akothen / PIMeval-PIMbench

PIMeval simulator and PIMbench suite

PIMeval Simulator and PIMbench Suite

Description

• PIMeval
  • A PIM simulation and evaluation framework implemented as a C++ library
  • Support various subarray-level bit-serial, subarray-level bit-parallel and bank-level PIM architectures
  • Support both vertical and horizontal data layouts
  • Support multi-PIM-core programming model and resource management
  • Support high-level functional programming with a set of general APIs common to all PIM architectures
  • Support low-level micro-ops programming for modeling architecture details
  • Support performance and energy modeling with detailed stats tracking
  • Support multi-threaded simulation for runtime
• PIMbench
  • A rich set of PIM benchmark applications on top of the PIMeval functional simulation and evaluation framework

Quick start

git clone <url_to_your_fork>
cd PIMeval/
make -j<n_proc>
cd /PIMbench/<application_dir>
./<application_executable_name>.out

Code Structure

• PIMeval: PIM similation framework - libpimeval
  • libpimeval/src: PIMeval simulator source code
  • libpimeval.h: PIMeval simulator library interface
  • libpimeval.a: PIMeval simulator library (after make)
• PIMbench: PIM benchmark suite
  • cpp-aes: AES encryption/decryption
  • cpp-axpy: aX+Y operation
  • cpp-filter-by-key: Filer by key
  • cpp-gemm: General matrix-matrix product
  • cpp-gemv: General matrix-vector product
  • cpp-histogram: Histogram
  • cpp-brightness: Image brightness
  • cpp-image-downsampling: Image downsampling
  • cpp-kmeans: K-means
  • cpp-knn: kNN
  • cpp-linear-regression: Linear regression
  • cpp-radix-sort: Radix sort
  • cpp-triangle-count: Triangle counting
  • cpp-vec-add: Vector addition
  • cpp-vgg13: VGG-13
  • cpp-vgg16: VGG-16
  • cpp-vgg19: VGG-19
• More applications
  • cpp-convolution: Convolution
  • cpp-db-filtering: DB filtering
  • cpp-dot-prod: Dot product
  • cpp-pooling: Max pooling
  • cpp-relu: ReLU
  • cpp-sad: Sum of absolute difference
  • cpp-vec-arithmetic: Vector arithmetic
  • cpp-vec-comp: Vector comparison
  • cpp-vec-div: Vector division
  • cpp-vec-logical: Vector logical operations
  • cpp-vec-mul: Vector multiplication
  • cpp-vec-popcount: Vector popcount
  • cpp-vec-broadcast-popcnt: Vector broadcast and pop count
• Bit-serial micro-program evaluation framework
  • bit-serial
• Functional tests
  • tests

How To Build

• Run make at root directory or subdirectories
  • make perf: Build with -Ofast for performance measurement (default)
  • make debug: Build with -g and -DDEBUG for debugging and printing verbose messages

• Multi-threaded building
  • make -j<n_proc>
• Specify simulation target
  • make PIM_SIM_TARGET=PIM_DEVICE_BITSIMD_V (default)
  • make PIM_SIM_TARGET=PIM_DEVICE_FULCRUM
  • make PIM_SIM_TARGET=PIM_DEVICE_BANK_LEVEL
• Build with OpenMP
  • make USE_OPENMP=1
  • Guard any -fopenmp with this flag in Makefile used by a few applications

Contributors

This repository is the result of a collaborative effort by many individuals, including Farzana Ahmed Siddique, Deyuan Guo, Zhenxing Fan, Mohammadhosein Gholamrezaei, Morteza Baradaran, Alif Ahmed, Hugo Abbot, Kyle Durrer, Ethan Ermovick, Kumaresh Nandagopal and Khyati Kiyawat. We are grateful to everyone who contributed to this repository. Special thanks go to Deyuan Guo for initially architecting the PIMeval simulator framework and bit-serial evaluation, and to Farzana Ahmed Siddique for her exceptional contributions to both the simulator and the PIMbench suite.

<citation recommendation to be updated>