Merge branch 'main' into annotate_rootOp

Signed-off-by: nithinsubbiah <[email protected]>

.github/CODEOWNERS

@@ -58,7 +58,7 @@
 /compiler/src/iree/compiler/Codegen/Common @hanhanW
 /compiler/src/iree/compiler/Codegen/Common/GPU @antiagainst @qedawkins
 /compiler/src/iree/compiler/Codegen/Dialect/GPU @antiagainst @qedawkins
-/compiler/src/iree/compiler/Codegen/LLVMCPU/ @hanhanW @MaheshRavishankar
+/compiler/src/iree/compiler/Codegen/LLVMCPU/ @hanhanW @MaheshRavishankar @pashu123
 /compiler/src/iree/compiler/Codegen/LLVMGPU/ @MaheshRavishankar @qedawkins @kuhar @Groverkss
 /compiler/src/iree/compiler/Codegen/SPIRV/ @antiagainst @MaheshRavishankar @kuhar
 /compiler/src/iree/compiler/ConstEval/ @hanhanW @stellaraccident

.github/workflows/ci.yml

@@ -41,21 +41,11 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.event.number || github.sha }}
   cancel-in-progress: true
 
-env:
-  # This needs to be in env instead of the outputs of setup because it contains
-  # the run attempt and we want that to be the current attempt, not whatever
-  # attempt the setup step last ran in.
-  GCS_URL: https://storage.googleapis.com/iree-github-actions-${{ github.event_name == 'pull_request' && 'presubmit' || 'postsubmit' }}-artifacts/${{ github.run_id }}/${{ github.run_attempt }}
-
 # Jobs are organized into groups and topologically sorted by dependencies
 jobs:
   setup:
     uses: ./.github/workflows/setup.yml
 
-  ############################### Configurations ###############################
-  # Jobs that build IREE in some non-default configuration
-  ##############################################################################
-
   runtime:
     needs: setup
     name: "runtime :: ${{ matrix.name }}"
@@ -206,86 +196,6 @@ jobs:
       - name: CMake - build
         run: cmake --build ${BUILD_DIR} -- -k 0
 
-  ############################## Crosscompilation ##############################
-  # Jobs that cross-compile IREE for other platforms
-  ##############################################################################
-
-  # Disabled to reduce self-hosted runners needed. See #17957
-  # TODO(#17957): migrate to pkgci
-  # cross_compile_and_test:
-  #   needs: [setup, build_all]
-  #   if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'cross_compile_and_test')
-  #   runs-on:
-  #     - self-hosted # must come first
-  #     - runner-group=${{ needs.setup.outputs.runner-group }}
-  #     - environment=${{ needs.setup.outputs.runner-env }}
-  #     - cpu
-  #     - os-family=Linux
-  #   strategy:
-  #     matrix:
-  #       target:
-  #         - platform: linux
-  #           arch: riscv_32
-  #           abi: ilp32d
-  #           docker_image: "gcr.io/iree-oss/riscv@sha256:62e87bad3405d691ddba6f9be0ef44eeb60461a467c8d86f0842c81a1f97da79"
-  #           build_script: "./build_tools/cmake/build_riscv.sh"
-  #           test_script: "./build_tools/cmake/test_riscv.sh"
-  #         - platform: generic
-  #           arch: riscv_32
-  #           abi: ilp32
-  #           docker_image: "gcr.io/iree-oss/riscv@sha256:62e87bad3405d691ddba6f9be0ef44eeb60461a467c8d86f0842c81a1f97da79"
-  #           build_script: "./build_tools/cmake/build_riscv.sh"
-  #           test_script: "./tests/riscv32/smoke.sh"
-  #         - platform: emscripten
-  #           arch: wasm32
-  #           abi: wasm32
-  #           docker_image: "gcr.io/iree-oss/emscripten@sha256:2dd4c52f1bb499ab365aad0111fe5538b685d88af38636b409b0cf6a576ab214"
-  #           build_script: "./build_tools/cmake/build_runtime_emscripten.sh"
-  #           # No test script
-  #   env:
-  #     PLATFORM: ${{ matrix.target.platform }}
-  #     ARCH: ${{ matrix.target.arch }}
-  #     ABI: ${{ matrix.target.abi }}
-  #     DOCKER_IMAGE: ${{ matrix.target.docker_image }}
-  #     BUILD_SCRIPT: ${{ matrix.target.build_script }}
-  #     TEST_SCRIPT: ${{ matrix.target.test_script }}
-  #     INSTALL_DIR: ${{ needs.build_all.outputs.install-dir }}
-  #     INSTALL_DIR_ARCHIVE: ${{ needs.build_all.outputs.install-dir-archive }}
-  #     INSTALL_DIR_GCS_ARTIFACT: ${{ needs.build_all.outputs.install-dir-gcs-artifact }}
-  #     TARGET_BUILD_DIR: build-${{ matrix.target.platform }}-${{ matrix.target.arch }}
-  #     IREE_WRITE_REMOTE_CCACHE: ${{ needs.setup.outputs.write-caches }}
-  #   steps:
-  #     - name: "Checking out repository"
-  #       uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
-  #     - name: "Checking out runtime submodules"
-  #       run: ./build_tools/scripts/git/update_runtime_submodules.sh
-  #     - name: "Downloading install dir archive"
-  #       run: gcloud storage cp "${INSTALL_DIR_GCS_ARTIFACT}" "${INSTALL_DIR_ARCHIVE}"
-  #     - name: "Extracting install directory"
-  #       run: tar -xf "${INSTALL_DIR_ARCHIVE}"
-  #     - name: "Build cross-compiling target"
-  #       run: |
-  #         ./build_tools/github_actions/docker_run.sh \
-  #           --env "IREE_CCACHE_GCP_TOKEN=$(gcloud auth application-default print-access-token)" \
-  #           --env "IREE_WRITE_REMOTE_CCACHE=${IREE_WRITE_REMOTE_CCACHE}" \
-  #           --env "CCACHE_NAMESPACE=${DOCKER_IMAGE}" \
-  #           --env "IREE_TARGET_PLATFORM=${PLATFORM}" \
-  #           --env "IREE_TARGET_ARCH=${ARCH}" \
-  #           --env "IREE_TARGET_ABI=${ABI}" \
-  #           --env "IREE_TARGET_BUILD_DIR=${TARGET_BUILD_DIR}" \
-  #           --env "IREE_HOST_BIN_DIR=${INSTALL_DIR}/bin" \
-  #           "${DOCKER_IMAGE}" \
-  #           "${BUILD_SCRIPT}"
-  #     - name: "Test cross-compiling target"
-  #       if: matrix.target.test_script
-  #       run: |
-  #         ./build_tools/github_actions/docker_run.sh \
-  #           --env "IREE_TARGET_PLATFORM=${PLATFORM}" \
-  #           --env "IREE_TARGET_ARCH=${ARCH}" \
-  #           --env "IREE_TARGET_BUILD_DIR=${TARGET_BUILD_DIR}" \
-  #           "${DOCKER_IMAGE}" \
-  #           "${TEST_SCRIPT}"
-
   ##############################################################################
 
   # Depends on all the other jobs to provide a single anchor that indicates the
@@ -298,17 +208,9 @@ jobs:
     runs-on: ubuntu-20.04
     needs:
       - setup
-
-      # Accelerators
-      # - test_nvidia_a100
-
-      # Runtime build variants
       - runtime
       - runtime_small
       - runtime_tracing
-
-      # Crosscompilation
-      # - cross_compile_and_test
     steps:
       - name: "Checking out repository"
         uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7

.github/workflows/pkgci.yml

@@ -104,6 +104,12 @@ jobs:
     if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'test_onnx')
     uses: ./.github/workflows/pkgci_test_onnx.yml
 
+  test_sharktank:
+    name: Test Sharktank
+    needs: [setup, build_packages]
+    if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'test_sharktank')
+    uses: ./.github/workflows/pkgci_test_sharktank.yml
+
   test_tensorflow:
     name: Test TensorFlow
     needs: [setup, build_packages]

.github/workflows/pkgci_regression_test.yml

@@ -220,7 +220,7 @@ jobs:
             --goldentime-rocm-unet-ms 419.0 \
             --goldentime-rocm-clip-ms 18.5 \
             --goldentime-rocm-vae-ms 337.0 \
-            --goldendispatch-rocm-unet 1531 \
+            --goldendispatch-rocm-unet 1602 \
             --goldendispatch-rocm-clip 1139 \
             --goldendispatch-rocm-vae 246 \
             --goldensize-rocm-unet-bytes 2280000  \
@@ -238,21 +238,21 @@ jobs:
         run: |
           source ${VENV_DIR}/bin/activate
           pytest ./experimental/benchmarks/sdxl/benchmark_sdxl_rocm.py \
-            --goldentime-rocm-e2e-ms 372.0 \
-            --goldentime-rocm-unet-ms 95.0 \
+            --goldentime-rocm-e2e-ms 330.0 \
+            --goldentime-rocm-unet-ms 80.0 \
             --goldentime-rocm-clip-ms 15.5 \
             --goldentime-rocm-vae-ms 80.0 \
-            --goldendispatch-rocm-unet 1531 \
+            --goldendispatch-rocm-unet 1602 \
             --goldendispatch-rocm-clip 1139 \
             --goldendispatch-rocm-vae 246 \
             --goldensize-rocm-unet-bytes 2270000 \
             --goldensize-rocm-clip-bytes 860000  \
             --goldensize-rocm-vae-bytes 840000 \
-            --goldentime-rocm-punet-int8-fp16-ms 55 \
-            --goldendispatch-rocm-punet-int8-fp16 1284 \
+            --goldentime-rocm-punet-int8-fp16-ms 53 \
+            --goldendispatch-rocm-punet-int8-fp16 1424 \
             --goldensize-rocm-punet-int8-fp16-bytes 2560000 \
-            --goldentime-rocm-punet-int8-fp8-ms 59 \
-            --goldendispatch-rocm-punet-int8-fp8 1564 \
+            --goldentime-rocm-punet-int8-fp8-ms 53 \
+            --goldendispatch-rocm-punet-int8-fp8 1704 \
             --goldensize-rocm-punet-int8-fp8-bytes 2800000 \
             --rocm-chip gfx942 \
             --log-cli-level=info \

.github/workflows/pkgci_test_sharktank.yml

@@ -0,0 +1,73 @@
+# Copyright 2024 The IREE Authors
+#
+# 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: PkgCI Test Sharktank
+on:
+  workflow_call:
+    inputs:
+      artifact_run_id:
+        type: string
+        default: ""
+  workflow_dispatch:
+    inputs:
+      artifact_run_id:
+        type: string
+        default: ""
+
+jobs:
+  test_sharktank_models:
+    name: "test_sharktank_models :: ${{ matrix.name }}"
+    runs-on: ${{ matrix.runs-on }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          # CPU
+          - name: cpu_llvm_task
+            runs-on: ubuntu-20.04
+
+    env:
+      VENV_DIR: ${{ github.workspace }}/venv
+    steps:
+      - name: Checking out IREE repository
+        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
+        with:
+          submodules: false
+      - uses: actions/setup-python@f677139bbe7f9c59b41e40162b753c062f5d49a3 # v5.1.0
+        with:
+          # Must match the subset of versions built in pkgci_build_packages.
+          python-version: "3.11"
+      - uses: actions/download-artifact@fa0a91b85d4f404e444e00e005971372dc801d16 # v4.1.8
+        with:
+          name: linux_x86_64_release_packages
+          path: ${{ env.PACKAGE_DOWNLOAD_DIR }}
+      - name: Setup venv
+        run: |
+          ./build_tools/pkgci/setup_venv.py ${VENV_DIR} \
+            --artifact-path=${PACKAGE_DOWNLOAD_DIR} \
+            --fetch-gh-workflow=${{ inputs.artifact_run_id }}
+
+      - name: Checkout test suites repository
+        uses: actions/checkout@692973e3d937129bcbf40652eb9f2f61becf3332 # v4.1.7
+        with:
+          repository: iree-org/iree-test-suites
+          ref: a0c84d59c4332463dd46a3c4877d8e0ab2e0a80d
+          path: iree-test-suites
+          lfs: true
+      - name: Install Sharktank models test suite requirements
+        run: |
+          source ${VENV_DIR}/bin/activate
+          python -m pip install -r iree-test-suites/sharktank_models/requirements.txt
+      - name: Run Sharktank models test suite
+        run: |
+          source ${VENV_DIR}/bin/activate
+          pytest iree-test-suites/sharktank_models/ \
+              -rA \
+              -m "target_cpu" \
+              --log-cli-level=info \
+              --override-ini=xfail_strict=false \
+              --timeout=120 \
+              --durations=0

.gitignore

@@ -27,6 +27,8 @@ Testing/
 
 # Bazel artifacts
 **/bazel-*
+MODULE.bazel
+MODULE.bazel.lock
 
 # Executables
 *.exe

.pre-commit-config.yaml

@@ -83,14 +83,21 @@ repos:
         # probably changed again in a future version, whatever.
         language_version: "1.16"
 
-      - id: bazel_to_cmake
-        name: Run bazel_to_cmake.py
+      # Convert BUILD.bazel files to CMakeLists.txt files in two passes to
+      # avoid potential race conditions.
+      # Note: this passes file names to the tool. The tool can also be run
+      # manually with no arguments specified to walk directories on its own.
+      # Keep the top level directories here in sync with .bazel_to_cmake.cfg.py.
+      - id: bazel_to_cmake_1
+        name: Run bazel_to_cmake.py on BUILD.bazel files
         language: python
-        # Note: this passes file names to the tool. The tool can also be run
-        # manually with no arguments specified to walk directories on its own.
         entry: ./build_tools/bazel_to_cmake/bazel_to_cmake.py
-        # Keep the top level directories here in sync with .bazel_to_cmake.cfg.py.
-        files: '^(compiler|runtime|samples|tests|tools)/(.*/)?(BUILD\.bazel|CMakeLists.txt)$'
+        files: '^(compiler|runtime|samples|tests|tools)/(.*/)?(BUILD\.bazel)$'
+      - id: bazel_to_cmake_2
+        name: Run bazel_to_cmake.py on CMakeLists.txt files
+        language: python
+        entry: ./build_tools/bazel_to_cmake/bazel_to_cmake.py
+        files: "^(compiler|runtime|samples|tests|tools)/(.*/)?(CMakeLists.txt)$"
 
       - id: check_path_lengths
         name: Check for excessively long path lengths

build_tools/bazel/iree_bitcode_library.bzl

@@ -255,6 +255,80 @@ def iree_cuda_bitcode_library(
         **kwargs
     )
 
+def iree_amdgpu_bitcode_library(
+        name,
+        gpu_arch,
+        srcs,
+        copts = [],
+        out = None,
+        **kwargs):
+    """Builds an AMDGPU LLVM bitcode library from an input file using clang.
+
+    Args:
+        name: Name of the target.
+        gpu_arch: Target AMDGPU architecture, e.g. gfx942.
+        srcs: Source files to pass to clang. Headers (*.h) are for dependency
+              tracking only. Current limitation: only one non-header source is
+              supported.
+        copts: Additional flags to pass to clang.
+        out: Output file name. Defaults to {source.c}.{gpu_arch}.bc.
+        **kwargs: any additional attributes to pass to the underlying rules.
+    """
+
+    clang_tool = "@llvm-project//clang:clang"
+
+    base_copts = [
+        # Language: C23.
+        "-std=c23",
+
+        # Avoid dependencies.
+        "-nogpulib",
+
+        # Avoid ABI issues.
+        "-fno-short-wchar",  # Shouldn't matter to us, but doesn't hurt.
+
+        # Target architecture/machine.
+        "-target",
+        "amdgcn-amd-amdhsa",
+        "-march=%s" % gpu_arch,
+        "-fgpu-rdc",  # NOTE: may not be required for all targets.
+
+        # Optimized.
+        "-O3",
+        "-fno-ident",
+        "-fvisibility=hidden",
+
+        # Object file only in bitcode format.
+        "-c",
+        "-emit-llvm",
+    ]
+
+    non_header_srcs = [src for src in srcs if not src.endswith(".h")]
+    if len(non_header_srcs) != 1:
+        fail("Expected exactly one non-header file in srcs, got srcs=[" + ", ".join(srcs) + "]")
+    src = non_header_srcs[0]
+
+    if not out:
+        out = "%s.%s.bc" % (src, gpu_arch)
+
+    native.genrule(
+        name = "gen_%s" % (out),
+        srcs = srcs,
+        outs = [out],
+        cmd = " ".join([
+            "$(location %s)" % (clang_tool),
+            "$(location %s)" % (src),
+            "-o $(location %s)" % (out),
+            "-I .",
+        ] + base_copts + copts),
+        tools = [
+            clang_tool,
+        ],
+        message = "Compiling %s to %s..." % (src, out),
+        output_to_bindir = 1,
+        **kwargs
+    )
+
 def iree_link_bitcode(
         name,
         bitcode_files,