| Philippos Mordohai Associate Professor Department of Computer Science Stevens Institute of Technology Office: North 209
|
CS 677: Parallel Programming for Many-core
Processors Spring 2020 |
Homepage |
Location Carnegie 315 Time Wednesday 6:30-9:00 PM. Office Hours Tuesday 5-6 and by appointment. Pre-requisites One of the following courses or demonstrated experience in C/C++.
Syllabus Textbook The required textbook is the following. I will also use notes outside the textbook, mostly in the second half of the semester. Programming Massively Parallel Processors: A Hands-on Approach by David Kirk and Wen-mei Hwu Morgan Kaufmann, 2016 (3rd edition) Evaluation The weights of the following instruments have changed when the course was moved online. See syllabus on Canvas. Homework assignments (40%) Homework assignments will be assigned almost every week up to Week 7 and will be due a week later. Quizzes (10%) Midterm (15%) The midterm is scheduled for Week 8. It will cover theoretical aspects of massively parallel programming to aid the implementation of the final projects. Project (35%) Each student will select a project, which has to be approved by me regarding relevance and feasibility. I will also provide suggestions for potential projects and pointers to relevant material. Students actively involved in research can select a project related to their research, but new work has to be done during the semester. Large projects can be performed by groups of two students. Each student will briefly present a proposal of his or her project, which will have to be approved by Week 8, in Week 9. Longer status updates will be given three weeks later and the final presentations will be given in the last week of classes. The written reports will be due on the date of the (non-existent) final exam.. Class Schedule Week 1: Introduction to massively parallel programming and CUDA (Kirk & Hwu Ch. 1, 2 and 3) Lecture 1 slides (pdf) Week 2: CUDA threads and atomics; CUDA memories (Kirk & Hwu Ch. 4 and 5) Lecture 2 slides (pdf) Week 3: Performance considerations (Kirk & Hwu Ch. 5) Lecture 3 slides (pdf) Week 4: More performance considerations; reduction trees; parallel patterns: prefix sum (Kirk & Hwu Ch. 5 and 8) Lecture 4 slides (pdf) Week 5: Project ideas; Convolution, constant memory and cache (Kirk & Hwu Ch. 7) Lecture 5 slides (pdf) Week 6: Homework 3 preliminaries; Case study: MRI reconstruction; Timers (Kirk & Hwu Ch. 14) Lecture 6 slides (pdf) Week 7: Case study: electrostatic potential calculation; Binning (Kirk & Hwu Ch. 15) Lecture 7 slides (pdf) Week 8: Computational thinking (Kirk & Hwu Ch. 17) Lecture 8 slides (pdf) Week 9: Pinned Memory; streams; Thurst (Kirk & Hwu Ch. 13 and notes) Lecture 9 slides (pdf) Week 10: (NEW) Sparse matrix and vector operations; Summed area tables; Parallel Sorting (Notes) Lecture 10 slides (pdf) Week 11: Project mid-point reports; Bitonic sorting; More libraries; OpenCL (Kirk & Hwu Ch. 14 and notes) Lecture 11 slides (pdf) Week 12: More OpenCL; OpenACC; OpenMP (Notes) Lecture 12 slides (pdf) Week 13: Deep learning; Latest GPU and CUDA features Lecture 13 slides (pdf) Week 14: Project presentations Week X: Midterm (cancelled) and Project Proposals (cancelled) Resources The most recent CUDA toolkit. The toolkit includes the NVIDIA CUDA Compiler, and other software necessary to develop CUDA applications. |