README.md

Matrix Multiplication Bench

Compare MatMul performance of various library implementations.

Setup

Note

Eigen and Google's Benchmark need to be located one directory up from gemm_microbench.

Install Eigen

#Get Eigen from the official website
wget https://gitlab.com/libeigen/eigen/-/archive/3.4.0/eigen-3.4.0.zip

#Unpack and set the correct name
unzip eigen-3.4.0.zip && rm eigen-3.4.0.zip && mv eigen-3.4.0 Eigen

If the above doesn't work, reference Eigen's Official Website

Install Google's Benchmark

# Check out the library.
$ git clone https://github.com/google/benchmark.git

# Go to the library root directory
$ cd benchmark

# Make a build directory to place the build output
$ cmake -E make_directory "build"

# Generate build system files with cmake, and download any dependencies
$ cmake -E chdir "build" cmake -DBENCHMARK_DOWNLOAD_DEPENDENCIES=on -DCMAKE_BUILD_TYPE=Release ../

# or, starting with CMake 3.13, use a simpler form:
# cmake -DBENCHMARK_DOWNLOAD_DEPENDENCIES=on -DCMAKE_BUILD_TYPE=Release -S . -B "build"
# Build the library
$ cmake --build "build" --config Release

# Next, you can run the tests to check the build
$ cmake -E chdir "build" ctest --build-config Release

# Install the library globally
& sudo cmake --build "build" --config Release --target install

If the above doesn't work, reference Google's Benchmark GitHub

Install Intel oneAPU Base Toolkit

# Download the installer
$ wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/4a5320d1-0b48-458d-9668-fd0e4501208c/intel-oneapi-base-toolkit-2025.1.3.7_offline.sh

# Run the install script
$ sudo sh ./intel-oneapi-base-toolkit-2025.1.3.7_offline.sh -a --silent --cli --eula accept

# Set the environment variables
$ . /opt/intel/oneapi/setvars.sh

If the above doesn't work, reference Intel oneAPI Base Toolkit's Official Website

Running the Benchmark

# Default setup where the max threads available on your system are used
make all

# This specifies how many threads should be used
make all THREADS=16

Results

The benchmark also checks that the result of the GEMM operation matches across all libraries and crates.

Single-thread

Test Case	Library	Language	Result
Input A	Eigen	C++	0.466s
	matrixmultiply	Rust	0.599s
	CBLAS	Rust	0.406s
Input B	Eigen	C++	0.495s
	matrixmultiply	Rust	0.607s
	CBLAS	Rust	0.409s
Input C	Eigen	C++	2.073s
	matrixmultiply	Rust	2.644s
	CBLAS	Rust	1.705s
Input D	Eigen	C++	40.823s
	matrixmultiply	Rust	42.953s
	CBLAS	Rust	29.670s

2 Threads

Test Case	Library	Language	Result
Input A	Eigen	C++	0.268s
	matrixmultiply	Rust	0.329s
	CBLAS	Rust	0.207s
Input B	Eigen	C++	0.281s
	matrixmultiply	Rust	0.343s
	CBLAS	Rust	0.211s
Input C	Eigen	C++	1.151s
	matrixmultiply	Rust	1.443s
	CBLAS	Rust	0.852s
Input D	Eigen	C++	28.048s
	matrixmultiply	Rust	25.668s
	CBLAS	Rust	15.485s

4 Threads (Maximum Threads Supported by matrixmultiply)

Test Case	Library	Language	Result
Input A	Eigen	C++	0.172s
	matrixmultiply	Rust	0.209s
	CBLAS	Rust	0.122s
Input B	Eigen	C++	0.173s
	matrixmultiply	Rust	0.191s
	CBLAS	Rust	0.124s
Input C	Eigen	C++	0.609s
	matrixmultiply	Rust	0.808s
	CBLAS	Rust	0.480s
Input D	Eigen	C++	14.786s
	matrixmultiply	Rust	15.655s
	CBLAS	Rust	9.329s

8 Threads

Test Case	Library	Language	Result
Input A	Eigen	C++	0.091s
	matrixmultiply	Rust	0.248s
	CBLAS	Rust	0.0695s
Input B	Eigen	C++	0.0959s
	matrixmultiply	Rust	0.253s
	CBLAS	Rust	0.0668s
Input C	Eigen	C++	0.401s
	matrixmultiply	Rust	0.976s
	CBLAS	Rust	0.258s
Input D	Eigen	C++	5.503s
	matrixmultiply	Rust	15.779s
	CBLAS	Rust	4.948s

16 Threads

Test Case	Library	Language	Result
Input A	Eigen	C++	0.0691s
	matrixmultiply	Rust	0.243s
	CBLAS	Rust	0.0498s
Input B	Eigen	C++	0.0767s
	matrixmultiply	Rust	0.246s
	CBLAS	Rust	0.0343s
Input C	Eigen	C++	0.259s
	matrixmultiply	Rust	1.024s
	CBLAS	Rust	0.165s
Input D	Eigen	C++	3.381s
	matrixmultiply	Rust	16.960s
	CBLAS	Rust	3.994s

32 Threads

Test Case	Library	Language	Result
Input A	Eigen	C++	0.0606s
	matrixmultiply	Rust	0.231s
	CBLAS	Rust	0.0462s
Input B	Eigen	C++	0.0725s
	matrixmultiply	Rust	0.240s
	CBLAS	Rust	0.0555s
Input C	Eigen	C++	0.235s
	matrixmultiply	Rust	0.966s
	CBLAS	Rust	0.187s
Input D	Eigen	C++	2.933s
	matrixmultiply	Rust	17.732s
	CBLAS	Rust	3.597s

48 Threads (Maximum Threads Supported by my testing machine - Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz)

Test Case	Library	Language	Result
Input A	Eigen	C++	0.0629s
	matrixmultiply	Rust	0.231s
	CBLAS	Rust	0.0307s
Input B	Eigen	C++	0.0744s
	matrixmultiply	Rust	0.234s
	CBLAS	Rust	0.0462s
Input C	Eigen	C++	0.161s
	matrixmultiply	Rust	1.009s
	CBLAS	Rust	0.121s
Input D	Eigen	C++	3.065s
	matrixmultiply	Rust	18.508s
	CBLAS	Rust	3.928s