Merge branch 'main' into add_tuner

.github/workflows/ci_linux_x64-libshortfin.yml

@@ -1,4 +1,3 @@
-#!/bin/bash
 # Copyright 2024 Advanced Micro Devices, Inc
 #
 # Licensed under the Apache License v2.0 with LLVM Exceptions.
@@ -22,7 +21,7 @@ permissions:
 
 env:
   IREE_REPO_DIR: ${{ github.workspace }}/iree
-  BUILD_DIR: ${{ github.workspace }}/libshortfin/build
+  LIBSHORTFIN_DIR: ${{ github.workspace }}/libshortfin/
 
 jobs:
   build-and-test:
@@ -47,16 +46,15 @@ jobs:
         repository: iree-org/iree
         path: ${{ env.IREE_REPO_DIR }}
         submodules: false
-        depth: 1
 
     - name: Initalize IREE submodules
       run : |
         cd ${{ env.IREE_REPO_DIR }}
-        git submodule update --init -- third_party/benchmark
-        git submodule update --init -- third_party/cpuinfo/
-        git submodule update --init -- third_party/flatcc
-        git submodule update --init -- third_party/googletest
-        git submodule update --init -- third_party/hip-build-deps/
+        git submodule update --init --depth 1 -- third_party/benchmark
+        git submodule update --init --depth 1 -- third_party/cpuinfo/
+        git submodule update --init --depth 1 -- third_party/flatcc
+        git submodule update --init --depth 1 -- third_party/googletest
+        git submodule update --init --depth 1 -- third_party/hip-build-deps/
 
     - name: Build IREE runtime
       run: |
@@ -80,16 +78,18 @@ jobs:
     - name: Setup Python
       uses: actions/setup-python@39cd14951b08e74b54015e9e001cdefcf80e669f # v5.1.1
       with:
-        python-version: "3.11"
+        python-version: "3.12"
         cache: "pip"
     - name: Install Python packages
       # TODO: Switch to `pip install -r requirements.txt -e libshortfin/`.
-      run: pip install nanobind typing_extensions
+      run: |
+        pip install nanobind
+        pip install -r ${{ env.LIBSHORTFIN_DIR }}/requirements-tests.txt
 
-    - name: Build libshortfin
+    - name: Build libshortfin (full)
       run: |
-        mkdir ${{ env.BUILD_DIR }}
-        cd ${{ env.BUILD_DIR }}
+        mkdir ${{ env.LIBSHORTFIN_DIR }}/build
+        cd ${{ env.LIBSHORTFIN_DIR }}/build
         cmake -GNinja \
           -DCMAKE_C_COMPILER=clang-18 \
           -DCMAKE_CXX_COMPILER=clang++-18 \
@@ -98,8 +98,25 @@ jobs:
           -DSHORTFIN_BUILD_PYTHON_BINDINGS=ON \
           ..
         cmake --build . --target all
+        pip install -v -e .
+
+    - name: Test libshortfin (full)
+      run: |
+        cd ${{ env.LIBSHORTFIN_DIR }}/build
+        ctest --timeout 30 --output-on-failure
+        cd ${{ env.LIBSHORTFIN_DIR }}
+        pytest -s -v -m "not requires_amd_gpu"
 
-    - name: Test libshortfin
+    - name: Build libshortfin (host-only)
       run: |
-        cd ${{ env.BUILD_DIR }}
-        cmake --build . --target test
+        mkdir ${{ env.LIBSHORTFIN_DIR }}/build-host-only
+        cd ${{ env.LIBSHORTFIN_DIR }}/build-host-only
+        cmake -GNinja \
+          -DCMAKE_C_COMPILER=clang-18 \
+          -DCMAKE_CXX_COMPILER=clang++-18 \
+          -DCMAKE_LINKER_TYPE=LLD \
+          -DCMAKE_PREFIX_PATH=${{ env.IREE_REPO_DIR }}/build/lib/cmake/IREE \
+          -DSHORTFIN_BUILD_PYTHON_BINDINGS=ON \
+          -DSHORTFIN_HAVE_AMDGPU=OFF \
+          ..
+        cmake --build . --target all

.github/workflows/ci_linux_x64_asan-libshortfin.yml

@@ -0,0 +1,170 @@
+# Copyright 2024 Advanced Micro Devices, Inc
+#
+# Licensed under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+name: CI - libshortfin - ASan
+
+on:
+  workflow_dispatch:
+  pull_request:
+  push:
+    branches:
+      - main
+    paths:
+      - '.github/workflows/ci_linux_x64_asan-libshortfin.yml'
+      - 'libshortfin/**'
+
+permissions:
+  contents: read
+
+env:
+  PYENV_ROOT: ${{ github.workspace }}/pyenv
+  PYENV_REF: 9ecd803bffaffb949fbdd8c70cb086227f6a3202 # v2.4.10
+  PYTHON_VER: 3.12.3
+  CACHE_ASAN_VER: 1
+  CACHE_DEPS_VER: 1
+  IREE_SOURCE_DIR: ${{ github.workspace }}/iree
+  LIBSHORTFIN_DIR: ${{ github.workspace }}/libshortfin/
+
+
+jobs:
+  setup-python-asan:
+    name: Setup Python ASan
+    runs-on: ubuntu-24.04
+
+    steps:
+    - name: Cache Python ASan
+      id: cache-python-asan
+      uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+      with:
+        path: ${{ env.PYENV_ROOT }}
+        key: ${{ runner.os }}-python-asan-${{ env.PYENV_REF }}-${{ env.PYTHON_VER }}-v${{ env.CACHE_ASAN_VER }}
+        lookup-only: 'true'
+
+    - name: Install dependencies
+      if: steps.cache-python-asan.outputs.cache-hit != 'true'
+      run: |
+        sudo apt update
+        sudo apt install clang lld cmake ninja-build
+        sudo apt install build-essential libssl-dev zlib1g-dev libbz2-dev libreadline-dev libsqlite3-dev curl git libncursesw5-dev xz-utils tk-dev libxml2-dev libxmlsec1-dev libffi-dev liblzma-dev
+
+    - name: Checkout pyenv
+      if: steps.cache-python-asan.outputs.cache-hit != 'true'
+      uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
+      with:
+        repository: pyenv/pyenv
+        ref: ${{ env.PYENV_REF }}
+        path: ${{ env.PYENV_ROOT }}
+
+    - name: Install pyenv & Python
+      if: steps.cache-python-asan.outputs.cache-hit != 'true'
+      run: |
+        cd ${{ env.PYENV_ROOT }}
+        src/configure && make -C src
+        export PATH=${{ env.PYENV_ROOT }}/bin:$PATH && eval "$(pyenv init -)"
+        CC=clang-18 CXX=clang++-18 LDFLAGS="-lstdc++" PYTHON_CONFIGURE_OPTS="--with-address-sanitizer" pyenv install -v -g ${{ env.PYTHON_VER }}
+        pyenv global ${{ env.PYTHON_VER }}-debug
+
+
+  build-and-test:
+    name: Build and test libshortfin
+    needs: [setup-python-asan]
+    runs-on: ubuntu-24.04
+
+    steps:
+    - name: Install dependencies
+      run: |
+        sudo apt update
+        sudo apt install clang lld cmake ninja-build
+
+    - name: Checkout repository
+      uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
+      with:
+        submodules: false
+
+    - name: Checkout IREE repo
+      uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
+      with:
+        repository: iree-org/iree
+        path: ${{ env.IREE_SOURCE_DIR }}
+        submodules: false
+
+    - name: Initalize IREE submodules
+      run : |
+        cd ${{ env.IREE_SOURCE_DIR }}
+        git submodule update --init --depth 1 -- third_party/benchmark
+        git submodule update --init --depth 1 -- third_party/cpuinfo/
+        git submodule update --init --depth 1 -- third_party/flatcc
+        git submodule update --init --depth 1 -- third_party/googletest
+        git submodule update --init --depth 1 -- third_party/hip-build-deps/
+
+    - name: Restore Python dependencies cache
+      id: cache-python-deps-restore
+      uses: actions/cache/restore@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+      with:
+        path: ${{ env.PYENV_ROOT }}
+        key: ${{ runner.os }}-python-deps-${{ hashFiles('libshortfin/requirements-tests.txt') }}-v${{ env.CACHE_DEPS_VER }}
+
+    - name: Restore Python ASan cache
+      id: cache-python-asan
+      if: steps.cache-python-deps-restore.outputs.cache-hit != 'true'
+      uses: actions/cache/restore@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+      with:
+        path: ${{ env.PYENV_ROOT }}
+        key: ${{ runner.os }}-python-asan-${{ env.PYENV_REF }}-${{ env.PYTHON_VER }}-v${{ env.CACHE_ASAN_VER }}
+
+    - name: Set path
+      run:
+        echo "${{ env.PYENV_ROOT }}/bin" >> $GITHUB_PATH
+
+    - name: Install Python dependencies
+      if: steps.cache-python-deps-restore.outputs.cache-hit != 'true'
+      run: |
+        eval "$(pyenv init -)"
+        pip install -r ${{ env.LIBSHORTFIN_DIR }}/requirements-tests.txt
+
+    - name: Save Python dependencies cache
+      if: steps.cache-python-deps-restore.outputs.cache-hit != 'true'
+      id: cache-python-deps-save
+      uses: actions/cache/save@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+      with:
+        path: ${{ env.PYENV_ROOT }}
+        key: ${{ steps.cache-python-deps-restore.outputs.cache-primary-key }}
+
+    - name: Build libshortfin
+      env:
+        # TODO(#151): Don't ignore ODR violations
+        ASAN_OPTIONS=detect_odr_violation: 0
+      run: |
+        eval "$(pyenv init -)"
+        mkdir ${{ env.LIBSHORTFIN_DIR }}/build
+        cd ${{ env.LIBSHORTFIN_DIR }}/build
+        cmake -GNinja \
+          -DCMAKE_BUILD_TYPE=Debug \
+          -DCMAKE_C_COMPILER=clang-18 \
+          -DCMAKE_CXX_COMPILER=clang++-18 \
+          -DCMAKE_LINKER_TYPE=LLD \
+          -DSHORTFIN_BUNDLE_DEPS=ON \
+          -DSHORTFIN_IREE_SOURCE_DIR=${{ env.IREE_SOURCE_DIR }} \
+          -DSHORTFIN_BUILD_PYTHON_BINDINGS=ON \
+          -DSHORTFIN_ENABLE_ASAN=ON \
+          ..
+        cmake --build . --target all
+        pip install -v -e .
+
+    - name: Run ctest
+      if: ${{ !cancelled() }}
+      env:
+        CTEST_OUTPUT_ON_FAILURE: 1
+      run: |
+        cd ${{ env.LIBSHORTFIN_DIR }}/build
+        ctest --timeout 30 --output-on-failure
+
+    - name: Run pytest
+      if: ${{ !cancelled() }}
+      run: |
+        eval "$(pyenv init -)"
+        cd ${{ env.LIBSHORTFIN_DIR }}
+        pytest -m "not requires_amd_gpu"

docs/amdgpu_kernel_optimization_guide.md

@@ -4,7 +4,7 @@ Author: Jakub Kuderski @kuhar
 
 Date: 2024-06-24
 
-Last Update: 2024-08-14
+Last Update: 2024-08-22
 
 ## Introduction
 
@@ -280,6 +280,11 @@ at once.
 A sequence of up to 4 adjacent `global_load_dwordx4` instructions (implicitly)
 forms a *clause* that translates to a single data fabric transaction.
 
+> [!TIP]
+> To achieve peak L1 bandwidth, make sure that your memory access engages all
+> four L1 cache sets. That is, at the level of the workgroup, you should be
+> loading 4 cache lines (128 B) that each map to a different cache set.
+
 > [!TIP]
 > For data that is 'streamed' and does not need to be cached, consider
 > using *non-temporal* loads/stores. This disables coherency and invalidates

libshortfin/CMakeLists.txt

@@ -18,6 +18,8 @@ project(
   VERSION 0.9
   LANGUAGES C CXX)
 
+set(SOVERSION 1)
+
 set(CMAKE_C_STANDARD 11)
 set(CMAKE_CXX_STANDARD 20)
 # https://discourse.cmake.org/t/cmake-3-28-cmake-cxx-compiler-clang-scan-deps-notfound-not-found/9244/3
@@ -33,8 +35,30 @@ endif()
 option(SHORTFIN_BUILD_PYTHON_BINDINGS "Builds Python Bindings" OFF)
 option(SHORTFIN_BUILD_TESTS "Builds C++ tests" ON)
 option(SHORTFIN_BUNDLE_DEPS "Download dependencies instead of using system libraries" OFF)
+
 set(SHORTFIN_IREE_SOURCE_DIR "" CACHE FILEPATH "Path to IREE source")
 
+# Enabling ASAN. Note that this will work best if building in a completely
+# bundled fashion and with an ASAN rigged CPython. Otherwise, various LD_PRELOAD
+# hacks are needed. This is merely a develope convenience: people are more
+# than welcome to set flags themselves.
+option(SHORTFIN_ENABLE_ASAN "Enable ASAN" OFF)
+if(SHORTFIN_ENABLE_ASAN)
+  add_compile_options(-fsanitize=address)
+  add_link_options(-fsanitize=address)
+
+  # Enable more ASAN checks.
+  add_compile_definitions(IREE_SANITIZER_ADDRESS)
+endif()
+
+option(SHORTFIN_SYSTEMS_AMDGPU "Builds for AMD GPU systems" ON)
+message(STATUS "libshortfin supported systems:")
+if(SHORTFIN_SYSTEMS_AMDGPU)
+  message(STATUS "  - AMD GPU")
+  add_compile_definitions("SHORTFIN_HAVE_AMDGPU")
+endif()
+message(STATUS "  - Host")
+
 include(FetchContent)
 
 # Includes.
@@ -92,6 +116,7 @@ if (NOT SHORTFIN_IREE_SOURCE_DIR AND SHORTFIN_BUNDLE_DEPS)
     # TODO: We shouldn't have to pull googletest when we are not building tests.
     #       This needs to be fixed with IREE.
     GIT_SUBMODULES "third_party/benchmark third_party/cpuinfo third_party/flatcc third_party/hip-build-deps third_party/googletest"
+    GIT_SHALLOW TRUE
   )
   FetchContent_GetProperties(iree)
   if(NOT iree_POPULATED)
@@ -110,7 +135,9 @@ if(SHORTFIN_IREE_SOURCE_DIR)
   set(IREE_HAL_DRIVER_DEFAULTS OFF)
   set(IREE_HAL_DRIVER_LOCAL_SYNC ON)
   set(IREE_HAL_DRIVER_LOCAL_TASK ON)
-  set(IREE_HAL_DRIVER_HIP ON)
+  if(SHORTFIN_SYSTEMS_AMDGPU)
+    set(IREE_HAL_DRIVER_HIP ON)
+  endif()
   add_subdirectory(${SHORTFIN_IREE_SOURCE_DIR} shortfin_iree SYSTEM EXCLUDE_FROM_ALL)
 else()
   # Try to find iree using find_package

libshortfin/bindings/python/_shortfin/asyncio_bridge.py

@@ -5,9 +5,6 @@
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 import asyncio
-from collections.abc import Callable
-from contextvars import Context
-from typing_extensions import Unpack
 
 from . import lib as sfl