From a5ff7ee0578ba1ede040d7ec623b9306d115209e Mon Sep 17 00:00:00 2001 From: Huanyi Chen Date: Wed, 31 Jan 2024 11:39:17 -0500 Subject: [PATCH] Prepare L12 & L13 flipped note --- lectures/flipped/L12.md | 8 ++++++++ lectures/flipped/L13.md | 12 +++++------- 2 files changed, 13 insertions(+), 7 deletions(-) diff --git a/lectures/flipped/L12.md b/lectures/flipped/L12.md index a0773ab..dbca9a7 100644 --- a/lectures/flipped/L12.md +++ b/lectures/flipped/L12.md @@ -1,9 +1,17 @@ # Lecture 12 — Dependencies and Speculation +## Roadmap + +We will talk about how to break dependencies to make some computations which +appear to be "inherently sequential" parallelizable. + ## Dependencies Talk about loop-carried and memory-carried dependencies. +Maybe talk about the initial implementation of lab 2, which is an example of the +loop-carried dependency. + ## Breaking Dependencies with Speculation ### Speculative Execution for Threads diff --git a/lectures/flipped/L13.md b/lectures/flipped/L13.md index f0421dc..6c2d089 100644 --- a/lectures/flipped/L13.md +++ b/lectures/flipped/L13.md @@ -1,16 +1,14 @@ # Lecture 13 — Early Termination, Reduced-Resource Computation -## Two basic ideas +## Roadmap -- skip some parts of work -- intentionally reduce accuracy to speed things up - -Activity: think of some examples? +We will talk about two ideas: 1) to skip some parts of work; and 2) +intentionally reduce accuracy to speed things up. ## N-body problem -Simple ideas are computing the forces in parallel and using `float` instead of -`double`. But what if we want more? +Simple ideas: computing the forces in parallel, using `float` instead of +`double`, etc. ### live-coding