Millad Ghane

Millad Ghane
PhD
Department of Computer Science
University of Houston
Email: mghane2 AT uh.edu, mghane AT cs.uh.edu

For a copy of my resume and CV,
please contant me through my email or social media.

News

2020: Our paper in PMAM, in conjunction with PPoPP, San Diego, CA, got accepted.
2019: Our paper on arXiv: Assessing Performance Implications of Deep Copy Operations via Microbenchmarking
2019: Our pointerchain paper has been accepted in the ParCo Journal.
2019: Attending PMAM, in conjunction with PPoPP, Washington, DC.
2018: Attending SC 2018 as a student volunteer.
2017: Joining NVIDIA for the summer as a graduate intern (developing proof-of-concept of HiHAT).

Skills

Technical Knowledge: Parallel Computing, GPU Development, Embedded System (Jetson TX2), Numerical Analysis, Scientific Computing, Software Engineering, Performance-critical Systems, Computer Architecture
C/C++, Java, Python, and .Net Framework (C#), OpenCL, CUDA, OpenACC, OpenMP, Pthreads
Google Benchmark, Google Test
Distributed/Heterogeneous/Parallel Computing
SQL and NoSQL (e.g., Redis)
Hadoop and HBase

Research

I am conducting my research under supervision of Prof. Sunita Chandrasekaran and Prof. Margaret S. Cheung.

Research Projects

Gecko

A hierarchical model to represent the memory hierarchy in current architectures. Based on this model, a programming model with similar name is developed.

#pragma gecko

Gecko helps applications improve their

Portability
Usability
Maintainability

A sample source code with Gecko

Implementation languages:

C/C++, CUDA, OpenACC, OpenMP: 4,000+ LOC
Python: 1,000+ LOC

Please refer to [the github repo] and our PMAM'19 and PMAM'20 papers for more information.

The STREAM benchmark in Gecko


#pragma gecko config file("gecko.conf")
T *a, *b, *c;
#pragma gecko memory allocate(a[0:array_size]) type(T) location("LocH") 
#pragma gecko memory allocate(b[0:array_size]) type(T) location("LocH") 
#pragma gecko memory allocate(c[0:array_size]) type(T) location("LocH") 

#pragma gecko region gang vector exec_pol("runtime") variable_list(a,b,c)
for (int i = 0; i < array_size; i++) {
    a[i] = initA;
    b[i] = initB;
    c[i] = initC;
}
#pragma gecko region end
#pragma gecko region pause 

std::chrono::high_resolution_clock::time_point t1, t2;
// Run triad in loop
t1 = std::chrono::high_resolution_clock::now();
for (unsigned int k = 0; k < num_times; k++) {
    #pragma gecko region gang vector exec_pol("runtime") \\
                variable_list(a,b,c)
    for (int i = 0; i < array_size; i++) {
        a[i] = b[i] + scalar * c[i];
    }
    #pragma gecko region end
    #pragma gecko region pause 
}
t2 = std::chrono::high_resolution_clock::now();
double runtime = duration_cast>(t2 - t1).count();

#pragma gecko memory free(a,b,c)

HiHAT

A proprietary project at NVIDIA. HiHAT is an intermediate software stack to expose the underlying hardware to developers without any significant code modifications through a software library. My proof-of-concept code became basis for the production code.

Implementation languages:

C/C++, and CUDA: 1,000+ LOC

Deep copy benchmarking

Proposing a set of microbenchmarks to assess the effect of nested data structures on the data transfer in a heterogeneous platform.

Implementation languages:

C/C++
Python
Google Benchmark

Please refer to the [the github repo] and the arXiv'19 paper for more information.

Phoenix

A header file to measure power/energy consumption of source codes on Intel processors and NVIDIA GPU devices.

Implementation languages:

C/C++, and CUDA: 400+ LOC

Please refer to the [the github repo] and our arXiv'18 paper for more information.

OpenMP Tool Interfaces (OMPT) in OpenUH compiler

Implementation of OMPT in the OpenMP 3.0 runtime library of the OpenUH compiler. OMPT provides an API interface to measure the performance of OpenMP runtime library.

Implementation languages:

C/C++

Please refer to the [the github repo] and the IWOMP'15 paper for more information.

pointerchain

A directive-based approach to deal with a chain of pointers in heterogeneous platforms to tackle deep copy implications in heterogeneous platforms.

#pragma pointerchain

Implementation languages:

C/C++, OpenACC, Python: 400+ LOC

Please refer to the [the github repo] and ParCo journal paper for more information. For a complete assessment of pointerchain, please refer to our arXiv paper, Assessing Performance Implications of Deep Copy Operations via Microbenchmarking.

Publications

Towards A Portable Hierarchical View of Distributed Shared Memory Systems: Challenges and Solutions
Millad Ghane, Sunita Chandrasekaran, Margaret S. Cheung
The 11th International Workshop on Programming Models and Applications for Multicores and Manycores (PMAM 2020), in conjunction with PPoPP
Feb. 2020
San Diego, CA, US.
[pdf] [source code]
Assessing Performance Implications of Deep Copy Operations via Microbenchmarking
Millad Ghane, Sunita Chandrasekaran, Margaret S. Cheung
arXiv 2019
Jun. 2019
[pdf] [source code]
pointerchain: Tracing Pointers to Their Roots – A Case Study in Molecular Dynamics Simulations
Millad Ghane, Sunita Chandrasekaran, Margaret S. Cheung
Parallel Computing (ParCo) Journal
Apr. 2019
[Elsevier link] [pdf] [suppl.]
Gecko: Hierarchical Distributed View of Heterogeneous Shared Memory Architectures
Millad Ghane, Sunita Chandrasekaran, Margaret S. Cheung
The 10th International Workshop on Programming Models and Applications for Multicores and Manycores (PMAM 2019), in conjunction with PPoPP
Feb. 2019
Washington, DC, US.
[pdf] [pptx] [source code]
Power and Energy-efficiency Roofline Model for GPUs
Millad Ghane, Jeff Larkin, Larry Shi, Sunita Chandrasekaran, Margaret S. Cheung
arXiv 2018 preprint arXiv:1809.09206
Sep. 2018
[pdf]
Path forward for softwarization to tackle evolving hardware
Millad Ghane, Sunita Chandrasekaran, Robert Searles, Margaret Cheung, Oscar Hernandez
Proceeding SPIE, Disruptive Technologies in Information Sciences (SPIE 2018)
May 2018
Orlando, Florida, US.
[link] [pdf] [pdf]
False Sharing Detection in OpenMP Applications Using OMPT API
Millad Ghane, A. M. Malik, Barbara Chapman, and Ahmad Qawasmeh
11th International Workshop on OpenMP (IWOMP 2015)
October 2015
Aachen, Germany.
An Opto-electrical NoC with Traffic Flow Prediction in Chip Multiprocessors
Millad Ghane, Mohammad Arjomand, and Hamid Sarbazi-azad
22nd Euromicro International Conference on Parallel, Distributed and network-based Processing (PDP 2014)
February 2014
Turin, Italy.
Remaining Energy Based Routing Protocol for Wireless Sensor Networks
Millad Ghane and Amir Rajabzadeh
15th Symposium on Computer Architecture and Digital Systems (CADS 2010)
September 2010
Tehran, Iran.
Energy Efficient Antcolony-based Routing Protocol in Wireless Sensor Network
Millad Ghane and Amir Rajabzadeh
15th CSI Computer Conference (CSICC 2010)
February 2010
Tehran, Iran.

Reviewer

SC (Supercomputing)
ICPP
Cluster
SCA (ACM SIGGRAPH/Eurographics Symposium on Computer Animation)

Tips

The things that I regularly forget, so I put them in here for my convenience.