tt-umd 0.9.6

User Mode Driver for tenstorrent

TT-UMD

User Mode Driver

Quickstart

Software Dependencies

UMD requires Tenstorrent's kernel-mode driver

Required Ubuntu dependencies:

sudo apt install -y libhwloc-dev cmake ninja-build

UMD currently supports gcc-11 and newer gcc versions, and clang-13 and newer clang versions.

IOMMU and Hugepage requirements

To determine whether your system requires hugepage configuration, run the provided script:

./scripts/iommu_detect.sh

Wormhole and Blackhole

If your system IOMMU is enabled, no hugepage setup is required. If you don't have IOMMU enabled, than hugepages might be required for some of the driver functionality. 1G hugepages are required for shared device/host memory. Techniques for setup:

• Recommended: the tt-system-tools repository contains a .deb package which will configure your system
  • sudo dpkg -i tenstorrent-tools_1.1-5_all.deb
• Alternative: Metal project provides instructions and a script.
• For experts:
  • Put system IOMMU in passthrough mode or disable it
  • Allocate 1 or more 1G hugepages
  • Mount the hugetlbfs at /dev/hugepages-1G (e.g. mount -t hugetlbfs hugetlbfs /dev/hugepages-1G -o mode=777,pagesize=1024M)

Install and use UMD

Python bindings

You can just run the following command, and you'll have tt_umd python package available in your environment:

pip install git+https://github.com/tenstorrent/tt-umd.git

Or if you have UMD downloaded locally you can install from local source:

pip install .

Build flow for C++ lib

To build libtt-umd.so:

cmake -B build -G Ninja
cmake --build build

To build all components (some are turned off by default, like tests), you can run these commands:

cmake -B build -G Ninja -DTT_UMD_BUILD_ALL=ON
cmake --build build

To build with GCC, set these environment variables before invoking cmake:

export CC=gcc
export CXX=g++

Disabling -Werror

By default, all warnings are treated as errors. This is controlled via the standard CMake variable CMAKE_COMPILE_WARNING_AS_ERROR.

Systems with recent libc may cause the python package building to fail due python redefining some *_SOURCES macros to higher versions and GCC has no mean to disable the macro redefinition diagnostic.

# Plain CMake
cmake -B build -G Ninja -DCMAKE_COMPILE_WARNING_AS_ERROR=OFF

# Python build (scikit-build-core passes CMAKE_ARGS through to CMake)
CMAKE_ARGS="-DCMAKE_COMPILE_WARNING_AS_ERROR=OFF" pip install .

Build debian dev package

cmake --build build --target package

# Generates umd-dev-x.y.z-Linux.deb

Enabling Logging

UMD uses a two-level logging system with compile-time and runtime controls.

Compile-Time Logging Control

By default, log_debug and log_trace statements are compiled out of release builds for performance. To include them in the binary:

Option 1: Enable logging explicitly

cmake -B build -G Ninja -DTT_UMD_ENABLE_LOGGING=ON
cmake --build build

Option 2: Use Debug build type (enables logging automatically)

cmake -B build -G Ninja -DCMAKE_BUILD_TYPE=Debug
cmake --build build

Runtime Logging Control

At runtime, control the logging level using the TT_LOGGER_LEVEL environment variable:

export TT_LOGGER_LEVEL=debug  # Show debug and above
export TT_LOGGER_LEVEL=trace  # Show all log messages (most verbose)
export TT_LOGGER_LEVEL=info   # Default level

Available log levels (from most to least verbose):

• trace - Most detailed logging, traces program execution
• debug - Debugging information useful during development
• info - General informational messages (default)
• warn - Warning messages for potentially harmful situations
• error - Error messages for serious problems
• critical - Critical errors that may lead to program termination
• off - Disables all logging

Example: Running with debug logging

# Build with logging enabled
cmake -B build -G Ninja -DTT_UMD_ENABLE_LOGGING=ON
cmake --build build

# Run with debug level
TT_LOGGER_LEVEL=debug ./build/bin/your_program

Tracy Profiling

UMD supports Tracy profiling via the TT_UMD_ENABLE_TRACY build option. When disabled (the default), Tracy has zero footprint — no binary overhead, no runtime cost.

Building with Tracy

cmake -B build -G Ninja -DTT_UMD_ENABLE_TRACY=ON
cmake --build build

Capturing a trace

Launch the Tracy GUI locally, then on remote configure Port forwarding in VS Code for port 8086, start the application you want to profile on remote, then click Connect from GUI started locally. Alternatively, use tracy-capture -o trace.tracy to capture trace from the command line, then open the resulting file in the Tracy GUI.

Integration

UMD can be consumed by downstream projects in multiple ways.

From Source (Python)

You can use tt_umd module by installing it in your current python environment

From Source (CMake)

You can link libtt-umd.so by linking against the umd::tt-umd target.

Using CPM Package Manager

CPMAddPackage(
  NAME umd
  GITHUB_REPOSITORY tenstorrent/tt-umd
  GIT_TAG v0.1.0
  VERSION 0.1.0
)

As a submodule/external project

add_subdirectory(<path to umd>)

From Prebuilt Binaries

Ubuntu

apt install ./umd-dev-x.y.z-Linux.deb

Simulator Integration

You can run UMD tests without silicon by following setup instructions here.

Development workflow

For developing tt-umd, you can see the full set of dependencies in docker_install_common.sh

After that you can look at the section defined above Install and use UMD

Pre-commit Hook Integration for Formatting and Linting

As part of maintaining consistent code formatting across the project, we have integrated the pre-commit framework into our workflow. The pre-commit hooks will help automatically check and format code before commits are made, ensuring that we adhere to the project's coding standards.

What is Pre-commit?

Pre-commit is a framework for managing and maintaining multi-language pre-commit hooks. It helps catch common issues early by running a set of hooks before code is committed, automating tasks like:

• Formatting code (e.g., fixing trailing whitespace, enforcing end-of-file newlines)
• Running linters (e.g., clang-format, black, flake8)
• Checking for merge conflicts or other common issues.

For more details on pre-commit, you can visit the official documentation.

How to Set Up Pre-commit Locally

To set up pre-commit on your local machine, follow these steps:

1. Install Pre-commit: Ensure you have Python installed, then run:

   pip install pre-commit
   

2. Install the Git Hook Scripts: In your local repository, run the following command to install the pre-commit hooks:

   pre-commit install
   

   This command will configure your local Git to run the defined hooks automatically before each commit.
3. Run Pre-commit Hooks Manually: You can also run the hooks manually against all files at any time with:

   pre-commit run --all-files
   

Why You Should Use Pre-commit

By setting up pre-commit locally, you can help maintain the quality of the codebase and ensure that commits consistently meet the project's formatting standards. This saves time during code reviews and reduces the likelihood of code formatting issues slipping into the repository.

Since the hooks run automatically before each commit, you don't need to remember to manually format or check your code, making it easier to maintain consistency.

We strongly encourage all developers to integrate pre-commit into their workflow.

Formatting C++ code

Installing clang-format

If you're using an IRD docker, clang-format should be already available. If you don't have clang-format in your working environment, follow the instructions on llvm website for installing it.

Formatting files

If working with VSCode, you can copy the provided default settings:

cp .vscode/default.settings.json .vscode/settings.json

From now on, c++ files will be formatted on save (given that clang-format is available).

Note that if you setup pre-commit hook, the files will be automatically formatted when you commit changes. You can also manually auto format the whole repo using mentioned pre-commit:

   pre-commit run --all-files

Bumping the UMD version

There is an automated workflow for creating releases. It is triggered by merging a PR to main which changes the VERSION file.

You can change the VERSION as part of another PR or as an isolated PR. Please also update the CHANGELOG with the exact version you are changeing to.

Once the PR is merged, a draft Release will be created with the generated changelog and artifacts. Please review it and publish it using the tag which exactly matches the version of the release.