Skip to content

Updated README.md with instructions for profiling #155

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
hhkit wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Updated README.md with instructions for profiling #155

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a575c7a
updated readme with profile instructions
hhkit May 14, 2026
858e5ff
typo
hhkit May 14, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 30 additions & 2 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,10 +43,8 @@ source ./venv/bin/activate

mkdir build && cd build
cmake .. -GNinja \
# Optional but highly recommended \
-DCMAKE_C_COMPILER=clang \
-DCMAKE_CXX_COMPILER=clang++ \
# Optional for improved caching, requires ccache to be installed. \

@szerdick szerdick May 14, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think I would keep a note for this somewhere. Otherwise I am left wondering whether I need this and why it is commented out.

# -DCMAKE_C_COMPILER_LAUNCHER=ccache \
# -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
-DQUIDDITCH_TOOLCHAIN_FILE=../toolchain/ToolchainFile.cmake
Expand All @@ -67,6 +65,36 @@ independently.
> It should be possible to decouple the compilation infrastructure from the runtime, letting users compile the neural networks on their local machines, to then execute it on Linux.
> Everything to do with Verilator is only supported on Linux, so that part cannot be made platform-independent, but the `runtime/toolchain/Dockerfile` can be massaged to exclude everything Verilator-related, to build MLIR+IREE+Quidditch, and be run as a script to produce a compiler that emits everything necessary to just run it on a Linux machine.

## Executing and Profiling without Snitch hardware

*For this example, we will use the built executable located at `build/runtime/samples/nsnet/NsNet2`.*
*All commands are executed from the root `quidditch` directory*

The compilation process produces a Snitch executable, which can only be run on Snitch hardware.
The Snitch executable can be simulated on the host using the Snitch toolchain's Verilator simulator.

```shell
export TARGET_EXE=build/runtime/sampels/nsnet/NsNet2
Comment thread
hhkit marked this conversation as resolved.
Outdated
toolchain/bin/snitch_cluster.vlt $TARGET_EXE
```

@szerdick szerdick May 14, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here I am very confused about when to use test as specified below and when to use this (or what the difference is). Also, is this TARGET_EXE macro already set automatically in the script above or do I explicitly set it here?


The Verilator execution emits `dasm` traces in a `logs` folder (one will be created if its does not exist).
They will be labeled `trace_hart_0000000X.dasm` from `X` = 0 to 8.
`X` = 8 is the DMA core, whereas cores 0 to 7 are the compute cores.

Each `dasm` trace should then be converted to human-readable assembly, which can then be analyzed for metrics

```shell
cat logs/trace_hart_00000001.dasm | toolchain/bin/spike-dasm > logs/trace_hart_00000001.asm
python3 snitch_cluster/util/trace/gen_trace.py logs/trace_hart_00000001.asm -d hart_1_perf.json > metrics_hart_1.txt
```

A [Perfetto](http://ui.perfetto.dev) `events.json` trace can then be generated using the helper script, taking the hart_X_perf.json and metrics_hart_X.txt as input.

```shell
python3 runtime/util/snitch_trace_to_perfetto.py -i hart_1_perf.json --trace logs/trace_hart_00000001.dasm --elf $TARGET_EXE --addr2line toolchain/bin/llvm-addr2line -o events.json
```

## Compiling Neural Networks

Any information you can find on how to use `iree-compile` is applicable, with the only major difference bing the `--iree-hal-target-backends=llvm-cpu`, which also supports our `quidditch` value for the flag, which leverages our flow.
Expand Down