd tmv && make
  • If Test PASSED is printed, then test passes. Otherwise, if test fails, Test FAILED will be printed.
    
• Clean individial test outputs (take tmv for example): cd tmv && make clean
• Run test case with the original naive kernel (take tmv for example): cd tmv && make original
• Profile the execution time for naive, coalesced, and prefetched kernels (take tmv for example): cd tmv && make profile
• To generate intermeidate ll files after each pass (take tmv for example): cd tmv && make debug
  • The ll for original naive kernel can be found in tmv.nvvm.ll
  • The ll for coalesced kernel can be found in tmv.coal.ll
  • The ll for prefetched kernel can be found in tmv.opt.ll