Merge remote-tracking branch 'origin/main'

# Conflicts:
#	Project.toml

Author: glou-nes

.github/workflows/CI-localjll.yml

@@ -39,6 +39,13 @@ jobs:
     env:
       TMPDIR: ${{ github.workspace }}/tmp
     steps:
+      - name: Free Disk Space
+        uses: jlumbroso/free-disk-space@main
+        with:
+          tool-cache: false
+        if: ${{ startsWith(matrix.os, 'ubuntu-') }}
+      - name: Clean `/opt`
+        run: sudo rm -rf /opt/*
       - uses: actions/checkout@v4
       - name: Create TMPDIR
         run: |

.github/workflows/CI.yml

@@ -33,6 +33,8 @@ jobs:
     timeout-minutes: 90
     name: Julia ${{ matrix.version }} - ${{ matrix.test_group }} - ${{ matrix.os }} - ${{ matrix.runtime }} - assertions=${{ matrix.assertions }} - ${{ github.event_name }}
     runs-on: ${{ matrix.os }}
+    container:
+      image: ${{ contains(matrix.os, 'linux') && 'ghcr.io/enzymead/reactant-docker-images:main' || '' }}
     strategy:
       fail-fast: false
       matrix:
@@ -59,6 +61,11 @@ jobs:
         assertions:
           - false
         include:
+          - os: linux-x86-ct6e-180-4tpu
+            version: "1.11"
+            assertions: false
+            test_group: core
+            runtime: "IFRT"
           - os: ubuntu-24.04
             version: "1.10"
             assertions: true
@@ -86,9 +93,13 @@ jobs:
           #   libReactant: packaged
           #   version: '1.10'
           #   test_group: integration
-    env:
-      TMPDIR: ${{ github.workspace }}/tmp
     steps:
+      - name: Set TMPDIR
+        # We have to use `${GITHUB_WORKSPACE}` instead of `github.workspace` because GitHub
+        # is terrible and the two don't match inside containers:
+        # https://github.com/actions/runner/issues/2058
+        run: |
+          echo "TMPDIR=${GITHUB_WORKSPACE}/tmp" >> ${GITHUB_ENV}
       - uses: actions/checkout@v4
       - name: Create TMPDIR
         run: |

Project.toml

@@ -1,7 +1,7 @@
 name = "Reactant"
 uuid = "3c362404-f566-11ee-1572-e11a4b42c853"
 authors = ["William Moses <[email protected]>", "Valentin Churavy <[email protected]>", "Sergio Sánchez Ramírez <[email protected]>", "Paul Berg <[email protected]>", "Avik Pal <[email protected]>", "Mosè Giordano <[email protected]>"]
-version = "0.2.144"
+version = "0.2.149"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
@@ -27,11 +27,15 @@ Reactant_jll = "0192cb87-2b54-54ad-80e0-3be72ad8a3c0"
 ScopedValues = "7e506255-f358-4e82-b7e4-beb19740aa63"
 Scratch = "6c6a2e73-6563-6170-7368-637461726353"
 Sockets = "6462fe0b-24de-5631-8697-dd941f90decc"
+unzip_jll = "88f77b66-78eb-5ed0-bc16-ebba0796830d"
 
 [weakdeps]
 AbstractFFTs = "621f4979-c628-5d54-868e-fcf4e3e8185c"
 ArrayInterface = "4fba245c-0d91-5ea0-9b3e-6abc04ee57a9"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
+DLFP8Types = "f4c16678-4a16-415b-82ef-ed337c5d6c7c"
+Float8s = "81dfefd7-55b0-40c6-a251-db853704e186"
+GPUCompiler = "61eb1bfa-7361-4325-ad38-22787b887f55"
 KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 MPI = "da04e1cc-30fd-572f-bb4f-1f8673147195"
 NNlib = "872c559c-99b0-510c-b3b7-b6c96a88d5cd"
@@ -50,6 +54,8 @@ ReactantCore = {path = "lib/ReactantCore"}
 ReactantAbstractFFTsExt = "AbstractFFTs"
 ReactantArrayInterfaceExt = "ArrayInterface"
 ReactantCUDAExt = ["CUDA", "GPUCompiler", "KernelAbstractions", "LLVM"]
+ReactantDLFP8TypesExt = "DLFP8Types"
+ReactantFloat8sExt = "Float8s"
 ReactantKernelAbstractionsExt = "KernelAbstractions"
 ReactantMPIExt = "MPI"
 ReactantNNlibExt = ["NNlib", "Statistics"]
@@ -67,10 +73,12 @@ Adapt = "4.1"
 ArrayInterface = "7.17.1"
 CEnum = "0.5"
 CUDA = "5.6"
+DLFP8Types = "0.1"
 Downloads = "1.6"
 EnumX = "1"
 Enzyme = "0.13.49"
 EnzymeCore = "0.8.11"
+Float8s = "0.1"
 Functors = "0.5"
 GPUArraysCore = "0.2"
 GPUCompiler = "1.3"
@@ -90,13 +98,14 @@ PythonCall = "0.9.25"
 Random = "1.10"
 Random123 = "1.7"
 ReactantCore = "0.1.15"
-Reactant_jll = "0.0.216"
+Reactant_jll = "0.0.224"
 ScopedValues = "1.3.0"
 Scratch = "1.2"
 Sockets = "1.10"
 SpecialFunctions = "2.4"
 Statistics = "1.10"
-YaoBlocks = "0.13"
+unzip_jll = "6"
+YaoBlocks = "0.13, 0.14"
 julia = "1.10"
 
 [extras]

deps/ReactantExtra/.bazelrc

@@ -2,12 +2,14 @@ build --announce_rc
 
 # TODO: Migrate for https://github.com/bazelbuild/bazel/issues/7260
 common --noincompatible_enable_cc_toolchain_resolution
+common --repo_env USE_HERMETIC_CC_TOOLCHAIN=0
 common --experimental_repo_remote_exec
 common --cxxopt=-std=c++17 --host_cxxopt=-std=c++17
 common --cxxopt=-w --host_cxxopt=-w
 common --define=grpc_no_ares=true
 common --noenable_bzlmod
 
+
 build --repo_env=USE_PYWRAP_RULES=True
 build --copt=-DGRPC_BAZEL_BUILD
 build --host_copt=-DGRPC_BAZEL_BUILD
@@ -27,6 +29,7 @@ build:cuda --repo_env TF_NVCC_CLANG=1
 build:cuda --repo_env TF_NCCL_USE_STUB=1
 build:cuda --repo_env=HERMETIC_CUDA_VERSION="12.8.1"
 build:cuda --repo_env=HERMETIC_CUDNN_VERSION="9.8.0"
+build:cuda --repo_env=HERMETIC_NVSHMEM_VERSION="3.2.5"
 # "sm" means we emit only cubin, which is forward compatible within a GPU generation.
 # "compute" means we emit both cubin and PTX, which is larger but also forward compatible to future GPU generations.
 build:cuda --repo_env HERMETIC_CUDA_COMPUTE_CAPABILITIES="sm_60,sm_70,sm_80,compute_90"
@@ -35,6 +38,8 @@ build:cuda --@local_config_cuda//:enable_cuda
 # Default hermetic CUDA and CUDNN versions.
 build:cuda --@local_config_cuda//cuda:include_cuda_libs=true
 build:cuda --@local_config_cuda//:cuda_compiler=nvcc
+# build:cuda --@local_config_nvshmem//:override_include_nvshmem_libs=true
+# build:cuda --@local_config_nvshmem//cuda:include_nvshmem_libs=true
 
 build:rocm --repo_env TF_NEED_ROCM=1
 build:rocm --define=using_rocm=true