The latest set of lectures are here (which may be this
page). Slides, examples, and most other textual material is freely
available. For privacy purposes Zoom lecture recordings are passcode
protected. Zoom lecture pass codes will be E-mailed to registered
students. Do not share the pass codes with those not taking the class.
(This is a University rule. Lectures cannot be shared with others
unless there is unanimous written consent.) The set of links below
will not be complete until the end of the semester. For a preview of
what is to come see a previous semester's lectures.
Midterm Exam Review and Solution Covered 24, 26 March, 5, 7 April 2021
[Slides, Normal (107 kB PDF)]
Zoom Recording of 12 March 2021 LectureEnd of RISC comparisons. Demo of pi program with and without optimization. Disappointing attempt to predict optimized execution time based on optimized dynamic instruction count. Explanation of why optimization reduced instruction count by a greater percentage than execution time.
Set 2: Benchmarks Covered 19, 29, 31 March 2021
[Slides, Normal (139 kB PDF)]
Zoom Recording of 31 Mar 2021 LectureSPECcpu benchmarks: The config file. Why it's important for the user to compile the benchmarks, how users might compile in such a way to exaggerate the results, and how the SPEC rules deal with such possible exaggeration.
Set 8: Interrupts, Exceptions, and Traps Covered 7,9 April 2021
[Slides, Normal (245 kB PDF)]
Set 9: Long Latency Operations (floating point, etc) Covered 9, 10, 12 April 2021
[Slides, Normal (295 kB PDF)]
Zoom Recording of 10 Apr 2021 LectureFully pipelined, unpipelined, and partially pipelined FP functional units. Definition of initiation interval and operation latency. How our baseline FP illustration could be modified to implement the mfc1 instruction.
Zoom Recording of 12 Apr 2021 LecturePossible ways of implementing FP instructions that raise precise exceptions. Sample problem in which FP hardware has a comparison unit and FCC (condition code) register, from 2019 Final Exam Problem 2.
Set 11: Multiple Issue (VLIW and Superscalar) Covered 14, 16 April 2021
[Slides, Normal (276 kB PDF)]
Zoom Recording of 14 Apr 2021 LectureWhat to do when the best 5-stage (scalar, statically scheduled) implementation isn't fast enough: deep pipelining, superscalar, SIMD insn, dynamic scheduling, branch prediction, parallel implementations. Each is briefly described. Then, start of detailed coverage of superscalar implementations.
Zoom Recording of 16 Apr 2021 LectureSuperscalar implementations of ordinary ISAs and their costs, VLIW ISAs and their unfulfilled promise to dramatically reduce superscalar implementation cost, and vector instructions.
Set 12: CTI (Branch, etc) Prediction (Under construction) Covered 19, 21 April 2021
[Slides, Normal (381 kB PDF)]
Zoom Recording of 26 Apr 2021 LectureSolution to Homework 8 (Branch predictor problems.) Direct mapped, set-associative caches. Miscellaneous cache topics. A bit about virtual memory and its applications.