Merge branch 'pulp-platform:devel' into devel

Author: lee2716

.github/workflows/publish.yml

@@ -0,0 +1,65 @@
+# SPDX-FileCopyrightText: 2025 ETH Zurich and University of Bologna
+#
+# SPDX-License-Identifier: Apache-2.0
+
+name: Publish
+
+on:
+  push:
+    tags:
+      - "v*"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  id-token: write
+
+env:
+  UV_EXTRA_INDEX_URL: https://pypi.ngc.nvidia.com
+
+jobs:
+  publish-pypi:
+    name: Publish to PyPI
+    if: github.event_name == 'push'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v5
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
+
+      - name: Build artifacts
+        run: uv build
+
+      - name: Test wheel installation
+        run: uv run --isolated --no-project --with dist/*.whl python -c "import Deeploy"
+
+      - name: Test sdist installation
+        run: uv run --isolated --no-project --with dist/*.tar.gz python -c "import Deeploy"
+
+      - name: Publish to PyPI
+        run: uv publish
+
+  publish-test-pypi:
+    name: Publish to Test PyPI
+    if: github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v5
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
+
+      - name: Build artifacts
+        run: uv build
+
+      - name: Test wheel installation
+        run: uv run --isolated --no-project --with dist/*.whl python -c "import Deeploy"
+
+      - name: Test sdist installation
+        run: uv run --isolated --no-project --with dist/*.tar.gz python -c "import Deeploy"
+
+      - name: Publish to Test PyPI
+        run: uv publish --publish-url https://test.pypi.org/legacy/

CHANGELOG.md

@@ -4,6 +4,7 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 ## Unreleased (Planned Release Target: v0.2.1)
 
 ### List of Pull Requests
+- PyPi Package Deployment + Remove Banshee Dept [#154](https://github.com/pulp-platform/Deeploy/pull/154)
 - PyTest Migration [#144](https://github.com/pulp-platform/Deeploy/pull/144)
 - Update submodule `pulp-nn-mixed` [#145](https://github.com/pulp-platform/Deeploy/pull/145)
 - Improve Profiling [#138](https://github.com/pulp-platform/Deeploy/pull/138)
@@ -31,6 +32,10 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 - Fix bias hoisting in generic GEMM with no bias [#126](https://github.com/pulp-platform/Deeploy/pull/126)
 
 ### Added
+- The `publish.yml` action to build a branch and push it to PyPi. The action is automatically triggered when a tag with the "v*" format is emitted. 
+- I created a release of [Banshee](https://github.com/pulp-platform/banshee/releases/tag/v0.5.0-prebuilt) so we don't need to rebuild it over and over. The `Makefile` now pulls that release depending on the platform.
+- I bumped the onnx-graphsurgeon version such that we don't need to use NVIDIA's PyPi index anymore.
+- `_export_graph` assigns their export type to the tensors before export.
 - `pytest` and `pytest-xdist` as dependencies of Deeploy.
 - A `pytest.ini` for the global configuration of PyTest for the project.
 - `conftest.py` to define CLI args for PyTest for the whole project, it also defines a set of global fixtures and markers.
@@ -93,6 +98,7 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 - Calculate non-kernel overhead and show total time spent during profiling
 
 ### Changed
+- Rename package name from `PULP-Deeploy` to `deeploy-pulp`.
 - Each CI workflow has been simplified to call the pytest suite with certain markers.
 - Structure of Tests subdir for improved ordering
 - Structure of .gitignore file for improved ordering
@@ -137,6 +143,8 @@ This file contains the changelog for the Deeploy project. The changelog is divid
 - Print kernel profiling information for all memory levels
 
 ### Fixed
+- Update `install.md` to remove rust mention and fix test command.
+- Update `README.md` to remove reference to NVIDIA's PyPi index.
 - `nvidia-pyindex` was broken as it now tries to build the wheel to respect the new policy on packages using `pyproject`. Instead of installing this package, we just add the `https://pypi.ngc.nvidia.com` channel to the pip config file.
 - Pin versions of broken dependencies of Banshee.
 - Fixed ReduceMean parallelization and tiling issues described in Issue [#134](https://github.com/pulp-platform/Deeploy/issues/134).

Container/Dockerfile.deeploy

@@ -8,13 +8,13 @@ FROM ${BASE_IMAGE} AS toolchain
 
 # Intermediate Stage
 ARG DEBIAN_FRONTEND=noninteractive
-ARG BENDER_VERSION=0.28.1
+ARG BENDER_VERSION=0.29.1
 ARG UBUNTU_VERSION=22.04
 ARG TARGETPLATFORM
 
 ENV DEBIAN_FRONTEND=noninteractive
 ENV TZ=Etc/UTC
-ENV PATH="/root/.cargo/bin:/app/bender:${PATH}"
+ENV PATH="/app/install/bender:${PATH}"
 ENV LLVM_INSTALL_DIR=/app/install/llvm
 
 WORKDIR /app
@@ -42,14 +42,13 @@ RUN --mount=type=cache,target=/ccache \
     ccache -s
 
 # Install Bender
-RUN if [ "$TARGETPLATFORM" = "linux/amd64" ]; then \
+RUN mkdir -p /app/install/bender && \
+if [ "$TARGETPLATFORM" = "linux/amd64" ]; then \
     wget https://github.com/pulp-platform/bender/releases/download/v${BENDER_VERSION}/bender-${BENDER_VERSION}-x86_64-linux-gnu-ubuntu${UBUNTU_VERSION}.tar.gz && \
-    tar xzf bender-${BENDER_VERSION}-x86_64-linux-gnu-ubuntu${UBUNTU_VERSION}.tar.gz && \
-    cp  /app/bender /bin; \
+    tar xzf bender-${BENDER_VERSION}-x86_64-linux-gnu-ubuntu${UBUNTU_VERSION}.tar.gz -C /app/install/bender; \
 elif [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
-    wget https://github.com/Xeratec/bender/releases/download/v0.28.3-rc1/bender-0.28.3-rc1-arm64-linux-gnu-ubuntu${UBUNTU_VERSION}.tar.gz && \
-    tar xzf bender-0.28.3-rc1-arm64-linux-gnu-ubuntu${UBUNTU_VERSION}.tar.gz && \
-    cp  /app/bender /bin; \
+    wget https://github.com/pulp-platform/bender/releases/download/v${BENDER_VERSION}/bender-${BENDER_VERSION}-arm64-linux-gnu-ubuntu${UBUNTU_VERSION}.tar.gz && \
+    tar xzf bender-${BENDER_VERSION}-arm64-linux-gnu-ubuntu${UBUNTU_VERSION}.tar.gz -C /app/install/bender; \
 fi
 
 # Compile Snitch Runtime
@@ -77,10 +76,12 @@ ENV SNITCH_HOME=/app/install/snitch_cluster
 ENV CHIMERA_SDK_HOME=/app/install/chimera-sdk
 ENV LLVM_INSTALL_DIR=/app/install/llvm
 ENV GVSOC_INSTALL_DIR=/app/install/gvsoc
+ENV BANSHEE_INSTALL_DIR=/app/install/banshee
 ENV SOFTHIER_INSTALL_DIR=/app/install/softhier
 ENV MINIMALLOC_INSTALL_DIR=/app/install/minimalloc
 ENV MEMPOOL_HOME=/app/install/mempool
-ENV PATH=/root/.cargo/bin:/app/install/qemu/bin:/app/install/banshee:$PATH
+ENV BENDER_INSTALL_DIR=/app/install/bender
+ENV PATH=/app/install/qemu/bin:/app/install/banshee:/app/install/bender:$PATH 
 
 WORKDIR /app
 
@@ -90,8 +91,7 @@ COPY pyproject.toml ./
 RUN mkdir -p /etc && printf "[global]\nextra-index-url = https://pypi.ngc.nvidia.com\n" > /etc/pip.conf
 
 # Install dependencies
-RUN mkdir -p /root/.cargo/bin/ && \
-    apt-get update && \
+RUN apt-get update && \
     apt-get install -y git-lfs \
     wget \
     ccache \
@@ -128,4 +128,3 @@ RUN pip install -r requirements-dev.txt -r core-requirements.txt -r gapy-require
 
 # Copy pre-built toolchains and emulators
 COPY --from=toolchain /app/install ./install
-COPY --from=toolchain /root/.cargo/bin/banshee /root/.cargo/bin/banshee

Container/Dockerfile.toolchain

@@ -13,7 +13,6 @@ ENV TZ=Etc/UTC
 ENV CC="ccache gcc"
 ENV CXX="ccache g++"
 ENV CCACHE_DIR=/ccache
-ENV PATH="/root/.cargo/bin:${PATH}"
 
 # Change the working directory
 WORKDIR /app
@@ -68,14 +67,6 @@ RUN curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py && \
     python get-pip.py && \
     rm get-pip.py
 
-# Build Rust tools
-RUN apt remove cargo -y && \
-    apt autoremove -y && \
-    curl https://sh.rustup.rs -sSf | bash -s -- -y && \
-    rustup install 1.63.0 && \
-    rustup default 1.63.0 && \
-    rustup component add rust-src
-
 # Install meson
 RUN pip install meson
 

Deeploy/DeeployTypes.py

@@ -3104,6 +3104,12 @@ def _exportGraph(self, folderPath, fileName):
         if not os.path.isabs(absoluteOnnxPath) or not os.path.isabs(absoluteDataPath):
             raise OSError(f"Error exporting the context to: {absoluteOnnxPath}")
 
+        # VJUNG: ONNX-Graphsurgeon needs tensors to be in their export types
+        constTensors = [tensor for tensor in self.graph.tensors().values() if isinstance(tensor, gs.Constant)]
+        for tensor in constTensors:
+            if tensor.dtype != tensor.export_dtype:
+                tensor.values = tensor.values.astype(tensor.export_dtype)
+
         model = gs.export_onnx(self.graph)
 
         # Annotate additional information in doc_string of tensors

Makefile

@@ -42,7 +42,6 @@ CMAKE ?= cmake
 LLVM_COMMIT_HASH ?= 1ccb97ef1789b8c574e3fcab0de674e11b189b96
 PICOLIBC_COMMIT_HASH ?= 31ff1b3601b379e4cab63837f253f59729ce1fef
 PULP_SDK_COMMIT_HASH ?= 7f4f22516157a1b7c55bcbbc72ca81326180b3b4
-BANSHEE_COMMIT_HASH ?= 0e105921e77796e83d01c2aa4f4cadfa2005b4d9
 MEMPOOL_COMMIT_HASH ?= affd45d94e05e375a6966af6a762deeb182a7bd6
 SNITCH_COMMIT_HASH ?= e02cc9e3f24b92d4607455d5345caba3eb6273b2
 SOFTHIER_COMMIT_HASH ?= 0       # bowwang: to be updated
@@ -53,7 +52,22 @@ XTL_VERSION ?= 0.7.5
 XSIMD_VERSION ?= 13.2.0
 XTENSOR_VERSION ?= 0.25.0
 
-RUSTUP_CARGO ?= $$(rustup which cargo)
+OS  := $(shell uname -s)
+ARCH:= $(shell uname -m)
+
+ifeq ($(OS),Linux)
+	ifeq ($(ARCH),x86_64)
+		TARGET := x86_64-unknown-linux-gnu
+	else ifeq ($(ARCH),aarch64)
+		TARGET := aarch64-unknown-linux-gnu
+	else
+		$(error unsupported Linux architecture $(ARCH))
+	endif
+else ifeq ($(OS),Darwin)
+	TARGET := aarch64-apple-darwin
+else
+	$(error unsupported platform $(OS))
+endif
 
 all: toolchain emulators docs echo-bash
 
@@ -503,11 +517,11 @@ ${TOOLCHAIN_DIR}/banshee:
 	git submodule update --init --recursive && \
 	git apply ${TOOLCHAIN_DIR}/banshee.patch
 
-${BANSHEE_INSTALL_DIR}: ${TOOLCHAIN_DIR}/banshee
+${BANSHEE_INSTALL_DIR}:
 	export LLVM_SYS_150_PREFIX=${LLVM_INSTALL_DIR} && \
-	cd ${TOOLCHAIN_DIR}/banshee/ && \
-	${RUSTUP_CARGO} clean && \
-	${RUSTUP_CARGO} install --path . -f
+	mkdir -p ${BANSHEE_INSTALL_DIR} && cd ${BANSHEE_INSTALL_DIR} && \
+	curl -LO https://github.com/pulp-platform/banshee/releases/download/v0.5.0-prebuilt/banshee-0.5.0-$(TARGET).tar.gz && \
+	tar -xzf banshee-0.5.0-$(TARGET).tar.gz --strip-components=1 -C .
 
 banshee: ${BANSHEE_INSTALL_DIR}
 

README.md

@@ -5,7 +5,7 @@
 # Deeploy: DNN Compiler for Heterogeneous SoCs
 </div>
 
-[![Documentation Status](https://img.shields.io/github/deployments/pulp-platform/Deeploy/github-pages?logo=readthedocs&logoColor=white&label=Docs
+![PyPI](https://img.shields.io/pypi/v/deeploy-pulp)[![Documentation Status](https://img.shields.io/github/deployments/pulp-platform/Deeploy/github-pages?logo=readthedocs&logoColor=white&label=Docs
 )](https://pulp-platform.github.io/Deeploy/)
 [![CI](https://github.com/pulp-platform/Deeploy/actions/workflows/ci-deeploy.yml/badge.svg)](https://github.com/pulp-platform/Deeploy/actions/workflows/ci-deeploy.yml)
 [![Deeploy Docker](https://github.com/pulp-platform/Deeploy/actions/workflows/docker-build-deeploy.yml/badge.svg)](https://github.com/pulp-platform/Deeploy/actions/workflows/docker-build-deeploy.yml)
@@ -35,7 +35,7 @@ git submodule update --init --recursive
 
 Installing Deeploy is as simple as running:
 ```
-pip install -e . --extra-index-url=https://pypi.ngc.nvidia.com
+pip install -e .
 ```
 However, to run the code generated by Deeploy on a certain target, you need the toolchains and the simulators associated with this platform.
 
@@ -50,7 +50,7 @@ docker run -it --name deeploy_main -v $(pwd):/app/Deeploy ghcr.io/pulp-platform/
 Install Deeploy inside the container in editable mode:
 ```
 cd Deeploy
-pip install -e . --extra-index-url=https://pypi.ngc.nvidia.com
+pip install -e .
 ```
 Congratulations, you installed Deeploy and its dependencies! Now, to test your installation let's run one simple test on each platform with the following commands:
 ```

cmake/simulation.cmake

@@ -42,9 +42,13 @@ macro(print_simulation_config)
 endmacro()
 
 macro(add_banshee_simulation name)
+	if(NOT DEFINED ENV{BANSHEE_INSTALL_DIR})
+		message(FATAL_ERROR "Environment variable BANSHEE_INSTALL_DIR not set")
+	endif()
+	set(BANSHEE_EXECUTABLE "$ENV{BANSHEE_INSTALL_DIR}/banshee")
     add_custom_target(banshee_${name}
 	DEPENDS ${name}
-	COMMAND RUST_MIN_STACK=${banshee_stack_size} banshee
+	COMMAND RUST_MIN_STACK=${banshee_stack_size} ${BANSHEE_EXECUTABLE}
 	--num-cores=${num_threads}
 	--num-clusters=1
 	--latency

docs/index.rst

@@ -24,3 +24,4 @@ Deeploy is developed as part of the PULP project, a joint effort between ETH Zur
    tutorials/overview
    structure
    apidocs
+   releasing

docs/install.md

@@ -47,14 +47,6 @@ pip install -e .
 
 ## Testing Framework Installation
 
-Please make sure to use a Rust version that is compatible with LLVM 15, like 1.63.0:
-
-```
-sudo snap install rustup --classic
-rustup install 1.63.0
-rustup default 1.63.0
-```
-
 The Makefile expects the environemt variable `CMAKE` to be defined. In case you have no strong preferences, you may run
 
 ```
@@ -84,7 +76,5 @@ For example, you can run
 
 ```
 cd DeeployTest
-python testRunner_generic.py -t Tests/Others/SimpleRegression
+python testRunner_generic.py -t ./Tests/Kernels/Integer/Add/Regular
 ```
-
-to run the `simpleRegression` test on your workstation. Various other tests are available and compatibility between tests and platforms is tested in the `.gitlab-ci.yml` file.