Final Project #542
Final Project #542
Conversation
sampsyo
left a comment
sampsyo
left a comment
There was a problem hiding this comment.
It's too bad that the original plan did not work out! This was, however, a clear explanation of the small change that you did manage to apply. There are a few places where some additional detail would be very useful.
| +++ | ||
| title = "Welcome to CS 6120!" | ||
| [extra] | ||
| bio = """ | ||
| Grace Hopper made the first compiler. [Adrian Sampson](https://www.cs.cornell.edu/~asampson/) is an associate professor of computer science, so that's pretty cool too I guess. | ||
| """ | ||
| [[extra.authors]] | ||
| name = "Adrian Sampson" | ||
| link = "https://www.cs.cornell.edu/~asampson/" # Links are optional. | ||
| [[extra.authors]] | ||
| name = "Grace Hopper" | ||
| +++ |
There was a problem hiding this comment.
Please include your own title, author name, and bio.
| name = "Grace Hopper" | ||
| +++ | ||
|
|
||
| My project was based on Dr.Jit codebase. Here is the [paper](https://dl.acm.org/doi/10.1145/3528223.3530099) that describes the compiler. In short, Dr.Jit traces the program to compute an AST, performs some optimizations on this representation, then manually assembles either LLVM IR or PTX code depending on the used backend, and finally compiles it into a kernel. |
There was a problem hiding this comment.
What does "manually" mean in this context?
| <img src="./2025-05-13-Final_Project/graph_old.png" alt="drawing" height="300"/> | ||
| <img src="./2025-05-13-Final_Project/graph_new.png" alt="drawing" height="300"/> | ||
|
|
||
| To better test the optimization and evaluate the performance, I planned to render the three scenes shown in Fig. 6 of the Dr.Jit paper. However, I noticed that during rendering, the optimization was never invoked. To address this, I modified the renderer code to make it less efficient, ensuring that there will be nodes to which the optimization can be applied. I rendered all the scenes with and without my optimization five times to get average times. The results are reported below. |
There was a problem hiding this comment.
Can you say a tiny bit more about that change you applied? Did it involve manually going the "opposite direction" from your transformation (i.e., replacing a cos expression with a sin expression?
|
|
||
| To better test the optimization and evaluate the performance, I planned to render the three scenes shown in Fig. 6 of the Dr.Jit paper. However, I noticed that during rendering, the optimization was never invoked. To address this, I modified the renderer code to make it less efficient, ensuring that there will be nodes to which the optimization can be applied. I rendered all the scenes with and without my optimization five times to get average times. The results are reported below. | ||
|
|
||
| <img src="./2025-05-13-Final_Project/evaluation.png" alt="drawing" width="300"/> |
There was a problem hiding this comment.
Before showing your results, can you briefly say something about your experimental setup (hardware/OS, Dr.Jit version, etc.) and your approach to measurement (how did you measure the execution time, how many replicas did you use, etc.)?
|
|
||
| Well, my optimization did not improve the performance, but at least I know that the trace modification was correct since produced images were identical. | ||
|
|
||
| The second part of the project was much less straightforward, and I was not able to figure it out. As described in the project proposal, the idea was to trace functions that lack hardware support and cannot be represented by a single node into a separate trace, and then redirect the main AST to that newly created trace. I was unable to find a way to achieve this using the existing tools in the codebase. Implementing this optimization would require introducing a new type of node (e.g., `call`) and writing the corresponding PTX or LLVM IR code to support it. |
There was a problem hiding this comment.
As described in the project proposal
Maybe it would be a good idea to link to the proposal, so people can read it if they want.
|
Looks good; I'll publish this now! |
2 participants
Closes #514