MIMDRAM: An End-to-End Processing-Using-DRAM System for High-Throughput, Energy-Efficient and Programmer-Transparent Multiple-Instruction Multiple-Data Processing

MIMDRAM is a hardware/software co-designed processing-using-DRAM (PUD) system that introduces new mechanisms to allocate and control only the necessary resources for a given PUD operation. The key idea of MIMDRAM is to leverage fine-grained DRAM (i.e., the ability to independently access smaller segments of a large DRAM row) for PUD computation. MIMDRAM exploits this key idea to enable a multiple-instruction multiple-data (MIMD) execution model in each DRAM subarray (and SIMD execution within each DRAM row segment).

This repository contains the simulation infrastructure used to evaluate MIMDRAM. We plan to release the compiler infrastructure soon.

Citation

Please cite the following if you find this repository useful:

Geraldo F. Oliveira, Ataberk Olgun, Abdullah Giray Yağlıkçı, F. Nisa Bostancı, Juan Gómez-Luna, Saugata Ghose, Onur Mutlu, "MIMDRAM: An End-to-End Processing-Using-DRAM System for High-Throughput, Energy-Efficient and Programmer-Transparent Multiple-Instruction Multiple-Data Computing", in HPCA, 2024.

Bibtex entry for citation:

@inproceedings{mimdram,
    author = "Oliveira, Geraldo F and Olgun, Ataberk and Yaglik{\c{c}}i, A. Giray and Bostanci, Nisa and G{\'o}mez-Luna, Juan and Ghose, Saugata and Mutlu, Onur",
    title = "{MIMDRAM: An End-to-End Processing-Using-DRAM System for High-Throughput, Energy-Efficient and Programmer-Transparent Multiple-Instruction Multiple-Data Computing}",
    booktitle = "HPCA",
    year = "2024"
}

Setting up MIMDRAM

Repository Structure and Installation

We point out next to the repository structure and some important folders and files.

.
+-- README.md
+-- gem5/
|   +-- build_opts/
|   +-- configs/
|   +-- configs/
|   +-- ext/
|   +-- protoc3/
|   +-- src/
|   +-- system/
|   +-- tests/
|   +-- util/
+-- microworkloads/

Step 0: Prerequisites

Our framework requires all gem5 dependencies and was tested using Ubuntu 14.04.

To install gem5 dependencies:

sudo apt install build-essential git m4 scons zlib1g zlib1g-dev \
                 libprotobuf-dev protobuf-compiler libprotoc-dev libgoogle-perftools-dev \
                 python-dev python swig

You may need to include the following line into gem5/SConstruct at line 555:

main.Append(CCFLAGS=['-DPROTOBUF_INLINE_NOT_IN_HEADERS=0'])

Step 1: Installing the Simulator

To install the simulator:

cd gem5
scons build/X86/gem5.opt -j8
cd ..

Step 2: Compiling the Workloads

To compile a workload, you need to link the source code with the ../gem5/util/m5 folder, as follows (we use the bitweave-buddy.c file as an example):

cd microworkloads/
gcc -**static -std=c99 -O3 -msse2 -I ../gem5/util/m5 m5op_x86.S rowop.S** bitweave-buddy.c -o bitweave-buddy.exe
cd ..

You can compile all workloads at once by using the compile.sh script

cd microworkloads/
./compile.sh
cd ..

Step 3: Running a Simulation

After building gem5 and the workloads, you can execute a simulation as follows:

cd gem5/
./build/X86/gem5.opt configs/example/se.py --cpu-type=detailed --caches --l2cache --mem-type=DDR4_2400_x64 --mem-size=8192MB -c ABSOLUTE_PATH/bitweave-buddy.exe -o "10 1"

Note that you must specify the absolute path (ABSOLUTE_PATH) to the folder containing the compiled workloads.

Getting Help

If you have any suggestions for improvement, please contact geraldo dot deoliveira at safari dot ethz dot ch. If you find any bugs or have further questions or requests, please post an issue at the issue page.

Acknowledgments

We acknowledge the generous gifts from our industrial partners; including Google, Huawei, Intel, Microsoft. This work is supported in part by the Semiconductor Research Corporation, the ETH Future Computing Laboratory, the AI Chip Center for Emerging Smart Systems (ACCESS), and the UIUC/ZJU HYBRID Center.

CMU-SAFARI/MIMDRAM