diff options
| -rw-r--r-- | README.md | 32 |
1 files changed, 17 insertions, 15 deletions
@@ -17,9 +17,9 @@ As a quick introduction, here's an excerpt from the `ejit` README: `ejit` is primarily inteded to be used as the main execution engine for experimental/toy scripting languages, so the **prospero** challenge is arguably -a bit outside the intended use case of relatively heavy computation. Still, -this was a good opportunity to see just how well/poorly the library can handle -such tasks. +a bit outside the intended use case of `ejit`, it not really being designed with +relatively heavy computation in mind. Still, this was a good opportunity to see +just how well/poorly the library can handle such tasks. ## Running @@ -30,10 +30,12 @@ make The produced binary has a few command line parameters: -> usage: prospero [-j threads] [-r resolution] [-c] -> -j how many threads to run -> -r length of one side -> -c enable JIT +``` +usage: prospero [-j threads] [-r resolution] [-c] + -j how many threads to run + -r length of one side + -c enable JIT +``` By default, the binary runs a single worker thread, produces an image of size 1024x1024 and does not enable the JIT part of `ejit`. @@ -44,10 +46,10 @@ Here are some runtime results for when the JIT compiler is enabled: | Params | Runtime for JIT (s) | |---------|---------------------| -| -j1 -c | 8.943 | -| -j2 -c | 4.357 | -| -j6 -c | 1.678 | -| -j12 -c | 1.617 | +|`-j1 -c`| 8.943 | +|`-j2 -c`| 4.357 | +|`-j6 -c`| 1.678 | +|`-j12 -c`| 1.617 | I have an AMD Ryzen 5 5600X with 6 cores and 12 threads. I was kind of surprised that `-j12` wasn't any faster than `-j6`, but if I had to guess as to why, @@ -63,10 +65,10 @@ JIT compiler is disabled: | Params | Runtime for bytecode (s) | |--------|--------------------------| -| -j1 | 35.311 | -| -j2 | 18.514 | -| -j6 | 6.155 | -| -j12 | 3.760 | +|`-j1` | 35.311 | +|`-j2` | 18.514 | +|`-j6` | 6.155 | +|`-j12` | 3.760 | In this case, I'm guessing that there's enough overhead during interpretation that adding threads beyond the number of physical cores still improves |